xref: /titanic_41/usr/src/uts/sun4v/io/vsw_ldc.c (revision c2b6e926ea57d0ba055f91471cfc9772c7fbacd0)
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 /*
23  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/types.h>
30 #include <sys/errno.h>
31 #include <sys/debug.h>
32 #include <sys/time.h>
33 #include <sys/sysmacros.h>
34 #include <sys/systm.h>
35 #include <sys/user.h>
36 #include <sys/stropts.h>
37 #include <sys/stream.h>
38 #include <sys/strlog.h>
39 #include <sys/strsubr.h>
40 #include <sys/cmn_err.h>
41 #include <sys/cpu.h>
42 #include <sys/kmem.h>
43 #include <sys/conf.h>
44 #include <sys/ddi.h>
45 #include <sys/sunddi.h>
46 #include <sys/ksynch.h>
47 #include <sys/stat.h>
48 #include <sys/kstat.h>
49 #include <sys/vtrace.h>
50 #include <sys/strsun.h>
51 #include <sys/dlpi.h>
52 #include <sys/ethernet.h>
53 #include <net/if.h>
54 #include <sys/varargs.h>
55 #include <sys/machsystm.h>
56 #include <sys/modctl.h>
57 #include <sys/modhash.h>
58 #include <sys/mac.h>
59 #include <sys/mac_ether.h>
60 #include <sys/taskq.h>
61 #include <sys/note.h>
62 #include <sys/mach_descrip.h>
63 #include <sys/mac.h>
64 #include <sys/mdeg.h>
65 #include <sys/ldc.h>
66 #include <sys/vsw_fdb.h>
67 #include <sys/vsw.h>
68 #include <sys/vio_mailbox.h>
69 #include <sys/vnet_mailbox.h>
70 #include <sys/vnet_common.h>
71 #include <sys/vio_util.h>
72 #include <sys/sdt.h>
73 #include <sys/atomic.h>
74 #include <sys/callb.h>
75 
76 /* Port add/deletion/etc routines */
77 static	int vsw_port_delete(vsw_port_t *port);
78 static	int vsw_ldc_attach(vsw_port_t *port, uint64_t ldc_id);
79 static	int vsw_ldc_detach(vsw_port_t *port, uint64_t ldc_id);
80 static	int vsw_init_ldcs(vsw_port_t *port);
81 static	int vsw_uninit_ldcs(vsw_port_t *port);
82 static	int vsw_ldc_init(vsw_ldc_t *ldcp);
83 static	int vsw_ldc_uninit(vsw_ldc_t *ldcp);
84 static	int vsw_drain_ldcs(vsw_port_t *port);
85 static	int vsw_drain_port_taskq(vsw_port_t *port);
86 static	void vsw_marker_task(void *);
87 static	int vsw_plist_del_node(vsw_t *, vsw_port_t *port);
88 int vsw_detach_ports(vsw_t *vswp);
89 int vsw_port_add(vsw_t *vswp, md_t *mdp, mde_cookie_t *node);
90 mcst_addr_t *vsw_del_addr(uint8_t devtype, void *arg, uint64_t addr);
91 int vsw_port_detach(vsw_t *vswp, int p_instance);
92 int vsw_portsend(vsw_port_t *port, mblk_t *mp, mblk_t *mpt);
93 int vsw_port_attach(vsw_t *vswp, int p_instance,
94 	uint64_t *ldcids, int nids, struct ether_addr *macaddr);
95 vsw_port_t *vsw_lookup_port(vsw_t *vswp, int p_instance);
96 
97 
98 /* Interrupt routines */
99 static	uint_t vsw_ldc_cb(uint64_t cb, caddr_t arg);
100 
101 /* Handshake routines */
102 static	void vsw_ldc_reinit(vsw_ldc_t *);
103 static	void vsw_process_conn_evt(vsw_ldc_t *, uint16_t);
104 static	void vsw_conn_task(void *);
105 static	int vsw_check_flag(vsw_ldc_t *, int, uint64_t);
106 static	void vsw_next_milestone(vsw_ldc_t *);
107 static	int vsw_supported_version(vio_ver_msg_t *);
108 
109 /* Data processing routines */
110 static void vsw_process_pkt(void *);
111 static void vsw_dispatch_ctrl_task(vsw_ldc_t *, void *, vio_msg_tag_t);
112 static void vsw_process_ctrl_pkt(void *);
113 static void vsw_process_ctrl_ver_pkt(vsw_ldc_t *, void *);
114 static void vsw_process_ctrl_attr_pkt(vsw_ldc_t *, void *);
115 static void vsw_process_ctrl_mcst_pkt(vsw_ldc_t *, void *);
116 static void vsw_process_ctrl_dring_reg_pkt(vsw_ldc_t *, void *);
117 static void vsw_process_ctrl_dring_unreg_pkt(vsw_ldc_t *, void *);
118 static void vsw_process_ctrl_rdx_pkt(vsw_ldc_t *, void *);
119 static void vsw_process_data_pkt(vsw_ldc_t *, void *, vio_msg_tag_t);
120 static void vsw_process_data_dring_pkt(vsw_ldc_t *, void *);
121 static void vsw_process_data_raw_pkt(vsw_ldc_t *, void *);
122 static void vsw_process_data_ibnd_pkt(vsw_ldc_t *, void *);
123 static void vsw_process_err_pkt(vsw_ldc_t *, void *, vio_msg_tag_t);
124 
125 /* Switching/data transmit routines */
126 static	int vsw_dringsend(vsw_ldc_t *, mblk_t *);
127 static	int vsw_descrsend(vsw_ldc_t *, mblk_t *);
128 static int vsw_ldcsend(vsw_ldc_t *ldcp, mblk_t *mp, int retries);
129 
130 /* Packet creation routines */
131 static void vsw_send_ver(void *);
132 static void vsw_send_attr(vsw_ldc_t *);
133 static vio_dring_reg_msg_t *vsw_create_dring_info_pkt(vsw_ldc_t *);
134 static void vsw_send_dring_info(vsw_ldc_t *);
135 static void vsw_send_rdx(vsw_ldc_t *);
136 static int vsw_send_msg(vsw_ldc_t *, void *, int, boolean_t);
137 
138 /* Dring routines */
139 static dring_info_t *vsw_create_dring(vsw_ldc_t *);
140 static void vsw_create_privring(vsw_ldc_t *);
141 static int vsw_setup_ring(vsw_ldc_t *ldcp, dring_info_t *dp);
142 static int vsw_dring_find_free_desc(dring_info_t *, vsw_private_desc_t **,
143     int *);
144 static dring_info_t *vsw_ident2dring(lane_t *, uint64_t);
145 static int vsw_reclaim_dring(dring_info_t *dp, int start);
146 
147 static void vsw_set_lane_attr(vsw_t *, lane_t *);
148 static int vsw_check_attr(vnet_attr_msg_t *, vsw_port_t *);
149 static int vsw_dring_match(dring_info_t *dp, vio_dring_reg_msg_t *msg);
150 static int vsw_mem_cookie_match(ldc_mem_cookie_t *, ldc_mem_cookie_t *);
151 static int vsw_check_dring_info(vio_dring_reg_msg_t *);
152 
153 /* Rcv/Tx thread routines */
154 static void vsw_stop_tx_thread(vsw_ldc_t *ldcp);
155 static void vsw_ldc_tx_worker(void *arg);
156 static uint_t vsw_rx_softintr(caddr_t arg1, caddr_t arg2);
157 static void vsw_stop_rx_thread(vsw_ldc_t *ldcp);
158 static void vsw_ldc_rx_worker(void *arg);
159 
160 /* Misc support routines */
161 static	caddr_t vsw_print_ethaddr(uint8_t *addr, char *ebuf);
162 static void vsw_free_lane_resources(vsw_ldc_t *, uint64_t);
163 static int vsw_free_ring(dring_info_t *);
164 static void vsw_save_lmacaddr(vsw_t *vswp, uint64_t macaddr);
165 static int vsw_get_same_dest_list(struct ether_header *ehp,
166     mblk_t **rhead, mblk_t **rtail, mblk_t **mpp);
167 static mblk_t *vsw_dupmsgchain(mblk_t *mp);
168 static void vsw_mac_rx(vsw_t *vswp, int caller, mac_resource_handle_t mrh,
169     mblk_t *mp, mblk_t *mpt, vsw_macrx_flags_t flags);
170 
171 /* Debugging routines */
172 static void dump_flags(uint64_t);
173 static void display_state(void);
174 static void display_lane(lane_t *);
175 static void display_ring(dring_info_t *);
176 
177 /*
178  * Functions imported from other files.
179  */
180 extern int vsw_set_hw(vsw_t *, vsw_port_t *, int);
181 extern int vsw_unset_hw(vsw_t *, vsw_port_t *, int);
182 extern void vsw_reconfig_hw(vsw_t *);
183 extern int vsw_add_fdb(vsw_t *vswp, vsw_port_t *port);
184 extern int vsw_del_fdb(vsw_t *vswp, vsw_port_t *port);
185 extern int vsw_add_rem_mcst(vnet_mcast_msg_t *mcst_pkt, vsw_port_t *port);
186 extern void vsw_del_mcst_port(vsw_port_t *port);
187 extern int vsw_add_mcst(vsw_t *vswp, uint8_t devtype, uint64_t addr, void *arg);
188 extern int vsw_del_mcst(vsw_t *vswp, uint8_t devtype, uint64_t addr, void *arg);
189 
190 #define	VSW_NUM_VMPOOLS		3	/* number of vio mblk pools */
191 #define	VSW_PORT_REF_DELAY	30	/* delay for port ref_cnt to become 0 */
192 
193 /*
194  * Tunables used in this file.
195  */
196 extern int vsw_num_handshakes;
197 extern int vsw_wretries;
198 extern int vsw_desc_delay;
199 extern int vsw_read_attempts;
200 extern int vsw_ldc_tx_delay;
201 extern int vsw_ldc_tx_retries;
202 extern int vsw_ldc_tx_max_failures;
203 extern boolean_t vsw_ldc_rxthr_enabled;
204 extern boolean_t vsw_ldc_txthr_enabled;
205 extern uint32_t vsw_chain_len;
206 extern uint32_t vsw_mblk_size1;
207 extern uint32_t vsw_mblk_size2;
208 extern uint32_t vsw_mblk_size3;
209 extern uint32_t vsw_num_mblks1;
210 extern uint32_t vsw_num_mblks2;
211 extern uint32_t vsw_num_mblks3;
212 
213 
214 #define	LDC_ENTER_LOCK(ldcp)	\
215 				mutex_enter(&((ldcp)->ldc_cblock));\
216 				mutex_enter(&((ldcp)->ldc_rxlock));\
217 				mutex_enter(&((ldcp)->ldc_txlock));
218 #define	LDC_EXIT_LOCK(ldcp)	\
219 				mutex_exit(&((ldcp)->ldc_txlock));\
220 				mutex_exit(&((ldcp)->ldc_rxlock));\
221 				mutex_exit(&((ldcp)->ldc_cblock));
222 
223 
224 /* supported versions */
225 static	ver_sup_t	vsw_versions[] = { {1, 0} };
226 
227 /*
228  * For the moment the state dump routines have their own
229  * private flag.
230  */
231 #define	DUMP_STATE	0
232 
233 #if DUMP_STATE
234 
235 #define	DUMP_TAG(tag) \
236 {			\
237 	D1(NULL, "DUMP_TAG: type 0x%llx", (tag).vio_msgtype); \
238 	D1(NULL, "DUMP_TAG: stype 0x%llx", (tag).vio_subtype);	\
239 	D1(NULL, "DUMP_TAG: senv 0x%llx", (tag).vio_subtype_env);	\
240 }
241 
242 #define	DUMP_TAG_PTR(tag) \
243 {			\
244 	D1(NULL, "DUMP_TAG: type 0x%llx", (tag)->vio_msgtype); \
245 	D1(NULL, "DUMP_TAG: stype 0x%llx", (tag)->vio_subtype);	\
246 	D1(NULL, "DUMP_TAG: senv 0x%llx", (tag)->vio_subtype_env);	\
247 }
248 
249 #define	DUMP_FLAGS(flags) dump_flags(flags);
250 #define	DISPLAY_STATE()	display_state()
251 
252 #else
253 
254 #define	DUMP_TAG(tag)
255 #define	DUMP_TAG_PTR(tag)
256 #define	DUMP_FLAGS(state)
257 #define	DISPLAY_STATE()
258 
259 #endif	/* DUMP_STATE */
260 
261 /*
262  * Attach the specified port.
263  *
264  * Returns 0 on success, 1 on failure.
265  */
266 int
267 vsw_port_attach(vsw_t *vswp, int p_instance, uint64_t *ldcids, int nids,
268 struct ether_addr *macaddr)
269 {
270 	vsw_port_list_t		*plist = &vswp->plist;
271 	vsw_port_t		*port, **prev_port;
272 	int			i;
273 
274 	D1(vswp, "%s: enter : port %d", __func__, p_instance);
275 
276 	/* port already exists? */
277 	READ_ENTER(&plist->lockrw);
278 	for (port = plist->head; port != NULL; port = port->p_next) {
279 		if (port->p_instance == p_instance) {
280 			DWARN(vswp, "%s: port instance %d already attached",
281 			    __func__, p_instance);
282 			RW_EXIT(&plist->lockrw);
283 			return (1);
284 		}
285 	}
286 	RW_EXIT(&plist->lockrw);
287 
288 	port = kmem_zalloc(sizeof (vsw_port_t), KM_SLEEP);
289 	port->p_vswp = vswp;
290 	port->p_instance = p_instance;
291 	port->p_ldclist.num_ldcs = 0;
292 	port->p_ldclist.head = NULL;
293 	port->addr_set = VSW_ADDR_UNSET;
294 
295 	rw_init(&port->p_ldclist.lockrw, NULL, RW_DRIVER, NULL);
296 
297 	mutex_init(&port->tx_lock, NULL, MUTEX_DRIVER, NULL);
298 	mutex_init(&port->mca_lock, NULL, MUTEX_DRIVER, NULL);
299 
300 	mutex_init(&port->state_lock, NULL, MUTEX_DRIVER, NULL);
301 	cv_init(&port->state_cv, NULL, CV_DRIVER, NULL);
302 	port->state = VSW_PORT_INIT;
303 
304 	if (nids > VSW_PORT_MAX_LDCS) {
305 		D2(vswp, "%s: using first of %d ldc ids",
306 		    __func__, nids);
307 		nids = VSW_PORT_MAX_LDCS;
308 	}
309 
310 	D2(vswp, "%s: %d nids", __func__, nids);
311 	for (i = 0; i < nids; i++) {
312 		D2(vswp, "%s: ldcid (%llx)", __func__, (uint64_t)ldcids[i]);
313 		if (vsw_ldc_attach(port, (uint64_t)ldcids[i]) != 0) {
314 			DERR(vswp, "%s: ldc_attach failed", __func__);
315 
316 			rw_destroy(&port->p_ldclist.lockrw);
317 
318 			cv_destroy(&port->state_cv);
319 			mutex_destroy(&port->state_lock);
320 
321 			mutex_destroy(&port->tx_lock);
322 			mutex_destroy(&port->mca_lock);
323 			kmem_free(port, sizeof (vsw_port_t));
324 			return (1);
325 		}
326 	}
327 
328 	ether_copy(macaddr, &port->p_macaddr);
329 
330 	if (vswp->switching_setup_done == B_TRUE) {
331 		/*
332 		 * If the underlying physical device has been setup,
333 		 * program the mac address of this port in it.
334 		 * Otherwise, port macaddr will be set after the physical
335 		 * device is successfully setup by the timeout handler.
336 		 */
337 		mutex_enter(&vswp->hw_lock);
338 		(void) vsw_set_hw(vswp, port, VSW_VNETPORT);
339 		mutex_exit(&vswp->hw_lock);
340 	}
341 
342 	WRITE_ENTER(&plist->lockrw);
343 
344 	/* create the fdb entry for this port/mac address */
345 	(void) vsw_add_fdb(vswp, port);
346 
347 	/* link it into the list of ports for this vsw instance */
348 	prev_port = (vsw_port_t **)(&plist->head);
349 	port->p_next = *prev_port;
350 	*prev_port = port;
351 	plist->num_ports++;
352 
353 	RW_EXIT(&plist->lockrw);
354 
355 	/*
356 	 * Initialise the port and any ldc's under it.
357 	 */
358 	(void) vsw_init_ldcs(port);
359 
360 	D1(vswp, "%s: exit", __func__);
361 	return (0);
362 }
363 
364 /*
365  * Detach the specified port.
366  *
367  * Returns 0 on success, 1 on failure.
368  */
369 int
370 vsw_port_detach(vsw_t *vswp, int p_instance)
371 {
372 	vsw_port_t	*port = NULL;
373 	vsw_port_list_t	*plist = &vswp->plist;
374 
375 	D1(vswp, "%s: enter: port id %d", __func__, p_instance);
376 
377 	WRITE_ENTER(&plist->lockrw);
378 
379 	if ((port = vsw_lookup_port(vswp, p_instance)) == NULL) {
380 		RW_EXIT(&plist->lockrw);
381 		return (1);
382 	}
383 
384 	if (vsw_plist_del_node(vswp, port)) {
385 		RW_EXIT(&plist->lockrw);
386 		return (1);
387 	}
388 
389 	/* Remove the fdb entry for this port/mac address */
390 	(void) vsw_del_fdb(vswp, port);
391 
392 	/* Remove any multicast addresses.. */
393 	vsw_del_mcst_port(port);
394 
395 	/*
396 	 * No longer need to hold writer lock on port list now
397 	 * that we have unlinked the target port from the list.
398 	 */
399 	RW_EXIT(&plist->lockrw);
400 
401 	/* Remove address if was programmed into HW. */
402 	mutex_enter(&vswp->hw_lock);
403 
404 	/*
405 	 * Port's address may not have been set in hardware. This could
406 	 * happen if the underlying physical device is not yet available and
407 	 * vsw_setup_switching_timeout() may be in progress.
408 	 * We remove its addr from hardware only if it has been set before.
409 	 */
410 	if (port->addr_set != VSW_ADDR_UNSET)
411 		(void) vsw_unset_hw(vswp, port, VSW_VNETPORT);
412 
413 	if (vswp->recfg_reqd)
414 		vsw_reconfig_hw(vswp);
415 
416 	mutex_exit(&vswp->hw_lock);
417 
418 	if (vsw_port_delete(port)) {
419 		return (1);
420 	}
421 
422 	D1(vswp, "%s: exit: p_instance(%d)", __func__, p_instance);
423 	return (0);
424 }
425 
426 /*
427  * Detach all active ports.
428  *
429  * Returns 0 on success, 1 on failure.
430  */
431 int
432 vsw_detach_ports(vsw_t *vswp)
433 {
434 	vsw_port_list_t 	*plist = &vswp->plist;
435 	vsw_port_t		*port = NULL;
436 
437 	D1(vswp, "%s: enter", __func__);
438 
439 	WRITE_ENTER(&plist->lockrw);
440 
441 	while ((port = plist->head) != NULL) {
442 		if (vsw_plist_del_node(vswp, port)) {
443 			DERR(vswp, "%s: Error deleting port %d"
444 			    " from port list", __func__, port->p_instance);
445 			RW_EXIT(&plist->lockrw);
446 			return (1);
447 		}
448 
449 		/* Remove address if was programmed into HW. */
450 		mutex_enter(&vswp->hw_lock);
451 		(void) vsw_unset_hw(vswp, port, VSW_VNETPORT);
452 		mutex_exit(&vswp->hw_lock);
453 
454 		/* Remove the fdb entry for this port/mac address */
455 		(void) vsw_del_fdb(vswp, port);
456 
457 		/* Remove any multicast addresses.. */
458 		vsw_del_mcst_port(port);
459 
460 		/*
461 		 * No longer need to hold the lock on the port list
462 		 * now that we have unlinked the target port from the
463 		 * list.
464 		 */
465 		RW_EXIT(&plist->lockrw);
466 		if (vsw_port_delete(port)) {
467 			DERR(vswp, "%s: Error deleting port %d",
468 			    __func__, port->p_instance);
469 			return (1);
470 		}
471 		WRITE_ENTER(&plist->lockrw);
472 	}
473 	RW_EXIT(&plist->lockrw);
474 
475 	D1(vswp, "%s: exit", __func__);
476 
477 	return (0);
478 }
479 
480 /*
481  * Delete the specified port.
482  *
483  * Returns 0 on success, 1 on failure.
484  */
485 static int
486 vsw_port_delete(vsw_port_t *port)
487 {
488 	vsw_ldc_list_t 		*ldcl;
489 	vsw_t			*vswp = port->p_vswp;
490 
491 	D1(vswp, "%s: enter : port id %d", __func__, port->p_instance);
492 
493 	(void) vsw_uninit_ldcs(port);
494 
495 	/*
496 	 * Wait for any pending ctrl msg tasks which reference this
497 	 * port to finish.
498 	 */
499 	if (vsw_drain_port_taskq(port))
500 		return (1);
501 
502 	/*
503 	 * Wait for port reference count to hit zero.
504 	 */
505 	while (port->ref_cnt != 0) {
506 		delay(drv_usectohz(VSW_PORT_REF_DELAY));
507 	}
508 
509 	/*
510 	 * Wait for any active callbacks to finish
511 	 */
512 	if (vsw_drain_ldcs(port))
513 		return (1);
514 
515 	ldcl = &port->p_ldclist;
516 	WRITE_ENTER(&ldcl->lockrw);
517 	while (ldcl->num_ldcs > 0) {
518 		if (vsw_ldc_detach(port, ldcl->head->ldc_id) != 0) {
519 			cmn_err(CE_WARN, "!vsw%d: unable to detach ldc %ld",
520 			    vswp->instance, ldcl->head->ldc_id);
521 			RW_EXIT(&ldcl->lockrw);
522 			return (1);
523 		}
524 	}
525 	RW_EXIT(&ldcl->lockrw);
526 
527 	rw_destroy(&port->p_ldclist.lockrw);
528 
529 	mutex_destroy(&port->mca_lock);
530 	mutex_destroy(&port->tx_lock);
531 	cv_destroy(&port->state_cv);
532 	mutex_destroy(&port->state_lock);
533 
534 	kmem_free(port, sizeof (vsw_port_t));
535 
536 	D1(vswp, "%s: exit", __func__);
537 
538 	return (0);
539 }
540 
541 /*
542  * Attach a logical domain channel (ldc) under a specified port.
543  *
544  * Returns 0 on success, 1 on failure.
545  */
546 static int
547 vsw_ldc_attach(vsw_port_t *port, uint64_t ldc_id)
548 {
549 	vsw_t 		*vswp = port->p_vswp;
550 	vsw_ldc_list_t *ldcl = &port->p_ldclist;
551 	vsw_ldc_t 	*ldcp = NULL;
552 	ldc_attr_t 	attr;
553 	ldc_status_t	istatus;
554 	int 		status = DDI_FAILURE;
555 	int		rv;
556 	char		kname[MAXNAMELEN];
557 	enum		{ PROG_init = 0x0, PROG_mblks = 0x1,
558 			    PROG_callback = 0x2, PROG_rx_thread = 0x4,
559 			    PROG_tx_thread = 0x8}
560 			progress;
561 
562 	progress = PROG_init;
563 
564 	D1(vswp, "%s: enter", __func__);
565 
566 	ldcp = kmem_zalloc(sizeof (vsw_ldc_t), KM_NOSLEEP);
567 	if (ldcp == NULL) {
568 		DERR(vswp, "%s: kmem_zalloc failed", __func__);
569 		return (1);
570 	}
571 	ldcp->ldc_id = ldc_id;
572 
573 	/* Allocate pools of receive mblks */
574 	rv = vio_init_multipools(&ldcp->vmp, VSW_NUM_VMPOOLS,
575 	    vsw_mblk_size1, vsw_mblk_size2, vsw_mblk_size3,
576 	    vsw_num_mblks1, vsw_num_mblks2, vsw_num_mblks3);
577 	if (rv) {
578 		DWARN(vswp, "%s: unable to create free mblk pools for"
579 		    " channel %ld (rv %d)", __func__, ldc_id, rv);
580 		kmem_free(ldcp, sizeof (vsw_ldc_t));
581 		return (1);
582 	}
583 
584 	progress |= PROG_mblks;
585 
586 	mutex_init(&ldcp->ldc_txlock, NULL, MUTEX_DRIVER, NULL);
587 	mutex_init(&ldcp->ldc_rxlock, NULL, MUTEX_DRIVER, NULL);
588 	mutex_init(&ldcp->ldc_cblock, NULL, MUTEX_DRIVER, NULL);
589 	mutex_init(&ldcp->drain_cv_lock, NULL, MUTEX_DRIVER, NULL);
590 	cv_init(&ldcp->drain_cv, NULL, CV_DRIVER, NULL);
591 	rw_init(&ldcp->lane_in.dlistrw, NULL, RW_DRIVER, NULL);
592 	rw_init(&ldcp->lane_out.dlistrw, NULL, RW_DRIVER, NULL);
593 
594 	/* required for handshake with peer */
595 	ldcp->local_session = (uint64_t)ddi_get_lbolt();
596 	ldcp->peer_session = 0;
597 	ldcp->session_status = 0;
598 	ldcp->hss_id = 1;	/* Initial handshake session id */
599 
600 	/* only set for outbound lane, inbound set by peer */
601 	vsw_set_lane_attr(vswp, &ldcp->lane_out);
602 
603 	attr.devclass = LDC_DEV_NT_SVC;
604 	attr.instance = ddi_get_instance(vswp->dip);
605 	attr.mode = LDC_MODE_UNRELIABLE;
606 	attr.mtu = VSW_LDC_MTU;
607 	status = ldc_init(ldc_id, &attr, &ldcp->ldc_handle);
608 	if (status != 0) {
609 		DERR(vswp, "%s(%lld): ldc_init failed, rv (%d)",
610 		    __func__, ldc_id, status);
611 		goto ldc_attach_fail;
612 	}
613 
614 	if (vsw_ldc_rxthr_enabled) {
615 		ldcp->rx_thr_flags = 0;
616 
617 		mutex_init(&ldcp->rx_thr_lock, NULL, MUTEX_DRIVER, NULL);
618 		cv_init(&ldcp->rx_thr_cv, NULL, CV_DRIVER, NULL);
619 		ldcp->rx_thread = thread_create(NULL, 2 * DEFAULTSTKSZ,
620 		    vsw_ldc_rx_worker, ldcp, 0, &p0, TS_RUN, maxclsyspri);
621 
622 		progress |= PROG_rx_thread;
623 		if (ldcp->rx_thread == NULL) {
624 			DWARN(vswp, "%s(%lld): Failed to create worker thread",
625 			    __func__, ldc_id);
626 			goto ldc_attach_fail;
627 		}
628 	}
629 
630 	if (vsw_ldc_txthr_enabled) {
631 		ldcp->tx_thr_flags = 0;
632 		ldcp->tx_mhead = ldcp->tx_mtail = NULL;
633 
634 		mutex_init(&ldcp->tx_thr_lock, NULL, MUTEX_DRIVER, NULL);
635 		cv_init(&ldcp->tx_thr_cv, NULL, CV_DRIVER, NULL);
636 		ldcp->tx_thread = thread_create(NULL, 2 * DEFAULTSTKSZ,
637 		    vsw_ldc_tx_worker, ldcp, 0, &p0, TS_RUN, maxclsyspri);
638 
639 		progress |= PROG_tx_thread;
640 		if (ldcp->tx_thread == NULL) {
641 			DWARN(vswp, "%s(%lld): Failed to create worker thread",
642 			    __func__, ldc_id);
643 			goto ldc_attach_fail;
644 		}
645 	}
646 
647 	status = ldc_reg_callback(ldcp->ldc_handle, vsw_ldc_cb, (caddr_t)ldcp);
648 	if (status != 0) {
649 		DERR(vswp, "%s(%lld): ldc_reg_callback failed, rv (%d)",
650 		    __func__, ldc_id, status);
651 		(void) ldc_fini(ldcp->ldc_handle);
652 		goto ldc_attach_fail;
653 	}
654 
655 	progress |= PROG_callback;
656 
657 	mutex_init(&ldcp->status_lock, NULL, MUTEX_DRIVER, NULL);
658 
659 	if (ldc_status(ldcp->ldc_handle, &istatus) != 0) {
660 		DERR(vswp, "%s: ldc_status failed", __func__);
661 		mutex_destroy(&ldcp->status_lock);
662 		goto ldc_attach_fail;
663 	}
664 
665 	ldcp->ldc_status = istatus;
666 	ldcp->ldc_port = port;
667 	ldcp->ldc_vswp = vswp;
668 
669 	(void) sprintf(kname, "%sldc0x%lx", DRV_NAME, ldcp->ldc_id);
670 	ldcp->ksp = vgen_setup_kstats(DRV_NAME, vswp->instance,
671 	    kname, &ldcp->ldc_stats);
672 	if (ldcp->ksp == NULL) {
673 		DERR(vswp, "%s: kstats setup failed", __func__);
674 		goto ldc_attach_fail;
675 	}
676 
677 	/* link it into the list of channels for this port */
678 	WRITE_ENTER(&ldcl->lockrw);
679 	ldcp->ldc_next = ldcl->head;
680 	ldcl->head = ldcp;
681 	ldcl->num_ldcs++;
682 	RW_EXIT(&ldcl->lockrw);
683 
684 	D1(vswp, "%s: exit", __func__);
685 	return (0);
686 
687 ldc_attach_fail:
688 
689 	if (progress & PROG_callback) {
690 		(void) ldc_unreg_callback(ldcp->ldc_handle);
691 	}
692 
693 	if (progress & PROG_rx_thread) {
694 		if (ldcp->rx_thread != NULL) {
695 			vsw_stop_rx_thread(ldcp);
696 		}
697 		mutex_destroy(&ldcp->rx_thr_lock);
698 		cv_destroy(&ldcp->rx_thr_cv);
699 	}
700 
701 	if (progress & PROG_tx_thread) {
702 		if (ldcp->tx_thread != NULL) {
703 			vsw_stop_tx_thread(ldcp);
704 		}
705 		mutex_destroy(&ldcp->tx_thr_lock);
706 		cv_destroy(&ldcp->tx_thr_cv);
707 	}
708 	if (ldcp->ksp != NULL) {
709 		vgen_destroy_kstats(ldcp->ksp);
710 	}
711 	mutex_destroy(&ldcp->ldc_txlock);
712 	mutex_destroy(&ldcp->ldc_rxlock);
713 	mutex_destroy(&ldcp->ldc_cblock);
714 	mutex_destroy(&ldcp->drain_cv_lock);
715 
716 	cv_destroy(&ldcp->drain_cv);
717 
718 	rw_destroy(&ldcp->lane_in.dlistrw);
719 	rw_destroy(&ldcp->lane_out.dlistrw);
720 
721 	if (progress & PROG_mblks) {
722 		vio_destroy_multipools(&ldcp->vmp, &vswp->rxh);
723 	}
724 	kmem_free(ldcp, sizeof (vsw_ldc_t));
725 
726 	return (1);
727 }
728 
729 /*
730  * Detach a logical domain channel (ldc) belonging to a
731  * particular port.
732  *
733  * Returns 0 on success, 1 on failure.
734  */
735 static int
736 vsw_ldc_detach(vsw_port_t *port, uint64_t ldc_id)
737 {
738 	vsw_t 		*vswp = port->p_vswp;
739 	vsw_ldc_t 	*ldcp, *prev_ldcp;
740 	vsw_ldc_list_t	*ldcl = &port->p_ldclist;
741 	int 		rv;
742 
743 	prev_ldcp = ldcl->head;
744 	for (; (ldcp = prev_ldcp) != NULL; prev_ldcp = ldcp->ldc_next) {
745 		if (ldcp->ldc_id == ldc_id) {
746 			break;
747 		}
748 	}
749 
750 	/* specified ldc id not found */
751 	if (ldcp == NULL) {
752 		DERR(vswp, "%s: ldcp = NULL", __func__);
753 		return (1);
754 	}
755 
756 	D2(vswp, "%s: detaching channel %lld", __func__, ldcp->ldc_id);
757 
758 	/* Stop the receive thread */
759 	if (ldcp->rx_thread != NULL) {
760 		vsw_stop_rx_thread(ldcp);
761 		mutex_destroy(&ldcp->rx_thr_lock);
762 		cv_destroy(&ldcp->rx_thr_cv);
763 	}
764 
765 	/* Stop the tx thread */
766 	if (ldcp->tx_thread != NULL) {
767 		vsw_stop_tx_thread(ldcp);
768 		mutex_destroy(&ldcp->tx_thr_lock);
769 		cv_destroy(&ldcp->tx_thr_cv);
770 		if (ldcp->tx_mhead != NULL) {
771 			freemsgchain(ldcp->tx_mhead);
772 			ldcp->tx_mhead = ldcp->tx_mtail = NULL;
773 		}
774 	}
775 
776 	/* Destory kstats */
777 	vgen_destroy_kstats(ldcp->ksp);
778 
779 	/*
780 	 * Before we can close the channel we must release any mapped
781 	 * resources (e.g. drings).
782 	 */
783 	vsw_free_lane_resources(ldcp, INBOUND);
784 	vsw_free_lane_resources(ldcp, OUTBOUND);
785 
786 	/*
787 	 * If the close fails we are in serious trouble, as won't
788 	 * be able to delete the parent port.
789 	 */
790 	if ((rv = ldc_close(ldcp->ldc_handle)) != 0) {
791 		DERR(vswp, "%s: error %d closing channel %lld",
792 		    __func__, rv, ldcp->ldc_id);
793 		return (1);
794 	}
795 
796 	(void) ldc_fini(ldcp->ldc_handle);
797 
798 	ldcp->ldc_status = LDC_INIT;
799 	ldcp->ldc_handle = NULL;
800 	ldcp->ldc_vswp = NULL;
801 
802 
803 	/*
804 	 * Most likely some mblks are still in use and
805 	 * have not been returned to the pool. These mblks are
806 	 * added to the pool that is maintained in the device instance.
807 	 * Another attempt will be made to destroy the pool
808 	 * when the device detaches.
809 	 */
810 	vio_destroy_multipools(&ldcp->vmp, &vswp->rxh);
811 
812 	/* unlink it from the list */
813 	prev_ldcp = ldcp->ldc_next;
814 	ldcl->num_ldcs--;
815 
816 	mutex_destroy(&ldcp->ldc_txlock);
817 	mutex_destroy(&ldcp->ldc_rxlock);
818 	mutex_destroy(&ldcp->ldc_cblock);
819 	cv_destroy(&ldcp->drain_cv);
820 	mutex_destroy(&ldcp->drain_cv_lock);
821 	mutex_destroy(&ldcp->status_lock);
822 	rw_destroy(&ldcp->lane_in.dlistrw);
823 	rw_destroy(&ldcp->lane_out.dlistrw);
824 
825 	kmem_free(ldcp, sizeof (vsw_ldc_t));
826 
827 	return (0);
828 }
829 
830 /*
831  * Open and attempt to bring up the channel. Note that channel
832  * can only be brought up if peer has also opened channel.
833  *
834  * Returns 0 if can open and bring up channel, otherwise
835  * returns 1.
836  */
837 static int
838 vsw_ldc_init(vsw_ldc_t *ldcp)
839 {
840 	vsw_t 		*vswp = ldcp->ldc_vswp;
841 	ldc_status_t	istatus = 0;
842 	int		rv;
843 
844 	D1(vswp, "%s: enter", __func__);
845 
846 	LDC_ENTER_LOCK(ldcp);
847 
848 	/* don't start at 0 in case clients don't like that */
849 	ldcp->next_ident = 1;
850 
851 	rv = ldc_open(ldcp->ldc_handle);
852 	if (rv != 0) {
853 		DERR(vswp, "%s: ldc_open failed: id(%lld) rv(%d)",
854 		    __func__, ldcp->ldc_id, rv);
855 		LDC_EXIT_LOCK(ldcp);
856 		return (1);
857 	}
858 
859 	if (ldc_status(ldcp->ldc_handle, &istatus) != 0) {
860 		DERR(vswp, "%s: unable to get status", __func__);
861 		LDC_EXIT_LOCK(ldcp);
862 		return (1);
863 
864 	} else if (istatus != LDC_OPEN && istatus != LDC_READY) {
865 		DERR(vswp, "%s: id (%lld) status(%d) is not OPEN/READY",
866 		    __func__, ldcp->ldc_id, istatus);
867 		LDC_EXIT_LOCK(ldcp);
868 		return (1);
869 	}
870 
871 	mutex_enter(&ldcp->status_lock);
872 	ldcp->ldc_status = istatus;
873 	mutex_exit(&ldcp->status_lock);
874 
875 	rv = ldc_up(ldcp->ldc_handle);
876 	if (rv != 0) {
877 		/*
878 		 * Not a fatal error for ldc_up() to fail, as peer
879 		 * end point may simply not be ready yet.
880 		 */
881 		D2(vswp, "%s: ldc_up err id(%lld) rv(%d)", __func__,
882 		    ldcp->ldc_id, rv);
883 		LDC_EXIT_LOCK(ldcp);
884 		return (1);
885 	}
886 
887 	/*
888 	 * ldc_up() call is non-blocking so need to explicitly
889 	 * check channel status to see if in fact the channel
890 	 * is UP.
891 	 */
892 	mutex_enter(&ldcp->status_lock);
893 	if (ldc_status(ldcp->ldc_handle, &ldcp->ldc_status) != 0) {
894 		DERR(vswp, "%s: unable to get status", __func__);
895 		mutex_exit(&ldcp->status_lock);
896 		LDC_EXIT_LOCK(ldcp);
897 		return (1);
898 
899 	}
900 
901 	if (ldcp->ldc_status == LDC_UP) {
902 		D2(vswp, "%s: channel %ld now UP (%ld)", __func__,
903 		    ldcp->ldc_id, istatus);
904 		mutex_exit(&ldcp->status_lock);
905 		LDC_EXIT_LOCK(ldcp);
906 
907 		vsw_process_conn_evt(ldcp, VSW_CONN_UP);
908 		return (0);
909 	}
910 
911 	mutex_exit(&ldcp->status_lock);
912 	LDC_EXIT_LOCK(ldcp);
913 
914 	D1(vswp, "%s: exit", __func__);
915 	return (0);
916 }
917 
918 /* disable callbacks on the channel */
919 static int
920 vsw_ldc_uninit(vsw_ldc_t *ldcp)
921 {
922 	vsw_t	*vswp = ldcp->ldc_vswp;
923 	int	rv;
924 
925 	D1(vswp, "vsw_ldc_uninit: enter: id(%lx)\n", ldcp->ldc_id);
926 
927 	LDC_ENTER_LOCK(ldcp);
928 
929 	rv = ldc_set_cb_mode(ldcp->ldc_handle, LDC_CB_DISABLE);
930 	if (rv != 0) {
931 		DERR(vswp, "vsw_ldc_uninit(%lld): error disabling "
932 		    "interrupts (rv = %d)\n", ldcp->ldc_id, rv);
933 		LDC_EXIT_LOCK(ldcp);
934 		return (1);
935 	}
936 
937 	mutex_enter(&ldcp->status_lock);
938 	ldcp->ldc_status = LDC_INIT;
939 	mutex_exit(&ldcp->status_lock);
940 
941 	LDC_EXIT_LOCK(ldcp);
942 
943 	D1(vswp, "vsw_ldc_uninit: exit: id(%lx)", ldcp->ldc_id);
944 
945 	return (0);
946 }
947 
948 static int
949 vsw_init_ldcs(vsw_port_t *port)
950 {
951 	vsw_ldc_list_t	*ldcl = &port->p_ldclist;
952 	vsw_ldc_t	*ldcp;
953 
954 	READ_ENTER(&ldcl->lockrw);
955 	ldcp =  ldcl->head;
956 	for (; ldcp  != NULL; ldcp = ldcp->ldc_next) {
957 		(void) vsw_ldc_init(ldcp);
958 	}
959 	RW_EXIT(&ldcl->lockrw);
960 
961 	return (0);
962 }
963 
964 static int
965 vsw_uninit_ldcs(vsw_port_t *port)
966 {
967 	vsw_ldc_list_t	*ldcl = &port->p_ldclist;
968 	vsw_ldc_t	*ldcp;
969 
970 	D1(NULL, "vsw_uninit_ldcs: enter\n");
971 
972 	READ_ENTER(&ldcl->lockrw);
973 	ldcp =  ldcl->head;
974 	for (; ldcp  != NULL; ldcp = ldcp->ldc_next) {
975 		(void) vsw_ldc_uninit(ldcp);
976 	}
977 	RW_EXIT(&ldcl->lockrw);
978 
979 	D1(NULL, "vsw_uninit_ldcs: exit\n");
980 
981 	return (0);
982 }
983 
984 /*
985  * Wait until the callback(s) associated with the ldcs under the specified
986  * port have completed.
987  *
988  * Prior to this function being invoked each channel under this port
989  * should have been quiesced via ldc_set_cb_mode(DISABLE).
990  *
991  * A short explaination of what we are doing below..
992  *
993  * The simplest approach would be to have a reference counter in
994  * the ldc structure which is increment/decremented by the callbacks as
995  * they use the channel. The drain function could then simply disable any
996  * further callbacks and do a cv_wait for the ref to hit zero. Unfortunately
997  * there is a tiny window here - before the callback is able to get the lock
998  * on the channel it is interrupted and this function gets to execute. It
999  * sees that the ref count is zero and believes its free to delete the
1000  * associated data structures.
1001  *
1002  * We get around this by taking advantage of the fact that before the ldc
1003  * framework invokes a callback it sets a flag to indicate that there is a
1004  * callback active (or about to become active). If when we attempt to
1005  * unregister a callback when this active flag is set then the unregister
1006  * will fail with EWOULDBLOCK.
1007  *
1008  * If the unregister fails we do a cv_timedwait. We will either be signaled
1009  * by the callback as it is exiting (note we have to wait a short period to
1010  * allow the callback to return fully to the ldc framework and it to clear
1011  * the active flag), or by the timer expiring. In either case we again attempt
1012  * the unregister. We repeat this until we can succesfully unregister the
1013  * callback.
1014  *
1015  * The reason we use a cv_timedwait rather than a simple cv_wait is to catch
1016  * the case where the callback has finished but the ldc framework has not yet
1017  * cleared the active flag. In this case we would never get a cv_signal.
1018  */
1019 static int
1020 vsw_drain_ldcs(vsw_port_t *port)
1021 {
1022 	vsw_ldc_list_t	*ldcl = &port->p_ldclist;
1023 	vsw_ldc_t	*ldcp;
1024 	vsw_t		*vswp = port->p_vswp;
1025 
1026 	D1(vswp, "%s: enter", __func__);
1027 
1028 	READ_ENTER(&ldcl->lockrw);
1029 
1030 	ldcp = ldcl->head;
1031 
1032 	for (; ldcp  != NULL; ldcp = ldcp->ldc_next) {
1033 		/*
1034 		 * If we can unregister the channel callback then we
1035 		 * know that there is no callback either running or
1036 		 * scheduled to run for this channel so move on to next
1037 		 * channel in the list.
1038 		 */
1039 		mutex_enter(&ldcp->drain_cv_lock);
1040 
1041 		/* prompt active callbacks to quit */
1042 		ldcp->drain_state = VSW_LDC_DRAINING;
1043 
1044 		if ((ldc_unreg_callback(ldcp->ldc_handle)) == 0) {
1045 			D2(vswp, "%s: unreg callback for chan %ld", __func__,
1046 			    ldcp->ldc_id);
1047 			mutex_exit(&ldcp->drain_cv_lock);
1048 			continue;
1049 		} else {
1050 			/*
1051 			 * If we end up here we know that either 1) a callback
1052 			 * is currently executing, 2) is about to start (i.e.
1053 			 * the ldc framework has set the active flag but
1054 			 * has not actually invoked the callback yet, or 3)
1055 			 * has finished and has returned to the ldc framework
1056 			 * but the ldc framework has not yet cleared the
1057 			 * active bit.
1058 			 *
1059 			 * Wait for it to finish.
1060 			 */
1061 			while (ldc_unreg_callback(ldcp->ldc_handle)
1062 			    == EWOULDBLOCK)
1063 				(void) cv_timedwait(&ldcp->drain_cv,
1064 				    &ldcp->drain_cv_lock, lbolt + hz);
1065 
1066 			mutex_exit(&ldcp->drain_cv_lock);
1067 			D2(vswp, "%s: unreg callback for chan %ld after "
1068 			    "timeout", __func__, ldcp->ldc_id);
1069 		}
1070 	}
1071 	RW_EXIT(&ldcl->lockrw);
1072 
1073 	D1(vswp, "%s: exit", __func__);
1074 	return (0);
1075 }
1076 
1077 /*
1078  * Wait until all tasks which reference this port have completed.
1079  *
1080  * Prior to this function being invoked each channel under this port
1081  * should have been quiesced via ldc_set_cb_mode(DISABLE).
1082  */
1083 static int
1084 vsw_drain_port_taskq(vsw_port_t *port)
1085 {
1086 	vsw_t		*vswp = port->p_vswp;
1087 
1088 	D1(vswp, "%s: enter", __func__);
1089 
1090 	/*
1091 	 * Mark the port as in the process of being detached, and
1092 	 * dispatch a marker task to the queue so we know when all
1093 	 * relevant tasks have completed.
1094 	 */
1095 	mutex_enter(&port->state_lock);
1096 	port->state = VSW_PORT_DETACHING;
1097 
1098 	if ((vswp->taskq_p == NULL) ||
1099 	    (ddi_taskq_dispatch(vswp->taskq_p, vsw_marker_task,
1100 	    port, DDI_NOSLEEP) != DDI_SUCCESS)) {
1101 		DERR(vswp, "%s: unable to dispatch marker task",
1102 		    __func__);
1103 		mutex_exit(&port->state_lock);
1104 		return (1);
1105 	}
1106 
1107 	/*
1108 	 * Wait for the marker task to finish.
1109 	 */
1110 	while (port->state != VSW_PORT_DETACHABLE)
1111 		cv_wait(&port->state_cv, &port->state_lock);
1112 
1113 	mutex_exit(&port->state_lock);
1114 
1115 	D1(vswp, "%s: exit", __func__);
1116 
1117 	return (0);
1118 }
1119 
1120 static void
1121 vsw_marker_task(void *arg)
1122 {
1123 	vsw_port_t	*port = arg;
1124 	vsw_t		*vswp = port->p_vswp;
1125 
1126 	D1(vswp, "%s: enter", __func__);
1127 
1128 	mutex_enter(&port->state_lock);
1129 
1130 	/*
1131 	 * No further tasks should be dispatched which reference
1132 	 * this port so ok to mark it as safe to detach.
1133 	 */
1134 	port->state = VSW_PORT_DETACHABLE;
1135 
1136 	cv_signal(&port->state_cv);
1137 
1138 	mutex_exit(&port->state_lock);
1139 
1140 	D1(vswp, "%s: exit", __func__);
1141 }
1142 
1143 vsw_port_t *
1144 vsw_lookup_port(vsw_t *vswp, int p_instance)
1145 {
1146 	vsw_port_list_t *plist = &vswp->plist;
1147 	vsw_port_t	*port;
1148 
1149 	for (port = plist->head; port != NULL; port = port->p_next) {
1150 		if (port->p_instance == p_instance) {
1151 			D2(vswp, "vsw_lookup_port: found p_instance\n");
1152 			return (port);
1153 		}
1154 	}
1155 
1156 	return (NULL);
1157 }
1158 
1159 /*
1160  * Search for and remove the specified port from the port
1161  * list. Returns 0 if able to locate and remove port, otherwise
1162  * returns 1.
1163  */
1164 static int
1165 vsw_plist_del_node(vsw_t *vswp, vsw_port_t *port)
1166 {
1167 	vsw_port_list_t *plist = &vswp->plist;
1168 	vsw_port_t	*curr_p, *prev_p;
1169 
1170 	if (plist->head == NULL)
1171 		return (1);
1172 
1173 	curr_p = prev_p = plist->head;
1174 
1175 	while (curr_p != NULL) {
1176 		if (curr_p == port) {
1177 			if (prev_p == curr_p) {
1178 				plist->head = curr_p->p_next;
1179 			} else {
1180 				prev_p->p_next = curr_p->p_next;
1181 			}
1182 			plist->num_ports--;
1183 			break;
1184 		} else {
1185 			prev_p = curr_p;
1186 			curr_p = curr_p->p_next;
1187 		}
1188 	}
1189 	return (0);
1190 }
1191 
1192 /*
1193  * Interrupt handler for ldc messages.
1194  */
1195 static uint_t
1196 vsw_ldc_cb(uint64_t event, caddr_t arg)
1197 {
1198 	vsw_ldc_t	*ldcp = (vsw_ldc_t  *)arg;
1199 	vsw_t 		*vswp = ldcp->ldc_vswp;
1200 
1201 	D1(vswp, "%s: enter: ldcid (%lld)\n", __func__, ldcp->ldc_id);
1202 
1203 	mutex_enter(&ldcp->ldc_cblock);
1204 	ldcp->ldc_stats.callbacks++;
1205 
1206 	mutex_enter(&ldcp->status_lock);
1207 	if ((ldcp->ldc_status == LDC_INIT) || (ldcp->ldc_handle == NULL)) {
1208 		mutex_exit(&ldcp->status_lock);
1209 		mutex_exit(&ldcp->ldc_cblock);
1210 		return (LDC_SUCCESS);
1211 	}
1212 	mutex_exit(&ldcp->status_lock);
1213 
1214 	if (event & LDC_EVT_UP) {
1215 		/*
1216 		 * Channel has come up.
1217 		 */
1218 		D2(vswp, "%s: id(%ld) event(%llx) UP: status(%ld)",
1219 		    __func__, ldcp->ldc_id, event, ldcp->ldc_status);
1220 
1221 		vsw_process_conn_evt(ldcp, VSW_CONN_UP);
1222 
1223 		ASSERT((event & (LDC_EVT_RESET | LDC_EVT_DOWN)) == 0);
1224 	}
1225 
1226 	if (event & LDC_EVT_READ) {
1227 		/*
1228 		 * Data available for reading.
1229 		 */
1230 		D2(vswp, "%s: id(ld) event(%llx) data READ",
1231 		    __func__, ldcp->ldc_id, event);
1232 
1233 		if (ldcp->rx_thread != NULL) {
1234 			/*
1235 			 * If the receive thread is enabled, then
1236 			 * wakeup the receive thread to process the
1237 			 * LDC messages.
1238 			 */
1239 			mutex_exit(&ldcp->ldc_cblock);
1240 			mutex_enter(&ldcp->rx_thr_lock);
1241 			if (!(ldcp->rx_thr_flags & VSW_WTHR_DATARCVD)) {
1242 				ldcp->rx_thr_flags |= VSW_WTHR_DATARCVD;
1243 				cv_signal(&ldcp->rx_thr_cv);
1244 			}
1245 			mutex_exit(&ldcp->rx_thr_lock);
1246 			mutex_enter(&ldcp->ldc_cblock);
1247 		} else {
1248 			vsw_process_pkt(ldcp);
1249 		}
1250 
1251 		ASSERT((event & (LDC_EVT_RESET | LDC_EVT_DOWN)) == 0);
1252 
1253 		goto vsw_cb_exit;
1254 	}
1255 
1256 	if (event & (LDC_EVT_DOWN | LDC_EVT_RESET)) {
1257 		D2(vswp, "%s: id(%ld) event (%lx) DOWN/RESET: status(%ld)",
1258 		    __func__, ldcp->ldc_id, event, ldcp->ldc_status);
1259 
1260 		vsw_process_conn_evt(ldcp, VSW_CONN_RESET);
1261 	}
1262 
1263 	/*
1264 	 * Catch either LDC_EVT_WRITE which we don't support or any
1265 	 * unknown event.
1266 	 */
1267 	if (event &
1268 	    ~(LDC_EVT_UP | LDC_EVT_RESET | LDC_EVT_DOWN | LDC_EVT_READ)) {
1269 		DERR(vswp, "%s: id(%ld) Unexpected event=(%llx) status(%ld)",
1270 		    __func__, ldcp->ldc_id, event, ldcp->ldc_status);
1271 	}
1272 
1273 vsw_cb_exit:
1274 	mutex_exit(&ldcp->ldc_cblock);
1275 
1276 	/*
1277 	 * Let the drain function know we are finishing if it
1278 	 * is waiting.
1279 	 */
1280 	mutex_enter(&ldcp->drain_cv_lock);
1281 	if (ldcp->drain_state == VSW_LDC_DRAINING)
1282 		cv_signal(&ldcp->drain_cv);
1283 	mutex_exit(&ldcp->drain_cv_lock);
1284 
1285 	return (LDC_SUCCESS);
1286 }
1287 
1288 /*
1289  * Reinitialise data structures associated with the channel.
1290  */
1291 static void
1292 vsw_ldc_reinit(vsw_ldc_t *ldcp)
1293 {
1294 	vsw_t		*vswp = ldcp->ldc_vswp;
1295 	vsw_port_t	*port;
1296 	vsw_ldc_list_t	*ldcl;
1297 
1298 	D1(vswp, "%s: enter", __func__);
1299 
1300 	port = ldcp->ldc_port;
1301 	ldcl = &port->p_ldclist;
1302 
1303 	READ_ENTER(&ldcl->lockrw);
1304 
1305 	D2(vswp, "%s: in 0x%llx : out 0x%llx", __func__,
1306 	    ldcp->lane_in.lstate, ldcp->lane_out.lstate);
1307 
1308 	vsw_free_lane_resources(ldcp, INBOUND);
1309 	vsw_free_lane_resources(ldcp, OUTBOUND);
1310 	RW_EXIT(&ldcl->lockrw);
1311 
1312 	ldcp->lane_in.lstate = 0;
1313 	ldcp->lane_out.lstate = 0;
1314 
1315 	/*
1316 	 * Remove parent port from any multicast groups
1317 	 * it may have registered with. Client must resend
1318 	 * multicast add command after handshake completes.
1319 	 */
1320 	(void) vsw_del_fdb(vswp, port);
1321 
1322 	vsw_del_mcst_port(port);
1323 
1324 	ldcp->peer_session = 0;
1325 	ldcp->session_status = 0;
1326 	ldcp->hcnt = 0;
1327 	ldcp->hphase = VSW_MILESTONE0;
1328 	ldcp->tx_failures = 0;
1329 
1330 	D1(vswp, "%s: exit", __func__);
1331 }
1332 
1333 /*
1334  * Process a connection event.
1335  *
1336  * Note - care must be taken to ensure that this function is
1337  * not called with the dlistrw lock held.
1338  */
1339 static void
1340 vsw_process_conn_evt(vsw_ldc_t *ldcp, uint16_t evt)
1341 {
1342 	vsw_t		*vswp = ldcp->ldc_vswp;
1343 	vsw_conn_evt_t	*conn = NULL;
1344 
1345 	D1(vswp, "%s: enter", __func__);
1346 
1347 	/*
1348 	 * Check if either a reset or restart event is pending
1349 	 * or in progress. If so just return.
1350 	 *
1351 	 * A VSW_CONN_RESET event originates either with a LDC_RESET_EVT
1352 	 * being received by the callback handler, or a ECONNRESET error
1353 	 * code being returned from a ldc_read() or ldc_write() call.
1354 	 *
1355 	 * A VSW_CONN_RESTART event occurs when some error checking code
1356 	 * decides that there is a problem with data from the channel,
1357 	 * and that the handshake should be restarted.
1358 	 */
1359 	if (((evt == VSW_CONN_RESET) || (evt == VSW_CONN_RESTART)) &&
1360 	    (ldstub((uint8_t *)&ldcp->reset_active)))
1361 		return;
1362 
1363 	/*
1364 	 * If it is an LDC_UP event we first check the recorded
1365 	 * state of the channel. If this is UP then we know that
1366 	 * the channel moving to the UP state has already been dealt
1367 	 * with and don't need to dispatch a  new task.
1368 	 *
1369 	 * The reason for this check is that when we do a ldc_up(),
1370 	 * depending on the state of the peer, we may or may not get
1371 	 * a LDC_UP event. As we can't depend on getting a LDC_UP evt
1372 	 * every time we do ldc_up() we explicitly check the channel
1373 	 * status to see has it come up (ldc_up() is asynch and will
1374 	 * complete at some undefined time), and take the appropriate
1375 	 * action.
1376 	 *
1377 	 * The flip side of this is that we may get a LDC_UP event
1378 	 * when we have already seen that the channel is up and have
1379 	 * dealt with that.
1380 	 */
1381 	mutex_enter(&ldcp->status_lock);
1382 	if (evt == VSW_CONN_UP) {
1383 		if ((ldcp->ldc_status == LDC_UP) || (ldcp->reset_active != 0)) {
1384 			mutex_exit(&ldcp->status_lock);
1385 			return;
1386 		}
1387 	}
1388 	mutex_exit(&ldcp->status_lock);
1389 
1390 	/*
1391 	 * The transaction group id allows us to identify and discard
1392 	 * any tasks which are still pending on the taskq and refer
1393 	 * to the handshake session we are about to restart or reset.
1394 	 * These stale messages no longer have any real meaning.
1395 	 */
1396 	(void) atomic_inc_32(&ldcp->hss_id);
1397 
1398 	ASSERT(vswp->taskq_p != NULL);
1399 
1400 	if ((conn = kmem_zalloc(sizeof (vsw_conn_evt_t), KM_NOSLEEP)) == NULL) {
1401 		cmn_err(CE_WARN, "!vsw%d: unable to allocate memory for"
1402 		    " connection event", vswp->instance);
1403 		goto err_exit;
1404 	}
1405 
1406 	conn->evt = evt;
1407 	conn->ldcp = ldcp;
1408 
1409 	if (ddi_taskq_dispatch(vswp->taskq_p, vsw_conn_task, conn,
1410 	    DDI_NOSLEEP) != DDI_SUCCESS) {
1411 		cmn_err(CE_WARN, "!vsw%d: Can't dispatch connection task",
1412 		    vswp->instance);
1413 
1414 		kmem_free(conn, sizeof (vsw_conn_evt_t));
1415 		goto err_exit;
1416 	}
1417 
1418 	D1(vswp, "%s: exit", __func__);
1419 	return;
1420 
1421 err_exit:
1422 	/*
1423 	 * Have mostly likely failed due to memory shortage. Clear the flag so
1424 	 * that future requests will at least be attempted and will hopefully
1425 	 * succeed.
1426 	 */
1427 	if ((evt == VSW_CONN_RESET) || (evt == VSW_CONN_RESTART))
1428 		ldcp->reset_active = 0;
1429 }
1430 
1431 /*
1432  * Deal with events relating to a connection. Invoked from a taskq.
1433  */
1434 static void
1435 vsw_conn_task(void *arg)
1436 {
1437 	vsw_conn_evt_t	*conn = (vsw_conn_evt_t *)arg;
1438 	vsw_ldc_t	*ldcp = NULL;
1439 	vsw_t		*vswp = NULL;
1440 	uint16_t	evt;
1441 	ldc_status_t	curr_status;
1442 
1443 	ldcp = conn->ldcp;
1444 	evt = conn->evt;
1445 	vswp = ldcp->ldc_vswp;
1446 
1447 	D1(vswp, "%s: enter", __func__);
1448 
1449 	/* can safely free now have copied out data */
1450 	kmem_free(conn, sizeof (vsw_conn_evt_t));
1451 
1452 	mutex_enter(&ldcp->status_lock);
1453 	if (ldc_status(ldcp->ldc_handle, &curr_status) != 0) {
1454 		cmn_err(CE_WARN, "!vsw%d: Unable to read status of "
1455 		    "channel %ld", vswp->instance, ldcp->ldc_id);
1456 		mutex_exit(&ldcp->status_lock);
1457 		return;
1458 	}
1459 
1460 	/*
1461 	 * If we wish to restart the handshake on this channel, then if
1462 	 * the channel is UP we bring it DOWN to flush the underlying
1463 	 * ldc queue.
1464 	 */
1465 	if ((evt == VSW_CONN_RESTART) && (curr_status == LDC_UP))
1466 		(void) ldc_down(ldcp->ldc_handle);
1467 
1468 	/*
1469 	 * re-init all the associated data structures.
1470 	 */
1471 	vsw_ldc_reinit(ldcp);
1472 
1473 	/*
1474 	 * Bring the channel back up (note it does no harm to
1475 	 * do this even if the channel is already UP, Just
1476 	 * becomes effectively a no-op).
1477 	 */
1478 	(void) ldc_up(ldcp->ldc_handle);
1479 
1480 	/*
1481 	 * Check if channel is now UP. This will only happen if
1482 	 * peer has also done a ldc_up().
1483 	 */
1484 	if (ldc_status(ldcp->ldc_handle, &curr_status) != 0) {
1485 		cmn_err(CE_WARN, "!vsw%d: Unable to read status of "
1486 		    "channel %ld", vswp->instance, ldcp->ldc_id);
1487 		mutex_exit(&ldcp->status_lock);
1488 		return;
1489 	}
1490 
1491 	ldcp->ldc_status = curr_status;
1492 
1493 	/* channel UP so restart handshake by sending version info */
1494 	if (curr_status == LDC_UP) {
1495 		if (ldcp->hcnt++ > vsw_num_handshakes) {
1496 			cmn_err(CE_WARN, "!vsw%d: exceeded number of permitted"
1497 			    " handshake attempts (%d) on channel %ld",
1498 			    vswp->instance, ldcp->hcnt, ldcp->ldc_id);
1499 			mutex_exit(&ldcp->status_lock);
1500 			return;
1501 		}
1502 
1503 		if (ddi_taskq_dispatch(vswp->taskq_p, vsw_send_ver, ldcp,
1504 		    DDI_NOSLEEP) != DDI_SUCCESS) {
1505 			cmn_err(CE_WARN, "!vsw%d: Can't dispatch version task",
1506 			    vswp->instance);
1507 
1508 			/*
1509 			 * Don't count as valid restart attempt if couldn't
1510 			 * send version msg.
1511 			 */
1512 			if (ldcp->hcnt > 0)
1513 				ldcp->hcnt--;
1514 		}
1515 	}
1516 
1517 	/*
1518 	 * Mark that the process is complete by clearing the flag.
1519 	 *
1520 	 * Note is it possible that the taskq dispatch above may have failed,
1521 	 * most likely due to memory shortage. We still clear the flag so
1522 	 * future attempts will at least be attempted and will hopefully
1523 	 * succeed.
1524 	 */
1525 	if ((evt == VSW_CONN_RESET) || (evt == VSW_CONN_RESTART))
1526 		ldcp->reset_active = 0;
1527 
1528 	mutex_exit(&ldcp->status_lock);
1529 
1530 	D1(vswp, "%s: exit", __func__);
1531 }
1532 
1533 /*
1534  * returns 0 if legal for event signified by flag to have
1535  * occured at the time it did. Otherwise returns 1.
1536  */
1537 int
1538 vsw_check_flag(vsw_ldc_t *ldcp, int dir, uint64_t flag)
1539 {
1540 	vsw_t		*vswp = ldcp->ldc_vswp;
1541 	uint64_t	state;
1542 	uint64_t	phase;
1543 
1544 	if (dir == INBOUND)
1545 		state = ldcp->lane_in.lstate;
1546 	else
1547 		state = ldcp->lane_out.lstate;
1548 
1549 	phase = ldcp->hphase;
1550 
1551 	switch (flag) {
1552 	case VSW_VER_INFO_RECV:
1553 		if (phase > VSW_MILESTONE0) {
1554 			DERR(vswp, "vsw_check_flag (%d): VER_INFO_RECV"
1555 			    " when in state %d\n", ldcp->ldc_id, phase);
1556 			vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
1557 			return (1);
1558 		}
1559 		break;
1560 
1561 	case VSW_VER_ACK_RECV:
1562 	case VSW_VER_NACK_RECV:
1563 		if (!(state & VSW_VER_INFO_SENT)) {
1564 			DERR(vswp, "vsw_check_flag (%d): spurious VER_ACK or "
1565 			    "VER_NACK when in state %d\n", ldcp->ldc_id, phase);
1566 			vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
1567 			return (1);
1568 		} else
1569 			state &= ~VSW_VER_INFO_SENT;
1570 		break;
1571 
1572 	case VSW_ATTR_INFO_RECV:
1573 		if ((phase < VSW_MILESTONE1) || (phase >= VSW_MILESTONE2)) {
1574 			DERR(vswp, "vsw_check_flag (%d): ATTR_INFO_RECV"
1575 			    " when in state %d\n", ldcp->ldc_id, phase);
1576 			vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
1577 			return (1);
1578 		}
1579 		break;
1580 
1581 	case VSW_ATTR_ACK_RECV:
1582 	case VSW_ATTR_NACK_RECV:
1583 		if (!(state & VSW_ATTR_INFO_SENT)) {
1584 			DERR(vswp, "vsw_check_flag (%d): spurious ATTR_ACK"
1585 			    " or ATTR_NACK when in state %d\n",
1586 			    ldcp->ldc_id, phase);
1587 			vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
1588 			return (1);
1589 		} else
1590 			state &= ~VSW_ATTR_INFO_SENT;
1591 		break;
1592 
1593 	case VSW_DRING_INFO_RECV:
1594 		if (phase < VSW_MILESTONE1) {
1595 			DERR(vswp, "vsw_check_flag (%d): DRING_INFO_RECV"
1596 			    " when in state %d\n", ldcp->ldc_id, phase);
1597 			vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
1598 			return (1);
1599 		}
1600 		break;
1601 
1602 	case VSW_DRING_ACK_RECV:
1603 	case VSW_DRING_NACK_RECV:
1604 		if (!(state & VSW_DRING_INFO_SENT)) {
1605 			DERR(vswp, "vsw_check_flag (%d): spurious DRING_ACK "
1606 			    " or DRING_NACK when in state %d\n",
1607 			    ldcp->ldc_id, phase);
1608 			vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
1609 			return (1);
1610 		} else
1611 			state &= ~VSW_DRING_INFO_SENT;
1612 		break;
1613 
1614 	case VSW_RDX_INFO_RECV:
1615 		if (phase < VSW_MILESTONE3) {
1616 			DERR(vswp, "vsw_check_flag (%d): RDX_INFO_RECV"
1617 			    " when in state %d\n", ldcp->ldc_id, phase);
1618 			vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
1619 			return (1);
1620 		}
1621 		break;
1622 
1623 	case VSW_RDX_ACK_RECV:
1624 	case VSW_RDX_NACK_RECV:
1625 		if (!(state & VSW_RDX_INFO_SENT)) {
1626 			DERR(vswp, "vsw_check_flag (%d): spurious RDX_ACK or "
1627 			    "RDX_NACK when in state %d\n", ldcp->ldc_id, phase);
1628 			vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
1629 			return (1);
1630 		} else
1631 			state &= ~VSW_RDX_INFO_SENT;
1632 		break;
1633 
1634 	case VSW_MCST_INFO_RECV:
1635 		if (phase < VSW_MILESTONE3) {
1636 			DERR(vswp, "vsw_check_flag (%d): VSW_MCST_INFO_RECV"
1637 			    " when in state %d\n", ldcp->ldc_id, phase);
1638 			vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
1639 			return (1);
1640 		}
1641 		break;
1642 
1643 	default:
1644 		DERR(vswp, "vsw_check_flag (%lld): unknown flag (%llx)",
1645 		    ldcp->ldc_id, flag);
1646 		return (1);
1647 	}
1648 
1649 	if (dir == INBOUND)
1650 		ldcp->lane_in.lstate = state;
1651 	else
1652 		ldcp->lane_out.lstate = state;
1653 
1654 	D1(vswp, "vsw_check_flag (chan %lld): exit", ldcp->ldc_id);
1655 
1656 	return (0);
1657 }
1658 
1659 void
1660 vsw_next_milestone(vsw_ldc_t *ldcp)
1661 {
1662 	vsw_t		*vswp = ldcp->ldc_vswp;
1663 
1664 	D1(vswp, "%s (chan %lld): enter (phase %ld)", __func__,
1665 	    ldcp->ldc_id, ldcp->hphase);
1666 
1667 	DUMP_FLAGS(ldcp->lane_in.lstate);
1668 	DUMP_FLAGS(ldcp->lane_out.lstate);
1669 
1670 	switch (ldcp->hphase) {
1671 
1672 	case VSW_MILESTONE0:
1673 		/*
1674 		 * If we haven't started to handshake with our peer,
1675 		 * start to do so now.
1676 		 */
1677 		if (ldcp->lane_out.lstate == 0) {
1678 			D2(vswp, "%s: (chan %lld) starting handshake "
1679 			    "with peer", __func__, ldcp->ldc_id);
1680 			vsw_process_conn_evt(ldcp, VSW_CONN_UP);
1681 		}
1682 
1683 		/*
1684 		 * Only way to pass this milestone is to have successfully
1685 		 * negotiated version info.
1686 		 */
1687 		if ((ldcp->lane_in.lstate & VSW_VER_ACK_SENT) &&
1688 		    (ldcp->lane_out.lstate & VSW_VER_ACK_RECV)) {
1689 
1690 			D2(vswp, "%s: (chan %lld) leaving milestone 0",
1691 			    __func__, ldcp->ldc_id);
1692 
1693 			/*
1694 			 * Next milestone is passed when attribute
1695 			 * information has been successfully exchanged.
1696 			 */
1697 			ldcp->hphase = VSW_MILESTONE1;
1698 			vsw_send_attr(ldcp);
1699 
1700 		}
1701 		break;
1702 
1703 	case VSW_MILESTONE1:
1704 		/*
1705 		 * Only way to pass this milestone is to have successfully
1706 		 * negotiated attribute information.
1707 		 */
1708 		if (ldcp->lane_in.lstate & VSW_ATTR_ACK_SENT) {
1709 
1710 			ldcp->hphase = VSW_MILESTONE2;
1711 
1712 			/*
1713 			 * If the peer device has said it wishes to
1714 			 * use descriptor rings then we send it our ring
1715 			 * info, otherwise we just set up a private ring
1716 			 * which we use an internal buffer
1717 			 */
1718 			if (ldcp->lane_in.xfer_mode == VIO_DRING_MODE)
1719 				vsw_send_dring_info(ldcp);
1720 		}
1721 		break;
1722 
1723 	case VSW_MILESTONE2:
1724 		/*
1725 		 * If peer has indicated in its attribute message that
1726 		 * it wishes to use descriptor rings then the only way
1727 		 * to pass this milestone is for us to have received
1728 		 * valid dring info.
1729 		 *
1730 		 * If peer is not using descriptor rings then just fall
1731 		 * through.
1732 		 */
1733 		if ((ldcp->lane_in.xfer_mode == VIO_DRING_MODE) &&
1734 		    (!(ldcp->lane_in.lstate & VSW_DRING_ACK_SENT)))
1735 			break;
1736 
1737 		D2(vswp, "%s: (chan %lld) leaving milestone 2",
1738 		    __func__, ldcp->ldc_id);
1739 
1740 		ldcp->hphase = VSW_MILESTONE3;
1741 		vsw_send_rdx(ldcp);
1742 		break;
1743 
1744 	case VSW_MILESTONE3:
1745 		/*
1746 		 * Pass this milestone when all paramaters have been
1747 		 * successfully exchanged and RDX sent in both directions.
1748 		 *
1749 		 * Mark outbound lane as available to transmit data.
1750 		 */
1751 		if ((ldcp->lane_out.lstate & VSW_RDX_ACK_SENT) &&
1752 		    (ldcp->lane_in.lstate & VSW_RDX_ACK_RECV)) {
1753 
1754 			D2(vswp, "%s: (chan %lld) leaving milestone 3",
1755 			    __func__, ldcp->ldc_id);
1756 			D2(vswp, "%s: ** handshake complete (0x%llx : "
1757 			    "0x%llx) **", __func__, ldcp->lane_in.lstate,
1758 			    ldcp->lane_out.lstate);
1759 			ldcp->lane_out.lstate |= VSW_LANE_ACTIVE;
1760 			ldcp->hphase = VSW_MILESTONE4;
1761 			ldcp->hcnt = 0;
1762 			DISPLAY_STATE();
1763 		} else {
1764 			D2(vswp, "%s: still in milestone 3 (0x%llx : 0x%llx)",
1765 			    __func__, ldcp->lane_in.lstate,
1766 			    ldcp->lane_out.lstate);
1767 		}
1768 		break;
1769 
1770 	case VSW_MILESTONE4:
1771 		D2(vswp, "%s: (chan %lld) in milestone 4", __func__,
1772 		    ldcp->ldc_id);
1773 		break;
1774 
1775 	default:
1776 		DERR(vswp, "%s: (chan %lld) Unknown Phase %x", __func__,
1777 		    ldcp->ldc_id, ldcp->hphase);
1778 	}
1779 
1780 	D1(vswp, "%s (chan %lld): exit (phase %ld)", __func__, ldcp->ldc_id,
1781 	    ldcp->hphase);
1782 }
1783 
1784 /*
1785  * Check if major version is supported.
1786  *
1787  * Returns 0 if finds supported major number, and if necessary
1788  * adjusts the minor field.
1789  *
1790  * Returns 1 if can't match major number exactly. Sets mjor/minor
1791  * to next lowest support values, or to zero if no other values possible.
1792  */
1793 static int
1794 vsw_supported_version(vio_ver_msg_t *vp)
1795 {
1796 	int	i;
1797 
1798 	D1(NULL, "vsw_supported_version: enter");
1799 
1800 	for (i = 0; i < VSW_NUM_VER; i++) {
1801 		if (vsw_versions[i].ver_major == vp->ver_major) {
1802 			/*
1803 			 * Matching or lower major version found. Update
1804 			 * minor number if necessary.
1805 			 */
1806 			if (vp->ver_minor > vsw_versions[i].ver_minor) {
1807 				D2(NULL, "%s: adjusting minor value from %d "
1808 				    "to %d", __func__, vp->ver_minor,
1809 				    vsw_versions[i].ver_minor);
1810 				vp->ver_minor = vsw_versions[i].ver_minor;
1811 			}
1812 
1813 			return (0);
1814 		}
1815 
1816 		if (vsw_versions[i].ver_major < vp->ver_major) {
1817 			if (vp->ver_minor > vsw_versions[i].ver_minor) {
1818 				D2(NULL, "%s: adjusting minor value from %d "
1819 				    "to %d", __func__, vp->ver_minor,
1820 				    vsw_versions[i].ver_minor);
1821 				vp->ver_minor = vsw_versions[i].ver_minor;
1822 			}
1823 			return (1);
1824 		}
1825 	}
1826 
1827 	/* No match was possible, zero out fields */
1828 	vp->ver_major = 0;
1829 	vp->ver_minor = 0;
1830 
1831 	D1(NULL, "vsw_supported_version: exit");
1832 
1833 	return (1);
1834 }
1835 
1836 /*
1837  * Main routine for processing messages received over LDC.
1838  */
1839 static void
1840 vsw_process_pkt(void *arg)
1841 {
1842 	vsw_ldc_t	*ldcp = (vsw_ldc_t  *)arg;
1843 	vsw_t 		*vswp = ldcp->ldc_vswp;
1844 	size_t		msglen;
1845 	vio_msg_tag_t	tag;
1846 	def_msg_t	dmsg;
1847 	int 		rv = 0;
1848 
1849 
1850 	D1(vswp, "%s enter: ldcid (%lld)\n", __func__, ldcp->ldc_id);
1851 
1852 	ASSERT(MUTEX_HELD(&ldcp->ldc_cblock));
1853 
1854 	/*
1855 	 * If channel is up read messages until channel is empty.
1856 	 */
1857 	do {
1858 		msglen = sizeof (dmsg);
1859 		rv = ldc_read(ldcp->ldc_handle, (caddr_t)&dmsg, &msglen);
1860 
1861 		if (rv != 0) {
1862 			DERR(vswp, "%s :ldc_read err id(%lld) rv(%d) len(%d)\n",
1863 			    __func__, ldcp->ldc_id, rv, msglen);
1864 		}
1865 
1866 		/* channel has been reset */
1867 		if (rv == ECONNRESET) {
1868 			vsw_process_conn_evt(ldcp, VSW_CONN_RESET);
1869 			break;
1870 		}
1871 
1872 		if (msglen == 0) {
1873 			D2(vswp, "%s: ldc_read id(%lld) NODATA", __func__,
1874 			    ldcp->ldc_id);
1875 			break;
1876 		}
1877 
1878 		D2(vswp, "%s: ldc_read id(%lld): msglen(%d)", __func__,
1879 		    ldcp->ldc_id, msglen);
1880 
1881 		/*
1882 		 * Figure out what sort of packet we have gotten by
1883 		 * examining the msg tag, and then switch it appropriately.
1884 		 */
1885 		bcopy(&dmsg, &tag, sizeof (vio_msg_tag_t));
1886 
1887 		switch (tag.vio_msgtype) {
1888 		case VIO_TYPE_CTRL:
1889 			vsw_dispatch_ctrl_task(ldcp, &dmsg, tag);
1890 			break;
1891 		case VIO_TYPE_DATA:
1892 			vsw_process_data_pkt(ldcp, &dmsg, tag);
1893 			break;
1894 		case VIO_TYPE_ERR:
1895 			vsw_process_err_pkt(ldcp, &dmsg, tag);
1896 			break;
1897 		default:
1898 			DERR(vswp, "%s: Unknown tag(%lx) ", __func__,
1899 			    "id(%lx)\n", tag.vio_msgtype, ldcp->ldc_id);
1900 			break;
1901 		}
1902 	} while (msglen);
1903 
1904 	D1(vswp, "%s exit: ldcid (%lld)\n", __func__, ldcp->ldc_id);
1905 }
1906 
1907 /*
1908  * Dispatch a task to process a VIO control message.
1909  */
1910 static void
1911 vsw_dispatch_ctrl_task(vsw_ldc_t *ldcp, void *cpkt, vio_msg_tag_t tag)
1912 {
1913 	vsw_ctrl_task_t		*ctaskp = NULL;
1914 	vsw_port_t		*port = ldcp->ldc_port;
1915 	vsw_t			*vswp = port->p_vswp;
1916 
1917 	D1(vswp, "%s: enter", __func__);
1918 
1919 	/*
1920 	 * We need to handle RDX ACK messages in-band as once they
1921 	 * are exchanged it is possible that we will get an
1922 	 * immediate (legitimate) data packet.
1923 	 */
1924 	if ((tag.vio_subtype_env == VIO_RDX) &&
1925 	    (tag.vio_subtype == VIO_SUBTYPE_ACK)) {
1926 
1927 		if (vsw_check_flag(ldcp, INBOUND, VSW_RDX_ACK_RECV))
1928 			return;
1929 
1930 		ldcp->lane_in.lstate |= VSW_RDX_ACK_RECV;
1931 		D2(vswp, "%s (%ld) handling RDX_ACK in place "
1932 		    "(ostate 0x%llx : hphase %d)", __func__,
1933 		    ldcp->ldc_id, ldcp->lane_in.lstate, ldcp->hphase);
1934 		vsw_next_milestone(ldcp);
1935 		return;
1936 	}
1937 
1938 	ctaskp = kmem_alloc(sizeof (vsw_ctrl_task_t), KM_NOSLEEP);
1939 
1940 	if (ctaskp == NULL) {
1941 		DERR(vswp, "%s: unable to alloc space for ctrl msg", __func__);
1942 		vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
1943 		return;
1944 	}
1945 
1946 	ctaskp->ldcp = ldcp;
1947 	bcopy((def_msg_t *)cpkt, &ctaskp->pktp, sizeof (def_msg_t));
1948 	ctaskp->hss_id = ldcp->hss_id;
1949 
1950 	/*
1951 	 * Dispatch task to processing taskq if port is not in
1952 	 * the process of being detached.
1953 	 */
1954 	mutex_enter(&port->state_lock);
1955 	if (port->state == VSW_PORT_INIT) {
1956 		if ((vswp->taskq_p == NULL) ||
1957 		    (ddi_taskq_dispatch(vswp->taskq_p, vsw_process_ctrl_pkt,
1958 		    ctaskp, DDI_NOSLEEP) != DDI_SUCCESS)) {
1959 			DERR(vswp, "%s: unable to dispatch task to taskq",
1960 			    __func__);
1961 			kmem_free(ctaskp, sizeof (vsw_ctrl_task_t));
1962 			mutex_exit(&port->state_lock);
1963 			vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
1964 			return;
1965 		}
1966 	} else {
1967 		DWARN(vswp, "%s: port %d detaching, not dispatching "
1968 		    "task", __func__, port->p_instance);
1969 	}
1970 
1971 	mutex_exit(&port->state_lock);
1972 
1973 	D2(vswp, "%s: dispatched task to taskq for chan %d", __func__,
1974 	    ldcp->ldc_id);
1975 	D1(vswp, "%s: exit", __func__);
1976 }
1977 
1978 /*
1979  * Process a VIO ctrl message. Invoked from taskq.
1980  */
1981 static void
1982 vsw_process_ctrl_pkt(void *arg)
1983 {
1984 	vsw_ctrl_task_t	*ctaskp = (vsw_ctrl_task_t *)arg;
1985 	vsw_ldc_t	*ldcp = ctaskp->ldcp;
1986 	vsw_t 		*vswp = ldcp->ldc_vswp;
1987 	vio_msg_tag_t	tag;
1988 	uint16_t	env;
1989 
1990 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
1991 
1992 	bcopy(&ctaskp->pktp, &tag, sizeof (vio_msg_tag_t));
1993 	env = tag.vio_subtype_env;
1994 
1995 	/* stale pkt check */
1996 	if (ctaskp->hss_id < ldcp->hss_id) {
1997 		DWARN(vswp, "%s: discarding stale packet belonging to earlier"
1998 		    " (%ld) handshake session", __func__, ctaskp->hss_id);
1999 		return;
2000 	}
2001 
2002 	/* session id check */
2003 	if (ldcp->session_status & VSW_PEER_SESSION) {
2004 		if (ldcp->peer_session != tag.vio_sid) {
2005 			DERR(vswp, "%s (chan %d): invalid session id (%llx)",
2006 			    __func__, ldcp->ldc_id, tag.vio_sid);
2007 			kmem_free(ctaskp, sizeof (vsw_ctrl_task_t));
2008 			vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
2009 			return;
2010 		}
2011 	}
2012 
2013 	/*
2014 	 * Switch on vio_subtype envelope, then let lower routines
2015 	 * decide if its an INFO, ACK or NACK packet.
2016 	 */
2017 	switch (env) {
2018 	case VIO_VER_INFO:
2019 		vsw_process_ctrl_ver_pkt(ldcp, &ctaskp->pktp);
2020 		break;
2021 	case VIO_DRING_REG:
2022 		vsw_process_ctrl_dring_reg_pkt(ldcp, &ctaskp->pktp);
2023 		break;
2024 	case VIO_DRING_UNREG:
2025 		vsw_process_ctrl_dring_unreg_pkt(ldcp, &ctaskp->pktp);
2026 		break;
2027 	case VIO_ATTR_INFO:
2028 		vsw_process_ctrl_attr_pkt(ldcp, &ctaskp->pktp);
2029 		break;
2030 	case VNET_MCAST_INFO:
2031 		vsw_process_ctrl_mcst_pkt(ldcp, &ctaskp->pktp);
2032 		break;
2033 	case VIO_RDX:
2034 		vsw_process_ctrl_rdx_pkt(ldcp, &ctaskp->pktp);
2035 		break;
2036 	default:
2037 		DERR(vswp, "%s: unknown vio_subtype_env (%x)\n", __func__, env);
2038 	}
2039 
2040 	kmem_free(ctaskp, sizeof (vsw_ctrl_task_t));
2041 	D1(vswp, "%s(%lld): exit", __func__, ldcp->ldc_id);
2042 }
2043 
2044 /*
2045  * Version negotiation. We can end up here either because our peer
2046  * has responded to a handshake message we have sent it, or our peer
2047  * has initiated a handshake with us. If its the former then can only
2048  * be ACK or NACK, if its the later can only be INFO.
2049  *
2050  * If its an ACK we move to the next stage of the handshake, namely
2051  * attribute exchange. If its a NACK we see if we can specify another
2052  * version, if we can't we stop.
2053  *
2054  * If it is an INFO we reset all params associated with communication
2055  * in that direction over this channel (remember connection is
2056  * essentially 2 independent simplex channels).
2057  */
2058 void
2059 vsw_process_ctrl_ver_pkt(vsw_ldc_t *ldcp, void *pkt)
2060 {
2061 	vio_ver_msg_t	*ver_pkt;
2062 	vsw_t 		*vswp = ldcp->ldc_vswp;
2063 
2064 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
2065 
2066 	/*
2067 	 * We know this is a ctrl/version packet so
2068 	 * cast it into the correct structure.
2069 	 */
2070 	ver_pkt = (vio_ver_msg_t *)pkt;
2071 
2072 	switch (ver_pkt->tag.vio_subtype) {
2073 	case VIO_SUBTYPE_INFO:
2074 		D2(vswp, "vsw_process_ctrl_ver_pkt: VIO_SUBTYPE_INFO\n");
2075 
2076 		/*
2077 		 * Record the session id, which we will use from now
2078 		 * until we see another VER_INFO msg. Even then the
2079 		 * session id in most cases will be unchanged, execpt
2080 		 * if channel was reset.
2081 		 */
2082 		if ((ldcp->session_status & VSW_PEER_SESSION) &&
2083 		    (ldcp->peer_session != ver_pkt->tag.vio_sid)) {
2084 			DERR(vswp, "%s: updating session id for chan %lld "
2085 			    "from %llx to %llx", __func__, ldcp->ldc_id,
2086 			    ldcp->peer_session, ver_pkt->tag.vio_sid);
2087 		}
2088 
2089 		ldcp->peer_session = ver_pkt->tag.vio_sid;
2090 		ldcp->session_status |= VSW_PEER_SESSION;
2091 
2092 		/* Legal message at this time ? */
2093 		if (vsw_check_flag(ldcp, INBOUND, VSW_VER_INFO_RECV))
2094 			return;
2095 
2096 		/*
2097 		 * First check the device class. Currently only expect
2098 		 * to be talking to a network device. In the future may
2099 		 * also talk to another switch.
2100 		 */
2101 		if (ver_pkt->dev_class != VDEV_NETWORK) {
2102 			DERR(vswp, "%s: illegal device class %d", __func__,
2103 			    ver_pkt->dev_class);
2104 
2105 			ver_pkt->tag.vio_sid = ldcp->local_session;
2106 			ver_pkt->tag.vio_subtype = VIO_SUBTYPE_NACK;
2107 
2108 			DUMP_TAG_PTR((vio_msg_tag_t *)ver_pkt);
2109 
2110 			(void) vsw_send_msg(ldcp, (void *)ver_pkt,
2111 			    sizeof (vio_ver_msg_t), B_TRUE);
2112 
2113 			ldcp->lane_in.lstate |= VSW_VER_NACK_SENT;
2114 			vsw_next_milestone(ldcp);
2115 			return;
2116 		} else {
2117 			ldcp->dev_class = ver_pkt->dev_class;
2118 		}
2119 
2120 		/*
2121 		 * Now check the version.
2122 		 */
2123 		if (vsw_supported_version(ver_pkt) == 0) {
2124 			/*
2125 			 * Support this major version and possibly
2126 			 * adjusted minor version.
2127 			 */
2128 
2129 			D2(vswp, "%s: accepted ver %d:%d", __func__,
2130 			    ver_pkt->ver_major, ver_pkt->ver_minor);
2131 
2132 			/* Store accepted values */
2133 			ldcp->lane_in.ver_major = ver_pkt->ver_major;
2134 			ldcp->lane_in.ver_minor = ver_pkt->ver_minor;
2135 
2136 			ver_pkt->tag.vio_subtype = VIO_SUBTYPE_ACK;
2137 
2138 			ldcp->lane_in.lstate |= VSW_VER_ACK_SENT;
2139 		} else {
2140 			/*
2141 			 * NACK back with the next lower major/minor
2142 			 * pairing we support (if don't suuport any more
2143 			 * versions then they will be set to zero.
2144 			 */
2145 
2146 			D2(vswp, "%s: replying with ver %d:%d", __func__,
2147 			    ver_pkt->ver_major, ver_pkt->ver_minor);
2148 
2149 			/* Store updated values */
2150 			ldcp->lane_in.ver_major = ver_pkt->ver_major;
2151 			ldcp->lane_in.ver_minor = ver_pkt->ver_minor;
2152 
2153 			ver_pkt->tag.vio_subtype = VIO_SUBTYPE_NACK;
2154 
2155 			ldcp->lane_in.lstate |= VSW_VER_NACK_SENT;
2156 		}
2157 
2158 		DUMP_TAG_PTR((vio_msg_tag_t *)ver_pkt);
2159 		ver_pkt->tag.vio_sid = ldcp->local_session;
2160 		(void) vsw_send_msg(ldcp, (void *)ver_pkt,
2161 		    sizeof (vio_ver_msg_t), B_TRUE);
2162 
2163 		vsw_next_milestone(ldcp);
2164 		break;
2165 
2166 	case VIO_SUBTYPE_ACK:
2167 		D2(vswp, "%s: VIO_SUBTYPE_ACK\n", __func__);
2168 
2169 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_VER_ACK_RECV))
2170 			return;
2171 
2172 		/* Store updated values */
2173 		ldcp->lane_in.ver_major = ver_pkt->ver_major;
2174 		ldcp->lane_in.ver_minor = ver_pkt->ver_minor;
2175 
2176 		ldcp->lane_out.lstate |= VSW_VER_ACK_RECV;
2177 		vsw_next_milestone(ldcp);
2178 
2179 		break;
2180 
2181 	case VIO_SUBTYPE_NACK:
2182 		D2(vswp, "%s: VIO_SUBTYPE_NACK\n", __func__);
2183 
2184 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_VER_NACK_RECV))
2185 			return;
2186 
2187 		/*
2188 		 * If our peer sent us a NACK with the ver fields set to
2189 		 * zero then there is nothing more we can do. Otherwise see
2190 		 * if we support either the version suggested, or a lesser
2191 		 * one.
2192 		 */
2193 		if ((ver_pkt->ver_major == 0) && (ver_pkt->ver_minor == 0)) {
2194 			DERR(vswp, "%s: peer unable to negotiate any "
2195 			    "further.", __func__);
2196 			ldcp->lane_out.lstate |= VSW_VER_NACK_RECV;
2197 			vsw_next_milestone(ldcp);
2198 			return;
2199 		}
2200 
2201 		/*
2202 		 * Check to see if we support this major version or
2203 		 * a lower one. If we don't then maj/min will be set
2204 		 * to zero.
2205 		 */
2206 		(void) vsw_supported_version(ver_pkt);
2207 		if ((ver_pkt->ver_major == 0) && (ver_pkt->ver_minor == 0)) {
2208 			/* Nothing more we can do */
2209 			DERR(vswp, "%s: version negotiation failed.\n",
2210 			    __func__);
2211 			ldcp->lane_out.lstate |= VSW_VER_NACK_RECV;
2212 			vsw_next_milestone(ldcp);
2213 		} else {
2214 			/* found a supported major version */
2215 			ldcp->lane_out.ver_major = ver_pkt->ver_major;
2216 			ldcp->lane_out.ver_minor = ver_pkt->ver_minor;
2217 
2218 			D2(vswp, "%s: resending with updated values (%x, %x)",
2219 			    __func__, ver_pkt->ver_major, ver_pkt->ver_minor);
2220 
2221 			ldcp->lane_out.lstate |= VSW_VER_INFO_SENT;
2222 			ver_pkt->tag.vio_sid = ldcp->local_session;
2223 			ver_pkt->tag.vio_subtype = VIO_SUBTYPE_INFO;
2224 
2225 			DUMP_TAG_PTR((vio_msg_tag_t *)ver_pkt);
2226 
2227 			(void) vsw_send_msg(ldcp, (void *)ver_pkt,
2228 			    sizeof (vio_ver_msg_t), B_TRUE);
2229 
2230 			vsw_next_milestone(ldcp);
2231 
2232 		}
2233 		break;
2234 
2235 	default:
2236 		DERR(vswp, "%s: unknown vio_subtype %x\n", __func__,
2237 		    ver_pkt->tag.vio_subtype);
2238 	}
2239 
2240 	D1(vswp, "%s(%lld): exit\n", __func__, ldcp->ldc_id);
2241 }
2242 
2243 /*
2244  * Process an attribute packet. We can end up here either because our peer
2245  * has ACK/NACK'ed back to an earlier ATTR msg we had sent it, or our
2246  * peer has sent us an attribute INFO message
2247  *
2248  * If its an ACK we then move to the next stage of the handshake which
2249  * is to send our descriptor ring info to our peer. If its a NACK then
2250  * there is nothing more we can (currently) do.
2251  *
2252  * If we get a valid/acceptable INFO packet (and we have already negotiated
2253  * a version) we ACK back and set channel state to ATTR_RECV, otherwise we
2254  * NACK back and reset channel state to INACTIV.
2255  *
2256  * FUTURE: in time we will probably negotiate over attributes, but for
2257  * the moment unacceptable attributes are regarded as a fatal error.
2258  *
2259  */
2260 void
2261 vsw_process_ctrl_attr_pkt(vsw_ldc_t *ldcp, void *pkt)
2262 {
2263 	vnet_attr_msg_t		*attr_pkt;
2264 	vsw_t			*vswp = ldcp->ldc_vswp;
2265 	vsw_port_t		*port = ldcp->ldc_port;
2266 	uint64_t		macaddr = 0;
2267 	int			i;
2268 
2269 	D1(vswp, "%s(%lld) enter", __func__, ldcp->ldc_id);
2270 
2271 	/*
2272 	 * We know this is a ctrl/attr packet so
2273 	 * cast it into the correct structure.
2274 	 */
2275 	attr_pkt = (vnet_attr_msg_t *)pkt;
2276 
2277 	switch (attr_pkt->tag.vio_subtype) {
2278 	case VIO_SUBTYPE_INFO:
2279 		D2(vswp, "%s: VIO_SUBTYPE_INFO", __func__);
2280 
2281 		if (vsw_check_flag(ldcp, INBOUND, VSW_ATTR_INFO_RECV))
2282 			return;
2283 
2284 		/*
2285 		 * If the attributes are unacceptable then we NACK back.
2286 		 */
2287 		if (vsw_check_attr(attr_pkt, ldcp->ldc_port)) {
2288 
2289 			DERR(vswp, "%s (chan %d): invalid attributes",
2290 			    __func__, ldcp->ldc_id);
2291 
2292 			vsw_free_lane_resources(ldcp, INBOUND);
2293 
2294 			attr_pkt->tag.vio_sid = ldcp->local_session;
2295 			attr_pkt->tag.vio_subtype = VIO_SUBTYPE_NACK;
2296 
2297 			DUMP_TAG_PTR((vio_msg_tag_t *)attr_pkt);
2298 			ldcp->lane_in.lstate |= VSW_ATTR_NACK_SENT;
2299 			(void) vsw_send_msg(ldcp, (void *)attr_pkt,
2300 			    sizeof (vnet_attr_msg_t), B_TRUE);
2301 
2302 			vsw_next_milestone(ldcp);
2303 			return;
2304 		}
2305 
2306 		/*
2307 		 * Otherwise store attributes for this lane and update
2308 		 * lane state.
2309 		 */
2310 		ldcp->lane_in.mtu = attr_pkt->mtu;
2311 		ldcp->lane_in.addr = attr_pkt->addr;
2312 		ldcp->lane_in.addr_type = attr_pkt->addr_type;
2313 		ldcp->lane_in.xfer_mode = attr_pkt->xfer_mode;
2314 		ldcp->lane_in.ack_freq = attr_pkt->ack_freq;
2315 
2316 		macaddr = ldcp->lane_in.addr;
2317 		for (i = ETHERADDRL - 1; i >= 0; i--) {
2318 			port->p_macaddr.ether_addr_octet[i] = macaddr & 0xFF;
2319 			macaddr >>= 8;
2320 		}
2321 
2322 		/* create the fdb entry for this port/mac address */
2323 		(void) vsw_add_fdb(vswp, port);
2324 
2325 		/* setup device specifc xmit routines */
2326 		mutex_enter(&port->tx_lock);
2327 		if (ldcp->lane_in.xfer_mode == VIO_DRING_MODE) {
2328 			D2(vswp, "%s: mode = VIO_DRING_MODE", __func__);
2329 			port->transmit = vsw_dringsend;
2330 		} else if (ldcp->lane_in.xfer_mode == VIO_DESC_MODE) {
2331 			D2(vswp, "%s: mode = VIO_DESC_MODE", __func__);
2332 			vsw_create_privring(ldcp);
2333 			port->transmit = vsw_descrsend;
2334 		}
2335 		mutex_exit(&port->tx_lock);
2336 
2337 		attr_pkt->tag.vio_sid = ldcp->local_session;
2338 		attr_pkt->tag.vio_subtype = VIO_SUBTYPE_ACK;
2339 
2340 		DUMP_TAG_PTR((vio_msg_tag_t *)attr_pkt);
2341 
2342 		ldcp->lane_in.lstate |= VSW_ATTR_ACK_SENT;
2343 
2344 		(void) vsw_send_msg(ldcp, (void *)attr_pkt,
2345 		    sizeof (vnet_attr_msg_t), B_TRUE);
2346 
2347 		vsw_next_milestone(ldcp);
2348 		break;
2349 
2350 	case VIO_SUBTYPE_ACK:
2351 		D2(vswp, "%s: VIO_SUBTYPE_ACK", __func__);
2352 
2353 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_ATTR_ACK_RECV))
2354 			return;
2355 
2356 		ldcp->lane_out.lstate |= VSW_ATTR_ACK_RECV;
2357 		vsw_next_milestone(ldcp);
2358 		break;
2359 
2360 	case VIO_SUBTYPE_NACK:
2361 		D2(vswp, "%s: VIO_SUBTYPE_NACK", __func__);
2362 
2363 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_ATTR_NACK_RECV))
2364 			return;
2365 
2366 		ldcp->lane_out.lstate |= VSW_ATTR_NACK_RECV;
2367 		vsw_next_milestone(ldcp);
2368 		break;
2369 
2370 	default:
2371 		DERR(vswp, "%s: unknown vio_subtype %x\n", __func__,
2372 		    attr_pkt->tag.vio_subtype);
2373 	}
2374 
2375 	D1(vswp, "%s(%lld) exit", __func__, ldcp->ldc_id);
2376 }
2377 
2378 /*
2379  * Process a dring info packet. We can end up here either because our peer
2380  * has ACK/NACK'ed back to an earlier DRING msg we had sent it, or our
2381  * peer has sent us a dring INFO message.
2382  *
2383  * If we get a valid/acceptable INFO packet (and we have already negotiated
2384  * a version) we ACK back and update the lane state, otherwise we NACK back.
2385  *
2386  * FUTURE: nothing to stop client from sending us info on multiple dring's
2387  * but for the moment we will just use the first one we are given.
2388  *
2389  */
2390 void
2391 vsw_process_ctrl_dring_reg_pkt(vsw_ldc_t *ldcp, void *pkt)
2392 {
2393 	vio_dring_reg_msg_t	*dring_pkt;
2394 	vsw_t			*vswp = ldcp->ldc_vswp;
2395 	ldc_mem_info_t		minfo;
2396 	dring_info_t		*dp, *dbp;
2397 	int			dring_found = 0;
2398 
2399 	/*
2400 	 * We know this is a ctrl/dring packet so
2401 	 * cast it into the correct structure.
2402 	 */
2403 	dring_pkt = (vio_dring_reg_msg_t *)pkt;
2404 
2405 	D1(vswp, "%s(%lld) enter", __func__, ldcp->ldc_id);
2406 
2407 	switch (dring_pkt->tag.vio_subtype) {
2408 	case VIO_SUBTYPE_INFO:
2409 		D2(vswp, "%s: VIO_SUBTYPE_INFO", __func__);
2410 
2411 		if (vsw_check_flag(ldcp, INBOUND, VSW_DRING_INFO_RECV))
2412 			return;
2413 
2414 		/*
2415 		 * If the dring params are unacceptable then we NACK back.
2416 		 */
2417 		if (vsw_check_dring_info(dring_pkt)) {
2418 
2419 			DERR(vswp, "%s (%lld): invalid dring info",
2420 			    __func__, ldcp->ldc_id);
2421 
2422 			vsw_free_lane_resources(ldcp, INBOUND);
2423 
2424 			dring_pkt->tag.vio_sid = ldcp->local_session;
2425 			dring_pkt->tag.vio_subtype = VIO_SUBTYPE_NACK;
2426 
2427 			DUMP_TAG_PTR((vio_msg_tag_t *)dring_pkt);
2428 
2429 			ldcp->lane_in.lstate |= VSW_DRING_NACK_SENT;
2430 
2431 			(void) vsw_send_msg(ldcp, (void *)dring_pkt,
2432 			    sizeof (vio_dring_reg_msg_t), B_TRUE);
2433 
2434 			vsw_next_milestone(ldcp);
2435 			return;
2436 		}
2437 
2438 		/*
2439 		 * Otherwise, attempt to map in the dring using the
2440 		 * cookie. If that succeeds we send back a unique dring
2441 		 * identifier that the sending side will use in future
2442 		 * to refer to this descriptor ring.
2443 		 */
2444 		dp = kmem_zalloc(sizeof (dring_info_t), KM_SLEEP);
2445 
2446 		dp->num_descriptors = dring_pkt->num_descriptors;
2447 		dp->descriptor_size = dring_pkt->descriptor_size;
2448 		dp->options = dring_pkt->options;
2449 		dp->ncookies = dring_pkt->ncookies;
2450 
2451 		/*
2452 		 * Note: should only get one cookie. Enforced in
2453 		 * the ldc layer.
2454 		 */
2455 		bcopy(&dring_pkt->cookie[0], &dp->cookie[0],
2456 		    sizeof (ldc_mem_cookie_t));
2457 
2458 		D2(vswp, "%s: num_desc %ld : desc_size %ld", __func__,
2459 		    dp->num_descriptors, dp->descriptor_size);
2460 		D2(vswp, "%s: options 0x%lx: ncookies %ld", __func__,
2461 		    dp->options, dp->ncookies);
2462 
2463 		if ((ldc_mem_dring_map(ldcp->ldc_handle, &dp->cookie[0],
2464 		    dp->ncookies, dp->num_descriptors, dp->descriptor_size,
2465 		    LDC_SHADOW_MAP, &(dp->handle))) != 0) {
2466 
2467 			DERR(vswp, "%s: dring_map failed\n", __func__);
2468 
2469 			kmem_free(dp, sizeof (dring_info_t));
2470 			vsw_free_lane_resources(ldcp, INBOUND);
2471 
2472 			dring_pkt->tag.vio_sid = ldcp->local_session;
2473 			dring_pkt->tag.vio_subtype = VIO_SUBTYPE_NACK;
2474 
2475 			DUMP_TAG_PTR((vio_msg_tag_t *)dring_pkt);
2476 
2477 			ldcp->lane_in.lstate |= VSW_DRING_NACK_SENT;
2478 			(void) vsw_send_msg(ldcp, (void *)dring_pkt,
2479 			    sizeof (vio_dring_reg_msg_t), B_TRUE);
2480 
2481 			vsw_next_milestone(ldcp);
2482 			return;
2483 		}
2484 
2485 		if ((ldc_mem_dring_info(dp->handle, &minfo)) != 0) {
2486 
2487 			DERR(vswp, "%s: dring_addr failed\n", __func__);
2488 
2489 			kmem_free(dp, sizeof (dring_info_t));
2490 			vsw_free_lane_resources(ldcp, INBOUND);
2491 
2492 			dring_pkt->tag.vio_sid = ldcp->local_session;
2493 			dring_pkt->tag.vio_subtype = VIO_SUBTYPE_NACK;
2494 
2495 			DUMP_TAG_PTR((vio_msg_tag_t *)dring_pkt);
2496 
2497 			ldcp->lane_in.lstate |= VSW_DRING_NACK_SENT;
2498 			(void) vsw_send_msg(ldcp, (void *)dring_pkt,
2499 			    sizeof (vio_dring_reg_msg_t), B_TRUE);
2500 
2501 			vsw_next_milestone(ldcp);
2502 			return;
2503 		} else {
2504 			/* store the address of the pub part of ring */
2505 			dp->pub_addr = minfo.vaddr;
2506 		}
2507 
2508 		/* no private section as we are importing */
2509 		dp->priv_addr = NULL;
2510 
2511 		/*
2512 		 * Using simple mono increasing int for ident at
2513 		 * the moment.
2514 		 */
2515 		dp->ident = ldcp->next_ident;
2516 		ldcp->next_ident++;
2517 
2518 		dp->end_idx = 0;
2519 		dp->next = NULL;
2520 
2521 		/*
2522 		 * Link it onto the end of the list of drings
2523 		 * for this lane.
2524 		 */
2525 		if (ldcp->lane_in.dringp == NULL) {
2526 			D2(vswp, "%s: adding first INBOUND dring", __func__);
2527 			ldcp->lane_in.dringp = dp;
2528 		} else {
2529 			dbp = ldcp->lane_in.dringp;
2530 
2531 			while (dbp->next != NULL)
2532 				dbp = dbp->next;
2533 
2534 			dbp->next = dp;
2535 		}
2536 
2537 		/* acknowledge it */
2538 		dring_pkt->tag.vio_sid = ldcp->local_session;
2539 		dring_pkt->tag.vio_subtype = VIO_SUBTYPE_ACK;
2540 		dring_pkt->dring_ident = dp->ident;
2541 
2542 		(void) vsw_send_msg(ldcp, (void *)dring_pkt,
2543 		    sizeof (vio_dring_reg_msg_t), B_TRUE);
2544 
2545 		ldcp->lane_in.lstate |= VSW_DRING_ACK_SENT;
2546 		vsw_next_milestone(ldcp);
2547 		break;
2548 
2549 	case VIO_SUBTYPE_ACK:
2550 		D2(vswp, "%s: VIO_SUBTYPE_ACK", __func__);
2551 
2552 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_DRING_ACK_RECV))
2553 			return;
2554 
2555 		/*
2556 		 * Peer is acknowledging our dring info and will have
2557 		 * sent us a dring identifier which we will use to
2558 		 * refer to this ring w.r.t. our peer.
2559 		 */
2560 		dp = ldcp->lane_out.dringp;
2561 		if (dp != NULL) {
2562 			/*
2563 			 * Find the ring this ident should be associated
2564 			 * with.
2565 			 */
2566 			if (vsw_dring_match(dp, dring_pkt)) {
2567 				dring_found = 1;
2568 
2569 			} else while (dp != NULL) {
2570 				if (vsw_dring_match(dp, dring_pkt)) {
2571 					dring_found = 1;
2572 					break;
2573 				}
2574 				dp = dp->next;
2575 			}
2576 
2577 			if (dring_found == 0) {
2578 				DERR(NULL, "%s: unrecognised ring cookie",
2579 				    __func__);
2580 				vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
2581 				return;
2582 			}
2583 
2584 		} else {
2585 			DERR(vswp, "%s: DRING ACK received but no drings "
2586 			    "allocated", __func__);
2587 			vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
2588 			return;
2589 		}
2590 
2591 		/* store ident */
2592 		dp->ident = dring_pkt->dring_ident;
2593 		ldcp->lane_out.lstate |= VSW_DRING_ACK_RECV;
2594 		vsw_next_milestone(ldcp);
2595 		break;
2596 
2597 	case VIO_SUBTYPE_NACK:
2598 		D2(vswp, "%s: VIO_SUBTYPE_NACK", __func__);
2599 
2600 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_DRING_NACK_RECV))
2601 			return;
2602 
2603 		ldcp->lane_out.lstate |= VSW_DRING_NACK_RECV;
2604 		vsw_next_milestone(ldcp);
2605 		break;
2606 
2607 	default:
2608 		DERR(vswp, "%s: Unknown vio_subtype %x\n", __func__,
2609 		    dring_pkt->tag.vio_subtype);
2610 	}
2611 
2612 	D1(vswp, "%s(%lld) exit", __func__, ldcp->ldc_id);
2613 }
2614 
2615 /*
2616  * Process a request from peer to unregister a dring.
2617  *
2618  * For the moment we just restart the handshake if our
2619  * peer endpoint attempts to unregister a dring.
2620  */
2621 void
2622 vsw_process_ctrl_dring_unreg_pkt(vsw_ldc_t *ldcp, void *pkt)
2623 {
2624 	vsw_t			*vswp = ldcp->ldc_vswp;
2625 	vio_dring_unreg_msg_t	*dring_pkt;
2626 
2627 	/*
2628 	 * We know this is a ctrl/dring packet so
2629 	 * cast it into the correct structure.
2630 	 */
2631 	dring_pkt = (vio_dring_unreg_msg_t *)pkt;
2632 
2633 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
2634 
2635 	switch (dring_pkt->tag.vio_subtype) {
2636 	case VIO_SUBTYPE_INFO:
2637 		D2(vswp, "%s: VIO_SUBTYPE_INFO", __func__);
2638 
2639 		DWARN(vswp, "%s: restarting handshake..", __func__);
2640 		break;
2641 
2642 	case VIO_SUBTYPE_ACK:
2643 		D2(vswp, "%s: VIO_SUBTYPE_ACK", __func__);
2644 
2645 		DWARN(vswp, "%s: restarting handshake..", __func__);
2646 		break;
2647 
2648 	case VIO_SUBTYPE_NACK:
2649 		D2(vswp, "%s: VIO_SUBTYPE_NACK", __func__);
2650 
2651 		DWARN(vswp, "%s: restarting handshake..", __func__);
2652 		break;
2653 
2654 	default:
2655 		DERR(vswp, "%s: Unknown vio_subtype %x\n", __func__,
2656 		    dring_pkt->tag.vio_subtype);
2657 	}
2658 
2659 	vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
2660 
2661 	D1(vswp, "%s(%lld): exit", __func__, ldcp->ldc_id);
2662 }
2663 
2664 #define	SND_MCST_NACK(ldcp, pkt) \
2665 	pkt->tag.vio_subtype = VIO_SUBTYPE_NACK; \
2666 	pkt->tag.vio_sid = ldcp->local_session; \
2667 	(void) vsw_send_msg(ldcp, (void *)pkt, \
2668 			sizeof (vnet_mcast_msg_t), B_TRUE);
2669 
2670 /*
2671  * Process a multicast request from a vnet.
2672  *
2673  * Vnet's specify a multicast address that they are interested in. This
2674  * address is used as a key into the hash table which forms the multicast
2675  * forwarding database (mFDB).
2676  *
2677  * The table keys are the multicast addresses, while the table entries
2678  * are pointers to lists of ports which wish to receive packets for the
2679  * specified multicast address.
2680  *
2681  * When a multicast packet is being switched we use the address as a key
2682  * into the hash table, and then walk the appropriate port list forwarding
2683  * the pkt to each port in turn.
2684  *
2685  * If a vnet is no longer interested in a particular multicast grouping
2686  * we simply find the correct location in the hash table and then delete
2687  * the relevant port from the port list.
2688  *
2689  * To deal with the case whereby a port is being deleted without first
2690  * removing itself from the lists in the hash table, we maintain a list
2691  * of multicast addresses the port has registered an interest in, within
2692  * the port structure itself. We then simply walk that list of addresses
2693  * using them as keys into the hash table and remove the port from the
2694  * appropriate lists.
2695  */
2696 static void
2697 vsw_process_ctrl_mcst_pkt(vsw_ldc_t *ldcp, void *pkt)
2698 {
2699 	vnet_mcast_msg_t	*mcst_pkt;
2700 	vsw_port_t		*port = ldcp->ldc_port;
2701 	vsw_t			*vswp = ldcp->ldc_vswp;
2702 	int			i;
2703 
2704 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
2705 
2706 	/*
2707 	 * We know this is a ctrl/mcast packet so
2708 	 * cast it into the correct structure.
2709 	 */
2710 	mcst_pkt = (vnet_mcast_msg_t *)pkt;
2711 
2712 	switch (mcst_pkt->tag.vio_subtype) {
2713 	case VIO_SUBTYPE_INFO:
2714 		D2(vswp, "%s: VIO_SUBTYPE_INFO", __func__);
2715 
2716 		/*
2717 		 * Check if in correct state to receive a multicast
2718 		 * message (i.e. handshake complete). If not reset
2719 		 * the handshake.
2720 		 */
2721 		if (vsw_check_flag(ldcp, INBOUND, VSW_MCST_INFO_RECV))
2722 			return;
2723 
2724 		/*
2725 		 * Before attempting to add or remove address check
2726 		 * that they are valid multicast addresses.
2727 		 * If not, then NACK back.
2728 		 */
2729 		for (i = 0; i < mcst_pkt->count; i++) {
2730 			if ((mcst_pkt->mca[i].ether_addr_octet[0] & 01) != 1) {
2731 				DERR(vswp, "%s: invalid multicast address",
2732 				    __func__);
2733 				SND_MCST_NACK(ldcp, mcst_pkt);
2734 				return;
2735 			}
2736 		}
2737 
2738 		/*
2739 		 * Now add/remove the addresses. If this fails we
2740 		 * NACK back.
2741 		 */
2742 		if (vsw_add_rem_mcst(mcst_pkt, port) != 0) {
2743 			SND_MCST_NACK(ldcp, mcst_pkt);
2744 			return;
2745 		}
2746 
2747 		mcst_pkt->tag.vio_subtype = VIO_SUBTYPE_ACK;
2748 		mcst_pkt->tag.vio_sid = ldcp->local_session;
2749 
2750 		DUMP_TAG_PTR((vio_msg_tag_t *)mcst_pkt);
2751 
2752 		(void) vsw_send_msg(ldcp, (void *)mcst_pkt,
2753 		    sizeof (vnet_mcast_msg_t), B_TRUE);
2754 		break;
2755 
2756 	case VIO_SUBTYPE_ACK:
2757 		DWARN(vswp, "%s: VIO_SUBTYPE_ACK", __func__);
2758 
2759 		/*
2760 		 * We shouldn't ever get a multicast ACK message as
2761 		 * at the moment we never request multicast addresses
2762 		 * to be set on some other device. This may change in
2763 		 * the future if we have cascading switches.
2764 		 */
2765 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_MCST_ACK_RECV))
2766 			return;
2767 
2768 				/* Do nothing */
2769 		break;
2770 
2771 	case VIO_SUBTYPE_NACK:
2772 		DWARN(vswp, "%s: VIO_SUBTYPE_NACK", __func__);
2773 
2774 		/*
2775 		 * We shouldn't get a multicast NACK packet for the
2776 		 * same reasons as we shouldn't get a ACK packet.
2777 		 */
2778 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_MCST_NACK_RECV))
2779 			return;
2780 
2781 				/* Do nothing */
2782 		break;
2783 
2784 	default:
2785 		DERR(vswp, "%s: unknown vio_subtype %x\n", __func__,
2786 		    mcst_pkt->tag.vio_subtype);
2787 	}
2788 
2789 	D1(vswp, "%s(%lld): exit", __func__, ldcp->ldc_id);
2790 }
2791 
2792 static void
2793 vsw_process_ctrl_rdx_pkt(vsw_ldc_t *ldcp, void *pkt)
2794 {
2795 	vio_rdx_msg_t	*rdx_pkt;
2796 	vsw_t		*vswp = ldcp->ldc_vswp;
2797 
2798 	/*
2799 	 * We know this is a ctrl/rdx packet so
2800 	 * cast it into the correct structure.
2801 	 */
2802 	rdx_pkt = (vio_rdx_msg_t *)pkt;
2803 
2804 	D1(vswp, "%s(%lld) enter", __func__, ldcp->ldc_id);
2805 
2806 	switch (rdx_pkt->tag.vio_subtype) {
2807 	case VIO_SUBTYPE_INFO:
2808 		D2(vswp, "%s: VIO_SUBTYPE_INFO", __func__);
2809 
2810 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_RDX_INFO_RECV))
2811 			return;
2812 
2813 		rdx_pkt->tag.vio_sid = ldcp->local_session;
2814 		rdx_pkt->tag.vio_subtype = VIO_SUBTYPE_ACK;
2815 
2816 		DUMP_TAG_PTR((vio_msg_tag_t *)rdx_pkt);
2817 
2818 		ldcp->lane_out.lstate |= VSW_RDX_ACK_SENT;
2819 
2820 		(void) vsw_send_msg(ldcp, (void *)rdx_pkt,
2821 		    sizeof (vio_rdx_msg_t), B_TRUE);
2822 
2823 		vsw_next_milestone(ldcp);
2824 		break;
2825 
2826 	case VIO_SUBTYPE_ACK:
2827 		/*
2828 		 * Should be handled in-band by callback handler.
2829 		 */
2830 		DERR(vswp, "%s: Unexpected VIO_SUBTYPE_ACK", __func__);
2831 		vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
2832 		break;
2833 
2834 	case VIO_SUBTYPE_NACK:
2835 		D2(vswp, "%s: VIO_SUBTYPE_NACK", __func__);
2836 
2837 		if (vsw_check_flag(ldcp, INBOUND, VSW_RDX_NACK_RECV))
2838 			return;
2839 
2840 		ldcp->lane_in.lstate |= VSW_RDX_NACK_RECV;
2841 		vsw_next_milestone(ldcp);
2842 		break;
2843 
2844 	default:
2845 		DERR(vswp, "%s: Unknown vio_subtype %x\n", __func__,
2846 		    rdx_pkt->tag.vio_subtype);
2847 	}
2848 
2849 	D1(vswp, "%s(%lld): exit", __func__, ldcp->ldc_id);
2850 }
2851 
2852 static void
2853 vsw_process_data_pkt(vsw_ldc_t *ldcp, void *dpkt, vio_msg_tag_t tag)
2854 {
2855 	uint16_t	env = tag.vio_subtype_env;
2856 	vsw_t		*vswp = ldcp->ldc_vswp;
2857 
2858 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
2859 
2860 	/* session id check */
2861 	if (ldcp->session_status & VSW_PEER_SESSION) {
2862 		if (ldcp->peer_session != tag.vio_sid) {
2863 			DERR(vswp, "%s (chan %d): invalid session id (%llx)",
2864 			    __func__, ldcp->ldc_id, tag.vio_sid);
2865 			vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
2866 			return;
2867 		}
2868 	}
2869 
2870 	/*
2871 	 * It is an error for us to be getting data packets
2872 	 * before the handshake has completed.
2873 	 */
2874 	if (ldcp->hphase != VSW_MILESTONE4) {
2875 		DERR(vswp, "%s: got data packet before handshake complete "
2876 		    "hphase %d (%x: %x)", __func__, ldcp->hphase,
2877 		    ldcp->lane_in.lstate, ldcp->lane_out.lstate);
2878 		DUMP_FLAGS(ldcp->lane_in.lstate);
2879 		DUMP_FLAGS(ldcp->lane_out.lstate);
2880 		vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
2881 		return;
2882 	}
2883 
2884 	/*
2885 	 * To reduce the locking contention, release the
2886 	 * ldc_cblock here and re-acquire it once we are done
2887 	 * receiving packets.
2888 	 */
2889 	mutex_exit(&ldcp->ldc_cblock);
2890 	mutex_enter(&ldcp->ldc_rxlock);
2891 
2892 	/*
2893 	 * Switch on vio_subtype envelope, then let lower routines
2894 	 * decide if its an INFO, ACK or NACK packet.
2895 	 */
2896 	if (env == VIO_DRING_DATA) {
2897 		vsw_process_data_dring_pkt(ldcp, dpkt);
2898 	} else if (env == VIO_PKT_DATA) {
2899 		vsw_process_data_raw_pkt(ldcp, dpkt);
2900 	} else if (env == VIO_DESC_DATA) {
2901 		vsw_process_data_ibnd_pkt(ldcp, dpkt);
2902 	} else {
2903 		DERR(vswp, "%s: unknown vio_subtype_env (%x)\n", __func__, env);
2904 	}
2905 
2906 	mutex_exit(&ldcp->ldc_rxlock);
2907 	mutex_enter(&ldcp->ldc_cblock);
2908 
2909 	D1(vswp, "%s(%lld): exit", __func__, ldcp->ldc_id);
2910 }
2911 
2912 #define	SND_DRING_NACK(ldcp, pkt) \
2913 	pkt->tag.vio_subtype = VIO_SUBTYPE_NACK; \
2914 	pkt->tag.vio_sid = ldcp->local_session; \
2915 	(void) vsw_send_msg(ldcp, (void *)pkt, \
2916 			sizeof (vio_dring_msg_t), B_TRUE);
2917 
2918 static void
2919 vsw_process_data_dring_pkt(vsw_ldc_t *ldcp, void *dpkt)
2920 {
2921 	vio_dring_msg_t		*dring_pkt;
2922 	vnet_public_desc_t	*pub_addr = NULL;
2923 	vsw_private_desc_t	*priv_addr = NULL;
2924 	dring_info_t		*dp = NULL;
2925 	vsw_t			*vswp = ldcp->ldc_vswp;
2926 	mblk_t			*mp = NULL;
2927 	mblk_t			*bp = NULL;
2928 	mblk_t			*bpt = NULL;
2929 	size_t			nbytes = 0;
2930 	uint64_t		ncookies = 0;
2931 	uint64_t		chain = 0;
2932 	uint64_t		len;
2933 	uint32_t		pos, start, datalen;
2934 	uint32_t		range_start, range_end;
2935 	int32_t			end, num, cnt = 0;
2936 	int			i, rv, msg_rv = 0;
2937 	boolean_t		ack_needed = B_FALSE;
2938 	boolean_t		prev_desc_ack = B_FALSE;
2939 	int			read_attempts = 0;
2940 	struct ether_header	*ehp;
2941 
2942 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
2943 
2944 	/*
2945 	 * We know this is a data/dring packet so
2946 	 * cast it into the correct structure.
2947 	 */
2948 	dring_pkt = (vio_dring_msg_t *)dpkt;
2949 
2950 	/*
2951 	 * Switch on the vio_subtype. If its INFO then we need to
2952 	 * process the data. If its an ACK we need to make sure
2953 	 * it makes sense (i.e did we send an earlier data/info),
2954 	 * and if its a NACK then we maybe attempt a retry.
2955 	 */
2956 	switch (dring_pkt->tag.vio_subtype) {
2957 	case VIO_SUBTYPE_INFO:
2958 		D2(vswp, "%s(%lld): VIO_SUBTYPE_INFO", __func__, ldcp->ldc_id);
2959 
2960 		READ_ENTER(&ldcp->lane_in.dlistrw);
2961 		if ((dp = vsw_ident2dring(&ldcp->lane_in,
2962 		    dring_pkt->dring_ident)) == NULL) {
2963 			RW_EXIT(&ldcp->lane_in.dlistrw);
2964 
2965 			DERR(vswp, "%s(%lld): unable to find dring from "
2966 			    "ident 0x%llx", __func__, ldcp->ldc_id,
2967 			    dring_pkt->dring_ident);
2968 
2969 			SND_DRING_NACK(ldcp, dring_pkt);
2970 			return;
2971 		}
2972 
2973 		start = pos = dring_pkt->start_idx;
2974 		end = dring_pkt->end_idx;
2975 		len = dp->num_descriptors;
2976 
2977 		range_start = range_end = pos;
2978 
2979 		D2(vswp, "%s(%lld): start index %ld : end %ld\n",
2980 		    __func__, ldcp->ldc_id, start, end);
2981 
2982 		if (end == -1) {
2983 			num = -1;
2984 		} else if (end >= 0) {
2985 			num = end >= pos ? end - pos + 1: (len - pos + 1) + end;
2986 
2987 			/* basic sanity check */
2988 			if (end > len) {
2989 				RW_EXIT(&ldcp->lane_in.dlistrw);
2990 				DERR(vswp, "%s(%lld): endpoint %lld outside "
2991 				    "ring length %lld", __func__,
2992 				    ldcp->ldc_id, end, len);
2993 
2994 				SND_DRING_NACK(ldcp, dring_pkt);
2995 				return;
2996 			}
2997 		} else {
2998 			RW_EXIT(&ldcp->lane_in.dlistrw);
2999 			DERR(vswp, "%s(%lld): invalid endpoint %lld",
3000 			    __func__, ldcp->ldc_id, end);
3001 			SND_DRING_NACK(ldcp, dring_pkt);
3002 			return;
3003 		}
3004 
3005 		while (cnt != num) {
3006 vsw_recheck_desc:
3007 			if ((rv = ldc_mem_dring_acquire(dp->handle,
3008 			    pos, pos)) != 0) {
3009 				RW_EXIT(&ldcp->lane_in.dlistrw);
3010 				DERR(vswp, "%s(%lld): unable to acquire "
3011 				    "descriptor at pos %d: err %d",
3012 				    __func__, pos, ldcp->ldc_id, rv);
3013 				SND_DRING_NACK(ldcp, dring_pkt);
3014 				ldcp->ldc_stats.ierrors++;
3015 				return;
3016 			}
3017 
3018 			pub_addr = (vnet_public_desc_t *)dp->pub_addr + pos;
3019 
3020 			/*
3021 			 * When given a bounded range of descriptors
3022 			 * to process, its an error to hit a descriptor
3023 			 * which is not ready. In the non-bounded case
3024 			 * (end_idx == -1) this simply indicates we have
3025 			 * reached the end of the current active range.
3026 			 */
3027 			if (pub_addr->hdr.dstate != VIO_DESC_READY) {
3028 				/* unbound - no error */
3029 				if (end == -1) {
3030 					if (read_attempts == vsw_read_attempts)
3031 						break;
3032 
3033 					delay(drv_usectohz(vsw_desc_delay));
3034 					read_attempts++;
3035 					goto vsw_recheck_desc;
3036 				}
3037 
3038 				/* bounded - error - so NACK back */
3039 				RW_EXIT(&ldcp->lane_in.dlistrw);
3040 				DERR(vswp, "%s(%lld): descriptor not READY "
3041 				    "(%d)", __func__, ldcp->ldc_id,
3042 				    pub_addr->hdr.dstate);
3043 				SND_DRING_NACK(ldcp, dring_pkt);
3044 				return;
3045 			}
3046 
3047 			DTRACE_PROBE1(read_attempts, int, read_attempts);
3048 
3049 			range_end = pos;
3050 
3051 			/*
3052 			 * If we ACK'd the previous descriptor then now
3053 			 * record the new range start position for later
3054 			 * ACK's.
3055 			 */
3056 			if (prev_desc_ack) {
3057 				range_start = pos;
3058 
3059 				D2(vswp, "%s(%lld): updating range start to be "
3060 				    "%d", __func__, ldcp->ldc_id, range_start);
3061 
3062 				prev_desc_ack = B_FALSE;
3063 			}
3064 
3065 			/*
3066 			 * Data is padded to align on 8 byte boundary,
3067 			 * datalen is actual data length, i.e. minus that
3068 			 * padding.
3069 			 */
3070 			datalen = pub_addr->nbytes;
3071 
3072 			/*
3073 			 * Does peer wish us to ACK when we have finished
3074 			 * with this descriptor ?
3075 			 */
3076 			if (pub_addr->hdr.ack)
3077 				ack_needed = B_TRUE;
3078 
3079 			D2(vswp, "%s(%lld): processing desc %lld at pos"
3080 			    " 0x%llx : dstate 0x%lx : datalen 0x%lx",
3081 			    __func__, ldcp->ldc_id, pos, pub_addr,
3082 			    pub_addr->hdr.dstate, datalen);
3083 
3084 			/*
3085 			 * Mark that we are starting to process descriptor.
3086 			 */
3087 			pub_addr->hdr.dstate = VIO_DESC_ACCEPTED;
3088 
3089 			/*
3090 			 * Ensure that we ask ldc for an aligned
3091 			 * number of bytes.
3092 			 */
3093 			nbytes = (datalen + VNET_IPALIGN + 7) & ~7;
3094 
3095 			mp = vio_multipool_allocb(&ldcp->vmp, nbytes);
3096 			if (mp == NULL) {
3097 				ldcp->ldc_stats.rx_vio_allocb_fail++;
3098 				/*
3099 				 * No free receive buffers available, so
3100 				 * fallback onto allocb(9F). Make sure that
3101 				 * we get a data buffer which is a multiple
3102 				 * of 8 as this is required by ldc_mem_copy.
3103 				 */
3104 				DTRACE_PROBE(allocb);
3105 				if ((mp = allocb(datalen + VNET_IPALIGN + 8,
3106 				    BPRI_MED)) == NULL) {
3107 					DERR(vswp, "%s(%ld): allocb failed",
3108 					    __func__, ldcp->ldc_id);
3109 					pub_addr->hdr.dstate = VIO_DESC_DONE;
3110 					(void) ldc_mem_dring_release(dp->handle,
3111 					    pos, pos);
3112 					ldcp->ldc_stats.ierrors++;
3113 					ldcp->ldc_stats.rx_allocb_fail++;
3114 					break;
3115 				}
3116 			}
3117 
3118 			ncookies = pub_addr->ncookies;
3119 			rv = ldc_mem_copy(ldcp->ldc_handle,
3120 			    (caddr_t)mp->b_rptr, 0, &nbytes,
3121 			    pub_addr->memcookie, ncookies, LDC_COPY_IN);
3122 
3123 			if (rv != 0) {
3124 				DERR(vswp, "%s(%d): unable to copy in data "
3125 				    "from %d cookies in desc %d (rv %d)",
3126 				    __func__, ldcp->ldc_id, ncookies, pos, rv);
3127 				freemsg(mp);
3128 
3129 				pub_addr->hdr.dstate = VIO_DESC_DONE;
3130 				(void) ldc_mem_dring_release(dp->handle,
3131 				    pos, pos);
3132 				ldcp->ldc_stats.ierrors++;
3133 				break;
3134 			} else {
3135 				D2(vswp, "%s(%d): copied in %ld bytes"
3136 				    " using %d cookies", __func__,
3137 				    ldcp->ldc_id, nbytes, ncookies);
3138 			}
3139 
3140 			/* adjust the read pointer to skip over the padding */
3141 			mp->b_rptr += VNET_IPALIGN;
3142 
3143 			/* point to the actual end of data */
3144 			mp->b_wptr = mp->b_rptr + datalen;
3145 
3146 			/* update statistics */
3147 			ehp = (struct ether_header *)mp->b_rptr;
3148 			if (IS_BROADCAST(ehp))
3149 				ldcp->ldc_stats.brdcstrcv++;
3150 			else if (IS_MULTICAST(ehp))
3151 				ldcp->ldc_stats.multircv++;
3152 
3153 			ldcp->ldc_stats.ipackets++;
3154 			ldcp->ldc_stats.rbytes += datalen;
3155 
3156 			/* build a chain of received packets */
3157 			if (bp == NULL) {
3158 				/* first pkt */
3159 				bp = mp;
3160 				bp->b_next = bp->b_prev = NULL;
3161 				bpt = bp;
3162 				chain = 1;
3163 			} else {
3164 				mp->b_next = mp->b_prev = NULL;
3165 				bpt->b_next = mp;
3166 				bpt = mp;
3167 				chain++;
3168 			}
3169 
3170 			/* mark we are finished with this descriptor */
3171 			pub_addr->hdr.dstate = VIO_DESC_DONE;
3172 
3173 			(void) ldc_mem_dring_release(dp->handle, pos, pos);
3174 
3175 			/*
3176 			 * Send an ACK back to peer if requested.
3177 			 */
3178 			if (ack_needed) {
3179 				ack_needed = B_FALSE;
3180 
3181 				dring_pkt->start_idx = range_start;
3182 				dring_pkt->end_idx = range_end;
3183 
3184 				DERR(vswp, "%s(%lld): processed %d %d, ACK"
3185 				    " requested", __func__, ldcp->ldc_id,
3186 				    dring_pkt->start_idx, dring_pkt->end_idx);
3187 
3188 				dring_pkt->dring_process_state = VIO_DP_ACTIVE;
3189 				dring_pkt->tag.vio_subtype = VIO_SUBTYPE_ACK;
3190 				dring_pkt->tag.vio_sid = ldcp->local_session;
3191 
3192 				msg_rv = vsw_send_msg(ldcp, (void *)dring_pkt,
3193 				    sizeof (vio_dring_msg_t), B_FALSE);
3194 
3195 				/*
3196 				 * Check if ACK was successfully sent. If not
3197 				 * we break and deal with that below.
3198 				 */
3199 				if (msg_rv != 0)
3200 					break;
3201 
3202 				prev_desc_ack = B_TRUE;
3203 				range_start = pos;
3204 			}
3205 
3206 			/* next descriptor */
3207 			pos = (pos + 1) % len;
3208 			cnt++;
3209 
3210 			/*
3211 			 * Break out of loop here and stop processing to
3212 			 * allow some other network device (or disk) to
3213 			 * get access to the cpu.
3214 			 */
3215 			if (chain > vsw_chain_len) {
3216 				D3(vswp, "%s(%lld): switching chain of %d "
3217 				    "msgs", __func__, ldcp->ldc_id, chain);
3218 				break;
3219 			}
3220 		}
3221 		RW_EXIT(&ldcp->lane_in.dlistrw);
3222 
3223 		/*
3224 		 * If when we attempted to send the ACK we found that the
3225 		 * channel had been reset then now handle this. We deal with
3226 		 * it here as we cannot reset the channel while holding the
3227 		 * dlistrw lock, and we don't want to acquire/release it
3228 		 * continuously in the above loop, as a channel reset should
3229 		 * be a rare event.
3230 		 */
3231 		if (msg_rv == ECONNRESET) {
3232 			vsw_process_conn_evt(ldcp, VSW_CONN_RESET);
3233 			break;
3234 		}
3235 
3236 		/* send the chain of packets to be switched */
3237 		if (bp != NULL) {
3238 			DTRACE_PROBE1(vsw_rcv_msgs, int, chain);
3239 			D3(vswp, "%s(%lld): switching chain of %d msgs",
3240 			    __func__, ldcp->ldc_id, chain);
3241 			vswp->vsw_switch_frame(vswp, bp, VSW_VNETPORT,
3242 			    ldcp->ldc_port, NULL);
3243 		}
3244 
3245 		DTRACE_PROBE1(msg_cnt, int, cnt);
3246 
3247 		/*
3248 		 * We are now finished so ACK back with the state
3249 		 * set to STOPPING so our peer knows we are finished
3250 		 */
3251 		dring_pkt->tag.vio_subtype = VIO_SUBTYPE_ACK;
3252 		dring_pkt->tag.vio_sid = ldcp->local_session;
3253 
3254 		dring_pkt->dring_process_state = VIO_DP_STOPPED;
3255 
3256 		DTRACE_PROBE(stop_process_sent);
3257 
3258 		/*
3259 		 * We have not processed any more descriptors beyond
3260 		 * the last one we ACK'd.
3261 		 */
3262 		if (prev_desc_ack)
3263 			range_start = range_end;
3264 
3265 		dring_pkt->start_idx = range_start;
3266 		dring_pkt->end_idx = range_end;
3267 
3268 		D2(vswp, "%s(%lld) processed : %d : %d, now stopping",
3269 		    __func__, ldcp->ldc_id, dring_pkt->start_idx,
3270 		    dring_pkt->end_idx);
3271 
3272 		(void) vsw_send_msg(ldcp, (void *)dring_pkt,
3273 		    sizeof (vio_dring_msg_t), B_TRUE);
3274 		break;
3275 
3276 	case VIO_SUBTYPE_ACK:
3277 		D2(vswp, "%s(%lld): VIO_SUBTYPE_ACK", __func__, ldcp->ldc_id);
3278 		/*
3279 		 * Verify that the relevant descriptors are all
3280 		 * marked as DONE
3281 		 */
3282 		READ_ENTER(&ldcp->lane_out.dlistrw);
3283 		if ((dp = vsw_ident2dring(&ldcp->lane_out,
3284 		    dring_pkt->dring_ident)) == NULL) {
3285 			RW_EXIT(&ldcp->lane_out.dlistrw);
3286 			DERR(vswp, "%s: unknown ident in ACK", __func__);
3287 			return;
3288 		}
3289 
3290 		start = end = 0;
3291 		start = dring_pkt->start_idx;
3292 		end = dring_pkt->end_idx;
3293 		len = dp->num_descriptors;
3294 
3295 
3296 		mutex_enter(&dp->dlock);
3297 		dp->last_ack_recv = end;
3298 		ldcp->ldc_stats.dring_data_acks++;
3299 		mutex_exit(&dp->dlock);
3300 
3301 		(void) vsw_reclaim_dring(dp, start);
3302 
3303 		/*
3304 		 * If our peer is stopping processing descriptors then
3305 		 * we check to make sure it has processed all the descriptors
3306 		 * we have updated. If not then we send it a new message
3307 		 * to prompt it to restart.
3308 		 */
3309 		if (dring_pkt->dring_process_state == VIO_DP_STOPPED) {
3310 			DTRACE_PROBE(stop_process_recv);
3311 			D2(vswp, "%s(%lld): got stopping msg : %d : %d",
3312 			    __func__, ldcp->ldc_id, dring_pkt->start_idx,
3313 			    dring_pkt->end_idx);
3314 
3315 			/*
3316 			 * Check next descriptor in public section of ring.
3317 			 * If its marked as READY then we need to prompt our
3318 			 * peer to start processing the ring again.
3319 			 */
3320 			i = (end + 1) % len;
3321 			pub_addr = (vnet_public_desc_t *)dp->pub_addr + i;
3322 			priv_addr = (vsw_private_desc_t *)dp->priv_addr + i;
3323 
3324 			/*
3325 			 * Hold the restart lock across all of this to
3326 			 * make sure that its not possible for us to
3327 			 * decide that a msg needs to be sent in the future
3328 			 * but the sending code having already checked is
3329 			 * about to exit.
3330 			 */
3331 			mutex_enter(&dp->restart_lock);
3332 			ldcp->ldc_stats.dring_stopped_acks++;
3333 			mutex_enter(&priv_addr->dstate_lock);
3334 			if (pub_addr->hdr.dstate == VIO_DESC_READY) {
3335 
3336 				mutex_exit(&priv_addr->dstate_lock);
3337 
3338 				dring_pkt->tag.vio_subtype = VIO_SUBTYPE_INFO;
3339 				dring_pkt->tag.vio_sid = ldcp->local_session;
3340 
3341 				dring_pkt->seq_num =
3342 				    atomic_inc_64_nv(&ldcp->lane_out.seq_num);
3343 
3344 				dring_pkt->start_idx = (end + 1) % len;
3345 				dring_pkt->end_idx = -1;
3346 
3347 				D2(vswp, "%s(%lld) : sending restart msg:"
3348 				    " %d : %d", __func__, ldcp->ldc_id,
3349 				    dring_pkt->start_idx, dring_pkt->end_idx);
3350 
3351 				msg_rv = vsw_send_msg(ldcp, (void *)dring_pkt,
3352 				    sizeof (vio_dring_msg_t), B_FALSE);
3353 				ldcp->ldc_stats.dring_data_msgs++;
3354 
3355 			} else {
3356 				mutex_exit(&priv_addr->dstate_lock);
3357 				dp->restart_reqd = B_TRUE;
3358 			}
3359 			mutex_exit(&dp->restart_lock);
3360 		}
3361 		RW_EXIT(&ldcp->lane_out.dlistrw);
3362 
3363 		/* only do channel reset after dropping dlistrw lock */
3364 		if (msg_rv == ECONNRESET)
3365 			vsw_process_conn_evt(ldcp, VSW_CONN_RESET);
3366 
3367 		break;
3368 
3369 	case VIO_SUBTYPE_NACK:
3370 		DWARN(vswp, "%s(%lld): VIO_SUBTYPE_NACK",
3371 		    __func__, ldcp->ldc_id);
3372 		/*
3373 		 * Something is badly wrong if we are getting NACK's
3374 		 * for our data pkts. So reset the channel.
3375 		 */
3376 		vsw_process_conn_evt(ldcp, VSW_CONN_RESTART);
3377 
3378 		break;
3379 
3380 	default:
3381 		DERR(vswp, "%s(%lld): Unknown vio_subtype %x\n", __func__,
3382 		    ldcp->ldc_id, dring_pkt->tag.vio_subtype);
3383 	}
3384 
3385 	D1(vswp, "%s(%lld) exit", __func__, ldcp->ldc_id);
3386 }
3387 
3388 /*
3389  * VIO_PKT_DATA (a.k.a raw data mode )
3390  *
3391  * Note - currently not supported. Do nothing.
3392  */
3393 static void
3394 vsw_process_data_raw_pkt(vsw_ldc_t *ldcp, void *dpkt)
3395 {
3396 	_NOTE(ARGUNUSED(dpkt))
3397 
3398 	D1(NULL, "%s (%lld): enter\n", __func__, ldcp->ldc_id);
3399 	DERR(NULL, "%s (%lld): currently unsupported", __func__, ldcp->ldc_id);
3400 	D1(NULL, "%s (%lld): exit\n", __func__, ldcp->ldc_id);
3401 }
3402 
3403 /*
3404  * Process an in-band descriptor message (most likely from
3405  * OBP).
3406  */
3407 static void
3408 vsw_process_data_ibnd_pkt(vsw_ldc_t *ldcp, void *pkt)
3409 {
3410 	vnet_ibnd_desc_t	*ibnd_desc;
3411 	dring_info_t		*dp = NULL;
3412 	vsw_private_desc_t	*priv_addr = NULL;
3413 	vsw_t			*vswp = ldcp->ldc_vswp;
3414 	mblk_t			*mp = NULL;
3415 	size_t			nbytes = 0;
3416 	size_t			off = 0;
3417 	uint64_t		idx = 0;
3418 	uint32_t		num = 1, len, datalen = 0;
3419 	uint64_t		ncookies = 0;
3420 	int			i, rv;
3421 	int			j = 0;
3422 
3423 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
3424 
3425 	ibnd_desc = (vnet_ibnd_desc_t *)pkt;
3426 
3427 	switch (ibnd_desc->hdr.tag.vio_subtype) {
3428 	case VIO_SUBTYPE_INFO:
3429 		D1(vswp, "%s: VIO_SUBTYPE_INFO", __func__);
3430 
3431 		if (vsw_check_flag(ldcp, INBOUND, VSW_DRING_INFO_RECV))
3432 			return;
3433 
3434 		/*
3435 		 * Data is padded to align on a 8 byte boundary,
3436 		 * nbytes is actual data length, i.e. minus that
3437 		 * padding.
3438 		 */
3439 		datalen = ibnd_desc->nbytes;
3440 
3441 		D2(vswp, "%s(%lld): processing inband desc : "
3442 		    ": datalen 0x%lx", __func__, ldcp->ldc_id, datalen);
3443 
3444 		ncookies = ibnd_desc->ncookies;
3445 
3446 		/*
3447 		 * allocb(9F) returns an aligned data block. We
3448 		 * need to ensure that we ask ldc for an aligned
3449 		 * number of bytes also.
3450 		 */
3451 		nbytes = datalen;
3452 		if (nbytes & 0x7) {
3453 			off = 8 - (nbytes & 0x7);
3454 			nbytes += off;
3455 		}
3456 
3457 		mp = allocb(datalen, BPRI_MED);
3458 		if (mp == NULL) {
3459 			DERR(vswp, "%s(%lld): allocb failed",
3460 			    __func__, ldcp->ldc_id);
3461 			ldcp->ldc_stats.rx_allocb_fail++;
3462 			return;
3463 		}
3464 
3465 		rv = ldc_mem_copy(ldcp->ldc_handle, (caddr_t)mp->b_rptr,
3466 		    0, &nbytes, ibnd_desc->memcookie, (uint64_t)ncookies,
3467 		    LDC_COPY_IN);
3468 
3469 		if (rv != 0) {
3470 			DERR(vswp, "%s(%d): unable to copy in data from "
3471 			    "%d cookie(s)", __func__, ldcp->ldc_id, ncookies);
3472 			freemsg(mp);
3473 			ldcp->ldc_stats.ierrors++;
3474 			return;
3475 		}
3476 
3477 		D2(vswp, "%s(%d): copied in %ld bytes using %d cookies",
3478 		    __func__, ldcp->ldc_id, nbytes, ncookies);
3479 
3480 		/* point to the actual end of data */
3481 		mp->b_wptr = mp->b_rptr + datalen;
3482 		ldcp->ldc_stats.ipackets++;
3483 		ldcp->ldc_stats.rbytes += datalen;
3484 
3485 		/*
3486 		 * We ACK back every in-band descriptor message we process
3487 		 */
3488 		ibnd_desc->hdr.tag.vio_subtype = VIO_SUBTYPE_ACK;
3489 		ibnd_desc->hdr.tag.vio_sid = ldcp->local_session;
3490 		(void) vsw_send_msg(ldcp, (void *)ibnd_desc,
3491 		    sizeof (vnet_ibnd_desc_t), B_TRUE);
3492 
3493 		/* send the packet to be switched */
3494 		vswp->vsw_switch_frame(vswp, mp, VSW_VNETPORT,
3495 		    ldcp->ldc_port, NULL);
3496 
3497 		break;
3498 
3499 	case VIO_SUBTYPE_ACK:
3500 		D1(vswp, "%s: VIO_SUBTYPE_ACK", __func__);
3501 
3502 		/* Verify the ACK is valid */
3503 		idx = ibnd_desc->hdr.desc_handle;
3504 
3505 		if (idx >= VSW_RING_NUM_EL) {
3506 			cmn_err(CE_WARN, "!vsw%d: corrupted ACK received "
3507 			    "(idx %ld)", vswp->instance, idx);
3508 			return;
3509 		}
3510 
3511 		if ((dp = ldcp->lane_out.dringp) == NULL) {
3512 			DERR(vswp, "%s: no dring found", __func__);
3513 			return;
3514 		}
3515 
3516 		len = dp->num_descriptors;
3517 		/*
3518 		 * If the descriptor we are being ACK'ed for is not the
3519 		 * one we expected, then pkts were lost somwhere, either
3520 		 * when we tried to send a msg, or a previous ACK msg from
3521 		 * our peer. In either case we now reclaim the descriptors
3522 		 * in the range from the last ACK we received up to the
3523 		 * current ACK.
3524 		 */
3525 		if (idx != dp->last_ack_recv) {
3526 			DWARN(vswp, "%s: dropped pkts detected, (%ld, %ld)",
3527 			    __func__, dp->last_ack_recv, idx);
3528 			num = idx >= dp->last_ack_recv ?
3529 			    idx - dp->last_ack_recv + 1:
3530 			    (len - dp->last_ack_recv + 1) + idx;
3531 		}
3532 
3533 		/*
3534 		 * When we sent the in-band message to our peer we
3535 		 * marked the copy in our private ring as READY. We now
3536 		 * check that the descriptor we are being ACK'ed for is in
3537 		 * fact READY, i.e. it is one we have shared with our peer.
3538 		 *
3539 		 * If its not we flag an error, but still reset the descr
3540 		 * back to FREE.
3541 		 */
3542 		for (i = dp->last_ack_recv; j < num; i = (i + 1) % len, j++) {
3543 			priv_addr = (vsw_private_desc_t *)dp->priv_addr + i;
3544 			mutex_enter(&priv_addr->dstate_lock);
3545 			if (priv_addr->dstate != VIO_DESC_READY) {
3546 				DERR(vswp, "%s: (%ld) desc at index %ld not "
3547 				    "READY (0x%lx)", __func__,
3548 				    ldcp->ldc_id, idx, priv_addr->dstate);
3549 				DERR(vswp, "%s: bound %d: ncookies %ld : "
3550 				    "datalen %ld", __func__,
3551 				    priv_addr->bound, priv_addr->ncookies,
3552 				    priv_addr->datalen);
3553 			}
3554 			D2(vswp, "%s: (%lld) freeing descp at %lld", __func__,
3555 			    ldcp->ldc_id, idx);
3556 			/* release resources associated with sent msg */
3557 			bzero(priv_addr->datap, priv_addr->datalen);
3558 			priv_addr->datalen = 0;
3559 			priv_addr->dstate = VIO_DESC_FREE;
3560 			mutex_exit(&priv_addr->dstate_lock);
3561 		}
3562 		/* update to next expected value */
3563 		dp->last_ack_recv = (idx + 1) % dp->num_descriptors;
3564 
3565 		break;
3566 
3567 	case VIO_SUBTYPE_NACK:
3568 		DERR(vswp, "%s: VIO_SUBTYPE_NACK", __func__);
3569 
3570 		/*
3571 		 * We should only get a NACK if our peer doesn't like
3572 		 * something about a message we have sent it. If this
3573 		 * happens we just release the resources associated with
3574 		 * the message. (We are relying on higher layers to decide
3575 		 * whether or not to resend.
3576 		 */
3577 
3578 		/* limit check */
3579 		idx = ibnd_desc->hdr.desc_handle;
3580 
3581 		if (idx >= VSW_RING_NUM_EL) {
3582 			DERR(vswp, "%s: corrupted NACK received (idx %lld)",
3583 			    __func__, idx);
3584 			return;
3585 		}
3586 
3587 		if ((dp = ldcp->lane_out.dringp) == NULL) {
3588 			DERR(vswp, "%s: no dring found", __func__);
3589 			return;
3590 		}
3591 
3592 		priv_addr = (vsw_private_desc_t *)dp->priv_addr;
3593 
3594 		/* move to correct location in ring */
3595 		priv_addr += idx;
3596 
3597 		/* release resources associated with sent msg */
3598 		mutex_enter(&priv_addr->dstate_lock);
3599 		bzero(priv_addr->datap, priv_addr->datalen);
3600 		priv_addr->datalen = 0;
3601 		priv_addr->dstate = VIO_DESC_FREE;
3602 		mutex_exit(&priv_addr->dstate_lock);
3603 
3604 		break;
3605 
3606 	default:
3607 		DERR(vswp, "%s(%lld): Unknown vio_subtype %x\n", __func__,
3608 		    ldcp->ldc_id, ibnd_desc->hdr.tag.vio_subtype);
3609 	}
3610 
3611 	D1(vswp, "%s(%lld) exit", __func__, ldcp->ldc_id);
3612 }
3613 
3614 static void
3615 vsw_process_err_pkt(vsw_ldc_t *ldcp, void *epkt, vio_msg_tag_t tag)
3616 {
3617 	_NOTE(ARGUNUSED(epkt))
3618 
3619 	vsw_t		*vswp = ldcp->ldc_vswp;
3620 	uint16_t	env = tag.vio_subtype_env;
3621 
3622 	D1(vswp, "%s (%lld): enter\n", __func__, ldcp->ldc_id);
3623 
3624 	/*
3625 	 * Error vio_subtypes have yet to be defined. So for
3626 	 * the moment we can't do anything.
3627 	 */
3628 	D2(vswp, "%s: (%x) vio_subtype env", __func__, env);
3629 
3630 	D1(vswp, "%s (%lld): exit\n", __func__, ldcp->ldc_id);
3631 }
3632 
3633 /* transmit the packet over the given port */
3634 int
3635 vsw_portsend(vsw_port_t *port, mblk_t *mp, mblk_t *mpt)
3636 {
3637 	vsw_ldc_list_t 	*ldcl = &port->p_ldclist;
3638 	vsw_ldc_t 	*ldcp;
3639 	mblk_t		*tmp;
3640 	int		status = 0;
3641 
3642 	READ_ENTER(&ldcl->lockrw);
3643 	/*
3644 	 * Note for now, we have a single channel.
3645 	 */
3646 	ldcp = ldcl->head;
3647 	if (ldcp == NULL) {
3648 		DERR(port->p_vswp, "vsw_portsend: no ldc: dropping packet\n");
3649 		freemsg(mp);
3650 		RW_EXIT(&ldcl->lockrw);
3651 		return (1);
3652 	}
3653 
3654 	/*
3655 	 * If the TX thread is enabled, then queue the packets
3656 	 * and signal the tx thread.
3657 	 */
3658 	if (ldcp->tx_thread != NULL) {
3659 		mutex_enter(&ldcp->tx_thr_lock);
3660 		if (ldcp->tx_mhead == NULL) {
3661 			ldcp->tx_mhead = mp;
3662 			ldcp->tx_mtail = mpt;
3663 			cv_signal(&ldcp->tx_thr_cv);
3664 		} else {
3665 			ldcp->tx_mtail->b_next = mp;
3666 			ldcp->tx_mtail = mpt;
3667 		}
3668 		mutex_exit(&ldcp->tx_thr_lock);
3669 	} else {
3670 		while (mp != NULL) {
3671 			tmp = mp->b_next;
3672 			mp->b_next = mp->b_prev = NULL;
3673 			(void) vsw_ldcsend(ldcp, mp, 1);
3674 			mp = tmp;
3675 		}
3676 	}
3677 
3678 	RW_EXIT(&ldcl->lockrw);
3679 
3680 	return (status);
3681 }
3682 
3683 /*
3684  * Transmit the packet over the given LDC channel.
3685  *
3686  * The 'retries' argument indicates how many times a packet
3687  * is retried before it is dropped. Note, the retry is done
3688  * only for a resource related failure, for all other failures
3689  * the packet is dropped immediately.
3690  *
3691  * The 'tx_failure' counter is used as mechanism to track
3692  * continuous failures. Once these failures are more than
3693  * 'vsw_ldc_tx_max_failures' tunable, the packets are tried only
3694  * once and then they are dropped. This is done to avoid
3695  * buffering too many packets.
3696  */
3697 static int
3698 vsw_ldcsend(vsw_ldc_t *ldcp, mblk_t *mp, int retries)
3699 {
3700 	int i;
3701 	int rc;
3702 	int status = 0;
3703 	vsw_port_t *port = ldcp->ldc_port;
3704 	dring_info_t *dp = NULL;
3705 
3706 
3707 	for (i = 0; i < retries; ) {
3708 		/*
3709 		 * Send the message out using the appropriate
3710 		 * transmit function which will free mblock when it
3711 		 * is finished with it.
3712 		 */
3713 		mutex_enter(&port->tx_lock);
3714 		if (port->transmit != NULL) {
3715 			status = (*port->transmit)(ldcp, mp);
3716 		}
3717 		if (status == LDC_TX_SUCCESS) {
3718 			ldcp->tx_failures = 0;
3719 			mutex_exit(&port->tx_lock);
3720 			break;
3721 		} else if (ldcp->tx_failures > vsw_ldc_tx_max_failures) {
3722 			/*
3723 			 * If the failures crossed the threshold then
3724 			 * break here.
3725 			 */
3726 			ldcp->ldc_stats.oerrors++;
3727 			mutex_exit(&port->tx_lock);
3728 			break;
3729 		} else {
3730 			ldcp->tx_failures++;
3731 		}
3732 		i++;	/* increment the counter here */
3733 
3734 		/* If its the last retry, then update the oerror */
3735 		if ((i == retries) && (status == LDC_TX_NORESOURCES)) {
3736 			ldcp->ldc_stats.oerrors++;
3737 		}
3738 		mutex_exit(&port->tx_lock);
3739 
3740 		if (status != LDC_TX_NORESOURCES) {
3741 			/*
3742 			 * No retrying required for errors un-related
3743 			 * to resources.
3744 			 */
3745 			break;
3746 		}
3747 		READ_ENTER(&ldcp->lane_out.dlistrw);
3748 		if ((dp = ldcp->lane_out.dringp) == NULL) {
3749 			RW_EXIT(&ldcp->lane_out.dlistrw);
3750 			break;
3751 		}
3752 		rc = vsw_reclaim_dring(dp, dp->end_idx);
3753 		RW_EXIT(&ldcp->lane_out.dlistrw);
3754 
3755 		/*
3756 		 * Delay only if none were reclaimed
3757 		 * and its not the last retry.
3758 		 */
3759 		if ((rc == 0) && (i < retries)) {
3760 			delay(drv_usectohz(vsw_ldc_tx_delay));
3761 		}
3762 	}
3763 	freemsg(mp);
3764 	return (status);
3765 }
3766 
3767 /*
3768  * Send packet out via descriptor ring to a logical device.
3769  */
3770 static int
3771 vsw_dringsend(vsw_ldc_t *ldcp, mblk_t *mp)
3772 {
3773 	vio_dring_msg_t		dring_pkt;
3774 	dring_info_t		*dp = NULL;
3775 	vsw_private_desc_t	*priv_desc = NULL;
3776 	vnet_public_desc_t	*pub = NULL;
3777 	vsw_t			*vswp = ldcp->ldc_vswp;
3778 	mblk_t			*bp;
3779 	size_t			n, size;
3780 	caddr_t			bufp;
3781 	int			idx;
3782 	int			status = LDC_TX_SUCCESS;
3783 	struct ether_header	*ehp = (struct ether_header *)mp->b_rptr;
3784 
3785 	D1(vswp, "%s(%lld): enter\n", __func__, ldcp->ldc_id);
3786 
3787 	/* TODO: make test a macro */
3788 	if ((!(ldcp->lane_out.lstate & VSW_LANE_ACTIVE)) ||
3789 	    (ldcp->ldc_status != LDC_UP) || (ldcp->ldc_handle == NULL)) {
3790 		DWARN(vswp, "%s(%lld) status(%d) lstate(0x%llx), dropping "
3791 		    "packet\n", __func__, ldcp->ldc_id, ldcp->ldc_status,
3792 		    ldcp->lane_out.lstate);
3793 		ldcp->ldc_stats.oerrors++;
3794 		return (LDC_TX_FAILURE);
3795 	}
3796 
3797 	/*
3798 	 * Note - using first ring only, this may change
3799 	 * in the future.
3800 	 */
3801 	READ_ENTER(&ldcp->lane_out.dlistrw);
3802 	if ((dp = ldcp->lane_out.dringp) == NULL) {
3803 		RW_EXIT(&ldcp->lane_out.dlistrw);
3804 		DERR(vswp, "%s(%lld): no dring for outbound lane on"
3805 		    " channel %d", __func__, ldcp->ldc_id, ldcp->ldc_id);
3806 		ldcp->ldc_stats.oerrors++;
3807 		return (LDC_TX_FAILURE);
3808 	}
3809 
3810 	size = msgsize(mp);
3811 	if (size > (size_t)ETHERMAX) {
3812 		RW_EXIT(&ldcp->lane_out.dlistrw);
3813 		DERR(vswp, "%s(%lld) invalid size (%ld)\n", __func__,
3814 		    ldcp->ldc_id, size);
3815 		ldcp->ldc_stats.oerrors++;
3816 		return (LDC_TX_FAILURE);
3817 	}
3818 
3819 	/*
3820 	 * Find a free descriptor
3821 	 *
3822 	 * Note: for the moment we are assuming that we will only
3823 	 * have one dring going from the switch to each of its
3824 	 * peers. This may change in the future.
3825 	 */
3826 	if (vsw_dring_find_free_desc(dp, &priv_desc, &idx) != 0) {
3827 		D2(vswp, "%s(%lld): no descriptor available for ring "
3828 		    "at 0x%llx", __func__, ldcp->ldc_id, dp);
3829 
3830 		/* nothing more we can do */
3831 		status = LDC_TX_NORESOURCES;
3832 		ldcp->ldc_stats.tx_no_desc++;
3833 		goto vsw_dringsend_free_exit;
3834 	} else {
3835 		D2(vswp, "%s(%lld): free private descriptor found at pos %ld "
3836 		    "addr 0x%llx\n", __func__, ldcp->ldc_id, idx, priv_desc);
3837 	}
3838 
3839 	/* copy data into the descriptor */
3840 	bufp = priv_desc->datap;
3841 	bufp += VNET_IPALIGN;
3842 	for (bp = mp, n = 0; bp != NULL; bp = bp->b_cont) {
3843 		n = MBLKL(bp);
3844 		bcopy(bp->b_rptr, bufp, n);
3845 		bufp += n;
3846 	}
3847 
3848 	priv_desc->datalen = (size < (size_t)ETHERMIN) ? ETHERMIN : size;
3849 
3850 	pub = priv_desc->descp;
3851 	pub->nbytes = priv_desc->datalen;
3852 
3853 	/* update statistics */
3854 	if (IS_BROADCAST(ehp))
3855 		ldcp->ldc_stats.brdcstxmt++;
3856 	else if (IS_MULTICAST(ehp))
3857 		ldcp->ldc_stats.multixmt++;
3858 	ldcp->ldc_stats.opackets++;
3859 	ldcp->ldc_stats.obytes += priv_desc->datalen;
3860 
3861 	mutex_enter(&priv_desc->dstate_lock);
3862 	pub->hdr.dstate = VIO_DESC_READY;
3863 	mutex_exit(&priv_desc->dstate_lock);
3864 
3865 	/*
3866 	 * Determine whether or not we need to send a message to our
3867 	 * peer prompting them to read our newly updated descriptor(s).
3868 	 */
3869 	mutex_enter(&dp->restart_lock);
3870 	if (dp->restart_reqd) {
3871 		dp->restart_reqd = B_FALSE;
3872 		ldcp->ldc_stats.dring_data_msgs++;
3873 		mutex_exit(&dp->restart_lock);
3874 
3875 		/*
3876 		 * Send a vio_dring_msg to peer to prompt them to read
3877 		 * the updated descriptor ring.
3878 		 */
3879 		dring_pkt.tag.vio_msgtype = VIO_TYPE_DATA;
3880 		dring_pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
3881 		dring_pkt.tag.vio_subtype_env = VIO_DRING_DATA;
3882 		dring_pkt.tag.vio_sid = ldcp->local_session;
3883 
3884 		/* Note - for now using first ring */
3885 		dring_pkt.dring_ident = dp->ident;
3886 		dring_pkt.seq_num = atomic_inc_64_nv(&ldcp->lane_out.seq_num);
3887 
3888 		/*
3889 		 * If last_ack_recv is -1 then we know we've not
3890 		 * received any ack's yet, so this must be the first
3891 		 * msg sent, so set the start to the begining of the ring.
3892 		 */
3893 		mutex_enter(&dp->dlock);
3894 		if (dp->last_ack_recv == -1) {
3895 			dring_pkt.start_idx = 0;
3896 		} else {
3897 			dring_pkt.start_idx =
3898 			    (dp->last_ack_recv + 1) % dp->num_descriptors;
3899 		}
3900 		dring_pkt.end_idx = -1;
3901 		mutex_exit(&dp->dlock);
3902 
3903 		D3(vswp, "%s(%lld): dring 0x%llx : ident 0x%llx\n", __func__,
3904 		    ldcp->ldc_id, dp, dring_pkt.dring_ident);
3905 		D3(vswp, "%s(%lld): start %lld : end %lld : seq %lld\n",
3906 		    __func__, ldcp->ldc_id, dring_pkt.start_idx,
3907 		    dring_pkt.end_idx, dring_pkt.seq_num);
3908 
3909 		RW_EXIT(&ldcp->lane_out.dlistrw);
3910 
3911 		(void) vsw_send_msg(ldcp, (void *)&dring_pkt,
3912 		    sizeof (vio_dring_msg_t), B_TRUE);
3913 
3914 		return (status);
3915 
3916 	} else {
3917 		mutex_exit(&dp->restart_lock);
3918 		D2(vswp, "%s(%lld): updating descp %d", __func__,
3919 		    ldcp->ldc_id, idx);
3920 	}
3921 
3922 vsw_dringsend_free_exit:
3923 
3924 	RW_EXIT(&ldcp->lane_out.dlistrw);
3925 
3926 	D1(vswp, "%s(%lld): exit\n", __func__, ldcp->ldc_id);
3927 	return (status);
3928 }
3929 
3930 /*
3931  * Send an in-band descriptor message over ldc.
3932  */
3933 static int
3934 vsw_descrsend(vsw_ldc_t *ldcp, mblk_t *mp)
3935 {
3936 	vsw_t			*vswp = ldcp->ldc_vswp;
3937 	vnet_ibnd_desc_t	ibnd_msg;
3938 	vsw_private_desc_t	*priv_desc = NULL;
3939 	dring_info_t		*dp = NULL;
3940 	size_t			n, size = 0;
3941 	caddr_t			bufp;
3942 	mblk_t			*bp;
3943 	int			idx, i;
3944 	int			status = LDC_TX_SUCCESS;
3945 	static int		warn_msg = 1;
3946 
3947 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
3948 
3949 	ASSERT(mp != NULL);
3950 
3951 	if ((!(ldcp->lane_out.lstate & VSW_LANE_ACTIVE)) ||
3952 	    (ldcp->ldc_status != LDC_UP) || (ldcp->ldc_handle == NULL)) {
3953 		DERR(vswp, "%s(%lld) status(%d) state (0x%llx), dropping pkt",
3954 		    __func__, ldcp->ldc_id, ldcp->ldc_status,
3955 		    ldcp->lane_out.lstate);
3956 		ldcp->ldc_stats.oerrors++;
3957 		return (LDC_TX_FAILURE);
3958 	}
3959 
3960 	/*
3961 	 * only expect single dring to exist, which we use
3962 	 * as an internal buffer, rather than a transfer channel.
3963 	 */
3964 	READ_ENTER(&ldcp->lane_out.dlistrw);
3965 	if ((dp = ldcp->lane_out.dringp) == NULL) {
3966 		DERR(vswp, "%s(%lld): no dring for outbound lane",
3967 		    __func__, ldcp->ldc_id);
3968 		DERR(vswp, "%s(%lld) status(%d) state (0x%llx)", __func__,
3969 		    ldcp->ldc_id, ldcp->ldc_status, ldcp->lane_out.lstate);
3970 		RW_EXIT(&ldcp->lane_out.dlistrw);
3971 		ldcp->ldc_stats.oerrors++;
3972 		return (LDC_TX_FAILURE);
3973 	}
3974 
3975 	size = msgsize(mp);
3976 	if (size > (size_t)ETHERMAX) {
3977 		RW_EXIT(&ldcp->lane_out.dlistrw);
3978 		DERR(vswp, "%s(%lld) invalid size (%ld)\n", __func__,
3979 		    ldcp->ldc_id, size);
3980 		ldcp->ldc_stats.oerrors++;
3981 		return (LDC_TX_FAILURE);
3982 	}
3983 
3984 	/*
3985 	 * Find a free descriptor in our buffer ring
3986 	 */
3987 	if (vsw_dring_find_free_desc(dp, &priv_desc, &idx) != 0) {
3988 		RW_EXIT(&ldcp->lane_out.dlistrw);
3989 		if (warn_msg) {
3990 			DERR(vswp, "%s(%lld): no descriptor available for ring "
3991 			    "at 0x%llx", __func__, ldcp->ldc_id, dp);
3992 			warn_msg = 0;
3993 		}
3994 
3995 		/* nothing more we can do */
3996 		status = LDC_TX_NORESOURCES;
3997 		goto vsw_descrsend_free_exit;
3998 	} else {
3999 		D2(vswp, "%s(%lld): free private descriptor found at pos "
4000 		    "%ld addr 0x%x\n", __func__, ldcp->ldc_id, idx, priv_desc);
4001 		warn_msg = 1;
4002 	}
4003 
4004 	/* copy data into the descriptor */
4005 	bufp = priv_desc->datap;
4006 	for (bp = mp, n = 0; bp != NULL; bp = bp->b_cont) {
4007 		n = MBLKL(bp);
4008 		bcopy(bp->b_rptr, bufp, n);
4009 		bufp += n;
4010 	}
4011 
4012 	priv_desc->datalen = (size < (size_t)ETHERMIN) ? ETHERMIN : size;
4013 
4014 	/* create and send the in-band descp msg */
4015 	ibnd_msg.hdr.tag.vio_msgtype = VIO_TYPE_DATA;
4016 	ibnd_msg.hdr.tag.vio_subtype = VIO_SUBTYPE_INFO;
4017 	ibnd_msg.hdr.tag.vio_subtype_env = VIO_DESC_DATA;
4018 	ibnd_msg.hdr.tag.vio_sid = ldcp->local_session;
4019 
4020 	ibnd_msg.hdr.seq_num = atomic_inc_64_nv(&ldcp->lane_out.seq_num);
4021 
4022 	/*
4023 	 * Copy the mem cookies describing the data from the
4024 	 * private region of the descriptor ring into the inband
4025 	 * descriptor.
4026 	 */
4027 	for (i = 0; i < priv_desc->ncookies; i++) {
4028 		bcopy(&priv_desc->memcookie[i], &ibnd_msg.memcookie[i],
4029 		    sizeof (ldc_mem_cookie_t));
4030 	}
4031 
4032 	ibnd_msg.hdr.desc_handle = idx;
4033 	ibnd_msg.ncookies = priv_desc->ncookies;
4034 	ibnd_msg.nbytes = size;
4035 
4036 	ldcp->ldc_stats.opackets++;
4037 	ldcp->ldc_stats.obytes += size;
4038 
4039 	RW_EXIT(&ldcp->lane_out.dlistrw);
4040 
4041 	(void) vsw_send_msg(ldcp, (void *)&ibnd_msg,
4042 	    sizeof (vnet_ibnd_desc_t), B_TRUE);
4043 
4044 vsw_descrsend_free_exit:
4045 
4046 	D1(vswp, "%s(%lld): exit", __func__, ldcp->ldc_id);
4047 	return (status);
4048 }
4049 
4050 static void
4051 vsw_send_ver(void *arg)
4052 {
4053 	vsw_ldc_t	*ldcp = (vsw_ldc_t *)arg;
4054 	vsw_t		*vswp = ldcp->ldc_vswp;
4055 	lane_t		*lp = &ldcp->lane_out;
4056 	vio_ver_msg_t	ver_msg;
4057 
4058 	D1(vswp, "%s enter", __func__);
4059 
4060 	ver_msg.tag.vio_msgtype = VIO_TYPE_CTRL;
4061 	ver_msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
4062 	ver_msg.tag.vio_subtype_env = VIO_VER_INFO;
4063 	ver_msg.tag.vio_sid = ldcp->local_session;
4064 
4065 	ver_msg.ver_major = vsw_versions[0].ver_major;
4066 	ver_msg.ver_minor = vsw_versions[0].ver_minor;
4067 	ver_msg.dev_class = VDEV_NETWORK_SWITCH;
4068 
4069 	lp->lstate |= VSW_VER_INFO_SENT;
4070 	lp->ver_major = ver_msg.ver_major;
4071 	lp->ver_minor = ver_msg.ver_minor;
4072 
4073 	DUMP_TAG(ver_msg.tag);
4074 
4075 	(void) vsw_send_msg(ldcp, &ver_msg, sizeof (vio_ver_msg_t), B_TRUE);
4076 
4077 	D1(vswp, "%s (%d): exit", __func__, ldcp->ldc_id);
4078 }
4079 
4080 static void
4081 vsw_send_attr(vsw_ldc_t *ldcp)
4082 {
4083 	vsw_t			*vswp = ldcp->ldc_vswp;
4084 	lane_t			*lp = &ldcp->lane_out;
4085 	vnet_attr_msg_t		attr_msg;
4086 
4087 	D1(vswp, "%s (%ld) enter", __func__, ldcp->ldc_id);
4088 
4089 	/*
4090 	 * Subtype is set to INFO by default
4091 	 */
4092 	attr_msg.tag.vio_msgtype = VIO_TYPE_CTRL;
4093 	attr_msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
4094 	attr_msg.tag.vio_subtype_env = VIO_ATTR_INFO;
4095 	attr_msg.tag.vio_sid = ldcp->local_session;
4096 
4097 	/* payload copied from default settings for lane */
4098 	attr_msg.mtu = lp->mtu;
4099 	attr_msg.addr_type = lp->addr_type;
4100 	attr_msg.xfer_mode = lp->xfer_mode;
4101 	attr_msg.ack_freq = lp->xfer_mode;
4102 
4103 	READ_ENTER(&vswp->if_lockrw);
4104 	bcopy(&(vswp->if_addr), &(attr_msg.addr), ETHERADDRL);
4105 	RW_EXIT(&vswp->if_lockrw);
4106 
4107 	ldcp->lane_out.lstate |= VSW_ATTR_INFO_SENT;
4108 
4109 	DUMP_TAG(attr_msg.tag);
4110 
4111 	(void) vsw_send_msg(ldcp, &attr_msg, sizeof (vnet_attr_msg_t), B_TRUE);
4112 
4113 	D1(vswp, "%s (%ld) exit", __func__, ldcp->ldc_id);
4114 }
4115 
4116 /*
4117  * Create dring info msg (which also results in the creation of
4118  * a dring).
4119  */
4120 static vio_dring_reg_msg_t *
4121 vsw_create_dring_info_pkt(vsw_ldc_t *ldcp)
4122 {
4123 	vio_dring_reg_msg_t	*mp;
4124 	dring_info_t		*dp;
4125 	vsw_t			*vswp = ldcp->ldc_vswp;
4126 
4127 	D1(vswp, "vsw_create_dring_info_pkt enter\n");
4128 
4129 	/*
4130 	 * If we can't create a dring, obviously no point sending
4131 	 * a message.
4132 	 */
4133 	if ((dp = vsw_create_dring(ldcp)) == NULL)
4134 		return (NULL);
4135 
4136 	mp = kmem_zalloc(sizeof (vio_dring_reg_msg_t), KM_SLEEP);
4137 
4138 	mp->tag.vio_msgtype = VIO_TYPE_CTRL;
4139 	mp->tag.vio_subtype = VIO_SUBTYPE_INFO;
4140 	mp->tag.vio_subtype_env = VIO_DRING_REG;
4141 	mp->tag.vio_sid = ldcp->local_session;
4142 
4143 	/* payload */
4144 	mp->num_descriptors = dp->num_descriptors;
4145 	mp->descriptor_size = dp->descriptor_size;
4146 	mp->options = dp->options;
4147 	mp->ncookies = dp->ncookies;
4148 	bcopy(&dp->cookie[0], &mp->cookie[0], sizeof (ldc_mem_cookie_t));
4149 
4150 	mp->dring_ident = 0;
4151 
4152 	D1(vswp, "vsw_create_dring_info_pkt exit\n");
4153 
4154 	return (mp);
4155 }
4156 
4157 static void
4158 vsw_send_dring_info(vsw_ldc_t *ldcp)
4159 {
4160 	vio_dring_reg_msg_t	*dring_msg;
4161 	vsw_t			*vswp = ldcp->ldc_vswp;
4162 
4163 	D1(vswp, "%s: (%ld) enter", __func__, ldcp->ldc_id);
4164 
4165 	dring_msg = vsw_create_dring_info_pkt(ldcp);
4166 	if (dring_msg == NULL) {
4167 		cmn_err(CE_WARN, "!vsw%d: %s: error creating msg",
4168 		    vswp->instance, __func__);
4169 		return;
4170 	}
4171 
4172 	ldcp->lane_out.lstate |= VSW_DRING_INFO_SENT;
4173 
4174 	DUMP_TAG_PTR((vio_msg_tag_t *)dring_msg);
4175 
4176 	(void) vsw_send_msg(ldcp, dring_msg,
4177 	    sizeof (vio_dring_reg_msg_t), B_TRUE);
4178 
4179 	kmem_free(dring_msg, sizeof (vio_dring_reg_msg_t));
4180 
4181 	D1(vswp, "%s: (%ld) exit", __func__, ldcp->ldc_id);
4182 }
4183 
4184 static void
4185 vsw_send_rdx(vsw_ldc_t *ldcp)
4186 {
4187 	vsw_t		*vswp = ldcp->ldc_vswp;
4188 	vio_rdx_msg_t	rdx_msg;
4189 
4190 	D1(vswp, "%s (%ld) enter", __func__, ldcp->ldc_id);
4191 
4192 	rdx_msg.tag.vio_msgtype = VIO_TYPE_CTRL;
4193 	rdx_msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
4194 	rdx_msg.tag.vio_subtype_env = VIO_RDX;
4195 	rdx_msg.tag.vio_sid = ldcp->local_session;
4196 
4197 	ldcp->lane_in.lstate |= VSW_RDX_INFO_SENT;
4198 
4199 	DUMP_TAG(rdx_msg.tag);
4200 
4201 	(void) vsw_send_msg(ldcp, &rdx_msg, sizeof (vio_rdx_msg_t), B_TRUE);
4202 
4203 	D1(vswp, "%s (%ld) exit", __func__, ldcp->ldc_id);
4204 }
4205 
4206 /*
4207  * Generic routine to send message out over ldc channel.
4208  *
4209  * It is possible that when we attempt to write over the ldc channel
4210  * that we get notified that it has been reset. Depending on the value
4211  * of the handle_reset flag we either handle that event here or simply
4212  * notify the caller that the channel was reset.
4213  */
4214 static int
4215 vsw_send_msg(vsw_ldc_t *ldcp, void *msgp, int size, boolean_t handle_reset)
4216 {
4217 	int		rv;
4218 	size_t		msglen = size;
4219 	vio_msg_tag_t	*tag = (vio_msg_tag_t *)msgp;
4220 	vsw_t		*vswp = ldcp->ldc_vswp;
4221 
4222 	D1(vswp, "vsw_send_msg (%lld) enter : sending %d bytes",
4223 	    ldcp->ldc_id, size);
4224 
4225 	D2(vswp, "send_msg: type 0x%llx", tag->vio_msgtype);
4226 	D2(vswp, "send_msg: stype 0x%llx", tag->vio_subtype);
4227 	D2(vswp, "send_msg: senv 0x%llx", tag->vio_subtype_env);
4228 
4229 	mutex_enter(&ldcp->ldc_txlock);
4230 	do {
4231 		msglen = size;
4232 		rv = ldc_write(ldcp->ldc_handle, (caddr_t)msgp, &msglen);
4233 	} while (rv == EWOULDBLOCK && --vsw_wretries > 0);
4234 
4235 	if ((rv != 0) || (msglen != size)) {
4236 		DERR(vswp, "vsw_send_msg:ldc_write failed: chan(%lld) rv(%d) "
4237 		    "size (%d) msglen(%d)\n", ldcp->ldc_id, rv, size, msglen);
4238 		ldcp->ldc_stats.oerrors++;
4239 	}
4240 	mutex_exit(&ldcp->ldc_txlock);
4241 
4242 	/*
4243 	 * If channel has been reset we either handle it here or
4244 	 * simply report back that it has been reset and let caller
4245 	 * decide what to do.
4246 	 */
4247 	if (rv == ECONNRESET) {
4248 		DWARN(vswp, "%s (%lld) channel reset", __func__, ldcp->ldc_id);
4249 
4250 		/*
4251 		 * N.B - must never be holding the dlistrw lock when
4252 		 * we do a reset of the channel.
4253 		 */
4254 		if (handle_reset) {
4255 			vsw_process_conn_evt(ldcp, VSW_CONN_RESET);
4256 		}
4257 	}
4258 
4259 	return (rv);
4260 }
4261 
4262 /*
4263  * Remove the specified address from the list of address maintained
4264  * in this port node.
4265  */
4266 mcst_addr_t *
4267 vsw_del_addr(uint8_t devtype, void *arg, uint64_t addr)
4268 {
4269 	vsw_t		*vswp = NULL;
4270 	vsw_port_t	*port = NULL;
4271 	mcst_addr_t	*prev_p = NULL;
4272 	mcst_addr_t	*curr_p = NULL;
4273 
4274 	D1(NULL, "%s: enter : devtype %d : addr 0x%llx",
4275 	    __func__, devtype, addr);
4276 
4277 	if (devtype == VSW_VNETPORT) {
4278 		port = (vsw_port_t *)arg;
4279 		mutex_enter(&port->mca_lock);
4280 		prev_p = curr_p = port->mcap;
4281 	} else {
4282 		vswp = (vsw_t *)arg;
4283 		mutex_enter(&vswp->mca_lock);
4284 		prev_p = curr_p = vswp->mcap;
4285 	}
4286 
4287 	while (curr_p != NULL) {
4288 		if (curr_p->addr == addr) {
4289 			D2(NULL, "%s: address found", __func__);
4290 			/* match found */
4291 			if (prev_p == curr_p) {
4292 				/* list head */
4293 				if (devtype == VSW_VNETPORT)
4294 					port->mcap = curr_p->nextp;
4295 				else
4296 					vswp->mcap = curr_p->nextp;
4297 			} else {
4298 				prev_p->nextp = curr_p->nextp;
4299 			}
4300 			break;
4301 		} else {
4302 			prev_p = curr_p;
4303 			curr_p = curr_p->nextp;
4304 		}
4305 	}
4306 
4307 	if (devtype == VSW_VNETPORT)
4308 		mutex_exit(&port->mca_lock);
4309 	else
4310 		mutex_exit(&vswp->mca_lock);
4311 
4312 	D1(NULL, "%s: exit", __func__);
4313 
4314 	return (curr_p);
4315 }
4316 
4317 /*
4318  * Creates a descriptor ring (dring) and links it into the
4319  * link of outbound drings for this channel.
4320  *
4321  * Returns NULL if creation failed.
4322  */
4323 static dring_info_t *
4324 vsw_create_dring(vsw_ldc_t *ldcp)
4325 {
4326 	vsw_private_desc_t	*priv_addr = NULL;
4327 	vsw_t			*vswp = ldcp->ldc_vswp;
4328 	ldc_mem_info_t		minfo;
4329 	dring_info_t		*dp, *tp;
4330 	int			i;
4331 
4332 	dp = (dring_info_t *)kmem_zalloc(sizeof (dring_info_t), KM_SLEEP);
4333 
4334 	mutex_init(&dp->dlock, NULL, MUTEX_DRIVER, NULL);
4335 
4336 	/* create public section of ring */
4337 	if ((ldc_mem_dring_create(VSW_RING_NUM_EL,
4338 	    VSW_PUB_SIZE, &dp->handle)) != 0) {
4339 
4340 		DERR(vswp, "vsw_create_dring(%lld): ldc dring create "
4341 		    "failed", ldcp->ldc_id);
4342 		goto create_fail_exit;
4343 	}
4344 
4345 	ASSERT(dp->handle != NULL);
4346 
4347 	/*
4348 	 * Get the base address of the public section of the ring.
4349 	 */
4350 	if ((ldc_mem_dring_info(dp->handle, &minfo)) != 0) {
4351 		DERR(vswp, "vsw_create_dring(%lld): dring info failed\n",
4352 		    ldcp->ldc_id);
4353 		goto dring_fail_exit;
4354 	} else {
4355 		ASSERT(minfo.vaddr != 0);
4356 		dp->pub_addr = minfo.vaddr;
4357 	}
4358 
4359 	dp->num_descriptors = VSW_RING_NUM_EL;
4360 	dp->descriptor_size = VSW_PUB_SIZE;
4361 	dp->options = VIO_TX_DRING;
4362 	dp->ncookies = 1;	/* guaranteed by ldc */
4363 
4364 	/*
4365 	 * create private portion of ring
4366 	 */
4367 	dp->priv_addr = (vsw_private_desc_t *)kmem_zalloc(
4368 	    (sizeof (vsw_private_desc_t) * VSW_RING_NUM_EL), KM_SLEEP);
4369 
4370 	if (vsw_setup_ring(ldcp, dp)) {
4371 		DERR(vswp, "%s: unable to setup ring", __func__);
4372 		goto dring_fail_exit;
4373 	}
4374 
4375 	/* haven't used any descriptors yet */
4376 	dp->end_idx = 0;
4377 	dp->last_ack_recv = -1;
4378 
4379 	/* bind dring to the channel */
4380 	if ((ldc_mem_dring_bind(ldcp->ldc_handle, dp->handle,
4381 	    LDC_SHADOW_MAP, LDC_MEM_RW,
4382 	    &dp->cookie[0], &dp->ncookies)) != 0) {
4383 		DERR(vswp, "vsw_create_dring: unable to bind to channel "
4384 		    "%lld", ldcp->ldc_id);
4385 		goto dring_fail_exit;
4386 	}
4387 
4388 	mutex_init(&dp->restart_lock, NULL, MUTEX_DRIVER, NULL);
4389 	dp->restart_reqd = B_TRUE;
4390 
4391 	/*
4392 	 * Only ever create rings for outgoing lane. Link it onto
4393 	 * end of list.
4394 	 */
4395 	WRITE_ENTER(&ldcp->lane_out.dlistrw);
4396 	if (ldcp->lane_out.dringp == NULL) {
4397 		D2(vswp, "vsw_create_dring: adding first outbound ring");
4398 		ldcp->lane_out.dringp = dp;
4399 	} else {
4400 		tp = ldcp->lane_out.dringp;
4401 		while (tp->next != NULL)
4402 			tp = tp->next;
4403 
4404 		tp->next = dp;
4405 	}
4406 	RW_EXIT(&ldcp->lane_out.dlistrw);
4407 
4408 	return (dp);
4409 
4410 dring_fail_exit:
4411 	(void) ldc_mem_dring_destroy(dp->handle);
4412 
4413 create_fail_exit:
4414 	if (dp->priv_addr != NULL) {
4415 		priv_addr = dp->priv_addr;
4416 		for (i = 0; i < VSW_RING_NUM_EL; i++) {
4417 			if (priv_addr->memhandle != NULL)
4418 				(void) ldc_mem_free_handle(
4419 				    priv_addr->memhandle);
4420 			priv_addr++;
4421 		}
4422 		kmem_free(dp->priv_addr,
4423 		    (sizeof (vsw_private_desc_t) * VSW_RING_NUM_EL));
4424 	}
4425 	mutex_destroy(&dp->dlock);
4426 
4427 	kmem_free(dp, sizeof (dring_info_t));
4428 	return (NULL);
4429 }
4430 
4431 /*
4432  * Create a ring consisting of just a private portion and link
4433  * it into the list of rings for the outbound lane.
4434  *
4435  * These type of rings are used primarily for temporary data
4436  * storage (i.e. as data buffers).
4437  */
4438 void
4439 vsw_create_privring(vsw_ldc_t *ldcp)
4440 {
4441 	dring_info_t		*dp, *tp;
4442 	vsw_t			*vswp = ldcp->ldc_vswp;
4443 
4444 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
4445 
4446 	dp = kmem_zalloc(sizeof (dring_info_t), KM_SLEEP);
4447 
4448 	mutex_init(&dp->dlock, NULL, MUTEX_DRIVER, NULL);
4449 
4450 	/* no public section */
4451 	dp->pub_addr = NULL;
4452 
4453 	dp->priv_addr = kmem_zalloc(
4454 	    (sizeof (vsw_private_desc_t) * VSW_RING_NUM_EL), KM_SLEEP);
4455 
4456 	dp->num_descriptors = VSW_RING_NUM_EL;
4457 
4458 	if (vsw_setup_ring(ldcp, dp)) {
4459 		DERR(vswp, "%s: setup of ring failed", __func__);
4460 		kmem_free(dp->priv_addr,
4461 		    (sizeof (vsw_private_desc_t) * VSW_RING_NUM_EL));
4462 		mutex_destroy(&dp->dlock);
4463 		kmem_free(dp, sizeof (dring_info_t));
4464 		return;
4465 	}
4466 
4467 	/* haven't used any descriptors yet */
4468 	dp->end_idx = 0;
4469 
4470 	mutex_init(&dp->restart_lock, NULL, MUTEX_DRIVER, NULL);
4471 	dp->restart_reqd = B_TRUE;
4472 
4473 	/*
4474 	 * Only ever create rings for outgoing lane. Link it onto
4475 	 * end of list.
4476 	 */
4477 	WRITE_ENTER(&ldcp->lane_out.dlistrw);
4478 	if (ldcp->lane_out.dringp == NULL) {
4479 		D2(vswp, "%s: adding first outbound privring", __func__);
4480 		ldcp->lane_out.dringp = dp;
4481 	} else {
4482 		tp = ldcp->lane_out.dringp;
4483 		while (tp->next != NULL)
4484 			tp = tp->next;
4485 
4486 		tp->next = dp;
4487 	}
4488 	RW_EXIT(&ldcp->lane_out.dlistrw);
4489 
4490 	D1(vswp, "%s(%lld): exit", __func__, ldcp->ldc_id);
4491 }
4492 
4493 /*
4494  * Setup the descriptors in the dring. Returns 0 on success, 1 on
4495  * failure.
4496  */
4497 int
4498 vsw_setup_ring(vsw_ldc_t *ldcp, dring_info_t *dp)
4499 {
4500 	vnet_public_desc_t	*pub_addr = NULL;
4501 	vsw_private_desc_t	*priv_addr = NULL;
4502 	vsw_t			*vswp = ldcp->ldc_vswp;
4503 	uint64_t		*tmpp;
4504 	uint64_t		offset = 0;
4505 	uint32_t		ncookies = 0;
4506 	static char		*name = "vsw_setup_ring";
4507 	int			i, j, nc, rv;
4508 
4509 	priv_addr = dp->priv_addr;
4510 	pub_addr = dp->pub_addr;
4511 
4512 	/* public section may be null but private should never be */
4513 	ASSERT(priv_addr != NULL);
4514 
4515 	/*
4516 	 * Allocate the region of memory which will be used to hold
4517 	 * the data the descriptors will refer to.
4518 	 */
4519 	dp->data_sz = (VSW_RING_NUM_EL * VSW_RING_EL_DATA_SZ);
4520 	dp->data_addr = kmem_alloc(dp->data_sz, KM_SLEEP);
4521 
4522 	D2(vswp, "%s: allocated %lld bytes at 0x%llx\n", name,
4523 	    dp->data_sz, dp->data_addr);
4524 
4525 	tmpp = (uint64_t *)dp->data_addr;
4526 	offset = VSW_RING_EL_DATA_SZ / sizeof (tmpp);
4527 
4528 	/*
4529 	 * Initialise some of the private and public (if they exist)
4530 	 * descriptor fields.
4531 	 */
4532 	for (i = 0; i < VSW_RING_NUM_EL; i++) {
4533 		mutex_init(&priv_addr->dstate_lock, NULL, MUTEX_DRIVER, NULL);
4534 
4535 		if ((ldc_mem_alloc_handle(ldcp->ldc_handle,
4536 		    &priv_addr->memhandle)) != 0) {
4537 			DERR(vswp, "%s: alloc mem handle failed", name);
4538 			goto setup_ring_cleanup;
4539 		}
4540 
4541 		priv_addr->datap = (void *)tmpp;
4542 
4543 		rv = ldc_mem_bind_handle(priv_addr->memhandle,
4544 		    (caddr_t)priv_addr->datap, VSW_RING_EL_DATA_SZ,
4545 		    LDC_SHADOW_MAP, LDC_MEM_R|LDC_MEM_W,
4546 		    &(priv_addr->memcookie[0]), &ncookies);
4547 		if (rv != 0) {
4548 			DERR(vswp, "%s(%lld): ldc_mem_bind_handle failed "
4549 			    "(rv %d)", name, ldcp->ldc_id, rv);
4550 			goto setup_ring_cleanup;
4551 		}
4552 		priv_addr->bound = 1;
4553 
4554 		D2(vswp, "%s: %d: memcookie 0 : addr 0x%llx : size 0x%llx",
4555 		    name, i, priv_addr->memcookie[0].addr,
4556 		    priv_addr->memcookie[0].size);
4557 
4558 		if (ncookies >= (uint32_t)(VSW_MAX_COOKIES + 1)) {
4559 			DERR(vswp, "%s(%lld) ldc_mem_bind_handle returned "
4560 			    "invalid num of cookies (%d) for size 0x%llx",
4561 			    name, ldcp->ldc_id, ncookies, VSW_RING_EL_DATA_SZ);
4562 
4563 			goto setup_ring_cleanup;
4564 		} else {
4565 			for (j = 1; j < ncookies; j++) {
4566 				rv = ldc_mem_nextcookie(priv_addr->memhandle,
4567 				    &(priv_addr->memcookie[j]));
4568 				if (rv != 0) {
4569 					DERR(vswp, "%s: ldc_mem_nextcookie "
4570 					    "failed rv (%d)", name, rv);
4571 					goto setup_ring_cleanup;
4572 				}
4573 				D3(vswp, "%s: memcookie %d : addr 0x%llx : "
4574 				    "size 0x%llx", name, j,
4575 				    priv_addr->memcookie[j].addr,
4576 				    priv_addr->memcookie[j].size);
4577 			}
4578 
4579 		}
4580 		priv_addr->ncookies = ncookies;
4581 		priv_addr->dstate = VIO_DESC_FREE;
4582 
4583 		if (pub_addr != NULL) {
4584 
4585 			/* link pub and private sides */
4586 			priv_addr->descp = pub_addr;
4587 
4588 			pub_addr->ncookies = priv_addr->ncookies;
4589 
4590 			for (nc = 0; nc < pub_addr->ncookies; nc++) {
4591 				bcopy(&priv_addr->memcookie[nc],
4592 				    &pub_addr->memcookie[nc],
4593 				    sizeof (ldc_mem_cookie_t));
4594 			}
4595 
4596 			pub_addr->hdr.dstate = VIO_DESC_FREE;
4597 			pub_addr++;
4598 		}
4599 
4600 		/*
4601 		 * move to next element in the dring and the next
4602 		 * position in the data buffer.
4603 		 */
4604 		priv_addr++;
4605 		tmpp += offset;
4606 	}
4607 
4608 	return (0);
4609 
4610 setup_ring_cleanup:
4611 	priv_addr = dp->priv_addr;
4612 
4613 	for (j = 0; j < i; j++) {
4614 		(void) ldc_mem_unbind_handle(priv_addr->memhandle);
4615 		(void) ldc_mem_free_handle(priv_addr->memhandle);
4616 
4617 		mutex_destroy(&priv_addr->dstate_lock);
4618 
4619 		priv_addr++;
4620 	}
4621 	kmem_free(dp->data_addr, dp->data_sz);
4622 
4623 	return (1);
4624 }
4625 
4626 /*
4627  * Searches the private section of a ring for a free descriptor,
4628  * starting at the location of the last free descriptor found
4629  * previously.
4630  *
4631  * Returns 0 if free descriptor is available, and updates state
4632  * of private descriptor to VIO_DESC_READY,  otherwise returns 1.
4633  *
4634  * FUTURE: might need to return contiguous range of descriptors
4635  * as dring info msg assumes all will be contiguous.
4636  */
4637 static int
4638 vsw_dring_find_free_desc(dring_info_t *dringp,
4639 		vsw_private_desc_t **priv_p, int *idx)
4640 {
4641 	vsw_private_desc_t	*addr = NULL;
4642 	int			num = VSW_RING_NUM_EL;
4643 	int			ret = 1;
4644 
4645 	D1(NULL, "%s enter\n", __func__);
4646 
4647 	ASSERT(dringp->priv_addr != NULL);
4648 
4649 	D2(NULL, "%s: searching ring, dringp 0x%llx : start pos %lld",
4650 	    __func__, dringp, dringp->end_idx);
4651 
4652 	addr = (vsw_private_desc_t *)dringp->priv_addr + dringp->end_idx;
4653 
4654 	mutex_enter(&addr->dstate_lock);
4655 	if (addr->dstate == VIO_DESC_FREE) {
4656 		addr->dstate = VIO_DESC_READY;
4657 		*priv_p = addr;
4658 		*idx = dringp->end_idx;
4659 		dringp->end_idx = (dringp->end_idx + 1) % num;
4660 		ret = 0;
4661 
4662 	}
4663 	mutex_exit(&addr->dstate_lock);
4664 
4665 	/* ring full */
4666 	if (ret == 1) {
4667 		D2(NULL, "%s: no desp free: started at %d", __func__,
4668 		    dringp->end_idx);
4669 	}
4670 
4671 	D1(NULL, "%s: exit\n", __func__);
4672 
4673 	return (ret);
4674 }
4675 
4676 /*
4677  * Map from a dring identifier to the ring itself. Returns
4678  * pointer to ring or NULL if no match found.
4679  *
4680  * Should be called with dlistrw rwlock held as reader.
4681  */
4682 static dring_info_t *
4683 vsw_ident2dring(lane_t *lane, uint64_t ident)
4684 {
4685 	dring_info_t	*dp = NULL;
4686 
4687 	if ((dp = lane->dringp) == NULL) {
4688 		return (NULL);
4689 	} else {
4690 		if (dp->ident == ident)
4691 			return (dp);
4692 
4693 		while (dp != NULL) {
4694 			if (dp->ident == ident)
4695 				break;
4696 			dp = dp->next;
4697 		}
4698 	}
4699 
4700 	return (dp);
4701 }
4702 
4703 /*
4704  * Set the default lane attributes. These are copied into
4705  * the attr msg we send to our peer. If they are not acceptable
4706  * then (currently) the handshake ends.
4707  */
4708 static void
4709 vsw_set_lane_attr(vsw_t *vswp, lane_t *lp)
4710 {
4711 	bzero(lp, sizeof (lane_t));
4712 
4713 	READ_ENTER(&vswp->if_lockrw);
4714 	ether_copy(&(vswp->if_addr), &(lp->addr));
4715 	RW_EXIT(&vswp->if_lockrw);
4716 
4717 	lp->mtu = VSW_MTU;
4718 	lp->addr_type = ADDR_TYPE_MAC;
4719 	lp->xfer_mode = VIO_DRING_MODE;
4720 	lp->ack_freq = 0;	/* for shared mode */
4721 
4722 	/*
4723 	 * As the seq_num is incremented before sending,
4724 	 * initialize it with VNET_ISS - 1.
4725 	 */
4726 	atomic_swap_64(&lp->seq_num, (VNET_ISS - 1));
4727 }
4728 
4729 /*
4730  * Verify that the attributes are acceptable.
4731  *
4732  * FUTURE: If some attributes are not acceptable, change them
4733  * our desired values.
4734  */
4735 static int
4736 vsw_check_attr(vnet_attr_msg_t *pkt, vsw_port_t *port)
4737 {
4738 	int	ret = 0;
4739 
4740 	D1(NULL, "vsw_check_attr enter\n");
4741 
4742 	/*
4743 	 * Note we currently only support in-band descriptors
4744 	 * and descriptor rings, not packet based transfer (VIO_PKT_MODE)
4745 	 */
4746 	if ((pkt->xfer_mode != VIO_DESC_MODE) &&
4747 	    (pkt->xfer_mode != VIO_DRING_MODE)) {
4748 		D2(NULL, "vsw_check_attr: unknown mode %x\n", pkt->xfer_mode);
4749 		ret = 1;
4750 	}
4751 
4752 	/* Only support MAC addresses at moment. */
4753 	if ((pkt->addr_type != ADDR_TYPE_MAC) || (pkt->addr == 0)) {
4754 		D2(NULL, "vsw_check_attr: invalid addr_type %x, "
4755 		    "or address 0x%llx\n", pkt->addr_type, pkt->addr);
4756 		ret = 1;
4757 	}
4758 
4759 	/*
4760 	 * MAC address supplied by device should match that stored
4761 	 * in the vsw-port OBP node. Need to decide what to do if they
4762 	 * don't match, for the moment just warn but don't fail.
4763 	 */
4764 	if (bcmp(&pkt->addr, &port->p_macaddr, ETHERADDRL) != 0) {
4765 		DERR(NULL, "vsw_check_attr: device supplied address "
4766 		    "0x%llx doesn't match node address 0x%llx\n",
4767 		    pkt->addr, port->p_macaddr);
4768 	}
4769 
4770 	/*
4771 	 * Ack freq only makes sense in pkt mode, in shared
4772 	 * mode the ring descriptors say whether or not to
4773 	 * send back an ACK.
4774 	 */
4775 	if ((pkt->xfer_mode == VIO_DRING_MODE) &&
4776 	    (pkt->ack_freq > 0)) {
4777 		D2(NULL, "vsw_check_attr: non zero ack freq "
4778 		    " in SHM mode\n");
4779 		ret = 1;
4780 	}
4781 
4782 	/*
4783 	 * Note: for the moment we only support ETHER
4784 	 * frames. This may change in the future.
4785 	 */
4786 	if ((pkt->mtu > VSW_MTU) || (pkt->mtu <= 0)) {
4787 		D2(NULL, "vsw_check_attr: invalid MTU (0x%llx)\n",
4788 		    pkt->mtu);
4789 		ret = 1;
4790 	}
4791 
4792 	D1(NULL, "vsw_check_attr exit\n");
4793 
4794 	return (ret);
4795 }
4796 
4797 /*
4798  * Returns 1 if there is a problem, 0 otherwise.
4799  */
4800 static int
4801 vsw_check_dring_info(vio_dring_reg_msg_t *pkt)
4802 {
4803 	_NOTE(ARGUNUSED(pkt))
4804 
4805 	int	ret = 0;
4806 
4807 	D1(NULL, "vsw_check_dring_info enter\n");
4808 
4809 	if ((pkt->num_descriptors == 0) ||
4810 	    (pkt->descriptor_size == 0) ||
4811 	    (pkt->ncookies != 1)) {
4812 		DERR(NULL, "vsw_check_dring_info: invalid dring msg");
4813 		ret = 1;
4814 	}
4815 
4816 	D1(NULL, "vsw_check_dring_info exit\n");
4817 
4818 	return (ret);
4819 }
4820 
4821 /*
4822  * Returns 1 if two memory cookies match. Otherwise returns 0.
4823  */
4824 static int
4825 vsw_mem_cookie_match(ldc_mem_cookie_t *m1, ldc_mem_cookie_t *m2)
4826 {
4827 	if ((m1->addr != m2->addr) ||
4828 	    (m2->size != m2->size)) {
4829 		return (0);
4830 	} else {
4831 		return (1);
4832 	}
4833 }
4834 
4835 /*
4836  * Returns 1 if ring described in reg message matches that
4837  * described by dring_info structure. Otherwise returns 0.
4838  */
4839 static int
4840 vsw_dring_match(dring_info_t *dp, vio_dring_reg_msg_t *msg)
4841 {
4842 	if ((msg->descriptor_size != dp->descriptor_size) ||
4843 	    (msg->num_descriptors != dp->num_descriptors) ||
4844 	    (msg->ncookies != dp->ncookies) ||
4845 	    !(vsw_mem_cookie_match(&msg->cookie[0], &dp->cookie[0]))) {
4846 		return (0);
4847 	} else {
4848 		return (1);
4849 	}
4850 
4851 }
4852 
4853 static caddr_t
4854 vsw_print_ethaddr(uint8_t *a, char *ebuf)
4855 {
4856 	(void) sprintf(ebuf, "%x:%x:%x:%x:%x:%x",
4857 	    a[0], a[1], a[2], a[3], a[4], a[5]);
4858 	return (ebuf);
4859 }
4860 
4861 /*
4862  * Reset and free all the resources associated with
4863  * the channel.
4864  */
4865 static void
4866 vsw_free_lane_resources(vsw_ldc_t *ldcp, uint64_t dir)
4867 {
4868 	dring_info_t		*dp, *dpp;
4869 	lane_t			*lp = NULL;
4870 	int			rv = 0;
4871 
4872 	ASSERT(ldcp != NULL);
4873 
4874 	D1(ldcp->ldc_vswp, "%s (%lld): enter", __func__, ldcp->ldc_id);
4875 
4876 	if (dir == INBOUND) {
4877 		D2(ldcp->ldc_vswp, "%s: freeing INBOUND lane"
4878 		    " of channel %lld", __func__, ldcp->ldc_id);
4879 		lp = &ldcp->lane_in;
4880 	} else {
4881 		D2(ldcp->ldc_vswp, "%s: freeing OUTBOUND lane"
4882 		    " of channel %lld", __func__, ldcp->ldc_id);
4883 		lp = &ldcp->lane_out;
4884 	}
4885 
4886 	lp->lstate = VSW_LANE_INACTIV;
4887 
4888 	/*
4889 	 * As the seq_num is incremented before sending,
4890 	 * initialize it with VNET_ISS - 1.
4891 	 */
4892 	atomic_swap_64(&lp->seq_num, (VNET_ISS - 1));
4893 
4894 	if (lp->dringp) {
4895 		if (dir == INBOUND) {
4896 			WRITE_ENTER(&lp->dlistrw);
4897 			dp = lp->dringp;
4898 			while (dp != NULL) {
4899 				dpp = dp->next;
4900 				if (dp->handle != NULL)
4901 					(void) ldc_mem_dring_unmap(dp->handle);
4902 				kmem_free(dp, sizeof (dring_info_t));
4903 				dp = dpp;
4904 			}
4905 			RW_EXIT(&lp->dlistrw);
4906 		} else {
4907 			/*
4908 			 * unbind, destroy exported dring, free dring struct
4909 			 */
4910 			WRITE_ENTER(&lp->dlistrw);
4911 			dp = lp->dringp;
4912 			rv = vsw_free_ring(dp);
4913 			RW_EXIT(&lp->dlistrw);
4914 		}
4915 		if (rv == 0) {
4916 			lp->dringp = NULL;
4917 		}
4918 	}
4919 
4920 	D1(ldcp->ldc_vswp, "%s (%lld): exit", __func__, ldcp->ldc_id);
4921 }
4922 
4923 /*
4924  * Free ring and all associated resources.
4925  *
4926  * Should be called with dlistrw rwlock held as writer.
4927  */
4928 static int
4929 vsw_free_ring(dring_info_t *dp)
4930 {
4931 	vsw_private_desc_t	*paddr = NULL;
4932 	dring_info_t		*dpp;
4933 	int			i, rv = 1;
4934 
4935 	while (dp != NULL) {
4936 		mutex_enter(&dp->dlock);
4937 		dpp = dp->next;
4938 		if (dp->priv_addr != NULL) {
4939 			/*
4940 			 * First unbind and free the memory handles
4941 			 * stored in each descriptor within the ring.
4942 			 */
4943 			for (i = 0; i < VSW_RING_NUM_EL; i++) {
4944 				paddr = (vsw_private_desc_t *)
4945 				    dp->priv_addr + i;
4946 				if (paddr->memhandle != NULL) {
4947 					if (paddr->bound == 1) {
4948 						rv = ldc_mem_unbind_handle(
4949 						    paddr->memhandle);
4950 
4951 						if (rv != 0) {
4952 							DERR(NULL, "error "
4953 							"unbinding handle for "
4954 							"ring 0x%llx at pos %d",
4955 							    dp, i);
4956 							mutex_exit(&dp->dlock);
4957 							return (rv);
4958 						}
4959 						paddr->bound = 0;
4960 					}
4961 
4962 					rv = ldc_mem_free_handle(
4963 					    paddr->memhandle);
4964 					if (rv != 0) {
4965 						DERR(NULL, "error freeing "
4966 						    "handle for ring 0x%llx "
4967 						    "at pos %d", dp, i);
4968 						mutex_exit(&dp->dlock);
4969 						return (rv);
4970 					}
4971 					paddr->memhandle = NULL;
4972 				}
4973 				mutex_destroy(&paddr->dstate_lock);
4974 			}
4975 			kmem_free(dp->priv_addr,
4976 			    (sizeof (vsw_private_desc_t) * VSW_RING_NUM_EL));
4977 		}
4978 
4979 		/*
4980 		 * Now unbind and destroy the ring itself.
4981 		 */
4982 		if (dp->handle != NULL) {
4983 			(void) ldc_mem_dring_unbind(dp->handle);
4984 			(void) ldc_mem_dring_destroy(dp->handle);
4985 		}
4986 
4987 		if (dp->data_addr != NULL) {
4988 			kmem_free(dp->data_addr, dp->data_sz);
4989 		}
4990 
4991 		mutex_exit(&dp->dlock);
4992 		mutex_destroy(&dp->dlock);
4993 		mutex_destroy(&dp->restart_lock);
4994 		kmem_free(dp, sizeof (dring_info_t));
4995 
4996 		dp = dpp;
4997 	}
4998 	return (0);
4999 }
5000 
5001 /*
5002  * vsw_ldc_rx_worker -- A per LDC worker thread to receive data.
5003  * This thread is woken up by the LDC interrupt handler to process
5004  * LDC packets and receive data.
5005  */
5006 static void
5007 vsw_ldc_rx_worker(void *arg)
5008 {
5009 	callb_cpr_t	cprinfo;
5010 	vsw_ldc_t *ldcp = (vsw_ldc_t *)arg;
5011 	vsw_t *vswp = ldcp->ldc_vswp;
5012 
5013 	D1(vswp, "%s(%lld):enter\n", __func__, ldcp->ldc_id);
5014 	CALLB_CPR_INIT(&cprinfo, &ldcp->rx_thr_lock, callb_generic_cpr,
5015 	    "vsw_rx_thread");
5016 	mutex_enter(&ldcp->rx_thr_lock);
5017 	ldcp->rx_thr_flags |= VSW_WTHR_RUNNING;
5018 	while (!(ldcp->rx_thr_flags & VSW_WTHR_STOP)) {
5019 
5020 		CALLB_CPR_SAFE_BEGIN(&cprinfo);
5021 		/*
5022 		 * Wait until the data is received or a stop
5023 		 * request is received.
5024 		 */
5025 		while (!(ldcp->rx_thr_flags &
5026 		    (VSW_WTHR_DATARCVD | VSW_WTHR_STOP))) {
5027 			cv_wait(&ldcp->rx_thr_cv, &ldcp->rx_thr_lock);
5028 		}
5029 		CALLB_CPR_SAFE_END(&cprinfo, &ldcp->rx_thr_lock)
5030 
5031 		/*
5032 		 * First process the stop request.
5033 		 */
5034 		if (ldcp->rx_thr_flags & VSW_WTHR_STOP) {
5035 			D2(vswp, "%s(%lld):Rx thread stopped\n",
5036 			    __func__, ldcp->ldc_id);
5037 			break;
5038 		}
5039 		ldcp->rx_thr_flags &= ~VSW_WTHR_DATARCVD;
5040 		mutex_exit(&ldcp->rx_thr_lock);
5041 		D1(vswp, "%s(%lld):calling vsw_process_pkt\n",
5042 		    __func__, ldcp->ldc_id);
5043 		mutex_enter(&ldcp->ldc_cblock);
5044 		vsw_process_pkt(ldcp);
5045 		mutex_exit(&ldcp->ldc_cblock);
5046 		mutex_enter(&ldcp->rx_thr_lock);
5047 	}
5048 
5049 	/*
5050 	 * Update the run status and wakeup the thread that
5051 	 * has sent the stop request.
5052 	 */
5053 	ldcp->rx_thr_flags &= ~VSW_WTHR_RUNNING;
5054 	cv_signal(&ldcp->rx_thr_cv);
5055 	CALLB_CPR_EXIT(&cprinfo);
5056 	D1(vswp, "%s(%lld):exit\n", __func__, ldcp->ldc_id);
5057 	thread_exit();
5058 }
5059 
5060 /* vsw_stop_rx_thread -- Co-ordinate with receive thread to stop it */
5061 static void
5062 vsw_stop_rx_thread(vsw_ldc_t *ldcp)
5063 {
5064 	vsw_t *vswp = ldcp->ldc_vswp;
5065 
5066 	D1(vswp, "%s(%lld):enter\n", __func__, ldcp->ldc_id);
5067 	/*
5068 	 * Send a stop request by setting the stop flag and
5069 	 * wait until the receive thread stops.
5070 	 */
5071 	mutex_enter(&ldcp->rx_thr_lock);
5072 	if (ldcp->rx_thr_flags & VSW_WTHR_RUNNING) {
5073 		ldcp->rx_thr_flags |= VSW_WTHR_STOP;
5074 		cv_signal(&ldcp->rx_thr_cv);
5075 		while (ldcp->rx_thr_flags & VSW_WTHR_RUNNING) {
5076 			cv_wait(&ldcp->rx_thr_cv, &ldcp->rx_thr_lock);
5077 		}
5078 	}
5079 	mutex_exit(&ldcp->rx_thr_lock);
5080 	ldcp->rx_thread = NULL;
5081 	D1(vswp, "%s(%lld):exit\n", __func__, ldcp->ldc_id);
5082 }
5083 
5084 /*
5085  * vsw_ldc_tx_worker -- A per LDC worker thread to transmit data.
5086  * This thread is woken up by the vsw_portsend to transmit
5087  * packets.
5088  */
5089 static void
5090 vsw_ldc_tx_worker(void *arg)
5091 {
5092 	callb_cpr_t	cprinfo;
5093 	vsw_ldc_t *ldcp = (vsw_ldc_t *)arg;
5094 	vsw_t *vswp = ldcp->ldc_vswp;
5095 	mblk_t *mp;
5096 	mblk_t *tmp;
5097 
5098 	D1(vswp, "%s(%lld):enter\n", __func__, ldcp->ldc_id);
5099 	CALLB_CPR_INIT(&cprinfo, &ldcp->tx_thr_lock, callb_generic_cpr,
5100 	    "vnet_tx_thread");
5101 	mutex_enter(&ldcp->tx_thr_lock);
5102 	ldcp->tx_thr_flags |= VSW_WTHR_RUNNING;
5103 	while (!(ldcp->tx_thr_flags & VSW_WTHR_STOP)) {
5104 
5105 		CALLB_CPR_SAFE_BEGIN(&cprinfo);
5106 		/*
5107 		 * Wait until the data is received or a stop
5108 		 * request is received.
5109 		 */
5110 		while (!(ldcp->tx_thr_flags & VSW_WTHR_STOP) &&
5111 		    (ldcp->tx_mhead == NULL)) {
5112 			cv_wait(&ldcp->tx_thr_cv, &ldcp->tx_thr_lock);
5113 		}
5114 		CALLB_CPR_SAFE_END(&cprinfo, &ldcp->tx_thr_lock)
5115 
5116 		/*
5117 		 * First process the stop request.
5118 		 */
5119 		if (ldcp->tx_thr_flags & VSW_WTHR_STOP) {
5120 			D2(vswp, "%s(%lld):tx thread stopped\n",
5121 			    __func__, ldcp->ldc_id);
5122 			break;
5123 		}
5124 		mp = ldcp->tx_mhead;
5125 		ldcp->tx_mhead = ldcp->tx_mtail = NULL;
5126 		mutex_exit(&ldcp->tx_thr_lock);
5127 		D2(vswp, "%s(%lld):calling vsw_ldcsend\n",
5128 		    __func__, ldcp->ldc_id);
5129 		while (mp != NULL) {
5130 			tmp = mp->b_next;
5131 			mp->b_next = mp->b_prev = NULL;
5132 			(void) vsw_ldcsend(ldcp, mp, vsw_ldc_tx_retries);
5133 			mp = tmp;
5134 		}
5135 		mutex_enter(&ldcp->tx_thr_lock);
5136 	}
5137 
5138 	/*
5139 	 * Update the run status and wakeup the thread that
5140 	 * has sent the stop request.
5141 	 */
5142 	ldcp->tx_thr_flags &= ~VSW_WTHR_RUNNING;
5143 	cv_signal(&ldcp->tx_thr_cv);
5144 	CALLB_CPR_EXIT(&cprinfo);
5145 	D1(vswp, "%s(%lld):exit\n", __func__, ldcp->ldc_id);
5146 	thread_exit();
5147 }
5148 
5149 /* vsw_stop_tx_thread -- Co-ordinate with receive thread to stop it */
5150 static void
5151 vsw_stop_tx_thread(vsw_ldc_t *ldcp)
5152 {
5153 	vsw_t *vswp = ldcp->ldc_vswp;
5154 
5155 	D1(vswp, "%s(%lld):enter\n", __func__, ldcp->ldc_id);
5156 	/*
5157 	 * Send a stop request by setting the stop flag and
5158 	 * wait until the receive thread stops.
5159 	 */
5160 	mutex_enter(&ldcp->tx_thr_lock);
5161 	if (ldcp->tx_thr_flags & VSW_WTHR_RUNNING) {
5162 		ldcp->tx_thr_flags |= VSW_WTHR_STOP;
5163 		cv_signal(&ldcp->tx_thr_cv);
5164 		while (ldcp->tx_thr_flags & VSW_WTHR_RUNNING) {
5165 			cv_wait(&ldcp->tx_thr_cv, &ldcp->tx_thr_lock);
5166 		}
5167 	}
5168 	mutex_exit(&ldcp->tx_thr_lock);
5169 	ldcp->tx_thread = NULL;
5170 	D1(vswp, "%s(%lld):exit\n", __func__, ldcp->ldc_id);
5171 }
5172 
5173 /* vsw_reclaim_dring -- reclaim descriptors */
5174 static int
5175 vsw_reclaim_dring(dring_info_t *dp, int start)
5176 {
5177 	int i, j, len;
5178 	vsw_private_desc_t *priv_addr;
5179 	vnet_public_desc_t *pub_addr;
5180 
5181 	pub_addr = (vnet_public_desc_t *)dp->pub_addr;
5182 	priv_addr = (vsw_private_desc_t *)dp->priv_addr;
5183 	len = dp->num_descriptors;
5184 
5185 	D2(NULL, "%s: start index %ld\n", __func__, start);
5186 
5187 	j = 0;
5188 	for (i = start; j < len; i = (i + 1) % len, j++) {
5189 		pub_addr = (vnet_public_desc_t *)dp->pub_addr + i;
5190 		priv_addr = (vsw_private_desc_t *)dp->priv_addr + i;
5191 
5192 		mutex_enter(&priv_addr->dstate_lock);
5193 		if (pub_addr->hdr.dstate != VIO_DESC_DONE) {
5194 			mutex_exit(&priv_addr->dstate_lock);
5195 			DTRACE_PROBE1(vsw_reclaimed, int, j);
5196 			break;
5197 		}
5198 		pub_addr->hdr.dstate = VIO_DESC_FREE;
5199 		priv_addr->dstate = VIO_DESC_FREE;
5200 		/* clear all the fields */
5201 		priv_addr->datalen = 0;
5202 		pub_addr->hdr.ack = 0;
5203 		mutex_exit(&priv_addr->dstate_lock);
5204 
5205 		D3(NULL, "claiming descp:%d pub state:0x%llx priv state 0x%llx",
5206 		    i, pub_addr->hdr.dstate, priv_addr->dstate);
5207 	}
5208 	return (j);
5209 }
5210 
5211 /*
5212  * Debugging routines
5213  */
5214 static void
5215 display_state(void)
5216 {
5217 	vsw_t		*vswp;
5218 	vsw_port_list_t	*plist;
5219 	vsw_port_t 	*port;
5220 	vsw_ldc_list_t	*ldcl;
5221 	vsw_ldc_t 	*ldcp;
5222 	extern vsw_t 	*vsw_head;
5223 
5224 	cmn_err(CE_NOTE, "***** system state *****");
5225 
5226 	for (vswp = vsw_head; vswp; vswp = vswp->next) {
5227 		plist = &vswp->plist;
5228 		READ_ENTER(&plist->lockrw);
5229 		cmn_err(CE_CONT, "vsw instance %d has %d ports attached\n",
5230 		    vswp->instance, plist->num_ports);
5231 
5232 		for (port = plist->head; port != NULL; port = port->p_next) {
5233 			ldcl = &port->p_ldclist;
5234 			cmn_err(CE_CONT, "port %d : %d ldcs attached\n",
5235 			    port->p_instance, ldcl->num_ldcs);
5236 			READ_ENTER(&ldcl->lockrw);
5237 			ldcp = ldcl->head;
5238 			for (; ldcp != NULL; ldcp = ldcp->ldc_next) {
5239 				cmn_err(CE_CONT, "chan %lu : dev %d : "
5240 				    "status %d : phase %u\n",
5241 				    ldcp->ldc_id, ldcp->dev_class,
5242 				    ldcp->ldc_status, ldcp->hphase);
5243 				cmn_err(CE_CONT, "chan %lu : lsession %lu : "
5244 				    "psession %lu\n", ldcp->ldc_id,
5245 				    ldcp->local_session, ldcp->peer_session);
5246 
5247 				cmn_err(CE_CONT, "Inbound lane:\n");
5248 				display_lane(&ldcp->lane_in);
5249 				cmn_err(CE_CONT, "Outbound lane:\n");
5250 				display_lane(&ldcp->lane_out);
5251 			}
5252 			RW_EXIT(&ldcl->lockrw);
5253 		}
5254 		RW_EXIT(&plist->lockrw);
5255 	}
5256 	cmn_err(CE_NOTE, "***** system state *****");
5257 }
5258 
5259 static void
5260 display_lane(lane_t *lp)
5261 {
5262 	dring_info_t	*drp;
5263 
5264 	cmn_err(CE_CONT, "ver 0x%x:0x%x : state %lx : mtu 0x%lx\n",
5265 	    lp->ver_major, lp->ver_minor, lp->lstate, lp->mtu);
5266 	cmn_err(CE_CONT, "addr_type %d : addr 0x%lx : xmode %d\n",
5267 	    lp->addr_type, lp->addr, lp->xfer_mode);
5268 	cmn_err(CE_CONT, "dringp 0x%lx\n", (uint64_t)lp->dringp);
5269 
5270 	cmn_err(CE_CONT, "Dring info:\n");
5271 	for (drp = lp->dringp; drp != NULL; drp = drp->next) {
5272 		cmn_err(CE_CONT, "\tnum_desc %u : dsize %u\n",
5273 		    drp->num_descriptors, drp->descriptor_size);
5274 		cmn_err(CE_CONT, "\thandle 0x%lx\n", drp->handle);
5275 		cmn_err(CE_CONT, "\tpub_addr 0x%lx : priv_addr 0x%lx\n",
5276 		    (uint64_t)drp->pub_addr, (uint64_t)drp->priv_addr);
5277 		cmn_err(CE_CONT, "\tident 0x%lx : end_idx %lu\n",
5278 		    drp->ident, drp->end_idx);
5279 		display_ring(drp);
5280 	}
5281 }
5282 
5283 static void
5284 display_ring(dring_info_t *dringp)
5285 {
5286 	uint64_t		i;
5287 	uint64_t		priv_count = 0;
5288 	uint64_t		pub_count = 0;
5289 	vnet_public_desc_t	*pub_addr = NULL;
5290 	vsw_private_desc_t	*priv_addr = NULL;
5291 
5292 	for (i = 0; i < VSW_RING_NUM_EL; i++) {
5293 		if (dringp->pub_addr != NULL) {
5294 			pub_addr = (vnet_public_desc_t *)dringp->pub_addr + i;
5295 
5296 			if (pub_addr->hdr.dstate == VIO_DESC_FREE)
5297 				pub_count++;
5298 		}
5299 
5300 		if (dringp->priv_addr != NULL) {
5301 			priv_addr = (vsw_private_desc_t *)dringp->priv_addr + i;
5302 
5303 			if (priv_addr->dstate == VIO_DESC_FREE)
5304 				priv_count++;
5305 		}
5306 	}
5307 	cmn_err(CE_CONT, "\t%lu elements: %lu priv free: %lu pub free\n",
5308 	    i, priv_count, pub_count);
5309 }
5310 
5311 static void
5312 dump_flags(uint64_t state)
5313 {
5314 	int	i;
5315 
5316 	typedef struct flag_name {
5317 		int	flag_val;
5318 		char	*flag_name;
5319 	} flag_name_t;
5320 
5321 	flag_name_t	flags[] = {
5322 		VSW_VER_INFO_SENT, "VSW_VER_INFO_SENT",
5323 		VSW_VER_INFO_RECV, "VSW_VER_INFO_RECV",
5324 		VSW_VER_ACK_RECV, "VSW_VER_ACK_RECV",
5325 		VSW_VER_ACK_SENT, "VSW_VER_ACK_SENT",
5326 		VSW_VER_NACK_RECV, "VSW_VER_NACK_RECV",
5327 		VSW_VER_NACK_SENT, "VSW_VER_NACK_SENT",
5328 		VSW_ATTR_INFO_SENT, "VSW_ATTR_INFO_SENT",
5329 		VSW_ATTR_INFO_RECV, "VSW_ATTR_INFO_RECV",
5330 		VSW_ATTR_ACK_SENT, "VSW_ATTR_ACK_SENT",
5331 		VSW_ATTR_ACK_RECV, "VSW_ATTR_ACK_RECV",
5332 		VSW_ATTR_NACK_SENT, "VSW_ATTR_NACK_SENT",
5333 		VSW_ATTR_NACK_RECV, "VSW_ATTR_NACK_RECV",
5334 		VSW_DRING_INFO_SENT, "VSW_DRING_INFO_SENT",
5335 		VSW_DRING_INFO_RECV, "VSW_DRING_INFO_RECV",
5336 		VSW_DRING_ACK_SENT, "VSW_DRING_ACK_SENT",
5337 		VSW_DRING_ACK_RECV, "VSW_DRING_ACK_RECV",
5338 		VSW_DRING_NACK_SENT, "VSW_DRING_NACK_SENT",
5339 		VSW_DRING_NACK_RECV, "VSW_DRING_NACK_RECV",
5340 		VSW_RDX_INFO_SENT, "VSW_RDX_INFO_SENT",
5341 		VSW_RDX_INFO_RECV, "VSW_RDX_INFO_RECV",
5342 		VSW_RDX_ACK_SENT, "VSW_RDX_ACK_SENT",
5343 		VSW_RDX_ACK_RECV, "VSW_RDX_ACK_RECV",
5344 		VSW_RDX_NACK_SENT, "VSW_RDX_NACK_SENT",
5345 		VSW_RDX_NACK_RECV, "VSW_RDX_NACK_RECV",
5346 		VSW_MCST_INFO_SENT, "VSW_MCST_INFO_SENT",
5347 		VSW_MCST_INFO_RECV, "VSW_MCST_INFO_RECV",
5348 		VSW_MCST_ACK_SENT, "VSW_MCST_ACK_SENT",
5349 		VSW_MCST_ACK_RECV, "VSW_MCST_ACK_RECV",
5350 		VSW_MCST_NACK_SENT, "VSW_MCST_NACK_SENT",
5351 		VSW_MCST_NACK_RECV, "VSW_MCST_NACK_RECV",
5352 		VSW_LANE_ACTIVE, "VSW_LANE_ACTIVE"};
5353 
5354 	DERR(NULL, "DUMP_FLAGS: %llx\n", state);
5355 	for (i = 0; i < sizeof (flags)/sizeof (flag_name_t); i++) {
5356 		if (state & flags[i].flag_val)
5357 			DERR(NULL, "DUMP_FLAGS %s", flags[i].flag_name);
5358 	}
5359 }
5360