xref: /titanic_50/usr/src/uts/sun4v/io/vsw.c (revision ae115bc77f6fcde83175c75b4206dc2e50747966)
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 2006 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 
74 /*
75  * Function prototypes.
76  */
77 static	int vsw_attach(dev_info_t *, ddi_attach_cmd_t);
78 static	int vsw_detach(dev_info_t *, ddi_detach_cmd_t);
79 static	int vsw_getinfo(dev_info_t *, ddi_info_cmd_t, void *, void **);
80 static	int vsw_get_md_physname(vsw_t *, md_t *, mde_cookie_t, char *);
81 static	int vsw_get_md_smodes(vsw_t *, md_t *, mde_cookie_t, uint8_t *, int *);
82 static	int vsw_get_physaddr(vsw_t *);
83 static	int vsw_setup_switching(vsw_t *);
84 static	int vsw_setup_layer2(vsw_t *);
85 static	int vsw_setup_layer3(vsw_t *);
86 
87 /* MAC Ring table functions. */
88 static void vsw_mac_ring_tbl_init(vsw_t *vswp);
89 static void vsw_mac_ring_tbl_destroy(vsw_t *vswp);
90 static void vsw_queue_worker(vsw_mac_ring_t *rrp);
91 static void vsw_queue_stop(vsw_queue_t *vqp);
92 static vsw_queue_t *vsw_queue_create();
93 static void vsw_queue_destroy(vsw_queue_t *vqp);
94 
95 /* MAC layer routines */
96 static mac_resource_handle_t vsw_mac_ring_add_cb(void *arg,
97 		mac_resource_t *mrp);
98 static	int vsw_get_hw_maddr(vsw_t *);
99 static	int vsw_set_hw(vsw_t *, vsw_port_t *);
100 static	int vsw_set_hw_promisc(vsw_t *, vsw_port_t *);
101 static	int vsw_unset_hw(vsw_t *, vsw_port_t *);
102 static	int vsw_unset_hw_promisc(vsw_t *, vsw_port_t *);
103 static	int vsw_reconfig_hw(vsw_t *);
104 static int vsw_mac_attach(vsw_t *vswp);
105 static void vsw_mac_detach(vsw_t *vswp);
106 
107 static void vsw_rx_queue_cb(void *, mac_resource_handle_t, mblk_t *);
108 static void vsw_rx_cb(void *, mac_resource_handle_t, mblk_t *);
109 static mblk_t *vsw_tx_msg(vsw_t *, mblk_t *);
110 static int vsw_mac_register(vsw_t *);
111 static int vsw_mac_unregister(vsw_t *);
112 static int vsw_m_stat(void *, uint_t, uint64_t *);
113 static void vsw_m_stop(void *arg);
114 static int vsw_m_start(void *arg);
115 static int vsw_m_unicst(void *arg, const uint8_t *);
116 static int vsw_m_multicst(void *arg, boolean_t, const uint8_t *);
117 static int vsw_m_promisc(void *arg, boolean_t);
118 static mblk_t *vsw_m_tx(void *arg, mblk_t *);
119 
120 /* MDEG routines */
121 static	int vsw_mdeg_register(vsw_t *vswp);
122 static	void vsw_mdeg_unregister(vsw_t *vswp);
123 static	int vsw_mdeg_cb(void *cb_argp, mdeg_result_t *);
124 static	int vsw_port_mdeg_cb(void *cb_argp, mdeg_result_t *);
125 static	void vsw_get_initial_md_properties(vsw_t *vswp, md_t *, mde_cookie_t);
126 static	void vsw_update_md_prop(vsw_t *, md_t *, mde_cookie_t);
127 
128 /* Port add/deletion routines */
129 static	int vsw_port_add(vsw_t *vswp, md_t *mdp, mde_cookie_t *node);
130 static	int vsw_port_attach(vsw_t *vswp, int p_instance,
131 	uint64_t *ldcids, int nids, struct ether_addr *macaddr);
132 static	int vsw_detach_ports(vsw_t *vswp);
133 static	int vsw_port_detach(vsw_t *vswp, int p_instance);
134 static	int vsw_port_delete(vsw_port_t *port);
135 static	int vsw_ldc_attach(vsw_port_t *port, uint64_t ldc_id);
136 static	int vsw_ldc_detach(vsw_port_t *port, uint64_t ldc_id);
137 static	int vsw_init_ldcs(vsw_port_t *port);
138 static	int vsw_uninit_ldcs(vsw_port_t *port);
139 static	int vsw_ldc_init(vsw_ldc_t *ldcp);
140 static	int vsw_ldc_uninit(vsw_ldc_t *ldcp);
141 static	int vsw_drain_ldcs(vsw_port_t *port);
142 static	int vsw_drain_port_taskq(vsw_port_t *port);
143 static	void vsw_marker_task(void *);
144 static	vsw_port_t *vsw_lookup_port(vsw_t *vswp, int p_instance);
145 static	int vsw_plist_del_node(vsw_t *, vsw_port_t *port);
146 
147 /* Interrupt routines */
148 static	uint_t vsw_ldc_cb(uint64_t cb, caddr_t arg);
149 
150 /* Handshake routines */
151 static	void vsw_restart_ldc(vsw_ldc_t *);
152 static	void vsw_restart_handshake(vsw_ldc_t *);
153 static	void vsw_handle_reset(vsw_ldc_t *);
154 static	int vsw_check_flag(vsw_ldc_t *, int, uint64_t);
155 static	void vsw_next_milestone(vsw_ldc_t *);
156 static	int vsw_supported_version(vio_ver_msg_t *);
157 
158 /* Data processing routines */
159 static void vsw_process_pkt(void *);
160 static void vsw_dispatch_ctrl_task(vsw_ldc_t *, void *, vio_msg_tag_t);
161 static void vsw_process_ctrl_pkt(void *);
162 static void vsw_process_ctrl_ver_pkt(vsw_ldc_t *, void *);
163 static void vsw_process_ctrl_attr_pkt(vsw_ldc_t *, void *);
164 static void vsw_process_ctrl_mcst_pkt(vsw_ldc_t *, void *);
165 static void vsw_process_ctrl_dring_reg_pkt(vsw_ldc_t *, void *);
166 static void vsw_process_ctrl_dring_unreg_pkt(vsw_ldc_t *, void *);
167 static void vsw_process_ctrl_rdx_pkt(vsw_ldc_t *, void *);
168 static void vsw_process_data_pkt(vsw_ldc_t *, void *, vio_msg_tag_t);
169 static void vsw_process_data_dring_pkt(vsw_ldc_t *, void *);
170 static void vsw_process_data_raw_pkt(vsw_ldc_t *, void *);
171 static void vsw_process_data_ibnd_pkt(vsw_ldc_t *, void *);
172 static void vsw_process_err_pkt(vsw_ldc_t *, void *, vio_msg_tag_t);
173 
174 /* Switching/data transmit routines */
175 static	void vsw_switch_l2_frame(vsw_t *vswp, mblk_t *mp, int caller,
176 	    vsw_port_t *port, mac_resource_handle_t);
177 static	void vsw_switch_l3_frame(vsw_t *vswp, mblk_t *mp, int caller,
178 	    vsw_port_t *port, mac_resource_handle_t);
179 static	int vsw_forward_all(vsw_t *vswp, mblk_t *mp, int caller,
180 	    vsw_port_t *port);
181 static	int vsw_forward_grp(vsw_t *vswp, mblk_t *mp, int caller,
182 	    vsw_port_t *port);
183 static	int vsw_portsend(vsw_port_t *, mblk_t *);
184 static	int vsw_dringsend(vsw_ldc_t *, mblk_t *);
185 static	int vsw_descrsend(vsw_ldc_t *, mblk_t *);
186 
187 /* Packet creation routines */
188 static void vsw_send_ver(void *);
189 static void vsw_send_attr(vsw_ldc_t *);
190 static vio_dring_reg_msg_t *vsw_create_dring_info_pkt(vsw_ldc_t *);
191 static void vsw_send_dring_info(vsw_ldc_t *);
192 static void vsw_send_rdx(vsw_ldc_t *);
193 
194 static void vsw_send_msg(vsw_ldc_t *, void *, int);
195 
196 /* Forwarding database (FDB) routines */
197 static	int vsw_add_fdb(vsw_t *vswp, vsw_port_t *port);
198 static	int vsw_del_fdb(vsw_t *vswp, vsw_port_t *port);
199 static	vsw_port_t *vsw_lookup_fdb(vsw_t *vswp, struct ether_header *);
200 static	int vsw_add_rem_mcst(vnet_mcast_msg_t *, vsw_port_t *);
201 static	int vsw_add_mcst(vsw_t *, uint8_t, uint64_t, void *);
202 static	int vsw_del_mcst(vsw_t *, uint8_t, uint64_t, void *);
203 static	void vsw_del_addr(uint8_t, void *, uint64_t);
204 static	void vsw_del_mcst_port(vsw_port_t *);
205 static	void vsw_del_mcst_vsw(vsw_t *);
206 
207 /* Dring routines */
208 static dring_info_t *vsw_create_dring(vsw_ldc_t *);
209 static void vsw_create_privring(vsw_ldc_t *);
210 static int vsw_setup_ring(vsw_ldc_t *ldcp, dring_info_t *dp);
211 static int vsw_dring_find_free_desc(dring_info_t *, vsw_private_desc_t **,
212     int *);
213 static dring_info_t *vsw_ident2dring(lane_t *, uint64_t);
214 
215 static void vsw_set_lane_attr(vsw_t *, lane_t *);
216 static int vsw_check_attr(vnet_attr_msg_t *, vsw_port_t *);
217 static int vsw_dring_match(dring_info_t *dp, vio_dring_reg_msg_t *msg);
218 static int vsw_mem_cookie_match(ldc_mem_cookie_t *, ldc_mem_cookie_t *);
219 static int vsw_check_dring_info(vio_dring_reg_msg_t *);
220 
221 /* Misc support routines */
222 static	caddr_t vsw_print_ethaddr(uint8_t *addr, char *ebuf);
223 static void vsw_free_lane_resources(vsw_ldc_t *, uint64_t);
224 static int vsw_free_ring(dring_info_t *);
225 
226 /* Debugging routines */
227 static void dump_flags(uint64_t);
228 static void display_state(void);
229 static void display_lane(lane_t *);
230 static void display_ring(dring_info_t *);
231 
232 int	vsw_num_handshakes = VNET_NUM_HANDSHAKES; /* # of handshake attempts */
233 int	vsw_wretries = 100;		/* # of write attempts */
234 int	vsw_chain_len = 150;		/* max # of mblks in msg chain */
235 int	vsw_desc_delay = 0;		/* delay in us */
236 int	vsw_read_attempts = 5;		/* # of reads of descriptor */
237 
238 uint32_t	vsw_mblk_size = VSW_MBLK_SIZE;
239 uint32_t	vsw_num_mblks = VSW_NUM_MBLKS;
240 
241 static	mac_callbacks_t	vsw_m_callbacks = {
242 	0,
243 	vsw_m_stat,
244 	vsw_m_start,
245 	vsw_m_stop,
246 	vsw_m_promisc,
247 	vsw_m_multicst,
248 	vsw_m_unicst,
249 	vsw_m_tx,
250 	NULL,
251 	NULL,
252 	NULL
253 };
254 
255 static	struct	cb_ops	vsw_cb_ops = {
256 	nulldev,			/* cb_open */
257 	nulldev,			/* cb_close */
258 	nodev,				/* cb_strategy */
259 	nodev,				/* cb_print */
260 	nodev,				/* cb_dump */
261 	nodev,				/* cb_read */
262 	nodev,				/* cb_write */
263 	nodev,				/* cb_ioctl */
264 	nodev,				/* cb_devmap */
265 	nodev,				/* cb_mmap */
266 	nodev,				/* cb_segmap */
267 	nochpoll,			/* cb_chpoll */
268 	ddi_prop_op,			/* cb_prop_op */
269 	NULL,				/* cb_stream */
270 	D_MP,				/* cb_flag */
271 	CB_REV,				/* rev */
272 	nodev,				/* int (*cb_aread)() */
273 	nodev				/* int (*cb_awrite)() */
274 };
275 
276 static	struct	dev_ops	vsw_ops = {
277 	DEVO_REV,		/* devo_rev */
278 	0,			/* devo_refcnt */
279 	vsw_getinfo,		/* devo_getinfo */
280 	nulldev,		/* devo_identify */
281 	nulldev,		/* devo_probe */
282 	vsw_attach,		/* devo_attach */
283 	vsw_detach,		/* devo_detach */
284 	nodev,			/* devo_reset */
285 	&vsw_cb_ops,		/* devo_cb_ops */
286 	(struct bus_ops *)NULL,	/* devo_bus_ops */
287 	ddi_power		/* devo_power */
288 };
289 
290 extern	struct	mod_ops	mod_driverops;
291 static struct modldrv vswmodldrv = {
292 	&mod_driverops,
293 	"sun4v Virtual Switch Driver %I%",
294 	&vsw_ops,
295 };
296 
297 #define	LDC_ENTER_LOCK(ldcp)	\
298 				mutex_enter(&((ldcp)->ldc_cblock));\
299 				mutex_enter(&((ldcp)->ldc_txlock));
300 #define	LDC_EXIT_LOCK(ldcp)	\
301 				mutex_exit(&((ldcp)->ldc_txlock));\
302 				mutex_exit(&((ldcp)->ldc_cblock));
303 
304 /* Driver soft state ptr  */
305 static void	*vsw_state;
306 
307 /*
308  * Linked list of "vsw_t" structures - one per instance.
309  */
310 vsw_t		*vsw_head = NULL;
311 krwlock_t	vsw_rw;
312 
313 /*
314  * Property names
315  */
316 static char vdev_propname[] = "virtual-device";
317 static char vsw_propname[] = "virtual-network-switch";
318 static char physdev_propname[] = "vsw-phys-dev";
319 static char smode_propname[] = "vsw-switch-mode";
320 static char macaddr_propname[] = "local-mac-address";
321 static char remaddr_propname[] = "remote-mac-address";
322 static char ldcids_propname[] = "ldc-ids";
323 static char chan_propname[] = "channel-endpoint";
324 static char id_propname[] = "id";
325 static char reg_propname[] = "reg";
326 
327 /* supported versions */
328 static	ver_sup_t	vsw_versions[] = { {1, 0} };
329 
330 /*
331  * Matching criteria passed to the MDEG to register interest
332  * in changes to 'virtual-device-port' nodes identified by their
333  * 'id' property.
334  */
335 static md_prop_match_t vport_prop_match[] = {
336 	{ MDET_PROP_VAL,    "id"   },
337 	{ MDET_LIST_END,    NULL    }
338 };
339 
340 static mdeg_node_match_t vport_match = { "virtual-device-port",
341 						vport_prop_match };
342 
343 /*
344  * Matching criteria passed to the MDEG to register interest
345  * in changes to 'virtual-device' nodes (i.e. vsw nodes) identified
346  * by their 'name' and 'cfg-handle' properties.
347  */
348 static md_prop_match_t vdev_prop_match[] = {
349 	{ MDET_PROP_STR,    "name"   },
350 	{ MDET_PROP_VAL,    "cfg-handle" },
351 	{ MDET_LIST_END,    NULL    }
352 };
353 
354 static mdeg_node_match_t vdev_match = { "virtual-device",
355 						vdev_prop_match };
356 
357 
358 /*
359  * Specification of an MD node passed to the MDEG to filter any
360  * 'vport' nodes that do not belong to the specified node. This
361  * template is copied for each vsw instance and filled in with
362  * the appropriate 'cfg-handle' value before being passed to the MDEG.
363  */
364 static mdeg_prop_spec_t vsw_prop_template[] = {
365 	{ MDET_PROP_STR,    "name",		vsw_propname },
366 	{ MDET_PROP_VAL,    "cfg-handle",	NULL	},
367 	{ MDET_LIST_END,    NULL,		NULL	}
368 };
369 
370 #define	VSW_SET_MDEG_PROP_INST(specp, val)	(specp)[1].ps_val = (val);
371 
372 /*
373  * From /etc/system enable/disable thread per ring. This is a mode
374  * selection that is done a vsw driver attach time.
375  */
376 boolean_t vsw_multi_ring_enable = B_FALSE;
377 int vsw_mac_rx_rings = VSW_MAC_RX_RINGS;
378 
379 /*
380  * Print debug messages - set to 0x1f to enable all msgs
381  * or 0x0 to turn all off.
382  */
383 int vswdbg = 0x0;
384 
385 /*
386  * debug levels:
387  * 0x01:	Function entry/exit tracing
388  * 0x02:	Internal function messages
389  * 0x04:	Verbose internal messages
390  * 0x08:	Warning messages
391  * 0x10:	Error messages
392  */
393 
394 static void
395 vswdebug(vsw_t *vswp, const char *fmt, ...)
396 {
397 	char buf[512];
398 	va_list ap;
399 
400 	va_start(ap, fmt);
401 	(void) vsprintf(buf, fmt, ap);
402 	va_end(ap);
403 
404 	if (vswp == NULL)
405 		cmn_err(CE_CONT, "%s\n", buf);
406 	else
407 		cmn_err(CE_CONT, "vsw%d: %s\n", vswp->instance, buf);
408 }
409 
410 /*
411  * For the moment the state dump routines have their own
412  * private flag.
413  */
414 #define	DUMP_STATE	0
415 
416 #if DUMP_STATE
417 
418 #define	DUMP_TAG(tag) \
419 {			\
420 	D1(NULL, "DUMP_TAG: type 0x%llx", (tag).vio_msgtype); \
421 	D1(NULL, "DUMP_TAG: stype 0x%llx", (tag).vio_subtype);	\
422 	D1(NULL, "DUMP_TAG: senv 0x%llx", (tag).vio_subtype_env);	\
423 }
424 
425 #define	DUMP_TAG_PTR(tag) \
426 {			\
427 	D1(NULL, "DUMP_TAG: type 0x%llx", (tag)->vio_msgtype); \
428 	D1(NULL, "DUMP_TAG: stype 0x%llx", (tag)->vio_subtype);	\
429 	D1(NULL, "DUMP_TAG: senv 0x%llx", (tag)->vio_subtype_env);	\
430 }
431 
432 #define	DUMP_FLAGS(flags) dump_flags(flags);
433 #define	DISPLAY_STATE()	display_state()
434 
435 #else
436 
437 #define	DUMP_TAG(tag)
438 #define	DUMP_TAG_PTR(tag)
439 #define	DUMP_FLAGS(state)
440 #define	DISPLAY_STATE()
441 
442 #endif	/* DUMP_STATE */
443 
444 #ifdef DEBUG
445 
446 #define	D1		\
447 if (vswdbg & 0x01)	\
448 	vswdebug
449 
450 #define	D2		\
451 if (vswdbg & 0x02)	\
452 	vswdebug
453 
454 #define	D3		\
455 if (vswdbg & 0x04)	\
456 	vswdebug
457 
458 #define	DWARN		\
459 if (vswdbg & 0x08)	\
460 	vswdebug
461 
462 #define	DERR		\
463 if (vswdbg & 0x10)	\
464 	vswdebug
465 
466 #else
467 
468 #define	DERR		if (0)	vswdebug
469 #define	DWARN		if (0)	vswdebug
470 #define	D1		if (0)	vswdebug
471 #define	D2		if (0)	vswdebug
472 #define	D3		if (0)	vswdebug
473 
474 #endif	/* DEBUG */
475 
476 static struct modlinkage modlinkage = {
477 	MODREV_1,
478 	&vswmodldrv,
479 	NULL
480 };
481 
482 int
483 _init(void)
484 {
485 	int status;
486 
487 	rw_init(&vsw_rw, NULL, RW_DRIVER, NULL);
488 
489 	status = ddi_soft_state_init(&vsw_state, sizeof (vsw_t), 1);
490 	if (status != 0) {
491 		return (status);
492 	}
493 
494 	mac_init_ops(&vsw_ops, "vsw");
495 	status = mod_install(&modlinkage);
496 	if (status != 0) {
497 		ddi_soft_state_fini(&vsw_state);
498 	}
499 	return (status);
500 }
501 
502 int
503 _fini(void)
504 {
505 	int status;
506 
507 	status = mod_remove(&modlinkage);
508 	if (status != 0)
509 		return (status);
510 	mac_fini_ops(&vsw_ops);
511 	ddi_soft_state_fini(&vsw_state);
512 
513 	rw_destroy(&vsw_rw);
514 
515 	return (status);
516 }
517 
518 int
519 _info(struct modinfo *modinfop)
520 {
521 	return (mod_info(&modlinkage, modinfop));
522 }
523 
524 static int
525 vsw_attach(dev_info_t *dip, ddi_attach_cmd_t cmd)
526 {
527 	vsw_t		*vswp;
528 	int		instance;
529 	char		hashname[MAXNAMELEN];
530 	char		qname[TASKQ_NAMELEN];
531 	enum		{ PROG_init = 0x00,
532 				PROG_if_lock = 0x01,
533 				PROG_fdb = 0x02,
534 				PROG_mfdb = 0x04,
535 				PROG_report_dev = 0x08,
536 				PROG_plist = 0x10,
537 				PROG_taskq = 0x20}
538 			progress;
539 
540 	progress = PROG_init;
541 
542 	switch (cmd) {
543 	case DDI_ATTACH:
544 		break;
545 	case DDI_RESUME:
546 		/* nothing to do for this non-device */
547 		return (DDI_SUCCESS);
548 	case DDI_PM_RESUME:
549 	default:
550 		return (DDI_FAILURE);
551 	}
552 
553 	instance = ddi_get_instance(dip);
554 	if (ddi_soft_state_zalloc(vsw_state, instance) != DDI_SUCCESS) {
555 		DERR(NULL, "vsw%d: ddi_soft_state_zalloc failed", instance);
556 		return (DDI_FAILURE);
557 	}
558 	vswp = ddi_get_soft_state(vsw_state, instance);
559 
560 	if (vswp == NULL) {
561 		DERR(NULL, "vsw%d: ddi_get_soft_state failed", instance);
562 		goto vsw_attach_fail;
563 	}
564 
565 	vswp->dip = dip;
566 	vswp->instance = instance;
567 	ddi_set_driver_private(dip, (caddr_t)vswp);
568 
569 	mutex_init(&vswp->mac_lock, NULL, MUTEX_DRIVER, NULL);
570 	rw_init(&vswp->if_lockrw, NULL, RW_DRIVER, NULL);
571 	progress |= PROG_if_lock;
572 
573 	/* setup the unicast forwarding database  */
574 	(void) snprintf(hashname, MAXNAMELEN, "vsw_unicst_table-%d",
575 							vswp->instance);
576 	D2(vswp, "creating unicast hash table (%s)...", hashname);
577 	vswp->fdb = mod_hash_create_ptrhash(hashname, VSW_NCHAINS,
578 		mod_hash_null_valdtor, sizeof (void *));
579 
580 	progress |= PROG_fdb;
581 
582 	/* setup the multicast fowarding database */
583 	(void) snprintf(hashname, MAXNAMELEN, "vsw_mcst_table-%d",
584 							vswp->instance);
585 	D2(vswp, "creating multicast hash table %s)...", hashname);
586 	rw_init(&vswp->mfdbrw, NULL, RW_DRIVER, NULL);
587 	vswp->mfdb = mod_hash_create_ptrhash(hashname, VSW_NCHAINS,
588 			mod_hash_null_valdtor, sizeof (void *));
589 
590 	progress |= PROG_mfdb;
591 
592 	/*
593 	 * create lock protecting list of multicast addresses
594 	 * which could come via m_multicst() entry point when plumbed.
595 	 */
596 	mutex_init(&vswp->mca_lock, NULL, MUTEX_DRIVER, NULL);
597 	vswp->mcap = NULL;
598 
599 	ddi_report_dev(vswp->dip);
600 
601 	progress |= PROG_report_dev;
602 
603 	WRITE_ENTER(&vsw_rw);
604 	vswp->next = vsw_head;
605 	vsw_head = vswp;
606 	RW_EXIT(&vsw_rw);
607 
608 	/* setup the port list */
609 	rw_init(&vswp->plist.lockrw, NULL, RW_DRIVER, NULL);
610 	vswp->plist.head = NULL;
611 
612 	progress |= PROG_plist;
613 
614 	/*
615 	 * Create the taskq which will process all the VIO
616 	 * control messages.
617 	 */
618 	(void) snprintf(qname, TASKQ_NAMELEN, "vsw_taskq%d", vswp->instance);
619 	if ((vswp->taskq_p = ddi_taskq_create(vswp->dip, qname, 1,
620 					TASKQ_DEFAULTPRI, 0)) == NULL) {
621 		cmn_err(CE_WARN, "!vsw%d: Unable to create task queue",
622 			vswp->instance);
623 		goto vsw_attach_fail;
624 	}
625 
626 	progress |= PROG_taskq;
627 
628 	/* prevent auto-detaching */
629 	if (ddi_prop_update_int(DDI_DEV_T_NONE, vswp->dip,
630 				DDI_NO_AUTODETACH, 1) != DDI_SUCCESS) {
631 		cmn_err(CE_NOTE, "!Unable to set \"%s\" property for "
632 			"instance %u", DDI_NO_AUTODETACH, instance);
633 	}
634 
635 	/*
636 	 * Now we have everything setup, register an interest in
637 	 * specific MD nodes.
638 	 *
639 	 * The callback is invoked in 2 cases, firstly if upon mdeg
640 	 * registration there are existing nodes which match our specified
641 	 * criteria, and secondly if the MD is changed (and again, there
642 	 * are nodes which we are interested in present within it. Note
643 	 * that our callback will be invoked even if our specified nodes
644 	 * have not actually changed).
645 	 *
646 	 * Until the callback is invoked we cannot switch any pkts as
647 	 * we don't know basic information such as what mode we are
648 	 * operating in. However we expect the callback to be invoked
649 	 * immediately upon registration as this driver should only
650 	 * be attaching if there are vsw nodes in the MD.
651 	 */
652 	if (vsw_mdeg_register(vswp))
653 		goto vsw_attach_fail;
654 
655 	return (DDI_SUCCESS);
656 
657 vsw_attach_fail:
658 	DERR(NULL, "vsw_attach: failed");
659 
660 	if (progress & PROG_taskq)
661 		ddi_taskq_destroy(vswp->taskq_p);
662 
663 	if (progress & PROG_plist)
664 		rw_destroy(&vswp->plist.lockrw);
665 
666 	if (progress & PROG_report_dev) {
667 		ddi_remove_minor_node(dip, NULL);
668 		mutex_destroy(&vswp->mca_lock);
669 	}
670 
671 	if (progress & PROG_mfdb) {
672 		mod_hash_destroy_hash(vswp->mfdb);
673 		vswp->mfdb = NULL;
674 		rw_destroy(&vswp->mfdbrw);
675 	}
676 
677 	if (progress & PROG_fdb) {
678 		mod_hash_destroy_hash(vswp->fdb);
679 		vswp->fdb = NULL;
680 	}
681 
682 	if (progress & PROG_if_lock) {
683 		rw_destroy(&vswp->if_lockrw);
684 		mutex_destroy(&vswp->mac_lock);
685 	}
686 
687 	ddi_soft_state_free(vsw_state, instance);
688 	return (DDI_FAILURE);
689 }
690 
691 static int
692 vsw_detach(dev_info_t *dip, ddi_detach_cmd_t cmd)
693 {
694 	vio_mblk_pool_t		*poolp, *npoolp;
695 	vsw_t			**vswpp, *vswp;
696 	int 			instance;
697 
698 	instance = ddi_get_instance(dip);
699 	vswp = ddi_get_soft_state(vsw_state, instance);
700 
701 	if (vswp == NULL) {
702 		return (DDI_FAILURE);
703 	}
704 
705 	switch (cmd) {
706 	case DDI_DETACH:
707 		break;
708 	case DDI_SUSPEND:
709 	case DDI_PM_SUSPEND:
710 	default:
711 		return (DDI_FAILURE);
712 	}
713 
714 	D2(vswp, "detaching instance %d", instance);
715 
716 	if (vswp->if_state & VSW_IF_REG) {
717 		if (vsw_mac_unregister(vswp) != 0) {
718 			cmn_err(CE_WARN, "!vsw%d: Unable to detach from "
719 				"MAC layer", vswp->instance);
720 			return (DDI_FAILURE);
721 		}
722 	}
723 
724 	vsw_mdeg_unregister(vswp);
725 
726 	/* remove mac layer callback */
727 	mutex_enter(&vswp->mac_lock);
728 	if ((vswp->mh != NULL) && (vswp->mrh != NULL)) {
729 		mac_rx_remove(vswp->mh, vswp->mrh);
730 		vswp->mrh = NULL;
731 	}
732 	mutex_exit(&vswp->mac_lock);
733 
734 	if (vsw_detach_ports(vswp) != 0) {
735 		cmn_err(CE_WARN, "!vsw%d: Unable to detach ports",
736 							vswp->instance);
737 		return (DDI_FAILURE);
738 	}
739 
740 	rw_destroy(&vswp->if_lockrw);
741 
742 	/*
743 	 * Now that the ports have been deleted, stop and close
744 	 * the physical device.
745 	 */
746 	mutex_enter(&vswp->mac_lock);
747 	if (vswp->mh != NULL) {
748 		if (vswp->mstarted)
749 			mac_stop(vswp->mh);
750 		if (vswp->mresources)
751 			mac_resource_set(vswp->mh, NULL, NULL);
752 		mac_close(vswp->mh);
753 
754 		vswp->mh = NULL;
755 		vswp->txinfo = NULL;
756 	}
757 	mutex_exit(&vswp->mac_lock);
758 	mutex_destroy(&vswp->mac_lock);
759 
760 	/*
761 	 * Destroy any free pools that may still exist.
762 	 */
763 	poolp = vswp->rxh;
764 	while (poolp != NULL) {
765 		npoolp = vswp->rxh = poolp->nextp;
766 		if (vio_destroy_mblks(poolp) != 0) {
767 			vswp->rxh = poolp;
768 			return (DDI_FAILURE);
769 		}
770 		poolp = npoolp;
771 	}
772 
773 	/*
774 	 * Remove this instance from any entries it may be on in
775 	 * the hash table by using the list of addresses maintained
776 	 * in the vsw_t structure.
777 	 */
778 	vsw_del_mcst_vsw(vswp);
779 
780 	vswp->mcap = NULL;
781 	mutex_destroy(&vswp->mca_lock);
782 
783 	/*
784 	 * By now any pending tasks have finished and the underlying
785 	 * ldc's have been destroyed, so its safe to delete the control
786 	 * message taskq.
787 	 */
788 	if (vswp->taskq_p != NULL)
789 		ddi_taskq_destroy(vswp->taskq_p);
790 
791 	/*
792 	 * At this stage all the data pointers in the hash table
793 	 * should be NULL, as all the ports have been removed and will
794 	 * have deleted themselves from the port lists which the data
795 	 * pointers point to. Hence we can destroy the table using the
796 	 * default destructors.
797 	 */
798 	D2(vswp, "vsw_detach: destroying hash tables..");
799 	mod_hash_destroy_hash(vswp->fdb);
800 	vswp->fdb = NULL;
801 
802 	WRITE_ENTER(&vswp->mfdbrw);
803 	mod_hash_destroy_hash(vswp->mfdb);
804 	vswp->mfdb = NULL;
805 	RW_EXIT(&vswp->mfdbrw);
806 	rw_destroy(&vswp->mfdbrw);
807 
808 	ddi_remove_minor_node(dip, NULL);
809 
810 	rw_destroy(&vswp->plist.lockrw);
811 	WRITE_ENTER(&vsw_rw);
812 	for (vswpp = &vsw_head; *vswpp; vswpp = &(*vswpp)->next) {
813 		if (*vswpp == vswp) {
814 			*vswpp = vswp->next;
815 			break;
816 		}
817 	}
818 	RW_EXIT(&vsw_rw);
819 	ddi_soft_state_free(vsw_state, instance);
820 
821 	return (DDI_SUCCESS);
822 }
823 
824 static int
825 vsw_getinfo(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
826 {
827 	_NOTE(ARGUNUSED(dip))
828 
829 	vsw_t	*vswp = NULL;
830 	dev_t	dev = (dev_t)arg;
831 	int	instance;
832 
833 	instance = getminor(dev);
834 
835 	switch (infocmd) {
836 	case DDI_INFO_DEVT2DEVINFO:
837 		if ((vswp = ddi_get_soft_state(vsw_state, instance)) == NULL) {
838 			*result = NULL;
839 			return (DDI_FAILURE);
840 		}
841 		*result = vswp->dip;
842 		return (DDI_SUCCESS);
843 
844 	case DDI_INFO_DEVT2INSTANCE:
845 		*result = (void *)(uintptr_t)instance;
846 		return (DDI_SUCCESS);
847 
848 	default:
849 		*result = NULL;
850 		return (DDI_FAILURE);
851 	}
852 }
853 
854 /*
855  * Get the value of the "vsw-phys-dev" property in the specified
856  * node. This property is the name of the physical device that
857  * the virtual switch will use to talk to the outside world.
858  *
859  * Note it is valid for this property to be NULL (but the property
860  * itself must exist). Callers of this routine should verify that
861  * the value returned is what they expected (i.e. either NULL or non NULL).
862  *
863  * On success returns value of the property in region pointed to by
864  * the 'name' argument, and with return value of 0. Otherwise returns 1.
865  */
866 static int
867 vsw_get_md_physname(vsw_t *vswp, md_t *mdp, mde_cookie_t node, char *name)
868 {
869 	int	len = 0;
870 	char	*physname = NULL;
871 	char	*dev;
872 
873 	if (md_get_prop_data(mdp, node, physdev_propname,
874 				(uint8_t **)(&physname), &len) != 0) {
875 		cmn_err(CE_WARN, "!vsw%d: Unable to get name(s) of physical "
876 				"device(s) from MD", vswp->instance);
877 		return (1);
878 	} else if ((strlen(physname) + 1) > LIFNAMSIZ) {
879 		cmn_err(CE_WARN, "!vsw%d: %s is too long a device name",
880 			vswp->instance, physname);
881 		return (1);
882 	} else {
883 		(void) strncpy(name, physname, strlen(physname) + 1);
884 		D2(vswp, "%s: using first device specified (%s)",
885 			__func__, physname);
886 	}
887 
888 #ifdef DEBUG
889 	/*
890 	 * As a temporary measure to aid testing we check to see if there
891 	 * is a vsw.conf file present. If there is we use the value of the
892 	 * vsw_physname property in the file as the name of the physical
893 	 * device, overriding the value from the MD.
894 	 *
895 	 * There may be multiple devices listed, but for the moment
896 	 * we just use the first one.
897 	 */
898 	if (ddi_prop_lookup_string(DDI_DEV_T_ANY, vswp->dip, 0,
899 		"vsw_physname", &dev) == DDI_PROP_SUCCESS) {
900 		if ((strlen(dev) + 1) > LIFNAMSIZ) {
901 			cmn_err(CE_WARN, "vsw%d: %s is too long a device name",
902 				vswp->instance, dev);
903 			ddi_prop_free(dev);
904 			return (1);
905 		} else {
906 			cmn_err(CE_NOTE, "vsw%d: Using device name (%s) from "
907 				"config file", vswp->instance, dev);
908 
909 			(void) strncpy(name, dev, strlen(dev) + 1);
910 		}
911 
912 		ddi_prop_free(dev);
913 	}
914 #endif
915 
916 	return (0);
917 }
918 
919 /*
920  * Read the 'vsw-switch-mode' property from the specified MD node.
921  *
922  * Returns 0 on success and the number of modes found in 'found',
923  * otherwise returns 1.
924  */
925 static int
926 vsw_get_md_smodes(vsw_t *vswp, md_t *mdp, mde_cookie_t node,
927 						uint8_t *modes, int *found)
928 {
929 	int		len = 0;
930 	int		smode_num = 0;
931 	char		*smode = NULL;
932 	char		*curr_mode = NULL;
933 
934 	D1(vswp, "%s: enter", __func__);
935 
936 	/*
937 	 * Get the switch-mode property. The modes are listed in
938 	 * decreasing order of preference, i.e. prefered mode is
939 	 * first item in list.
940 	 */
941 	len = 0;
942 	smode_num = 0;
943 	if (md_get_prop_data(mdp, node, smode_propname,
944 				(uint8_t **)(&smode), &len) != 0) {
945 		/*
946 		 * Unable to get switch-mode property from MD, nothing
947 		 * more we can do.
948 		 */
949 		cmn_err(CE_WARN, "!vsw%d: Unable to get switch mode property"
950 			" from the MD", vswp->instance);
951 		*found = 0;
952 		return (1);
953 	}
954 
955 	curr_mode = smode;
956 	/*
957 	 * Modes of operation:
958 	 * 'switched'	 - layer 2 switching, underlying HW in
959 	 *			programmed mode.
960 	 * 'promiscuous' - layer 2 switching, underlying HW in
961 	 *			promiscuous mode.
962 	 * 'routed'	 - layer 3 (i.e. IP) routing, underlying HW
963 	 *			in non-promiscuous mode.
964 	 */
965 	while ((curr_mode < (smode + len)) && (smode_num < NUM_SMODES)) {
966 		D2(vswp, "%s: curr_mode = [%s]", __func__, curr_mode);
967 		if (strcmp(curr_mode, "switched") == 0) {
968 			modes[smode_num++] = VSW_LAYER2;
969 		} else if (strcmp(curr_mode, "promiscuous") == 0) {
970 			modes[smode_num++] = VSW_LAYER2_PROMISC;
971 		} else if (strcmp(curr_mode, "routed") == 0) {
972 			modes[smode_num++] = VSW_LAYER3;
973 		} else {
974 			cmn_err(CE_WARN, "!vsw%d: Unknown switch mode %s, "
975 				"setting to default switched mode",
976 				vswp->instance, curr_mode);
977 			modes[smode_num++] = VSW_LAYER2;
978 		}
979 		curr_mode += strlen(curr_mode) + 1;
980 	}
981 	*found = smode_num;
982 
983 	D2(vswp, "%s: %d modes found", __func__, smode_num);
984 
985 	D1(vswp, "%s: exit", __func__);
986 
987 	return (0);
988 }
989 
990 /*
991  * Get the mac address of the physical device.
992  *
993  * Returns 0 on success, 1 on failure.
994  */
995 static int
996 vsw_get_physaddr(vsw_t *vswp)
997 {
998 	mac_handle_t	mh;
999 	char		drv[LIFNAMSIZ];
1000 	uint_t		ddi_instance;
1001 
1002 	D1(vswp, "%s: enter", __func__);
1003 
1004 	if (ddi_parse(vswp->physname, drv, &ddi_instance) != DDI_SUCCESS)
1005 		return (1);
1006 
1007 	if (mac_open(vswp->physname, ddi_instance, &mh) != 0) {
1008 		cmn_err(CE_WARN, "!vsw%d: mac_open %s failed",
1009 				vswp->instance, vswp->physname);
1010 		return (1);
1011 	}
1012 
1013 	READ_ENTER(&vswp->if_lockrw);
1014 	mac_unicst_get(mh, vswp->if_addr.ether_addr_octet);
1015 	RW_EXIT(&vswp->if_lockrw);
1016 
1017 	mac_close(mh);
1018 
1019 	vswp->mdprops |= VSW_DEV_MACADDR;
1020 
1021 	D1(vswp, "%s: exit", __func__);
1022 
1023 	return (0);
1024 }
1025 
1026 /*
1027  * Check to see if the card supports the setting of multiple unicst
1028  * addresses.
1029  *
1030  * Returns 0 if card supports the programming of multiple unicast addresses
1031  * and there are free address slots available, otherwise returns 1.
1032  */
1033 static int
1034 vsw_get_hw_maddr(vsw_t *vswp)
1035 {
1036 	D1(vswp, "%s: enter", __func__);
1037 
1038 	mutex_enter(&vswp->mac_lock);
1039 	if (vswp->mh == NULL) {
1040 		mutex_exit(&vswp->mac_lock);
1041 		return (1);
1042 	}
1043 
1044 	if (!mac_capab_get(vswp->mh, MAC_CAPAB_MULTIADDRESS, &vswp->maddr)) {
1045 		DWARN(vswp, "Unable to get capabilities of"
1046 			" underlying device (%s)", vswp->physname);
1047 		mutex_exit(&vswp->mac_lock);
1048 		return (1);
1049 	}
1050 	mutex_exit(&vswp->mac_lock);
1051 
1052 	if (vswp->maddr.maddr_naddrfree == 0) {
1053 		cmn_err(CE_WARN,
1054 			"!vsw%d: device %s has no free unicast address slots",
1055 				vswp->instance, vswp->physname);
1056 		return (1);
1057 	}
1058 
1059 	D2(vswp, "%s: %d addrs : %d free", __func__,
1060 		vswp->maddr.maddr_naddr, vswp->maddr.maddr_naddrfree);
1061 
1062 	D1(vswp, "%s: exit", __func__);
1063 
1064 	return (0);
1065 }
1066 
1067 /*
1068  * Setup the required switching mode.
1069  *
1070  * Returns 0 on success, 1 on failure.
1071  */
1072 static int
1073 vsw_setup_switching(vsw_t *vswp)
1074 {
1075 	int	i, rv = 1;
1076 
1077 	D1(vswp, "%s: enter", __func__);
1078 
1079 	/* select best switching mode */
1080 	for (i = 0; i < vswp->smode_num; i++) {
1081 		vswp->smode_idx = i;
1082 		switch (vswp->smode[i]) {
1083 		case VSW_LAYER2:
1084 		case VSW_LAYER2_PROMISC:
1085 			rv = vsw_setup_layer2(vswp);
1086 			break;
1087 
1088 		case VSW_LAYER3:
1089 			rv = vsw_setup_layer3(vswp);
1090 			break;
1091 
1092 		default:
1093 			DERR(vswp, "unknown switch mode");
1094 			rv = 1;
1095 			break;
1096 		}
1097 
1098 		if (rv == 0)
1099 			break;
1100 	}
1101 
1102 	if (rv == 1) {
1103 		cmn_err(CE_WARN, "!vsw%d: Unable to setup specified "
1104 			"switching mode", vswp->instance);
1105 		return (rv);
1106 	}
1107 
1108 	D2(vswp, "%s: Operating in mode %d", __func__,
1109 					vswp->smode[vswp->smode_idx]);
1110 
1111 	D1(vswp, "%s: exit", __func__);
1112 
1113 	return (0);
1114 }
1115 
1116 /*
1117  * Setup for layer 2 switching.
1118  *
1119  * Returns 0 on success, 1 on failure.
1120  */
1121 static int
1122 vsw_setup_layer2(vsw_t *vswp)
1123 {
1124 	D1(vswp, "%s: enter", __func__);
1125 
1126 	vswp->vsw_switch_frame = vsw_switch_l2_frame;
1127 
1128 	/*
1129 	 * Attempt to link into the MAC layer so we can get
1130 	 * and send packets out over the physical adapter.
1131 	 */
1132 	if (vswp->mdprops & VSW_MD_PHYSNAME) {
1133 		if (vsw_mac_attach(vswp) != 0) {
1134 			/*
1135 			 * Registration with the MAC layer has failed,
1136 			 * so return 1 so that can fall back to next
1137 			 * prefered switching method.
1138 			 */
1139 			cmn_err(CE_WARN, "!vsw%d: Unable to join as MAC layer "
1140 				"client", vswp->instance);
1141 			return (1);
1142 		}
1143 
1144 		if (vswp->smode[vswp->smode_idx] == VSW_LAYER2) {
1145 			/*
1146 			 * Verify that underlying device can support multiple
1147 			 * unicast mac addresses, and has free capacity.
1148 			 */
1149 			if (vsw_get_hw_maddr(vswp) != 0) {
1150 				cmn_err(CE_WARN, "!vsw%d: Unable to setup "
1151 					"switching", vswp->instance);
1152 				vsw_mac_detach(vswp);
1153 				return (1);
1154 			}
1155 		}
1156 
1157 	} else {
1158 		/*
1159 		 * No physical device name found in MD which is
1160 		 * required for layer 2.
1161 		 */
1162 		cmn_err(CE_WARN, "!vsw%d: no physical device name specified",
1163 			vswp->instance);
1164 		return (1);
1165 	}
1166 
1167 	D1(vswp, "%s: exit", __func__);
1168 
1169 	return (0);
1170 }
1171 
1172 static int
1173 vsw_setup_layer3(vsw_t *vswp)
1174 {
1175 	D1(vswp, "%s: enter", __func__);
1176 
1177 	D2(vswp, "%s: operating in layer 3 mode", __func__);
1178 	vswp->vsw_switch_frame = vsw_switch_l3_frame;
1179 
1180 	D1(vswp, "%s: exit", __func__);
1181 
1182 	return (0);
1183 }
1184 
1185 /*
1186  * Link into the MAC layer to gain access to the services provided by
1187  * the underlying physical device driver (which should also have
1188  * registered with the MAC layer).
1189  *
1190  * Only when in layer 2 mode.
1191  */
1192 static int
1193 vsw_mac_attach(vsw_t *vswp)
1194 {
1195 	char	drv[LIFNAMSIZ];
1196 	uint_t	ddi_instance;
1197 
1198 	D1(vswp, "%s: enter", __func__);
1199 
1200 	ASSERT(vswp->mh == NULL);
1201 	ASSERT(vswp->mrh == NULL);
1202 	ASSERT(vswp->mstarted == B_FALSE);
1203 	ASSERT(vswp->mresources == B_FALSE);
1204 
1205 	ASSERT(vswp->mdprops & VSW_MD_PHYSNAME);
1206 
1207 	mutex_enter(&vswp->mac_lock);
1208 	if (ddi_parse(vswp->physname, drv, &ddi_instance) != DDI_SUCCESS) {
1209 		cmn_err(CE_WARN, "!vsw%d: invalid device name: %s",
1210 			vswp->instance, vswp->physname);
1211 		goto mac_fail_exit;
1212 	}
1213 
1214 	if ((mac_open(vswp->physname, ddi_instance, &vswp->mh)) != 0) {
1215 		cmn_err(CE_WARN, "!vsw%d: mac_open %s failed",
1216 			vswp->instance, vswp->physname);
1217 		goto mac_fail_exit;
1218 	}
1219 
1220 	ASSERT(vswp->mh != NULL);
1221 
1222 	D2(vswp, "vsw_mac_attach: using device %s", vswp->physname);
1223 
1224 	if (vsw_multi_ring_enable) {
1225 		/*
1226 		 * Initialize the ring table.
1227 		 */
1228 		vsw_mac_ring_tbl_init(vswp);
1229 
1230 		/*
1231 		 * Register our rx callback function.
1232 		 */
1233 		vswp->mrh = mac_rx_add(vswp->mh,
1234 			vsw_rx_queue_cb, (void *)vswp);
1235 		ASSERT(vswp->mrh != NULL);
1236 
1237 		/*
1238 		 * Register our mac resource callback.
1239 		 */
1240 		mac_resource_set(vswp->mh, vsw_mac_ring_add_cb, (void *)vswp);
1241 		vswp->mresources = B_TRUE;
1242 
1243 		/*
1244 		 * Get the ring resources available to us from
1245 		 * the mac below us.
1246 		 */
1247 		mac_resources(vswp->mh);
1248 	} else {
1249 		/*
1250 		 * Just register our rx callback function
1251 		 */
1252 		vswp->mrh = mac_rx_add(vswp->mh, vsw_rx_cb, (void *)vswp);
1253 		ASSERT(vswp->mrh != NULL);
1254 	}
1255 
1256 	/* Get the MAC tx fn */
1257 	vswp->txinfo = mac_tx_get(vswp->mh);
1258 
1259 	/* start the interface */
1260 	if (mac_start(vswp->mh) != 0) {
1261 		cmn_err(CE_WARN, "!vsw%d: Could not start mac interface",
1262 			vswp->instance);
1263 		goto mac_fail_exit;
1264 	}
1265 
1266 	mutex_exit(&vswp->mac_lock);
1267 
1268 	vswp->mstarted = B_TRUE;
1269 
1270 	D1(vswp, "%s: exit", __func__);
1271 	return (0);
1272 
1273 mac_fail_exit:
1274 	mutex_exit(&vswp->mac_lock);
1275 	vsw_mac_detach(vswp);
1276 
1277 	D1(vswp, "%s: exit", __func__);
1278 	return (1);
1279 }
1280 
1281 static void
1282 vsw_mac_detach(vsw_t *vswp)
1283 {
1284 	D1(vswp, "vsw_mac_detach: enter");
1285 
1286 	ASSERT(vswp != NULL);
1287 
1288 	if (vsw_multi_ring_enable) {
1289 		vsw_mac_ring_tbl_destroy(vswp);
1290 	}
1291 
1292 	mutex_enter(&vswp->mac_lock);
1293 
1294 	if (vswp->mh != NULL) {
1295 		if (vswp->mstarted)
1296 			mac_stop(vswp->mh);
1297 		if (vswp->mrh != NULL)
1298 			mac_rx_remove(vswp->mh, vswp->mrh);
1299 		if (vswp->mresources)
1300 			mac_resource_set(vswp->mh, NULL, NULL);
1301 		mac_close(vswp->mh);
1302 	}
1303 
1304 	vswp->mrh = NULL;
1305 	vswp->mh = NULL;
1306 	vswp->txinfo = NULL;
1307 	vswp->mstarted = B_FALSE;
1308 
1309 	mutex_exit(&vswp->mac_lock);
1310 
1311 	D1(vswp, "vsw_mac_detach: exit");
1312 }
1313 
1314 /*
1315  * Depending on the mode specified, the capabilites and capacity
1316  * of the underlying device setup the physical device.
1317  *
1318  * If in layer 3 mode, then do nothing.
1319  *
1320  * If in layer 2 programmed mode attempt to program the unicast address
1321  * associated with the port into the physical device. If this is not
1322  * possible due to resource exhaustion or simply because the device does
1323  * not support multiple unicast addresses then if required fallback onto
1324  * putting the card into promisc mode.
1325  *
1326  * If in promisc mode then simply set the card into promisc mode.
1327  *
1328  * Returns 0 success, 1 on failure.
1329  */
1330 static int
1331 vsw_set_hw(vsw_t *vswp, vsw_port_t *port)
1332 {
1333 	mac_multi_addr_t	mac_addr;
1334 	void			*mah;
1335 	int			err;
1336 
1337 	D1(vswp, "%s: enter", __func__);
1338 
1339 	if (vswp->smode[vswp->smode_idx] == VSW_LAYER3)
1340 		return (0);
1341 
1342 	if (vswp->smode[vswp->smode_idx] == VSW_LAYER2_PROMISC) {
1343 		return (vsw_set_hw_promisc(vswp, port));
1344 	}
1345 
1346 	if (vswp->maddr.maddr_handle == NULL)
1347 		return (1);
1348 
1349 	mah = vswp->maddr.maddr_handle;
1350 
1351 	/*
1352 	 * Attempt to program the unicast address into the HW.
1353 	 */
1354 	mac_addr.mma_addrlen = ETHERADDRL;
1355 	ether_copy(&port->p_macaddr, &mac_addr.mma_addr);
1356 
1357 	err = vswp->maddr.maddr_add(mah, &mac_addr);
1358 	if (err != 0) {
1359 		cmn_err(CE_WARN, "!vsw%d: failed to program addr "
1360 			"%x:%x:%x:%x:%x:%x for port %d into device %s "
1361 			": err %d", vswp->instance,
1362 			port->p_macaddr.ether_addr_octet[0],
1363 			port->p_macaddr.ether_addr_octet[1],
1364 			port->p_macaddr.ether_addr_octet[2],
1365 			port->p_macaddr.ether_addr_octet[3],
1366 			port->p_macaddr.ether_addr_octet[4],
1367 			port->p_macaddr.ether_addr_octet[5],
1368 			port->p_instance, vswp->physname, err);
1369 
1370 		/*
1371 		 * Mark that attempt should be made to re-config sometime
1372 		 * in future if a port is deleted.
1373 		 */
1374 		vswp->recfg_reqd = B_TRUE;
1375 
1376 		/*
1377 		 * Only 1 mode specified, nothing more to do.
1378 		 */
1379 		if (vswp->smode_num == 1)
1380 			return (err);
1381 
1382 		/*
1383 		 * If promiscuous was next mode specified try to
1384 		 * set the card into that mode.
1385 		 */
1386 		if ((vswp->smode_idx <= (vswp->smode_num - 2)) &&
1387 			(vswp->smode[vswp->smode_idx + 1]
1388 					== VSW_LAYER2_PROMISC)) {
1389 			vswp->smode_idx += 1;
1390 			return (vsw_set_hw_promisc(vswp, port));
1391 		}
1392 		return (err);
1393 	}
1394 
1395 	port->addr_slot = mac_addr.mma_slot;
1396 	port->addr_set = VSW_ADDR_HW;
1397 
1398 	D2(vswp, "programmed addr %x:%x:%x:%x:%x:%x for port %d "
1399 		"into slot %d of device %s",
1400 		port->p_macaddr.ether_addr_octet[0],
1401 		port->p_macaddr.ether_addr_octet[1],
1402 		port->p_macaddr.ether_addr_octet[2],
1403 		port->p_macaddr.ether_addr_octet[3],
1404 		port->p_macaddr.ether_addr_octet[4],
1405 		port->p_macaddr.ether_addr_octet[5],
1406 		port->p_instance, port->addr_slot, vswp->physname);
1407 
1408 	D1(vswp, "%s: exit", __func__);
1409 
1410 	return (0);
1411 }
1412 
1413 /*
1414  * If in layer 3 mode do nothing.
1415  *
1416  * If in layer 2 switched mode remove the address from the physical
1417  * device.
1418  *
1419  * If in layer 2 promiscuous mode disable promisc mode.
1420  *
1421  * Returns 0 on success.
1422  */
1423 static int
1424 vsw_unset_hw(vsw_t *vswp, vsw_port_t *port)
1425 {
1426 	int		err;
1427 	void		*mah;
1428 
1429 	D1(vswp, "%s: enter", __func__);
1430 
1431 	if (vswp->smode[vswp->smode_idx] == VSW_LAYER3)
1432 		return (0);
1433 
1434 	if (port->addr_set == VSW_ADDR_PROMISC) {
1435 		return (vsw_unset_hw_promisc(vswp, port));
1436 	}
1437 
1438 	if (port->addr_set == VSW_ADDR_HW) {
1439 		if (vswp->maddr.maddr_handle == NULL)
1440 			return (1);
1441 
1442 		mah = vswp->maddr.maddr_handle;
1443 
1444 		err = vswp->maddr.maddr_remove(mah, port->addr_slot);
1445 		if (err != 0) {
1446 			cmn_err(CE_WARN, "!vsw%d: Unable to remove addr "
1447 				"%x:%x:%x:%x:%x:%x for port %d from device %s"
1448 				" : (err %d)", vswp->instance,
1449 				port->p_macaddr.ether_addr_octet[0],
1450 				port->p_macaddr.ether_addr_octet[1],
1451 				port->p_macaddr.ether_addr_octet[2],
1452 				port->p_macaddr.ether_addr_octet[3],
1453 				port->p_macaddr.ether_addr_octet[4],
1454 				port->p_macaddr.ether_addr_octet[5],
1455 				port->p_instance, vswp->physname, err);
1456 			return (err);
1457 		}
1458 
1459 		port->addr_set = VSW_ADDR_UNSET;
1460 
1461 		D2(vswp, "removed addr %x:%x:%x:%x:%x:%x for "
1462 			"port %d from device %s",
1463 			port->p_macaddr.ether_addr_octet[0],
1464 			port->p_macaddr.ether_addr_octet[1],
1465 			port->p_macaddr.ether_addr_octet[2],
1466 			port->p_macaddr.ether_addr_octet[3],
1467 			port->p_macaddr.ether_addr_octet[4],
1468 			port->p_macaddr.ether_addr_octet[5],
1469 			port->p_instance, vswp->physname);
1470 	}
1471 
1472 	D1(vswp, "%s: exit", __func__);
1473 	return (0);
1474 }
1475 
1476 /*
1477  * Set network card into promisc mode.
1478  *
1479  * Returns 0 on success, 1 on failure.
1480  */
1481 static int
1482 vsw_set_hw_promisc(vsw_t *vswp, vsw_port_t *port)
1483 {
1484 	D1(vswp, "%s: enter", __func__);
1485 
1486 	mutex_enter(&vswp->mac_lock);
1487 	if (vswp->mh == NULL) {
1488 		mutex_exit(&vswp->mac_lock);
1489 		return (1);
1490 	}
1491 
1492 	if (vswp->promisc_cnt++ == 0) {
1493 		if (mac_promisc_set(vswp->mh, B_TRUE, MAC_DEVPROMISC) != 0) {
1494 			vswp->promisc_cnt--;
1495 			mutex_exit(&vswp->mac_lock);
1496 			return (1);
1497 		}
1498 		cmn_err(CE_NOTE, "!vsw%d: switching device %s into "
1499 			"promiscuous mode", vswp->instance, vswp->physname);
1500 	}
1501 	mutex_exit(&vswp->mac_lock);
1502 	port->addr_set = VSW_ADDR_PROMISC;
1503 
1504 	D1(vswp, "%s: exit", __func__);
1505 
1506 	return (0);
1507 }
1508 
1509 /*
1510  * Turn off promiscuous mode on network card.
1511  *
1512  * Returns 0 on success, 1 on failure.
1513  */
1514 static int
1515 vsw_unset_hw_promisc(vsw_t *vswp, vsw_port_t *port)
1516 {
1517 	vsw_port_list_t 	*plist = &vswp->plist;
1518 
1519 	D2(vswp, "%s: enter", __func__);
1520 
1521 	mutex_enter(&vswp->mac_lock);
1522 	if (vswp->mh == NULL) {
1523 		mutex_exit(&vswp->mac_lock);
1524 		return (1);
1525 	}
1526 
1527 	ASSERT(port->addr_set == VSW_ADDR_PROMISC);
1528 
1529 	if (--vswp->promisc_cnt == 0) {
1530 		if (mac_promisc_set(vswp->mh, B_FALSE, MAC_DEVPROMISC) != 0) {
1531 			vswp->promisc_cnt++;
1532 			mutex_exit(&vswp->mac_lock);
1533 			return (1);
1534 		}
1535 
1536 		/*
1537 		 * We are exiting promisc mode either because we were
1538 		 * only in promisc mode because we had failed over from
1539 		 * switched mode due to HW resource issues, or the user
1540 		 * wanted the card in promisc mode for all the ports and
1541 		 * the last port is now being deleted. Tweak the message
1542 		 * accordingly.
1543 		 */
1544 		if (plist->num_ports != 0) {
1545 			cmn_err(CE_NOTE, "!vsw%d: switching device %s back to "
1546 				"programmed mode", vswp->instance,
1547 				vswp->physname);
1548 		} else {
1549 			cmn_err(CE_NOTE, "!vsw%d: switching device %s out of "
1550 				"promiscuous mode", vswp->instance,
1551 				vswp->physname);
1552 		}
1553 	}
1554 	mutex_exit(&vswp->mac_lock);
1555 	port->addr_set = VSW_ADDR_UNSET;
1556 
1557 	D1(vswp, "%s: exit", __func__);
1558 	return (0);
1559 }
1560 
1561 /*
1562  * Determine whether or not we are operating in our prefered
1563  * mode and if not whether the physical resources now allow us
1564  * to operate in it.
1565  *
1566  * Should only be invoked after port which is being deleted has been
1567  * removed from the port list.
1568  */
1569 static int
1570 vsw_reconfig_hw(vsw_t *vswp)
1571 {
1572 	vsw_port_list_t 	*plist = &vswp->plist;
1573 	mac_multi_addr_t	mac_addr;
1574 	vsw_port_t		*tp;
1575 	void			*mah;
1576 	int			rv = 0;
1577 	int			s_idx;
1578 
1579 	D1(vswp, "%s: enter", __func__);
1580 
1581 	if (vswp->maddr.maddr_handle == NULL)
1582 		return (1);
1583 
1584 	/*
1585 	 * Check if there are now sufficient HW resources to
1586 	 * attempt a re-config.
1587 	 */
1588 	if (plist->num_ports > vswp->maddr.maddr_naddrfree)
1589 		return (1);
1590 
1591 	/*
1592 	 * If we are in layer 2 (i.e. switched) or would like to be
1593 	 * in layer 2 then check if any ports need to be programmed
1594 	 * into the HW.
1595 	 *
1596 	 * This can happen in two cases - switched was specified as
1597 	 * the prefered mode of operation but we exhausted the HW
1598 	 * resources and so failed over to the next specifed mode,
1599 	 * or switched was the only mode specified so after HW
1600 	 * resources were exhausted there was nothing more we
1601 	 * could do.
1602 	 */
1603 	if (vswp->smode_idx > 0)
1604 		s_idx = vswp->smode_idx - 1;
1605 	else
1606 		s_idx = vswp->smode_idx;
1607 
1608 	if (vswp->smode[s_idx] == VSW_LAYER2) {
1609 		mah = vswp->maddr.maddr_handle;
1610 
1611 		D2(vswp, "%s: attempting reconfig..", __func__);
1612 
1613 		/*
1614 		 * Scan the port list for any port whose address has not
1615 		 * be programmed in HW - there should be a max of one.
1616 		 */
1617 		for (tp = plist->head; tp != NULL; tp = tp->p_next) {
1618 			if (tp->addr_set != VSW_ADDR_HW) {
1619 				mac_addr.mma_addrlen = ETHERADDRL;
1620 				ether_copy(&tp->p_macaddr, &mac_addr.mma_addr);
1621 
1622 				rv = vswp->maddr.maddr_add(mah, &mac_addr);
1623 				if (rv != 0) {
1624 					DWARN(vswp, "Error setting addr in "
1625 						"HW for port %d err %d",
1626 						tp->p_instance, rv);
1627 					goto reconfig_err_exit;
1628 				}
1629 				tp->addr_slot = mac_addr.mma_slot;
1630 
1631 				D2(vswp, "re-programmed port %d "
1632 					"addr %x:%x:%x:%x:%x:%x into slot %d"
1633 					" of device %s", tp->p_instance,
1634 					tp->p_macaddr.ether_addr_octet[0],
1635 					tp->p_macaddr.ether_addr_octet[1],
1636 					tp->p_macaddr.ether_addr_octet[2],
1637 					tp->p_macaddr.ether_addr_octet[3],
1638 					tp->p_macaddr.ether_addr_octet[4],
1639 					tp->p_macaddr.ether_addr_octet[5],
1640 					tp->addr_slot, vswp->physname);
1641 
1642 				/*
1643 				 * If up to now we had to put the card into
1644 				 * promisc mode to see this address, we
1645 				 * can now safely disable promisc mode.
1646 				 */
1647 				if (tp->addr_set == VSW_ADDR_PROMISC)
1648 					(void) vsw_unset_hw_promisc(vswp, tp);
1649 
1650 				tp->addr_set = VSW_ADDR_HW;
1651 			}
1652 		}
1653 
1654 		/* no further re-config needed */
1655 		vswp->recfg_reqd = B_FALSE;
1656 
1657 		vswp->smode_idx = s_idx;
1658 
1659 		return (0);
1660 	}
1661 
1662 reconfig_err_exit:
1663 	return (rv);
1664 }
1665 
1666 static void
1667 vsw_mac_ring_tbl_entry_init(vsw_t *vswp, vsw_mac_ring_t *ringp)
1668 {
1669 	ringp->ring_state = VSW_MAC_RING_FREE;
1670 	ringp->ring_arg = NULL;
1671 	ringp->ring_blank = NULL;
1672 	ringp->ring_vqp = NULL;
1673 	ringp->ring_vswp = vswp;
1674 }
1675 
1676 static void
1677 vsw_mac_ring_tbl_init(vsw_t *vswp)
1678 {
1679 	int		i;
1680 
1681 	mutex_init(&vswp->mac_ring_lock, NULL, MUTEX_DRIVER, NULL);
1682 
1683 	vswp->mac_ring_tbl_sz = vsw_mac_rx_rings;
1684 	vswp->mac_ring_tbl  =
1685 		kmem_alloc(vsw_mac_rx_rings * sizeof (vsw_mac_ring_t),
1686 		KM_SLEEP);
1687 
1688 	for (i = 0; i < vswp->mac_ring_tbl_sz; i++)
1689 		vsw_mac_ring_tbl_entry_init(vswp, &vswp->mac_ring_tbl[i]);
1690 }
1691 
1692 static void
1693 vsw_mac_ring_tbl_destroy(vsw_t *vswp)
1694 {
1695 	int		i;
1696 	vsw_mac_ring_t	*ringp;
1697 
1698 	mutex_enter(&vswp->mac_ring_lock);
1699 	for (i = 0; i < vswp->mac_ring_tbl_sz; i++) {
1700 		ringp = &vswp->mac_ring_tbl[i];
1701 
1702 		if (ringp->ring_state != VSW_MAC_RING_FREE) {
1703 			/*
1704 			 * Destroy the queue.
1705 			 */
1706 			vsw_queue_stop(ringp->ring_vqp);
1707 			vsw_queue_destroy(ringp->ring_vqp);
1708 
1709 			/*
1710 			 * Re-initialize the structure.
1711 			 */
1712 			vsw_mac_ring_tbl_entry_init(vswp, ringp);
1713 		}
1714 	}
1715 	mutex_exit(&vswp->mac_ring_lock);
1716 
1717 	mutex_destroy(&vswp->mac_ring_lock);
1718 	kmem_free(vswp->mac_ring_tbl,
1719 		vswp->mac_ring_tbl_sz * sizeof (vsw_mac_ring_t));
1720 	vswp->mac_ring_tbl_sz = 0;
1721 }
1722 
1723 /*
1724  * Handle resource add callbacks from the driver below.
1725  */
1726 static mac_resource_handle_t
1727 vsw_mac_ring_add_cb(void *arg, mac_resource_t *mrp)
1728 {
1729 	vsw_t		*vswp = (vsw_t *)arg;
1730 	mac_rx_fifo_t	*mrfp = (mac_rx_fifo_t *)mrp;
1731 	vsw_mac_ring_t	*ringp;
1732 	vsw_queue_t	*vqp;
1733 	int		i;
1734 
1735 	ASSERT(vswp != NULL);
1736 	ASSERT(mrp != NULL);
1737 	ASSERT(vswp->mac_ring_tbl != NULL);
1738 
1739 	D1(vswp, "%s: enter", __func__);
1740 
1741 	/*
1742 	 * Check to make sure we have the correct resource type.
1743 	 */
1744 	if (mrp->mr_type != MAC_RX_FIFO)
1745 		return (NULL);
1746 
1747 	/*
1748 	 * Find a open entry in the ring table.
1749 	 */
1750 	mutex_enter(&vswp->mac_ring_lock);
1751 	for (i = 0; i < vswp->mac_ring_tbl_sz; i++) {
1752 		ringp = &vswp->mac_ring_tbl[i];
1753 
1754 		/*
1755 		 * Check for an empty slot, if found, then setup queue
1756 		 * and thread.
1757 		 */
1758 		if (ringp->ring_state == VSW_MAC_RING_FREE) {
1759 			/*
1760 			 * Create the queue for this ring.
1761 			 */
1762 			vqp = vsw_queue_create();
1763 
1764 			/*
1765 			 * Initialize the ring data structure.
1766 			 */
1767 			ringp->ring_vqp = vqp;
1768 			ringp->ring_arg = mrfp->mrf_arg;
1769 			ringp->ring_blank = mrfp->mrf_blank;
1770 			ringp->ring_state = VSW_MAC_RING_INUSE;
1771 
1772 			/*
1773 			 * Create the worker thread.
1774 			 */
1775 			vqp->vq_worker = thread_create(NULL, 0,
1776 				vsw_queue_worker, ringp, 0, &p0,
1777 				TS_RUN, minclsyspri);
1778 			if (vqp->vq_worker == NULL) {
1779 				vsw_queue_destroy(vqp);
1780 				vsw_mac_ring_tbl_entry_init(vswp, ringp);
1781 				ringp = NULL;
1782 			}
1783 
1784 			if (ringp != NULL) {
1785 				/*
1786 				 * Make sure thread get's running state for
1787 				 * this ring.
1788 				 */
1789 				mutex_enter(&vqp->vq_lock);
1790 				while ((vqp->vq_state != VSW_QUEUE_RUNNING) &&
1791 					(vqp->vq_state != VSW_QUEUE_DRAINED)) {
1792 					cv_wait(&vqp->vq_cv, &vqp->vq_lock);
1793 				}
1794 
1795 				/*
1796 				 * If the thread is not running, cleanup.
1797 				 */
1798 				if (vqp->vq_state == VSW_QUEUE_DRAINED) {
1799 					vsw_queue_destroy(vqp);
1800 					vsw_mac_ring_tbl_entry_init(vswp,
1801 						ringp);
1802 					ringp = NULL;
1803 				}
1804 				mutex_exit(&vqp->vq_lock);
1805 			}
1806 
1807 			mutex_exit(&vswp->mac_ring_lock);
1808 			D1(vswp, "%s: exit", __func__);
1809 			return ((mac_resource_handle_t)ringp);
1810 		}
1811 	}
1812 	mutex_exit(&vswp->mac_ring_lock);
1813 
1814 	/*
1815 	 * No slots in the ring table available.
1816 	 */
1817 	D1(vswp, "%s: exit", __func__);
1818 	return (NULL);
1819 }
1820 
1821 static void
1822 vsw_queue_stop(vsw_queue_t *vqp)
1823 {
1824 	mutex_enter(&vqp->vq_lock);
1825 
1826 	if (vqp->vq_state == VSW_QUEUE_RUNNING) {
1827 		vqp->vq_state = VSW_QUEUE_STOP;
1828 		cv_signal(&vqp->vq_cv);
1829 
1830 		while (vqp->vq_state != VSW_QUEUE_DRAINED)
1831 			cv_wait(&vqp->vq_cv, &vqp->vq_lock);
1832 	}
1833 
1834 	vqp->vq_state = VSW_QUEUE_STOPPED;
1835 
1836 	mutex_exit(&vqp->vq_lock);
1837 }
1838 
1839 static vsw_queue_t *
1840 vsw_queue_create()
1841 {
1842 	vsw_queue_t *vqp;
1843 
1844 	vqp = kmem_zalloc(sizeof (vsw_queue_t), KM_SLEEP);
1845 
1846 	mutex_init(&vqp->vq_lock, NULL, MUTEX_DRIVER, NULL);
1847 	cv_init(&vqp->vq_cv, NULL, CV_DRIVER, NULL);
1848 	vqp->vq_first = NULL;
1849 	vqp->vq_last = NULL;
1850 	vqp->vq_state = VSW_QUEUE_STOPPED;
1851 
1852 	return (vqp);
1853 }
1854 
1855 static void
1856 vsw_queue_destroy(vsw_queue_t *vqp)
1857 {
1858 	cv_destroy(&vqp->vq_cv);
1859 	mutex_destroy(&vqp->vq_lock);
1860 	kmem_free(vqp, sizeof (vsw_queue_t));
1861 }
1862 
1863 static void
1864 vsw_queue_worker(vsw_mac_ring_t *rrp)
1865 {
1866 	mblk_t		*mp;
1867 	vsw_queue_t	*vqp = rrp->ring_vqp;
1868 	vsw_t		*vswp = rrp->ring_vswp;
1869 
1870 	mutex_enter(&vqp->vq_lock);
1871 
1872 	ASSERT(vqp->vq_state == VSW_QUEUE_STOPPED);
1873 
1874 	/*
1875 	 * Set the state to running, since the thread is now active.
1876 	 */
1877 	vqp->vq_state = VSW_QUEUE_RUNNING;
1878 	cv_signal(&vqp->vq_cv);
1879 
1880 	while (vqp->vq_state == VSW_QUEUE_RUNNING) {
1881 		/*
1882 		 * Wait for work to do or the state has changed
1883 		 * to not running.
1884 		 */
1885 		while ((vqp->vq_state == VSW_QUEUE_RUNNING) &&
1886 				(vqp->vq_first == NULL)) {
1887 			cv_wait(&vqp->vq_cv, &vqp->vq_lock);
1888 		}
1889 
1890 		/*
1891 		 * Process packets that we received from the interface.
1892 		 */
1893 		if (vqp->vq_first != NULL) {
1894 			mp = vqp->vq_first;
1895 
1896 			vqp->vq_first = NULL;
1897 			vqp->vq_last = NULL;
1898 
1899 			mutex_exit(&vqp->vq_lock);
1900 
1901 			/* switch the chain of packets received */
1902 			vswp->vsw_switch_frame(vswp, mp,
1903 						VSW_PHYSDEV, NULL, NULL);
1904 
1905 			mutex_enter(&vqp->vq_lock);
1906 		}
1907 	}
1908 
1909 	/*
1910 	 * We are drained and signal we are done.
1911 	 */
1912 	vqp->vq_state = VSW_QUEUE_DRAINED;
1913 	cv_signal(&vqp->vq_cv);
1914 
1915 	/*
1916 	 * Exit lock and drain the remaining packets.
1917 	 */
1918 	mutex_exit(&vqp->vq_lock);
1919 
1920 	/*
1921 	 * Exit the thread
1922 	 */
1923 	thread_exit();
1924 }
1925 
1926 /*
1927  * static void
1928  * vsw_rx_queue_cb() - Receive callback routine when
1929  *	vsw_multi_ring_enable is non-zero.  Queue the packets
1930  *	to a packet queue for a worker thread to process.
1931  */
1932 static void
1933 vsw_rx_queue_cb(void *arg, mac_resource_handle_t mrh, mblk_t *mp)
1934 {
1935 	vsw_mac_ring_t	*ringp = (vsw_mac_ring_t *)mrh;
1936 	vsw_t		*vswp = (vsw_t *)arg;
1937 	vsw_queue_t	*vqp;
1938 	mblk_t		*bp, *last;
1939 
1940 	ASSERT(mrh != NULL);
1941 	ASSERT(vswp != NULL);
1942 	ASSERT(mp != NULL);
1943 
1944 	D1(vswp, "%s: enter", __func__);
1945 
1946 	/*
1947 	 * Find the last element in the mblk chain.
1948 	 */
1949 	bp = mp;
1950 	do {
1951 		last = bp;
1952 		bp = bp->b_next;
1953 	} while (bp != NULL);
1954 
1955 	/* Get the queue for the packets */
1956 	vqp = ringp->ring_vqp;
1957 
1958 	/*
1959 	 * Grab the lock such we can queue the packets.
1960 	 */
1961 	mutex_enter(&vqp->vq_lock);
1962 
1963 	if (vqp->vq_state != VSW_QUEUE_RUNNING) {
1964 		freemsg(mp);
1965 		mutex_exit(&vqp->vq_lock);
1966 		goto vsw_rx_queue_cb_exit;
1967 	}
1968 
1969 	/*
1970 	 * Add the mblk chain to the queue.  If there
1971 	 * is some mblks in the queue, then add the new
1972 	 * chain to the end.
1973 	 */
1974 	if (vqp->vq_first == NULL)
1975 		vqp->vq_first = mp;
1976 	else
1977 		vqp->vq_last->b_next = mp;
1978 
1979 	vqp->vq_last = last;
1980 
1981 	/*
1982 	 * Signal the worker thread that there is work to
1983 	 * do.
1984 	 */
1985 	cv_signal(&vqp->vq_cv);
1986 
1987 	/*
1988 	 * Let go of the lock and exit.
1989 	 */
1990 	mutex_exit(&vqp->vq_lock);
1991 
1992 vsw_rx_queue_cb_exit:
1993 	D1(vswp, "%s: exit", __func__);
1994 }
1995 
1996 /*
1997  * receive callback routine. Invoked by MAC layer when there
1998  * are pkts being passed up from physical device.
1999  *
2000  * PERF: It may be more efficient when the card is in promisc
2001  * mode to check the dest address of the pkts here (against
2002  * the FDB) rather than checking later. Needs to be investigated.
2003  */
2004 static void
2005 vsw_rx_cb(void *arg, mac_resource_handle_t mrh, mblk_t *mp)
2006 {
2007 	_NOTE(ARGUNUSED(mrh))
2008 
2009 	vsw_t		*vswp = (vsw_t *)arg;
2010 
2011 	ASSERT(vswp != NULL);
2012 
2013 	D1(vswp, "vsw_rx_cb: enter");
2014 
2015 	/* switch the chain of packets received */
2016 	vswp->vsw_switch_frame(vswp, mp, VSW_PHYSDEV, NULL, NULL);
2017 
2018 	D1(vswp, "vsw_rx_cb: exit");
2019 }
2020 
2021 /*
2022  * Send a message out over the physical device via the MAC layer.
2023  *
2024  * Returns any mblks that it was unable to transmit.
2025  */
2026 static mblk_t *
2027 vsw_tx_msg(vsw_t *vswp, mblk_t *mp)
2028 {
2029 	const mac_txinfo_t	*mtp;
2030 	mblk_t			*nextp;
2031 
2032 	mutex_enter(&vswp->mac_lock);
2033 	if (vswp->mh == NULL) {
2034 		DERR(vswp, "vsw_tx_msg: dropping pkts: no tx routine avail");
2035 		mutex_exit(&vswp->mac_lock);
2036 		return (mp);
2037 	} else {
2038 		for (;;) {
2039 			nextp = mp->b_next;
2040 			mp->b_next = NULL;
2041 
2042 			mtp = vswp->txinfo;
2043 
2044 			if ((mp = mtp->mt_fn(mtp->mt_arg, mp)) != NULL) {
2045 				mp->b_next = nextp;
2046 				break;
2047 			}
2048 
2049 			if ((mp = nextp) == NULL)
2050 				break;
2051 		}
2052 	}
2053 	mutex_exit(&vswp->mac_lock);
2054 
2055 	return (mp);
2056 }
2057 
2058 /*
2059  * Register with the MAC layer as a network device, so we
2060  * can be plumbed if necessary.
2061  */
2062 static int
2063 vsw_mac_register(vsw_t *vswp)
2064 {
2065 	mac_register_t	*macp;
2066 	int		rv;
2067 
2068 	D1(vswp, "%s: enter", __func__);
2069 
2070 	if ((macp = mac_alloc(MAC_VERSION)) == NULL)
2071 		return (EINVAL);
2072 	macp->m_type_ident = MAC_PLUGIN_IDENT_ETHER;
2073 	macp->m_driver = vswp;
2074 	macp->m_dip = vswp->dip;
2075 	macp->m_src_addr = (uint8_t *)&vswp->if_addr;
2076 	macp->m_callbacks = &vsw_m_callbacks;
2077 	macp->m_min_sdu = 0;
2078 	macp->m_max_sdu = ETHERMTU;
2079 	rv = mac_register(macp, &vswp->if_mh);
2080 	mac_free(macp);
2081 	if (rv == 0)
2082 		vswp->if_state |= VSW_IF_REG;
2083 
2084 	D1(vswp, "%s: exit", __func__);
2085 
2086 	return (rv);
2087 }
2088 
2089 static int
2090 vsw_mac_unregister(vsw_t *vswp)
2091 {
2092 	int		rv = 0;
2093 
2094 	D1(vswp, "%s: enter", __func__);
2095 
2096 	WRITE_ENTER(&vswp->if_lockrw);
2097 
2098 	if (vswp->if_state & VSW_IF_REG) {
2099 		rv = mac_unregister(vswp->if_mh);
2100 		if (rv != 0) {
2101 			DWARN(vswp, "%s: unable to unregister from MAC "
2102 				"framework", __func__);
2103 
2104 			RW_EXIT(&vswp->if_lockrw);
2105 			D1(vswp, "%s: fail exit", __func__);
2106 			return (rv);
2107 		}
2108 
2109 		/* mark i/f as down and unregistered */
2110 		vswp->if_state &= ~(VSW_IF_UP | VSW_IF_REG);
2111 	}
2112 	RW_EXIT(&vswp->if_lockrw);
2113 
2114 	D1(vswp, "%s: exit", __func__);
2115 
2116 	return (rv);
2117 }
2118 
2119 static int
2120 vsw_m_stat(void *arg, uint_t stat, uint64_t *val)
2121 {
2122 	vsw_t			*vswp = (vsw_t *)arg;
2123 
2124 	D1(vswp, "%s: enter", __func__);
2125 
2126 	mutex_enter(&vswp->mac_lock);
2127 	if (vswp->mh == NULL) {
2128 		mutex_exit(&vswp->mac_lock);
2129 		return (EINVAL);
2130 	}
2131 
2132 	/* return stats from underlying device */
2133 	*val = mac_stat_get(vswp->mh, stat);
2134 
2135 	mutex_exit(&vswp->mac_lock);
2136 
2137 	return (0);
2138 }
2139 
2140 static void
2141 vsw_m_stop(void *arg)
2142 {
2143 	vsw_t		*vswp = (vsw_t *)arg;
2144 
2145 	D1(vswp, "%s: enter", __func__);
2146 
2147 	WRITE_ENTER(&vswp->if_lockrw);
2148 	vswp->if_state &= ~VSW_IF_UP;
2149 	RW_EXIT(&vswp->if_lockrw);
2150 
2151 	D1(vswp, "%s: exit (state = %d)", __func__, vswp->if_state);
2152 }
2153 
2154 static int
2155 vsw_m_start(void *arg)
2156 {
2157 	vsw_t		*vswp = (vsw_t *)arg;
2158 
2159 	D1(vswp, "%s: enter", __func__);
2160 
2161 	WRITE_ENTER(&vswp->if_lockrw);
2162 	vswp->if_state |= VSW_IF_UP;
2163 	RW_EXIT(&vswp->if_lockrw);
2164 
2165 	D1(vswp, "%s: exit (state = %d)", __func__, vswp->if_state);
2166 	return (0);
2167 }
2168 
2169 /*
2170  * Change the local interface address.
2171  */
2172 static int
2173 vsw_m_unicst(void *arg, const uint8_t *macaddr)
2174 {
2175 	vsw_t		*vswp = (vsw_t *)arg;
2176 
2177 	D1(vswp, "%s: enter", __func__);
2178 
2179 	WRITE_ENTER(&vswp->if_lockrw);
2180 	ether_copy(macaddr, &vswp->if_addr);
2181 	RW_EXIT(&vswp->if_lockrw);
2182 
2183 	D1(vswp, "%s: exit", __func__);
2184 
2185 	return (0);
2186 }
2187 
2188 static int
2189 vsw_m_multicst(void *arg, boolean_t add, const uint8_t *mca)
2190 {
2191 	vsw_t		*vswp = (vsw_t *)arg;
2192 	mcst_addr_t	*mcst_p = NULL;
2193 	uint64_t	addr = 0x0;
2194 	int		i, ret = 0;
2195 
2196 	D1(vswp, "%s: enter", __func__);
2197 
2198 	/*
2199 	 * Convert address into form that can be used
2200 	 * as hash table key.
2201 	 */
2202 	for (i = 0; i < ETHERADDRL; i++) {
2203 		addr = (addr << 8) | mca[i];
2204 	}
2205 
2206 	D2(vswp, "%s: addr = 0x%llx", __func__, addr);
2207 
2208 	if (add) {
2209 		D2(vswp, "%s: adding multicast", __func__);
2210 		if (vsw_add_mcst(vswp, VSW_LOCALDEV, addr, NULL) == 0) {
2211 			/*
2212 			 * Update the list of multicast addresses
2213 			 * contained within the vsw_t structure to
2214 			 * include this new one.
2215 			 */
2216 			mcst_p = kmem_zalloc(sizeof (mcst_addr_t), KM_NOSLEEP);
2217 			if (mcst_p == NULL) {
2218 				DERR(vswp, "%s unable to alloc mem", __func__);
2219 				return (1);
2220 			}
2221 			mcst_p->addr = addr;
2222 
2223 			mutex_enter(&vswp->mca_lock);
2224 			mcst_p->nextp = vswp->mcap;
2225 			vswp->mcap = mcst_p;
2226 			mutex_exit(&vswp->mca_lock);
2227 
2228 			/*
2229 			 * Call into the underlying driver to program the
2230 			 * address into HW.
2231 			 */
2232 			mutex_enter(&vswp->mac_lock);
2233 			if (vswp->mh != NULL) {
2234 				ret = mac_multicst_add(vswp->mh, mca);
2235 				if (ret != 0) {
2236 					cmn_err(CE_WARN, "!vsw%d: unable to "
2237 						"add multicast address",
2238 						vswp->instance);
2239 					mutex_exit(&vswp->mac_lock);
2240 					goto vsw_remove_addr;
2241 				}
2242 			}
2243 			mutex_exit(&vswp->mac_lock);
2244 		} else {
2245 			cmn_err(CE_WARN, "!vsw%d: unable to add multicast "
2246 				"address", vswp->instance);
2247 		}
2248 		return (ret);
2249 	}
2250 
2251 vsw_remove_addr:
2252 
2253 	D2(vswp, "%s: removing multicast", __func__);
2254 	/*
2255 	 * Remove the address from the hash table..
2256 	 */
2257 	if (vsw_del_mcst(vswp, VSW_LOCALDEV, addr, NULL) == 0) {
2258 
2259 		/*
2260 		 * ..and then from the list maintained in the
2261 		 * vsw_t structure.
2262 		 */
2263 		vsw_del_addr(VSW_LOCALDEV, vswp, addr);
2264 
2265 		mutex_enter(&vswp->mac_lock);
2266 		if (vswp->mh != NULL)
2267 			(void) mac_multicst_remove(vswp->mh, mca);
2268 		mutex_exit(&vswp->mac_lock);
2269 	}
2270 
2271 	D1(vswp, "%s: exit", __func__);
2272 
2273 	return (0);
2274 }
2275 
2276 static int
2277 vsw_m_promisc(void *arg, boolean_t on)
2278 {
2279 	vsw_t		*vswp = (vsw_t *)arg;
2280 
2281 	D1(vswp, "%s: enter", __func__);
2282 
2283 	WRITE_ENTER(&vswp->if_lockrw);
2284 	if (on)
2285 		vswp->if_state |= VSW_IF_PROMISC;
2286 	else
2287 		vswp->if_state &= ~VSW_IF_PROMISC;
2288 	RW_EXIT(&vswp->if_lockrw);
2289 
2290 	D1(vswp, "%s: exit", __func__);
2291 
2292 	return (0);
2293 }
2294 
2295 static mblk_t *
2296 vsw_m_tx(void *arg, mblk_t *mp)
2297 {
2298 	vsw_t		*vswp = (vsw_t *)arg;
2299 
2300 	D1(vswp, "%s: enter", __func__);
2301 
2302 	vswp->vsw_switch_frame(vswp, mp, VSW_LOCALDEV, NULL, NULL);
2303 
2304 	D1(vswp, "%s: exit", __func__);
2305 
2306 	return (NULL);
2307 }
2308 
2309 /*
2310  * Register for machine description (MD) updates.
2311  *
2312  * Returns 0 on success, 1 on failure.
2313  */
2314 static int
2315 vsw_mdeg_register(vsw_t *vswp)
2316 {
2317 	mdeg_prop_spec_t	*pspecp;
2318 	mdeg_node_spec_t	*inst_specp;
2319 	mdeg_handle_t		mdeg_hdl, mdeg_port_hdl;
2320 	size_t			templatesz;
2321 	int			inst, rv;
2322 
2323 	D1(vswp, "%s: enter", __func__);
2324 
2325 	/*
2326 	 * In each 'virtual-device' node in the MD there is a
2327 	 * 'cfg-handle' property which is the MD's concept of
2328 	 * an instance number (this may be completely different from
2329 	 * the device drivers instance #). OBP reads that value and
2330 	 * stores it in the 'reg' property of the appropriate node in
2331 	 * the device tree. So we use the 'reg' value when registering
2332 	 * with the mdeg framework, to ensure we get events for the
2333 	 * correct nodes.
2334 	 */
2335 	inst = ddi_prop_get_int(DDI_DEV_T_ANY, vswp->dip,
2336 		DDI_PROP_DONTPASS, reg_propname, -1);
2337 	if (inst == -1) {
2338 		cmn_err(CE_WARN, "!vsw%d: Unable to read %s property from "
2339 			"OBP device tree", vswp->instance, reg_propname);
2340 		return (1);
2341 	}
2342 
2343 	D2(vswp, "%s: instance %d registering with mdeg", __func__, inst);
2344 
2345 	/*
2346 	 * Allocate and initialize a per-instance copy
2347 	 * of the global property spec array that will
2348 	 * uniquely identify this vsw instance.
2349 	 */
2350 	templatesz = sizeof (vsw_prop_template);
2351 	pspecp = kmem_zalloc(templatesz, KM_SLEEP);
2352 
2353 	bcopy(vsw_prop_template, pspecp, templatesz);
2354 
2355 	VSW_SET_MDEG_PROP_INST(pspecp, inst);
2356 
2357 	/* initialize the complete prop spec structure */
2358 	inst_specp = kmem_zalloc(sizeof (mdeg_node_spec_t), KM_SLEEP);
2359 	inst_specp->namep = "virtual-device";
2360 	inst_specp->specp = pspecp;
2361 
2362 	/*
2363 	 * Register an interest in 'virtual-device' nodes with a
2364 	 * 'name' property of 'virtual-network-switch'
2365 	 */
2366 	rv = mdeg_register(inst_specp, &vdev_match, vsw_mdeg_cb,
2367 	    (void *)vswp, &mdeg_hdl);
2368 	if (rv != MDEG_SUCCESS) {
2369 		DERR(vswp, "%s: mdeg_register failed (%d) for vsw node",
2370 			__func__, rv);
2371 		goto mdeg_reg_fail;
2372 	}
2373 
2374 	/*
2375 	 * Register an interest in 'vsw-port' nodes.
2376 	 */
2377 	rv = mdeg_register(inst_specp, &vport_match, vsw_port_mdeg_cb,
2378 	    (void *)vswp, &mdeg_port_hdl);
2379 	if (rv != MDEG_SUCCESS) {
2380 		DERR(vswp, "%s: mdeg_register failed (%d)\n", __func__, rv);
2381 		(void) mdeg_unregister(mdeg_hdl);
2382 		goto mdeg_reg_fail;
2383 	}
2384 
2385 	/* save off data that will be needed later */
2386 	vswp->inst_spec = inst_specp;
2387 	vswp->mdeg_hdl = mdeg_hdl;
2388 	vswp->mdeg_port_hdl = mdeg_port_hdl;
2389 
2390 	D1(vswp, "%s: exit", __func__);
2391 	return (0);
2392 
2393 mdeg_reg_fail:
2394 	cmn_err(CE_WARN, "!vsw%d: Unable to register MDEG callbacks",
2395 				vswp->instance);
2396 	kmem_free(pspecp, templatesz);
2397 	kmem_free(inst_specp, sizeof (mdeg_node_spec_t));
2398 
2399 	vswp->mdeg_hdl = NULL;
2400 	vswp->mdeg_port_hdl = NULL;
2401 
2402 	return (1);
2403 }
2404 
2405 static void
2406 vsw_mdeg_unregister(vsw_t *vswp)
2407 {
2408 	D1(vswp, "vsw_mdeg_unregister: enter");
2409 
2410 	if (vswp->mdeg_hdl != NULL)
2411 		(void) mdeg_unregister(vswp->mdeg_hdl);
2412 
2413 	if (vswp->mdeg_port_hdl != NULL)
2414 		(void) mdeg_unregister(vswp->mdeg_port_hdl);
2415 
2416 	if (vswp->inst_spec != NULL) {
2417 		if (vswp->inst_spec->specp != NULL) {
2418 			(void) kmem_free(vswp->inst_spec->specp,
2419 				sizeof (vsw_prop_template));
2420 			vswp->inst_spec->specp = NULL;
2421 		}
2422 
2423 		(void) kmem_free(vswp->inst_spec,
2424 			sizeof (mdeg_node_spec_t));
2425 		vswp->inst_spec = NULL;
2426 	}
2427 
2428 	D1(vswp, "vsw_mdeg_unregister: exit");
2429 }
2430 
2431 /*
2432  * Mdeg callback invoked for the vsw node itself.
2433  */
2434 static int
2435 vsw_mdeg_cb(void *cb_argp, mdeg_result_t *resp)
2436 {
2437 	vsw_t		*vswp;
2438 	int		idx;
2439 	md_t		*mdp;
2440 	mde_cookie_t	node;
2441 	uint64_t	inst;
2442 	char		*node_name = NULL;
2443 
2444 	if (resp == NULL)
2445 		return (MDEG_FAILURE);
2446 
2447 	vswp = (vsw_t *)cb_argp;
2448 
2449 	D1(vswp, "%s: added %d : removed %d : curr matched %d"
2450 		" : prev matched %d", __func__, resp->added.nelem,
2451 		resp->removed.nelem, resp->match_curr.nelem,
2452 		resp->match_prev.nelem);
2453 
2454 	/*
2455 	 * Expect 'added' to be non-zero if virtual-network-switch
2456 	 * nodes exist in the MD when the driver attaches.
2457 	 */
2458 	for (idx = 0; idx < resp->added.nelem; idx++) {
2459 		mdp = resp->added.mdp;
2460 		node = resp->added.mdep[idx];
2461 
2462 		if (md_get_prop_str(mdp, node, "name", &node_name) != 0) {
2463 			DERR(vswp, "%s: unable to get node name for "
2464 				"node(%d) 0x%lx", __func__, idx, node);
2465 			continue;
2466 		}
2467 
2468 		if (md_get_prop_val(mdp, node, "cfg-handle", &inst)) {
2469 			DERR(vswp, "%s: prop(cfg-handle) not found port(%d)",
2470 				__func__, idx);
2471 			continue;
2472 		}
2473 
2474 		D2(vswp, "%s: added node(%d) 0x%lx with name %s "
2475 			"and inst %d", __func__, idx, node, node_name, inst);
2476 
2477 		vsw_get_initial_md_properties(vswp, mdp, node);
2478 	}
2479 
2480 	/*
2481 	 * A non-zero 'match' value indicates that the MD has been
2482 	 * updated and that a virtual-network-switch node is present
2483 	 * which may or may not have been updated. It is up to the clients
2484 	 * to examine their own nodes and determine if they have changed.
2485 	 */
2486 	for (idx = 0; idx < resp->match_curr.nelem; idx++) {
2487 		mdp = resp->match_curr.mdp;
2488 		node = resp->match_curr.mdep[idx];
2489 
2490 		if (md_get_prop_str(mdp, node, "name", &node_name) != 0) {
2491 			DERR(vswp, "%s: unable to get node name for "
2492 				"node(%d) 0x%lx", __func__, idx, node);
2493 			continue;
2494 		}
2495 
2496 		if (md_get_prop_val(mdp, node, "cfg-handle", &inst)) {
2497 			DERR(vswp, "%s: prop(cfg-handle) not found port(%d)",
2498 				__func__, idx);
2499 			continue;
2500 		}
2501 
2502 		D2(vswp, "%s: changed node(%d) 0x%lx with name %s "
2503 			"and inst %d", __func__, idx, node, node_name, inst);
2504 
2505 		vsw_update_md_prop(vswp, mdp, node);
2506 	}
2507 
2508 	return (MDEG_SUCCESS);
2509 }
2510 
2511 /*
2512  * Mdeg callback invoked for changes to the vsw-port nodes
2513  * under the vsw node.
2514  */
2515 static int
2516 vsw_port_mdeg_cb(void *cb_argp, mdeg_result_t *resp)
2517 {
2518 	vsw_t		*vswp;
2519 	int		idx;
2520 	md_t		*mdp;
2521 	mde_cookie_t	node;
2522 	uint64_t	inst;
2523 
2524 	if ((resp == NULL) || (cb_argp == NULL))
2525 		return (MDEG_FAILURE);
2526 
2527 	vswp = (vsw_t *)cb_argp;
2528 
2529 	D2(vswp, "%s: added %d : removed %d : curr matched %d"
2530 		" : prev matched %d", __func__, resp->added.nelem,
2531 		resp->removed.nelem, resp->match_curr.nelem,
2532 		resp->match_prev.nelem);
2533 
2534 	/* process added ports */
2535 	for (idx = 0; idx < resp->added.nelem; idx++) {
2536 		mdp = resp->added.mdp;
2537 		node = resp->added.mdep[idx];
2538 
2539 		D2(vswp, "%s: adding node(%d) 0x%lx", __func__, idx, node);
2540 
2541 		if (vsw_port_add(vswp, mdp, &node) != 0) {
2542 			cmn_err(CE_WARN, "!vsw%d: Unable to add new port "
2543 				"(0x%lx)", vswp->instance, node);
2544 		}
2545 	}
2546 
2547 	/* process removed ports */
2548 	for (idx = 0; idx < resp->removed.nelem; idx++) {
2549 		mdp = resp->removed.mdp;
2550 		node = resp->removed.mdep[idx];
2551 
2552 		if (md_get_prop_val(mdp, node, id_propname, &inst)) {
2553 			DERR(vswp, "%s: prop(%s) not found in port(%d)",
2554 				__func__, id_propname, idx);
2555 			continue;
2556 		}
2557 
2558 		D2(vswp, "%s: removing node(%d) 0x%lx", __func__, idx, node);
2559 
2560 		if (vsw_port_detach(vswp, inst) != 0) {
2561 			cmn_err(CE_WARN, "!vsw%d: Unable to remove port %ld",
2562 				vswp->instance, inst);
2563 		}
2564 	}
2565 
2566 	/*
2567 	 * Currently no support for updating already active ports.
2568 	 * So, ignore the match_curr and match_priv arrays for now.
2569 	 */
2570 
2571 	D1(vswp, "%s: exit", __func__);
2572 
2573 	return (MDEG_SUCCESS);
2574 }
2575 
2576 /*
2577  * Read the initial start-of-day values from the specified MD node.
2578  */
2579 static void
2580 vsw_get_initial_md_properties(vsw_t *vswp, md_t *mdp, mde_cookie_t node)
2581 {
2582 	int		i;
2583 	uint64_t 	macaddr = 0;
2584 
2585 	D1(vswp, "%s: enter", __func__);
2586 
2587 	if (vsw_get_md_physname(vswp, mdp, node, vswp->physname) == 0) {
2588 		/*
2589 		 * Note it is valid for the physname property to
2590 		 * be NULL so check actual name length to determine
2591 		 * if we have a actual device name.
2592 		 */
2593 		if (strlen(vswp->physname) > 0)
2594 			vswp->mdprops |= VSW_MD_PHYSNAME;
2595 	} else {
2596 		cmn_err(CE_WARN, "!vsw%d: Unable to read name of physical "
2597 			"device from MD", vswp->instance);
2598 		return;
2599 	}
2600 
2601 	/* mac address for vswitch device itself */
2602 	if (md_get_prop_val(mdp, node, macaddr_propname, &macaddr) != 0) {
2603 		cmn_err(CE_WARN, "!vsw%d: Unable to get MAC address from MD",
2604 			vswp->instance);
2605 
2606 		/*
2607 		 * Fallback to using the mac address of the physical
2608 		 * device.
2609 		 */
2610 		if (vsw_get_physaddr(vswp) == 0) {
2611 			cmn_err(CE_NOTE, "!vsw%d: Using MAC address from "
2612 				"physical device (%s)", vswp->instance,
2613 				vswp->physname);
2614 		} else {
2615 			cmn_err(CE_WARN, "!vsw%d: Unable to get MAC address"
2616 				"from device %s", vswp->instance,
2617 				vswp->physname);
2618 		}
2619 	} else {
2620 		WRITE_ENTER(&vswp->if_lockrw);
2621 		for (i = ETHERADDRL - 1; i >= 0; i--) {
2622 			vswp->if_addr.ether_addr_octet[i] = macaddr & 0xFF;
2623 			macaddr >>= 8;
2624 		}
2625 		RW_EXIT(&vswp->if_lockrw);
2626 		vswp->mdprops |= VSW_MD_MACADDR;
2627 	}
2628 
2629 	if (vsw_get_md_smodes(vswp, mdp, node,
2630 				vswp->smode, &vswp->smode_num)) {
2631 		cmn_err(CE_WARN, "vsw%d: Unable to read %s property from "
2632 			"MD, defaulting to programmed mode", vswp->instance,
2633 			smode_propname);
2634 
2635 		for (i = 0; i < NUM_SMODES; i++)
2636 			vswp->smode[i] = VSW_LAYER2;
2637 
2638 		vswp->smode_num = NUM_SMODES;
2639 	} else {
2640 		ASSERT(vswp->smode_num != 0);
2641 		vswp->mdprops |= VSW_MD_SMODE;
2642 	}
2643 
2644 	/*
2645 	 * Unable to setup any switching mode, nothing more
2646 	 * we can do.
2647 	 */
2648 	if (vsw_setup_switching(vswp))
2649 		return;
2650 
2651 	WRITE_ENTER(&vswp->if_lockrw);
2652 	vswp->if_state &= ~VSW_IF_UP;
2653 	RW_EXIT(&vswp->if_lockrw);
2654 	if (vswp->mdprops & (VSW_MD_MACADDR | VSW_DEV_MACADDR)) {
2655 		if (vsw_mac_register(vswp) != 0) {
2656 			/*
2657 			 * Treat this as a non-fatal error as we may be
2658 			 * able to operate in some other mode.
2659 			 */
2660 			cmn_err(CE_WARN, "vsw%d: Unable to register as "
2661 				"provider with MAC layer", vswp->instance);
2662 		}
2663 	}
2664 
2665 	D1(vswp, "%s: exit", __func__);
2666 }
2667 
2668 /*
2669  * Check to see if the relevant properties in the specified node have
2670  * changed, and if so take the appropriate action.
2671  *
2672  * If any of the properties are missing or invalid we don't take
2673  * any action, as this function should only be invoked when modifications
2674  * have been made to what we assume is a working configuration, which
2675  * we leave active.
2676  *
2677  * Note it is legal for this routine to be invoked even if none of the
2678  * properties in the port node within the MD have actually changed.
2679  */
2680 static void
2681 vsw_update_md_prop(vsw_t *vswp, md_t *mdp, mde_cookie_t node)
2682 {
2683 	char		physname[LIFNAMSIZ];
2684 	char		drv[LIFNAMSIZ];
2685 	uint_t		ddi_instance;
2686 	uint8_t		new_smode[NUM_SMODES];
2687 	int		i, smode_num = 0;
2688 	uint64_t 	macaddr = 0;
2689 	vsw_port_list_t *plist = &vswp->plist;
2690 	vsw_port_t	*port = NULL;
2691 	enum		{MD_init = 0x1,
2692 				MD_physname = 0x2,
2693 				MD_macaddr = 0x4,
2694 				MD_smode = 0x8} updated;
2695 
2696 	updated = MD_init;
2697 
2698 	D1(vswp, "%s: enter", __func__);
2699 
2700 	/*
2701 	 * Check if name of physical device in MD has changed.
2702 	 */
2703 	if (vsw_get_md_physname(vswp, mdp, node, (char *)&physname) == 0) {
2704 		/*
2705 		 * Do basic sanity check on new device name/instance,
2706 		 * if its non NULL. It is valid for the device name to
2707 		 * have changed from a non NULL to a NULL value, i.e.
2708 		 * the vsw is being changed to 'routed' mode.
2709 		 */
2710 		if ((strlen(physname) != 0) &&
2711 			(ddi_parse(physname, drv,
2712 				&ddi_instance) != DDI_SUCCESS)) {
2713 			cmn_err(CE_WARN, "!vsw%d: new device name %s is not"
2714 				" a valid device name/instance",
2715 				vswp->instance, physname);
2716 			goto fail_reconf;
2717 		}
2718 
2719 		if (strcmp(physname, vswp->physname)) {
2720 			D2(vswp, "%s: device name changed from %s to %s",
2721 					__func__, vswp->physname, physname);
2722 
2723 			updated |= MD_physname;
2724 		} else {
2725 			D2(vswp, "%s: device name unchanged at %s",
2726 					__func__, vswp->physname);
2727 		}
2728 	} else {
2729 		cmn_err(CE_WARN, "!vsw%d: Unable to read name of physical "
2730 			"device from updated MD.", vswp->instance);
2731 		goto fail_reconf;
2732 	}
2733 
2734 	/*
2735 	 * Check if MAC address has changed.
2736 	 */
2737 	if (md_get_prop_val(mdp, node, macaddr_propname, &macaddr) != 0) {
2738 		cmn_err(CE_WARN, "!vsw%d: Unable to get MAC address from MD",
2739 			vswp->instance);
2740 		goto fail_reconf;
2741 	} else {
2742 		READ_ENTER(&vswp->if_lockrw);
2743 		for (i = ETHERADDRL - 1; i >= 0; i--) {
2744 			if (vswp->if_addr.ether_addr_octet[i]
2745 							!= (macaddr & 0xFF)) {
2746 				D2(vswp, "%s: octet[%d] 0x%x != 0x%x",
2747 					__func__, i,
2748 					vswp->if_addr.ether_addr_octet[i],
2749 					(macaddr & 0xFF));
2750 				updated |= MD_macaddr;
2751 				break;
2752 			}
2753 			macaddr >>= 8;
2754 		}
2755 		RW_EXIT(&vswp->if_lockrw);
2756 	}
2757 
2758 	/*
2759 	 * Check if switching modes have changed.
2760 	 */
2761 	if (vsw_get_md_smodes(vswp, mdp, node,
2762 				new_smode, &smode_num)) {
2763 		cmn_err(CE_WARN, "!vsw%d: Unable to read %s property from MD",
2764 					vswp->instance, smode_propname);
2765 		goto fail_reconf;
2766 	} else {
2767 		ASSERT(smode_num != 0);
2768 		if (smode_num != vswp->smode_num) {
2769 			D2(vswp, "%s: number of modes changed from %d to %d",
2770 				__func__, vswp->smode_num, smode_num);
2771 		}
2772 
2773 		for (i = 0; i < smode_num; i++) {
2774 			if (new_smode[i] != vswp->smode[i]) {
2775 				D2(vswp, "%s: mode changed from %d to %d",
2776 					__func__, vswp->smode[i], new_smode[i]);
2777 				updated |= MD_smode;
2778 				break;
2779 			}
2780 		}
2781 	}
2782 
2783 	/*
2784 	 * Now make any changes which are needed...
2785 	 */
2786 
2787 	if (updated & (MD_physname | MD_smode)) {
2788 		/*
2789 		 * Disconnect all ports from the current card
2790 		 */
2791 		WRITE_ENTER(&plist->lockrw);
2792 		for (port = plist->head; port != NULL; port = port->p_next) {
2793 			/* Remove address if was programmed into HW. */
2794 			if (vsw_unset_hw(vswp, port)) {
2795 				RW_EXIT(&plist->lockrw);
2796 				goto fail_update;
2797 			}
2798 		}
2799 		RW_EXIT(&plist->lockrw);
2800 
2801 		/*
2802 		 * Stop, detach the old device..
2803 		 */
2804 		vsw_mac_detach(vswp);
2805 
2806 		/*
2807 		 * Update phys name.
2808 		 */
2809 		if (updated & MD_physname) {
2810 			cmn_err(CE_NOTE, "!vsw%d: changing from %s to %s",
2811 				vswp->instance, vswp->physname, physname);
2812 			(void) strncpy(vswp->physname,
2813 					physname, strlen(physname) + 1);
2814 
2815 			if (strlen(vswp->physname) > 0)
2816 				vswp->mdprops |= VSW_MD_PHYSNAME;
2817 		}
2818 
2819 		/*
2820 		 * Update array with the new switch mode values.
2821 		 */
2822 		if (updated & MD_smode) {
2823 			for (i = 0; i < smode_num; i++)
2824 				vswp->smode[i] = new_smode[i];
2825 
2826 			vswp->smode_num = smode_num;
2827 			vswp->smode_idx = 0;
2828 		}
2829 
2830 		/*
2831 		 * ..and attach, start the new device.
2832 		 */
2833 		if (vsw_setup_switching(vswp))
2834 			goto fail_update;
2835 
2836 		/*
2837 		 * Connect ports to new card.
2838 		 */
2839 		WRITE_ENTER(&plist->lockrw);
2840 		for (port = plist->head; port != NULL; port = port->p_next) {
2841 			if (vsw_set_hw(vswp, port)) {
2842 				RW_EXIT(&plist->lockrw);
2843 				goto fail_update;
2844 			}
2845 		}
2846 		RW_EXIT(&plist->lockrw);
2847 	}
2848 
2849 	if (updated & MD_macaddr) {
2850 		cmn_err(CE_NOTE, "!vsw%d: changing mac address to 0x%lx",
2851 				vswp->instance, macaddr);
2852 
2853 		WRITE_ENTER(&vswp->if_lockrw);
2854 		for (i = ETHERADDRL - 1; i >= 0; i--) {
2855 			vswp->if_addr.ether_addr_octet[i] = macaddr & 0xFF;
2856 			macaddr >>= 8;
2857 		}
2858 		RW_EXIT(&vswp->if_lockrw);
2859 
2860 		/*
2861 		 * Notify the MAC layer of the changed address.
2862 		 */
2863 		mac_unicst_update(vswp->if_mh, (uint8_t *)&vswp->if_addr);
2864 	}
2865 
2866 	return;
2867 
2868 fail_reconf:
2869 	cmn_err(CE_WARN, "!vsw%d: configuration unchanged", vswp->instance);
2870 	return;
2871 
2872 fail_update:
2873 	cmn_err(CE_WARN, "!vsw%d: update of configuration failed",
2874 			vswp->instance);
2875 }
2876 
2877 /*
2878  * Add a new port to the system.
2879  *
2880  * Returns 0 on success, 1 on failure.
2881  */
2882 int
2883 vsw_port_add(vsw_t *vswp, md_t *mdp, mde_cookie_t *node)
2884 {
2885 	uint64_t		ldc_id;
2886 	uint8_t			*addrp;
2887 	int			i, addrsz;
2888 	int			num_nodes = 0, nchan = 0;
2889 	int			listsz = 0;
2890 	mde_cookie_t		*listp = NULL;
2891 	struct ether_addr	ea;
2892 	uint64_t		macaddr;
2893 	uint64_t		inst = 0;
2894 	vsw_port_t		*port;
2895 
2896 	if (md_get_prop_val(mdp, *node, id_propname, &inst)) {
2897 		DWARN(vswp, "%s: prop(%s) not found", __func__,
2898 			id_propname);
2899 		return (1);
2900 	}
2901 
2902 	/*
2903 	 * Find the channel endpoint node(s) (which should be under this
2904 	 * port node) which contain the channel id(s).
2905 	 */
2906 	if ((num_nodes = md_node_count(mdp)) <= 0) {
2907 		DERR(vswp, "%s: invalid number of nodes found (%d)",
2908 			__func__, num_nodes);
2909 		return (1);
2910 	}
2911 
2912 	D2(vswp, "%s: %d nodes found", __func__, num_nodes);
2913 
2914 	/* allocate enough space for node list */
2915 	listsz = num_nodes * sizeof (mde_cookie_t);
2916 	listp = kmem_zalloc(listsz, KM_SLEEP);
2917 
2918 	nchan = md_scan_dag(mdp, *node,
2919 		md_find_name(mdp, chan_propname),
2920 		md_find_name(mdp, "fwd"), listp);
2921 
2922 	if (nchan <= 0) {
2923 		DWARN(vswp, "%s: no %s nodes found", __func__, chan_propname);
2924 		kmem_free(listp, listsz);
2925 		return (1);
2926 	}
2927 
2928 	D2(vswp, "%s: %d %s nodes found", __func__, nchan, chan_propname);
2929 
2930 	/* use property from first node found */
2931 	if (md_get_prop_val(mdp, listp[0], id_propname, &ldc_id)) {
2932 		DWARN(vswp, "%s: prop(%s) not found\n", __func__,
2933 			id_propname);
2934 		kmem_free(listp, listsz);
2935 		return (1);
2936 	}
2937 
2938 	/* don't need list any more */
2939 	kmem_free(listp, listsz);
2940 
2941 	D2(vswp, "%s: ldc_id 0x%llx", __func__, ldc_id);
2942 
2943 	/* read mac-address property */
2944 	if (md_get_prop_data(mdp, *node, remaddr_propname,
2945 					&addrp, &addrsz)) {
2946 		DWARN(vswp, "%s: prop(%s) not found",
2947 				__func__, remaddr_propname);
2948 		return (1);
2949 	}
2950 
2951 	if (addrsz < ETHERADDRL) {
2952 		DWARN(vswp, "%s: invalid address size", __func__);
2953 		return (1);
2954 	}
2955 
2956 	macaddr = *((uint64_t *)addrp);
2957 	D2(vswp, "%s: remote mac address 0x%llx", __func__, macaddr);
2958 
2959 	for (i = ETHERADDRL - 1; i >= 0; i--) {
2960 		ea.ether_addr_octet[i] = macaddr & 0xFF;
2961 		macaddr >>= 8;
2962 	}
2963 
2964 	if (vsw_port_attach(vswp, (int)inst, &ldc_id, 1, &ea) != 0) {
2965 		DERR(vswp, "%s: failed to attach port", __func__);
2966 		return (1);
2967 	}
2968 
2969 	port = vsw_lookup_port(vswp, (int)inst);
2970 
2971 	/* just successfuly created the port, so it should exist */
2972 	ASSERT(port != NULL);
2973 
2974 	return (0);
2975 }
2976 
2977 /*
2978  * Attach the specified port.
2979  *
2980  * Returns 0 on success, 1 on failure.
2981  */
2982 static int
2983 vsw_port_attach(vsw_t *vswp, int p_instance, uint64_t *ldcids, int nids,
2984 struct ether_addr *macaddr)
2985 {
2986 	vsw_port_list_t		*plist = &vswp->plist;
2987 	vsw_port_t		*port, **prev_port;
2988 	int			i;
2989 
2990 	D1(vswp, "%s: enter : port %d", __func__, p_instance);
2991 
2992 	/* port already exists? */
2993 	READ_ENTER(&plist->lockrw);
2994 	for (port = plist->head; port != NULL; port = port->p_next) {
2995 		if (port->p_instance == p_instance) {
2996 			DWARN(vswp, "%s: port instance %d already attached",
2997 				__func__, p_instance);
2998 			RW_EXIT(&plist->lockrw);
2999 			return (1);
3000 		}
3001 	}
3002 	RW_EXIT(&plist->lockrw);
3003 
3004 	port = kmem_zalloc(sizeof (vsw_port_t), KM_SLEEP);
3005 	port->p_vswp = vswp;
3006 	port->p_instance = p_instance;
3007 	port->p_ldclist.num_ldcs = 0;
3008 	port->p_ldclist.head = NULL;
3009 	port->addr_set = VSW_ADDR_UNSET;
3010 
3011 	rw_init(&port->p_ldclist.lockrw, NULL, RW_DRIVER, NULL);
3012 
3013 	mutex_init(&port->tx_lock, NULL, MUTEX_DRIVER, NULL);
3014 	mutex_init(&port->mca_lock, NULL, MUTEX_DRIVER, NULL);
3015 
3016 	mutex_init(&port->ref_lock, NULL, MUTEX_DRIVER, NULL);
3017 	cv_init(&port->ref_cv, NULL, CV_DRIVER, NULL);
3018 
3019 	mutex_init(&port->state_lock, NULL, MUTEX_DRIVER, NULL);
3020 	cv_init(&port->state_cv, NULL, CV_DRIVER, NULL);
3021 	port->state = VSW_PORT_INIT;
3022 
3023 	if (nids > VSW_PORT_MAX_LDCS) {
3024 		D2(vswp, "%s: using first of %d ldc ids",
3025 			__func__, nids);
3026 		nids = VSW_PORT_MAX_LDCS;
3027 	}
3028 
3029 	D2(vswp, "%s: %d nids", __func__, nids);
3030 	for (i = 0; i < nids; i++) {
3031 		D2(vswp, "%s: ldcid (%llx)", __func__, (uint64_t)ldcids[i]);
3032 		if (vsw_ldc_attach(port, (uint64_t)ldcids[i]) != 0) {
3033 			DERR(vswp, "%s: ldc_attach failed", __func__);
3034 
3035 			rw_destroy(&port->p_ldclist.lockrw);
3036 
3037 			cv_destroy(&port->ref_cv);
3038 			mutex_destroy(&port->ref_lock);
3039 
3040 			cv_destroy(&port->state_cv);
3041 			mutex_destroy(&port->state_lock);
3042 
3043 			mutex_destroy(&port->tx_lock);
3044 			mutex_destroy(&port->mca_lock);
3045 			kmem_free(port, sizeof (vsw_port_t));
3046 			return (1);
3047 		}
3048 	}
3049 
3050 	ether_copy(macaddr, &port->p_macaddr);
3051 
3052 	WRITE_ENTER(&plist->lockrw);
3053 
3054 	/* create the fdb entry for this port/mac address */
3055 	(void) vsw_add_fdb(vswp, port);
3056 
3057 	(void) vsw_set_hw(vswp, port);
3058 
3059 	/* link it into the list of ports for this vsw instance */
3060 	prev_port = (vsw_port_t **)(&plist->head);
3061 	port->p_next = *prev_port;
3062 	*prev_port = port;
3063 	plist->num_ports++;
3064 	RW_EXIT(&plist->lockrw);
3065 
3066 	/*
3067 	 * Initialise the port and any ldc's under it.
3068 	 */
3069 	(void) vsw_init_ldcs(port);
3070 
3071 	D1(vswp, "%s: exit", __func__);
3072 	return (0);
3073 }
3074 
3075 /*
3076  * Detach the specified port.
3077  *
3078  * Returns 0 on success, 1 on failure.
3079  */
3080 static int
3081 vsw_port_detach(vsw_t *vswp, int p_instance)
3082 {
3083 	vsw_port_t	*port = NULL;
3084 	vsw_port_list_t	*plist = &vswp->plist;
3085 
3086 	D1(vswp, "%s: enter: port id %d", __func__, p_instance);
3087 
3088 	WRITE_ENTER(&plist->lockrw);
3089 
3090 	if ((port = vsw_lookup_port(vswp, p_instance)) == NULL) {
3091 		RW_EXIT(&plist->lockrw);
3092 		return (1);
3093 	}
3094 
3095 	if (vsw_plist_del_node(vswp, port)) {
3096 		RW_EXIT(&plist->lockrw);
3097 		return (1);
3098 	}
3099 
3100 	/* Remove address if was programmed into HW. */
3101 	(void) vsw_unset_hw(vswp, port);
3102 
3103 	/* Remove the fdb entry for this port/mac address */
3104 	(void) vsw_del_fdb(vswp, port);
3105 
3106 	/* Remove any multicast addresses.. */
3107 	vsw_del_mcst_port(port);
3108 
3109 	/*
3110 	 * No longer need to hold writer lock on port list now
3111 	 * that we have unlinked the target port from the list.
3112 	 */
3113 	RW_EXIT(&plist->lockrw);
3114 
3115 	READ_ENTER(&plist->lockrw);
3116 
3117 	if (vswp->recfg_reqd)
3118 		(void) vsw_reconfig_hw(vswp);
3119 
3120 	RW_EXIT(&plist->lockrw);
3121 
3122 	if (vsw_port_delete(port)) {
3123 		return (1);
3124 	}
3125 
3126 	D1(vswp, "%s: exit: p_instance(%d)", __func__, p_instance);
3127 	return (0);
3128 }
3129 
3130 /*
3131  * Detach all active ports.
3132  *
3133  * Returns 0 on success, 1 on failure.
3134  */
3135 static int
3136 vsw_detach_ports(vsw_t *vswp)
3137 {
3138 	vsw_port_list_t 	*plist = &vswp->plist;
3139 	vsw_port_t		*port = NULL;
3140 
3141 	D1(vswp, "%s: enter", __func__);
3142 
3143 	WRITE_ENTER(&plist->lockrw);
3144 
3145 	while ((port = plist->head) != NULL) {
3146 		if (vsw_plist_del_node(vswp, port)) {
3147 			DERR(vswp, "%s: Error deleting port %d"
3148 				" from port list", __func__,
3149 				port->p_instance);
3150 			RW_EXIT(&plist->lockrw);
3151 			return (1);
3152 		}
3153 
3154 		/* Remove address if was programmed into HW. */
3155 		(void) vsw_unset_hw(vswp, port);
3156 
3157 		/* Remove the fdb entry for this port/mac address */
3158 		(void) vsw_del_fdb(vswp, port);
3159 
3160 		/* Remove any multicast addresses.. */
3161 		vsw_del_mcst_port(port);
3162 
3163 		/*
3164 		 * No longer need to hold the lock on the port list
3165 		 * now that we have unlinked the target port from the
3166 		 * list.
3167 		 */
3168 		RW_EXIT(&plist->lockrw);
3169 		if (vsw_port_delete(port)) {
3170 			DERR(vswp, "%s: Error deleting port %d",
3171 				__func__, port->p_instance);
3172 			return (1);
3173 		}
3174 		WRITE_ENTER(&plist->lockrw);
3175 	}
3176 	RW_EXIT(&plist->lockrw);
3177 
3178 	D1(vswp, "%s: exit", __func__);
3179 
3180 	return (0);
3181 }
3182 
3183 /*
3184  * Delete the specified port.
3185  *
3186  * Returns 0 on success, 1 on failure.
3187  */
3188 static int
3189 vsw_port_delete(vsw_port_t *port)
3190 {
3191 	vsw_ldc_list_t 		*ldcl;
3192 	vsw_t			*vswp = port->p_vswp;
3193 
3194 	D1(vswp, "%s: enter : port id %d", __func__, port->p_instance);
3195 
3196 	(void) vsw_uninit_ldcs(port);
3197 
3198 	/*
3199 	 * Wait for any pending ctrl msg tasks which reference this
3200 	 * port to finish.
3201 	 */
3202 	if (vsw_drain_port_taskq(port))
3203 		return (1);
3204 
3205 	/*
3206 	 * Wait for port reference count to hit zero.
3207 	 */
3208 	mutex_enter(&port->ref_lock);
3209 	while (port->ref_cnt != 0)
3210 		cv_wait(&port->ref_cv, &port->ref_lock);
3211 	mutex_exit(&port->ref_lock);
3212 
3213 	/*
3214 	 * Wait for any active callbacks to finish
3215 	 */
3216 	if (vsw_drain_ldcs(port))
3217 		return (1);
3218 
3219 	ldcl = &port->p_ldclist;
3220 	WRITE_ENTER(&ldcl->lockrw);
3221 	while (ldcl->num_ldcs > 0) {
3222 		if (vsw_ldc_detach(port, ldcl->head->ldc_id) != 0) {;
3223 			cmn_err(CE_WARN, "!vsw%d: unable to detach ldc %ld",
3224 					vswp->instance, ldcl->head->ldc_id);
3225 			RW_EXIT(&ldcl->lockrw);
3226 			return (1);
3227 		}
3228 	}
3229 	RW_EXIT(&ldcl->lockrw);
3230 
3231 	rw_destroy(&port->p_ldclist.lockrw);
3232 
3233 	mutex_destroy(&port->mca_lock);
3234 	mutex_destroy(&port->tx_lock);
3235 	cv_destroy(&port->ref_cv);
3236 	mutex_destroy(&port->ref_lock);
3237 
3238 	cv_destroy(&port->state_cv);
3239 	mutex_destroy(&port->state_lock);
3240 
3241 	kmem_free(port, sizeof (vsw_port_t));
3242 
3243 	D1(vswp, "%s: exit", __func__);
3244 
3245 	return (0);
3246 }
3247 
3248 /*
3249  * Attach a logical domain channel (ldc) under a specified port.
3250  *
3251  * Returns 0 on success, 1 on failure.
3252  */
3253 static int
3254 vsw_ldc_attach(vsw_port_t *port, uint64_t ldc_id)
3255 {
3256 	vsw_t 		*vswp = port->p_vswp;
3257 	vsw_ldc_list_t *ldcl = &port->p_ldclist;
3258 	vsw_ldc_t 	*ldcp = NULL;
3259 	ldc_attr_t 	attr;
3260 	ldc_status_t	istatus;
3261 	int 		status = DDI_FAILURE;
3262 	int		rv;
3263 	enum		{ PROG_init = 0x0, PROG_mblks = 0x1,
3264 				PROG_callback = 0x2}
3265 			progress;
3266 
3267 	progress = PROG_init;
3268 
3269 	D1(vswp, "%s: enter", __func__);
3270 
3271 	ldcp = kmem_zalloc(sizeof (vsw_ldc_t), KM_NOSLEEP);
3272 	if (ldcp == NULL) {
3273 		DERR(vswp, "%s: kmem_zalloc failed", __func__);
3274 		return (1);
3275 	}
3276 	ldcp->ldc_id = ldc_id;
3277 
3278 	/* allocate pool of receive mblks */
3279 	rv = vio_create_mblks(vsw_num_mblks, vsw_mblk_size, &(ldcp->rxh));
3280 	if (rv) {
3281 		DWARN(vswp, "%s: unable to create free mblk pool for"
3282 			" channel %ld (rv %d)", __func__, ldc_id, rv);
3283 		kmem_free(ldcp, sizeof (vsw_ldc_t));
3284 		return (1);
3285 	}
3286 
3287 	progress |= PROG_mblks;
3288 
3289 	mutex_init(&ldcp->ldc_txlock, NULL, MUTEX_DRIVER, NULL);
3290 	mutex_init(&ldcp->ldc_cblock, NULL, MUTEX_DRIVER, NULL);
3291 	mutex_init(&ldcp->drain_cv_lock, NULL, MUTEX_DRIVER, NULL);
3292 	cv_init(&ldcp->drain_cv, NULL, CV_DRIVER, NULL);
3293 	rw_init(&ldcp->lane_in.dlistrw, NULL, RW_DRIVER, NULL);
3294 	rw_init(&ldcp->lane_out.dlistrw, NULL, RW_DRIVER, NULL);
3295 
3296 	/* required for handshake with peer */
3297 	ldcp->local_session = (uint64_t)ddi_get_lbolt();
3298 	ldcp->peer_session = 0;
3299 	ldcp->session_status = 0;
3300 
3301 	mutex_init(&ldcp->hss_lock, NULL, MUTEX_DRIVER, NULL);
3302 	ldcp->hss_id = 1;	/* Initial handshake session id */
3303 
3304 	/* only set for outbound lane, inbound set by peer */
3305 	mutex_init(&ldcp->lane_in.seq_lock, NULL, MUTEX_DRIVER, NULL);
3306 	mutex_init(&ldcp->lane_out.seq_lock, NULL, MUTEX_DRIVER, NULL);
3307 	vsw_set_lane_attr(vswp, &ldcp->lane_out);
3308 
3309 	attr.devclass = LDC_DEV_NT_SVC;
3310 	attr.instance = ddi_get_instance(vswp->dip);
3311 	attr.mode = LDC_MODE_UNRELIABLE;
3312 	attr.mtu = VSW_LDC_MTU;
3313 	status = ldc_init(ldc_id, &attr, &ldcp->ldc_handle);
3314 	if (status != 0) {
3315 		DERR(vswp, "%s(%lld): ldc_init failed, rv (%d)",
3316 		    __func__, ldc_id, status);
3317 		goto ldc_attach_fail;
3318 	}
3319 
3320 	status = ldc_reg_callback(ldcp->ldc_handle, vsw_ldc_cb, (caddr_t)ldcp);
3321 	if (status != 0) {
3322 		DERR(vswp, "%s(%lld): ldc_reg_callback failed, rv (%d)",
3323 		    __func__, ldc_id, status);
3324 		(void) ldc_fini(ldcp->ldc_handle);
3325 		goto ldc_attach_fail;
3326 	}
3327 
3328 	progress |= PROG_callback;
3329 
3330 	mutex_init(&ldcp->status_lock, NULL, MUTEX_DRIVER, NULL);
3331 
3332 	if (ldc_status(ldcp->ldc_handle, &istatus) != 0) {
3333 		DERR(vswp, "%s: ldc_status failed", __func__);
3334 		mutex_destroy(&ldcp->status_lock);
3335 		goto ldc_attach_fail;
3336 	}
3337 
3338 	ldcp->ldc_status = istatus;
3339 	ldcp->ldc_port = port;
3340 	ldcp->ldc_vswp = vswp;
3341 
3342 	/* link it into the list of channels for this port */
3343 	WRITE_ENTER(&ldcl->lockrw);
3344 	ldcp->ldc_next = ldcl->head;
3345 	ldcl->head = ldcp;
3346 	ldcl->num_ldcs++;
3347 	RW_EXIT(&ldcl->lockrw);
3348 
3349 	D1(vswp, "%s: exit", __func__);
3350 	return (0);
3351 
3352 ldc_attach_fail:
3353 	mutex_destroy(&ldcp->ldc_txlock);
3354 	mutex_destroy(&ldcp->ldc_cblock);
3355 
3356 	cv_destroy(&ldcp->drain_cv);
3357 
3358 	rw_destroy(&ldcp->lane_in.dlistrw);
3359 	rw_destroy(&ldcp->lane_out.dlistrw);
3360 
3361 	if (progress & PROG_callback) {
3362 		(void) ldc_unreg_callback(ldcp->ldc_handle);
3363 	}
3364 
3365 	if ((progress & PROG_mblks) && (ldcp->rxh != NULL)) {
3366 		if (vio_destroy_mblks(ldcp->rxh) != 0) {
3367 			/*
3368 			 * Something odd has happened, as the destroy
3369 			 * will only fail if some mblks have been allocated
3370 			 * from the pool already (which shouldn't happen)
3371 			 * and have not been returned.
3372 			 *
3373 			 * Add the pool pointer to a list maintained in
3374 			 * the device instance. Another attempt will be made
3375 			 * to free the pool when the device itself detaches.
3376 			 */
3377 			cmn_err(CE_WARN, "!vsw%d: Creation of ldc channel %ld "
3378 				"failed and cannot destroy associated mblk "
3379 				"pool", vswp->instance, ldc_id);
3380 			ldcp->rxh->nextp =  vswp->rxh;
3381 			vswp->rxh = ldcp->rxh;
3382 		}
3383 	}
3384 	mutex_destroy(&ldcp->drain_cv_lock);
3385 	mutex_destroy(&ldcp->hss_lock);
3386 
3387 	mutex_destroy(&ldcp->lane_in.seq_lock);
3388 	mutex_destroy(&ldcp->lane_out.seq_lock);
3389 	kmem_free(ldcp, sizeof (vsw_ldc_t));
3390 
3391 	return (1);
3392 }
3393 
3394 /*
3395  * Detach a logical domain channel (ldc) belonging to a
3396  * particular port.
3397  *
3398  * Returns 0 on success, 1 on failure.
3399  */
3400 static int
3401 vsw_ldc_detach(vsw_port_t *port, uint64_t ldc_id)
3402 {
3403 	vsw_t 		*vswp = port->p_vswp;
3404 	vsw_ldc_t 	*ldcp, *prev_ldcp;
3405 	vsw_ldc_list_t	*ldcl = &port->p_ldclist;
3406 	int 		rv;
3407 
3408 	prev_ldcp = ldcl->head;
3409 	for (; (ldcp = prev_ldcp) != NULL; prev_ldcp = ldcp->ldc_next) {
3410 		if (ldcp->ldc_id == ldc_id) {
3411 			break;
3412 		}
3413 	}
3414 
3415 	/* specified ldc id not found */
3416 	if (ldcp == NULL) {
3417 		DERR(vswp, "%s: ldcp = NULL", __func__);
3418 		return (1);
3419 	}
3420 
3421 	D2(vswp, "%s: detaching channel %lld", __func__, ldcp->ldc_id);
3422 
3423 	/*
3424 	 * Before we can close the channel we must release any mapped
3425 	 * resources (e.g. drings).
3426 	 */
3427 	vsw_free_lane_resources(ldcp, INBOUND);
3428 	vsw_free_lane_resources(ldcp, OUTBOUND);
3429 
3430 	/*
3431 	 * If the close fails we are in serious trouble, as won't
3432 	 * be able to delete the parent port.
3433 	 */
3434 	if ((rv = ldc_close(ldcp->ldc_handle)) != 0) {
3435 		DERR(vswp, "%s: error %d closing channel %lld",
3436 			__func__, rv, ldcp->ldc_id);
3437 		return (1);
3438 	}
3439 
3440 	(void) ldc_fini(ldcp->ldc_handle);
3441 
3442 	ldcp->ldc_status = LDC_INIT;
3443 	ldcp->ldc_handle = NULL;
3444 	ldcp->ldc_vswp = NULL;
3445 
3446 	if (ldcp->rxh != NULL) {
3447 		if (vio_destroy_mblks(ldcp->rxh)) {
3448 			/*
3449 			 * Mostly likely some mblks are still in use and
3450 			 * have not been returned to the pool. Add the pool
3451 			 * to the list maintained in the device instance.
3452 			 * Another attempt will be made to destroy the pool
3453 			 * when the device detaches.
3454 			 */
3455 			ldcp->rxh->nextp =  vswp->rxh;
3456 			vswp->rxh = ldcp->rxh;
3457 		}
3458 	}
3459 
3460 	/* unlink it from the list */
3461 	prev_ldcp = ldcp->ldc_next;
3462 	ldcl->num_ldcs--;
3463 
3464 	mutex_destroy(&ldcp->ldc_txlock);
3465 	mutex_destroy(&ldcp->ldc_cblock);
3466 	cv_destroy(&ldcp->drain_cv);
3467 	mutex_destroy(&ldcp->drain_cv_lock);
3468 	mutex_destroy(&ldcp->hss_lock);
3469 	mutex_destroy(&ldcp->lane_in.seq_lock);
3470 	mutex_destroy(&ldcp->lane_out.seq_lock);
3471 	mutex_destroy(&ldcp->status_lock);
3472 	rw_destroy(&ldcp->lane_in.dlistrw);
3473 	rw_destroy(&ldcp->lane_out.dlistrw);
3474 
3475 	kmem_free(ldcp, sizeof (vsw_ldc_t));
3476 
3477 	return (0);
3478 }
3479 
3480 /*
3481  * Open and attempt to bring up the channel. Note that channel
3482  * can only be brought up if peer has also opened channel.
3483  *
3484  * Returns 0 if can open and bring up channel, otherwise
3485  * returns 1.
3486  */
3487 static int
3488 vsw_ldc_init(vsw_ldc_t *ldcp)
3489 {
3490 	vsw_t 		*vswp = ldcp->ldc_vswp;
3491 	ldc_status_t	istatus = 0;
3492 	int		rv;
3493 
3494 	D1(vswp, "%s: enter", __func__);
3495 
3496 	LDC_ENTER_LOCK(ldcp);
3497 
3498 	/* don't start at 0 in case clients don't like that */
3499 	ldcp->next_ident = 1;
3500 
3501 	rv = ldc_open(ldcp->ldc_handle);
3502 	if (rv != 0) {
3503 		DERR(vswp, "%s: ldc_open failed: id(%lld) rv(%d)",
3504 		    __func__, ldcp->ldc_id, rv);
3505 		LDC_EXIT_LOCK(ldcp);
3506 		return (1);
3507 	}
3508 
3509 	if (ldc_status(ldcp->ldc_handle, &istatus) != 0) {
3510 		DERR(vswp, "%s: unable to get status", __func__);
3511 		LDC_EXIT_LOCK(ldcp);
3512 		return (1);
3513 
3514 	} else if (istatus != LDC_OPEN && istatus != LDC_READY) {
3515 		DERR(vswp, "%s: id (%lld) status(%d) is not OPEN/READY",
3516 		    __func__, ldcp->ldc_id, istatus);
3517 		LDC_EXIT_LOCK(ldcp);
3518 		return (1);
3519 	}
3520 
3521 	mutex_enter(&ldcp->status_lock);
3522 	ldcp->ldc_status = istatus;
3523 	mutex_exit(&ldcp->status_lock);
3524 
3525 	rv = ldc_up(ldcp->ldc_handle);
3526 	if (rv != 0) {
3527 		/*
3528 		 * Not a fatal error for ldc_up() to fail, as peer
3529 		 * end point may simply not be ready yet.
3530 		 */
3531 		D2(vswp, "%s: ldc_up err id(%lld) rv(%d)", __func__,
3532 			ldcp->ldc_id, rv);
3533 		LDC_EXIT_LOCK(ldcp);
3534 		return (1);
3535 	}
3536 
3537 	/*
3538 	 * ldc_up() call is non-blocking so need to explicitly
3539 	 * check channel status to see if in fact the channel
3540 	 * is UP.
3541 	 */
3542 	mutex_enter(&ldcp->status_lock);
3543 	istatus = ldcp->ldc_status;
3544 	if (ldc_status(ldcp->ldc_handle, &ldcp->ldc_status) != 0) {
3545 		DERR(vswp, "%s: unable to get status", __func__);
3546 		mutex_exit(&ldcp->status_lock);
3547 		LDC_EXIT_LOCK(ldcp);
3548 		return (1);
3549 
3550 	}
3551 	mutex_exit(&ldcp->status_lock);
3552 	LDC_EXIT_LOCK(ldcp);
3553 
3554 	if ((istatus != LDC_UP) && (ldcp->ldc_status == LDC_UP)) {
3555 		D2(vswp, "%s: channel %ld now UP (%ld)", __func__,
3556 			ldcp->ldc_id, istatus);
3557 		vsw_restart_handshake(ldcp);
3558 	}
3559 
3560 	D1(vswp, "%s: exit", __func__);
3561 	return (0);
3562 }
3563 
3564 /* disable callbacks on the channel */
3565 static int
3566 vsw_ldc_uninit(vsw_ldc_t *ldcp)
3567 {
3568 	vsw_t	*vswp = ldcp->ldc_vswp;
3569 	int	rv;
3570 
3571 	D1(vswp, "vsw_ldc_uninit: enter: id(%lx)\n", ldcp->ldc_id);
3572 
3573 	LDC_ENTER_LOCK(ldcp);
3574 
3575 	rv = ldc_set_cb_mode(ldcp->ldc_handle, LDC_CB_DISABLE);
3576 	if (rv != 0) {
3577 		DERR(vswp, "vsw_ldc_uninit(%lld): error disabling "
3578 			"interrupts (rv = %d)\n", ldcp->ldc_id, rv);
3579 		LDC_EXIT_LOCK(ldcp);
3580 		return (1);
3581 	}
3582 
3583 	mutex_enter(&ldcp->status_lock);
3584 	ldcp->ldc_status = LDC_INIT;
3585 	mutex_exit(&ldcp->status_lock);
3586 
3587 	LDC_EXIT_LOCK(ldcp);
3588 
3589 	D1(vswp, "vsw_ldc_uninit: exit: id(%lx)", ldcp->ldc_id);
3590 
3591 	return (0);
3592 }
3593 
3594 static int
3595 vsw_init_ldcs(vsw_port_t *port)
3596 {
3597 	vsw_ldc_list_t	*ldcl = &port->p_ldclist;
3598 	vsw_ldc_t	*ldcp;
3599 
3600 	READ_ENTER(&ldcl->lockrw);
3601 	ldcp =  ldcl->head;
3602 	for (; ldcp  != NULL; ldcp = ldcp->ldc_next) {
3603 		(void) vsw_ldc_init(ldcp);
3604 	}
3605 	RW_EXIT(&ldcl->lockrw);
3606 
3607 	return (0);
3608 }
3609 
3610 static int
3611 vsw_uninit_ldcs(vsw_port_t *port)
3612 {
3613 	vsw_ldc_list_t	*ldcl = &port->p_ldclist;
3614 	vsw_ldc_t	*ldcp;
3615 
3616 	D1(NULL, "vsw_uninit_ldcs: enter\n");
3617 
3618 	READ_ENTER(&ldcl->lockrw);
3619 	ldcp =  ldcl->head;
3620 	for (; ldcp  != NULL; ldcp = ldcp->ldc_next) {
3621 		(void) vsw_ldc_uninit(ldcp);
3622 	}
3623 	RW_EXIT(&ldcl->lockrw);
3624 
3625 	D1(NULL, "vsw_uninit_ldcs: exit\n");
3626 
3627 	return (0);
3628 }
3629 
3630 /*
3631  * Wait until the callback(s) associated with the ldcs under the specified
3632  * port have completed.
3633  *
3634  * Prior to this function being invoked each channel under this port
3635  * should have been quiesced via ldc_set_cb_mode(DISABLE).
3636  *
3637  * A short explaination of what we are doing below..
3638  *
3639  * The simplest approach would be to have a reference counter in
3640  * the ldc structure which is increment/decremented by the callbacks as
3641  * they use the channel. The drain function could then simply disable any
3642  * further callbacks and do a cv_wait for the ref to hit zero. Unfortunately
3643  * there is a tiny window here - before the callback is able to get the lock
3644  * on the channel it is interrupted and this function gets to execute. It
3645  * sees that the ref count is zero and believes its free to delete the
3646  * associated data structures.
3647  *
3648  * We get around this by taking advantage of the fact that before the ldc
3649  * framework invokes a callback it sets a flag to indicate that there is a
3650  * callback active (or about to become active). If when we attempt to
3651  * unregister a callback when this active flag is set then the unregister
3652  * will fail with EWOULDBLOCK.
3653  *
3654  * If the unregister fails we do a cv_timedwait. We will either be signaled
3655  * by the callback as it is exiting (note we have to wait a short period to
3656  * allow the callback to return fully to the ldc framework and it to clear
3657  * the active flag), or by the timer expiring. In either case we again attempt
3658  * the unregister. We repeat this until we can succesfully unregister the
3659  * callback.
3660  *
3661  * The reason we use a cv_timedwait rather than a simple cv_wait is to catch
3662  * the case where the callback has finished but the ldc framework has not yet
3663  * cleared the active flag. In this case we would never get a cv_signal.
3664  */
3665 static int
3666 vsw_drain_ldcs(vsw_port_t *port)
3667 {
3668 	vsw_ldc_list_t	*ldcl = &port->p_ldclist;
3669 	vsw_ldc_t	*ldcp;
3670 	vsw_t		*vswp = port->p_vswp;
3671 
3672 	D1(vswp, "%s: enter", __func__);
3673 
3674 	READ_ENTER(&ldcl->lockrw);
3675 
3676 	ldcp = ldcl->head;
3677 
3678 	for (; ldcp  != NULL; ldcp = ldcp->ldc_next) {
3679 		/*
3680 		 * If we can unregister the channel callback then we
3681 		 * know that there is no callback either running or
3682 		 * scheduled to run for this channel so move on to next
3683 		 * channel in the list.
3684 		 */
3685 		mutex_enter(&ldcp->drain_cv_lock);
3686 
3687 		/* prompt active callbacks to quit */
3688 		ldcp->drain_state = VSW_LDC_DRAINING;
3689 
3690 		if ((ldc_unreg_callback(ldcp->ldc_handle)) == 0) {
3691 			D2(vswp, "%s: unreg callback for chan %ld", __func__,
3692 				ldcp->ldc_id);
3693 			mutex_exit(&ldcp->drain_cv_lock);
3694 			continue;
3695 		} else {
3696 			/*
3697 			 * If we end up here we know that either 1) a callback
3698 			 * is currently executing, 2) is about to start (i.e.
3699 			 * the ldc framework has set the active flag but
3700 			 * has not actually invoked the callback yet, or 3)
3701 			 * has finished and has returned to the ldc framework
3702 			 * but the ldc framework has not yet cleared the
3703 			 * active bit.
3704 			 *
3705 			 * Wait for it to finish.
3706 			 */
3707 			while (ldc_unreg_callback(ldcp->ldc_handle)
3708 								== EWOULDBLOCK)
3709 				(void) cv_timedwait(&ldcp->drain_cv,
3710 					&ldcp->drain_cv_lock, lbolt + hz);
3711 
3712 			mutex_exit(&ldcp->drain_cv_lock);
3713 			D2(vswp, "%s: unreg callback for chan %ld after "
3714 				"timeout", __func__, ldcp->ldc_id);
3715 		}
3716 	}
3717 	RW_EXIT(&ldcl->lockrw);
3718 
3719 	D1(vswp, "%s: exit", __func__);
3720 	return (0);
3721 }
3722 
3723 /*
3724  * Wait until all tasks which reference this port have completed.
3725  *
3726  * Prior to this function being invoked each channel under this port
3727  * should have been quiesced via ldc_set_cb_mode(DISABLE).
3728  */
3729 static int
3730 vsw_drain_port_taskq(vsw_port_t *port)
3731 {
3732 	vsw_t		*vswp = port->p_vswp;
3733 
3734 	D1(vswp, "%s: enter", __func__);
3735 
3736 	/*
3737 	 * Mark the port as in the process of being detached, and
3738 	 * dispatch a marker task to the queue so we know when all
3739 	 * relevant tasks have completed.
3740 	 */
3741 	mutex_enter(&port->state_lock);
3742 	port->state = VSW_PORT_DETACHING;
3743 
3744 	if ((vswp->taskq_p == NULL) ||
3745 		(ddi_taskq_dispatch(vswp->taskq_p, vsw_marker_task,
3746 			port, DDI_NOSLEEP) != DDI_SUCCESS)) {
3747 		DERR(vswp, "%s: unable to dispatch marker task",
3748 			__func__);
3749 		mutex_exit(&port->state_lock);
3750 		return (1);
3751 	}
3752 
3753 	/*
3754 	 * Wait for the marker task to finish.
3755 	 */
3756 	while (port->state != VSW_PORT_DETACHABLE)
3757 		cv_wait(&port->state_cv, &port->state_lock);
3758 
3759 	mutex_exit(&port->state_lock);
3760 
3761 	D1(vswp, "%s: exit", __func__);
3762 
3763 	return (0);
3764 }
3765 
3766 static void
3767 vsw_marker_task(void *arg)
3768 {
3769 	vsw_port_t	*port = arg;
3770 	vsw_t		*vswp = port->p_vswp;
3771 
3772 	D1(vswp, "%s: enter", __func__);
3773 
3774 	mutex_enter(&port->state_lock);
3775 
3776 	/*
3777 	 * No further tasks should be dispatched which reference
3778 	 * this port so ok to mark it as safe to detach.
3779 	 */
3780 	port->state = VSW_PORT_DETACHABLE;
3781 
3782 	cv_signal(&port->state_cv);
3783 
3784 	mutex_exit(&port->state_lock);
3785 
3786 	D1(vswp, "%s: exit", __func__);
3787 }
3788 
3789 static vsw_port_t *
3790 vsw_lookup_port(vsw_t *vswp, int p_instance)
3791 {
3792 	vsw_port_list_t *plist = &vswp->plist;
3793 	vsw_port_t	*port;
3794 
3795 	for (port = plist->head; port != NULL; port = port->p_next) {
3796 		if (port->p_instance == p_instance) {
3797 			D2(vswp, "vsw_lookup_port: found p_instance\n");
3798 			return (port);
3799 		}
3800 	}
3801 
3802 	return (NULL);
3803 }
3804 
3805 /*
3806  * Search for and remove the specified port from the port
3807  * list. Returns 0 if able to locate and remove port, otherwise
3808  * returns 1.
3809  */
3810 static int
3811 vsw_plist_del_node(vsw_t *vswp, vsw_port_t *port)
3812 {
3813 	vsw_port_list_t *plist = &vswp->plist;
3814 	vsw_port_t	*curr_p, *prev_p;
3815 
3816 	if (plist->head == NULL)
3817 		return (1);
3818 
3819 	curr_p = prev_p = plist->head;
3820 
3821 	while (curr_p != NULL) {
3822 		if (curr_p == port) {
3823 			if (prev_p == curr_p) {
3824 				plist->head = curr_p->p_next;
3825 			} else {
3826 				prev_p->p_next = curr_p->p_next;
3827 			}
3828 			plist->num_ports--;
3829 			break;
3830 		} else {
3831 			prev_p = curr_p;
3832 			curr_p = curr_p->p_next;
3833 		}
3834 	}
3835 	return (0);
3836 }
3837 
3838 /*
3839  * Interrupt handler for ldc messages.
3840  */
3841 static uint_t
3842 vsw_ldc_cb(uint64_t event, caddr_t arg)
3843 {
3844 	vsw_ldc_t	*ldcp = (vsw_ldc_t  *)arg;
3845 	vsw_t 		*vswp = ldcp->ldc_vswp;
3846 	ldc_status_t	lstatus;
3847 	int		rv;
3848 
3849 	D1(vswp, "%s: enter: ldcid (%lld)\n", __func__, ldcp->ldc_id);
3850 
3851 	mutex_enter(&ldcp->ldc_cblock);
3852 
3853 	if ((ldcp->ldc_status == LDC_INIT) || (ldcp->ldc_handle == NULL)) {
3854 		mutex_exit(&ldcp->ldc_cblock);
3855 		return (LDC_SUCCESS);
3856 	}
3857 
3858 	mutex_enter(&ldcp->status_lock);
3859 	lstatus = ldcp->ldc_status;
3860 	rv = ldc_status(ldcp->ldc_handle, &ldcp->ldc_status);
3861 	mutex_exit(&ldcp->status_lock);
3862 	if (rv != 0) {
3863 		cmn_err(CE_WARN, "!vsw%d: Unable to read channel state",
3864 			vswp->instance);
3865 		goto vsw_cb_exit;
3866 	}
3867 
3868 	if (event & LDC_EVT_UP) {
3869 		/*
3870 		 * Channel has come up, get the state and then start
3871 		 * the handshake.
3872 		 */
3873 		D2(vswp, "%s: id(%ld) event(%llx) UP: status(%ld)",
3874 			__func__, ldcp->ldc_id, event, lstatus);
3875 		D2(vswp, "%s: UP: old status %ld : cur status %ld",
3876 			__func__, lstatus, ldcp->ldc_status);
3877 		if ((ldcp->ldc_status != lstatus) &&
3878 					(ldcp->ldc_status == LDC_UP)) {
3879 				ldcp->reset_active = 0;
3880 				vsw_restart_handshake(ldcp);
3881 		}
3882 
3883 		ASSERT((event & (LDC_EVT_RESET | LDC_EVT_DOWN)) == 0);
3884 	}
3885 
3886 	if (event & LDC_EVT_READ) {
3887 		/*
3888 		 * Data available for reading.
3889 		 */
3890 		D2(vswp, "%s: id(ld) event(%llx) data READ",
3891 				__func__, ldcp->ldc_id, event);
3892 
3893 		vsw_process_pkt(ldcp);
3894 
3895 		ASSERT((event & (LDC_EVT_RESET | LDC_EVT_DOWN)) == 0);
3896 
3897 		goto vsw_cb_exit;
3898 	}
3899 
3900 	if (event & (LDC_EVT_DOWN | LDC_EVT_RESET)) {
3901 		D2(vswp, "%s: id(%ld) event(%llx) DOWN/RESET",
3902 					__func__, ldcp->ldc_id, event);
3903 
3904 		/* attempt to restart the connection */
3905 		vsw_restart_ldc(ldcp);
3906 
3907 		/*
3908 		 * vsw_restart_ldc() will attempt to bring the channel
3909 		 * back up. Check here to see if that succeeded.
3910 		 */
3911 		mutex_enter(&ldcp->status_lock);
3912 		lstatus = ldcp->ldc_status;
3913 		rv = ldc_status(ldcp->ldc_handle, &ldcp->ldc_status);
3914 		mutex_exit(&ldcp->status_lock);
3915 		if (rv != 0) {
3916 			DERR(vswp, "%s: unable to read status for channel %ld",
3917 				__func__, ldcp->ldc_id);
3918 			goto vsw_cb_exit;
3919 		}
3920 
3921 		D2(vswp, "%s: id(%ld) event(%llx) DOWN/RESET event:"
3922 			" old status %ld : cur status %ld", __func__,
3923 			ldcp->ldc_id, event, lstatus, ldcp->ldc_status);
3924 
3925 		/*
3926 		 * If channel was not previously UP then (re)start the
3927 		 * handshake.
3928 		 */
3929 		if ((ldcp->ldc_status == LDC_UP) && (lstatus != LDC_UP)) {
3930 			D2(vswp, "%s: channel %ld now UP, restarting "
3931 				"handshake", __func__, ldcp->ldc_id);
3932 			ldcp->reset_active = 0;
3933 			vsw_restart_handshake(ldcp);
3934 		}
3935 	}
3936 
3937 	/*
3938 	 * Catch either LDC_EVT_WRITE which we don't support or any
3939 	 * unknown event.
3940 	 */
3941 	if (event & ~(LDC_EVT_UP | LDC_EVT_RESET
3942 					| LDC_EVT_DOWN | LDC_EVT_READ)) {
3943 
3944 		DERR(vswp, "%s: id(%ld) Unexpected event=(%llx) status(%ld)",
3945 			__func__, ldcp->ldc_id, event, ldcp->ldc_status);
3946 	}
3947 
3948 vsw_cb_exit:
3949 	mutex_exit(&ldcp->ldc_cblock);
3950 
3951 	/*
3952 	 * Let the drain function know we are finishing if it
3953 	 * is waiting.
3954 	 */
3955 	mutex_enter(&ldcp->drain_cv_lock);
3956 	if (ldcp->drain_state == VSW_LDC_DRAINING)
3957 		cv_signal(&ldcp->drain_cv);
3958 	mutex_exit(&ldcp->drain_cv_lock);
3959 
3960 	return (LDC_SUCCESS);
3961 }
3962 
3963 /*
3964  * Restart the connection with our peer. Free any existing
3965  * data structures and then attempt to bring channel back
3966  * up.
3967  */
3968 static void
3969 vsw_restart_ldc(vsw_ldc_t *ldcp)
3970 {
3971 	int		rv;
3972 	vsw_t		*vswp = ldcp->ldc_vswp;
3973 	vsw_port_t	*port;
3974 	vsw_ldc_list_t	*ldcl;
3975 
3976 	D1(vswp, "%s: enter", __func__);
3977 
3978 	/*
3979 	 * Check if reset already in progress for this channel.
3980 	 */
3981 	if (ldstub((uint8_t *)&ldcp->reset_active))
3982 		return;
3983 
3984 	port = ldcp->ldc_port;
3985 	ldcl = &port->p_ldclist;
3986 
3987 	READ_ENTER(&ldcl->lockrw);
3988 
3989 	D2(vswp, "%s: in 0x%llx : out 0x%llx", __func__,
3990 		ldcp->lane_in.lstate, ldcp->lane_out.lstate);
3991 
3992 	vsw_free_lane_resources(ldcp, INBOUND);
3993 	vsw_free_lane_resources(ldcp, OUTBOUND);
3994 	RW_EXIT(&ldcl->lockrw);
3995 
3996 	ldcp->lane_in.lstate = 0;
3997 	ldcp->lane_out.lstate = 0;
3998 
3999 	/*
4000 	 * Remove parent port from any multicast groups
4001 	 * it may have registered with. Client must resend
4002 	 * multicast add command after handshake completes.
4003 	 */
4004 	(void) vsw_del_fdb(vswp, port);
4005 
4006 	vsw_del_mcst_port(port);
4007 
4008 	ldcp->peer_session = 0;
4009 	ldcp->session_status = 0;
4010 	ldcp->hcnt = 0;
4011 	ldcp->hphase = VSW_MILESTONE0;
4012 
4013 	rv = ldc_up(ldcp->ldc_handle);
4014 	if (rv != 0) {
4015 		/*
4016 		 * Not a fatal error for ldc_up() to fail, as peer
4017 		 * end point may simply not be ready yet.
4018 		 */
4019 		D2(vswp, "%s: ldc_up err id(%lld) rv(%d)", __func__,
4020 			ldcp->ldc_id, rv);
4021 	}
4022 
4023 	D1(vswp, "%s: exit", __func__);
4024 }
4025 
4026 /*
4027  * (Re)start a handshake with our peer by sending them
4028  * our version info.
4029  */
4030 static void
4031 vsw_restart_handshake(vsw_ldc_t *ldcp)
4032 {
4033 	vsw_t		*vswp = ldcp->ldc_vswp;
4034 
4035 	D1(vswp, "vsw_restart_handshake: enter");
4036 
4037 	if (ldcp->hphase != VSW_MILESTONE0) {
4038 		vsw_restart_ldc(ldcp);
4039 	}
4040 
4041 	/*
4042 	 * We now increment the transaction group id. This allows
4043 	 * us to identify and disard any tasks which are still pending
4044 	 * on the taskq and refer to the handshake session we are about
4045 	 * to restart. These stale messages no longer have any real
4046 	 * meaning.
4047 	 */
4048 	mutex_enter(&ldcp->hss_lock);
4049 	ldcp->hss_id++;
4050 	mutex_exit(&ldcp->hss_lock);
4051 
4052 	if (ldcp->hcnt++ > vsw_num_handshakes) {
4053 		cmn_err(CE_WARN, "!vsw%d: exceeded number of permitted "
4054 			"handshake attempts (%d) on channel %ld",
4055 			vswp->instance, ldcp->hcnt, ldcp->ldc_id);
4056 		return;
4057 	}
4058 
4059 	if ((vswp->taskq_p == NULL) ||
4060 		(ddi_taskq_dispatch(vswp->taskq_p, vsw_send_ver, ldcp,
4061 			DDI_NOSLEEP) != DDI_SUCCESS)) {
4062 		cmn_err(CE_WARN, "!vsw%d: Can't dispatch version handshake "
4063 			"task", vswp->instance);
4064 	}
4065 
4066 	D1(vswp, "vsw_restart_handshake: exit");
4067 }
4068 
4069 /*
4070  * Deal appropriately with a ECONNRESET event encountered in a ldc_*
4071  * call.
4072  */
4073 static void
4074 vsw_handle_reset(vsw_ldc_t *ldcp)
4075 {
4076 	vsw_t		*vswp = ldcp->ldc_vswp;
4077 	ldc_status_t	lstatus;
4078 
4079 	D1(vswp, "%s: enter", __func__);
4080 
4081 	mutex_enter(&ldcp->status_lock);
4082 	lstatus = ldcp->ldc_status;
4083 	if (ldc_status(ldcp->ldc_handle, &ldcp->ldc_status) != 0) {
4084 		DERR(vswp, "%s: unable to read status for channel %ld",
4085 			__func__, ldcp->ldc_id);
4086 		mutex_exit(&ldcp->status_lock);
4087 		return;
4088 	}
4089 	mutex_exit(&ldcp->status_lock);
4090 
4091 	/*
4092 	 * Check the channel's previous recorded state to
4093 	 * determine if this is the first ECONNRESET event
4094 	 * we've gotten for this particular channel (i.e. was
4095 	 * previously up but is no longer). If so, terminate
4096 	 * the channel.
4097 	 */
4098 	if ((ldcp->ldc_status != LDC_UP) && (lstatus == LDC_UP)) {
4099 		vsw_restart_ldc(ldcp);
4100 	}
4101 
4102 	/*
4103 	 * vsw_restart_ldc() will also attempt to bring channel
4104 	 * back up. Check here if that succeeds.
4105 	 */
4106 	mutex_enter(&ldcp->status_lock);
4107 	lstatus = ldcp->ldc_status;
4108 	if (ldc_status(ldcp->ldc_handle, &ldcp->ldc_status) != 0) {
4109 		DERR(vswp, "%s: unable to read status for channel %ld",
4110 			__func__, ldcp->ldc_id);
4111 		mutex_exit(&ldcp->status_lock);
4112 		return;
4113 	}
4114 	mutex_exit(&ldcp->status_lock);
4115 
4116 	/*
4117 	 * If channel is now up and no one else (i.e. the callback routine)
4118 	 * has dealt with it then we restart the handshake here.
4119 	 */
4120 	if ((lstatus != LDC_UP) && (ldcp->ldc_status == LDC_UP)) {
4121 		ldcp->reset_active = 0;
4122 		vsw_restart_handshake(ldcp);
4123 	}
4124 
4125 	D1(vswp, "%s: exit", __func__);
4126 }
4127 
4128 /*
4129  * returns 0 if legal for event signified by flag to have
4130  * occured at the time it did. Otherwise returns 1.
4131  */
4132 int
4133 vsw_check_flag(vsw_ldc_t *ldcp, int dir, uint64_t flag)
4134 {
4135 	vsw_t		*vswp = ldcp->ldc_vswp;
4136 	uint64_t	state;
4137 	uint64_t	phase;
4138 
4139 	if (dir == INBOUND)
4140 		state = ldcp->lane_in.lstate;
4141 	else
4142 		state = ldcp->lane_out.lstate;
4143 
4144 	phase = ldcp->hphase;
4145 
4146 	switch (flag) {
4147 	case VSW_VER_INFO_RECV:
4148 		if (phase > VSW_MILESTONE0) {
4149 			DERR(vswp, "vsw_check_flag (%d): VER_INFO_RECV"
4150 				" when in state %d\n", ldcp->ldc_id, phase);
4151 			vsw_restart_handshake(ldcp);
4152 			return (1);
4153 		}
4154 		break;
4155 
4156 	case VSW_VER_ACK_RECV:
4157 	case VSW_VER_NACK_RECV:
4158 		if (!(state & VSW_VER_INFO_SENT)) {
4159 			DERR(vswp, "vsw_check_flag (%d): spurious VER_ACK"
4160 				" or VER_NACK when in state %d\n",
4161 				ldcp->ldc_id, phase);
4162 			vsw_restart_handshake(ldcp);
4163 			return (1);
4164 		} else
4165 			state &= ~VSW_VER_INFO_SENT;
4166 		break;
4167 
4168 	case VSW_ATTR_INFO_RECV:
4169 		if ((phase < VSW_MILESTONE1) || (phase >= VSW_MILESTONE2)) {
4170 			DERR(vswp, "vsw_check_flag (%d): ATTR_INFO_RECV"
4171 				" when in state %d\n", ldcp->ldc_id, phase);
4172 			vsw_restart_handshake(ldcp);
4173 			return (1);
4174 		}
4175 		break;
4176 
4177 	case VSW_ATTR_ACK_RECV:
4178 	case VSW_ATTR_NACK_RECV:
4179 		if (!(state & VSW_ATTR_INFO_SENT)) {
4180 			DERR(vswp, "vsw_check_flag (%d): spurious ATTR_ACK"
4181 				" or ATTR_NACK when in state %d\n",
4182 				ldcp->ldc_id, phase);
4183 			vsw_restart_handshake(ldcp);
4184 			return (1);
4185 		} else
4186 			state &= ~VSW_ATTR_INFO_SENT;
4187 		break;
4188 
4189 	case VSW_DRING_INFO_RECV:
4190 		if (phase < VSW_MILESTONE1) {
4191 			DERR(vswp, "vsw_check_flag (%d): DRING_INFO_RECV"
4192 				" when in state %d\n", ldcp->ldc_id, phase);
4193 			vsw_restart_handshake(ldcp);
4194 			return (1);
4195 		}
4196 		break;
4197 
4198 	case VSW_DRING_ACK_RECV:
4199 	case VSW_DRING_NACK_RECV:
4200 		if (!(state & VSW_DRING_INFO_SENT)) {
4201 			DERR(vswp, "vsw_check_flag (%d): spurious DRING_ACK"
4202 				" or DRING_NACK when in state %d\n",
4203 				ldcp->ldc_id, phase);
4204 			vsw_restart_handshake(ldcp);
4205 			return (1);
4206 		} else
4207 			state &= ~VSW_DRING_INFO_SENT;
4208 		break;
4209 
4210 	case VSW_RDX_INFO_RECV:
4211 		if (phase < VSW_MILESTONE3) {
4212 			DERR(vswp, "vsw_check_flag (%d): RDX_INFO_RECV"
4213 				" when in state %d\n", ldcp->ldc_id, phase);
4214 			vsw_restart_handshake(ldcp);
4215 			return (1);
4216 		}
4217 		break;
4218 
4219 	case VSW_RDX_ACK_RECV:
4220 	case VSW_RDX_NACK_RECV:
4221 		if (!(state & VSW_RDX_INFO_SENT)) {
4222 			DERR(vswp, "vsw_check_flag (%d): spurious RDX_ACK"
4223 				" or RDX_NACK when in state %d\n",
4224 				ldcp->ldc_id, phase);
4225 			vsw_restart_handshake(ldcp);
4226 			return (1);
4227 		} else
4228 			state &= ~VSW_RDX_INFO_SENT;
4229 		break;
4230 
4231 	case VSW_MCST_INFO_RECV:
4232 		if (phase < VSW_MILESTONE3) {
4233 			DERR(vswp, "vsw_check_flag (%d): VSW_MCST_INFO_RECV"
4234 				" when in state %d\n", ldcp->ldc_id, phase);
4235 			vsw_restart_handshake(ldcp);
4236 			return (1);
4237 		}
4238 		break;
4239 
4240 	default:
4241 		DERR(vswp, "vsw_check_flag (%lld): unknown flag (%llx)",
4242 				ldcp->ldc_id, flag);
4243 		return (1);
4244 	}
4245 
4246 	if (dir == INBOUND)
4247 		ldcp->lane_in.lstate = state;
4248 	else
4249 		ldcp->lane_out.lstate = state;
4250 
4251 	D1(vswp, "vsw_check_flag (chan %lld): exit", ldcp->ldc_id);
4252 
4253 	return (0);
4254 }
4255 
4256 void
4257 vsw_next_milestone(vsw_ldc_t *ldcp)
4258 {
4259 	vsw_t		*vswp = ldcp->ldc_vswp;
4260 
4261 	D1(vswp, "%s (chan %lld): enter (phase %ld)", __func__,
4262 		ldcp->ldc_id, ldcp->hphase);
4263 
4264 	DUMP_FLAGS(ldcp->lane_in.lstate);
4265 	DUMP_FLAGS(ldcp->lane_out.lstate);
4266 
4267 	switch (ldcp->hphase) {
4268 
4269 	case VSW_MILESTONE0:
4270 		/*
4271 		 * If we haven't started to handshake with our peer,
4272 		 * start to do so now.
4273 		 */
4274 		if (ldcp->lane_out.lstate == 0) {
4275 			D2(vswp, "%s: (chan %lld) starting handshake "
4276 				"with peer", __func__, ldcp->ldc_id);
4277 			vsw_restart_handshake(ldcp);
4278 		}
4279 
4280 		/*
4281 		 * Only way to pass this milestone is to have successfully
4282 		 * negotiated version info.
4283 		 */
4284 		if ((ldcp->lane_in.lstate & VSW_VER_ACK_SENT) &&
4285 			(ldcp->lane_out.lstate & VSW_VER_ACK_RECV)) {
4286 
4287 			D2(vswp, "%s: (chan %lld) leaving milestone 0",
4288 				__func__, ldcp->ldc_id);
4289 
4290 			/*
4291 			 * Next milestone is passed when attribute
4292 			 * information has been successfully exchanged.
4293 			 */
4294 			ldcp->hphase = VSW_MILESTONE1;
4295 			vsw_send_attr(ldcp);
4296 
4297 		}
4298 		break;
4299 
4300 	case VSW_MILESTONE1:
4301 		/*
4302 		 * Only way to pass this milestone is to have successfully
4303 		 * negotiated attribute information.
4304 		 */
4305 		if (ldcp->lane_in.lstate & VSW_ATTR_ACK_SENT) {
4306 
4307 			ldcp->hphase = VSW_MILESTONE2;
4308 
4309 			/*
4310 			 * If the peer device has said it wishes to
4311 			 * use descriptor rings then we send it our ring
4312 			 * info, otherwise we just set up a private ring
4313 			 * which we use an internal buffer
4314 			 */
4315 			if (ldcp->lane_in.xfer_mode == VIO_DRING_MODE)
4316 				vsw_send_dring_info(ldcp);
4317 		}
4318 		break;
4319 
4320 	case VSW_MILESTONE2:
4321 		/*
4322 		 * If peer has indicated in its attribute message that
4323 		 * it wishes to use descriptor rings then the only way
4324 		 * to pass this milestone is for us to have received
4325 		 * valid dring info.
4326 		 *
4327 		 * If peer is not using descriptor rings then just fall
4328 		 * through.
4329 		 */
4330 		if ((ldcp->lane_in.xfer_mode == VIO_DRING_MODE) &&
4331 			(!(ldcp->lane_in.lstate & VSW_DRING_ACK_SENT)))
4332 			break;
4333 
4334 		D2(vswp, "%s: (chan %lld) leaving milestone 2",
4335 				__func__, ldcp->ldc_id);
4336 
4337 		ldcp->hphase = VSW_MILESTONE3;
4338 		vsw_send_rdx(ldcp);
4339 		break;
4340 
4341 	case VSW_MILESTONE3:
4342 		/*
4343 		 * Pass this milestone when all paramaters have been
4344 		 * successfully exchanged and RDX sent in both directions.
4345 		 *
4346 		 * Mark outbound lane as available to transmit data.
4347 		 */
4348 		if ((ldcp->lane_in.lstate & VSW_RDX_ACK_SENT) &&
4349 			(ldcp->lane_out.lstate & VSW_RDX_ACK_RECV)) {
4350 
4351 			D2(vswp, "%s: (chan %lld) leaving milestone 3",
4352 				__func__, ldcp->ldc_id);
4353 			D2(vswp, "%s: ** handshake complete (0x%llx : "
4354 				"0x%llx) **", __func__, ldcp->lane_in.lstate,
4355 				ldcp->lane_out.lstate);
4356 			ldcp->lane_out.lstate |= VSW_LANE_ACTIVE;
4357 			ldcp->hphase = VSW_MILESTONE4;
4358 			ldcp->hcnt = 0;
4359 			DISPLAY_STATE();
4360 		} else {
4361 			D2(vswp, "%s: still in milestone 3 (0x%llx :"
4362 				" 0x%llx", __func__, ldcp->lane_in.lstate,
4363 				ldcp->lane_out.lstate);
4364 		}
4365 		break;
4366 
4367 	case VSW_MILESTONE4:
4368 		D2(vswp, "%s: (chan %lld) in milestone 4", __func__,
4369 							ldcp->ldc_id);
4370 		break;
4371 
4372 	default:
4373 		DERR(vswp, "%s: (chan %lld) Unknown Phase %x", __func__,
4374 			ldcp->ldc_id, ldcp->hphase);
4375 	}
4376 
4377 	D1(vswp, "%s (chan %lld): exit (phase %ld)", __func__, ldcp->ldc_id,
4378 		ldcp->hphase);
4379 }
4380 
4381 /*
4382  * Check if major version is supported.
4383  *
4384  * Returns 0 if finds supported major number, and if necessary
4385  * adjusts the minor field.
4386  *
4387  * Returns 1 if can't match major number exactly. Sets mjor/minor
4388  * to next lowest support values, or to zero if no other values possible.
4389  */
4390 static int
4391 vsw_supported_version(vio_ver_msg_t *vp)
4392 {
4393 	int	i;
4394 
4395 	D1(NULL, "vsw_supported_version: enter");
4396 
4397 	for (i = 0; i < VSW_NUM_VER; i++) {
4398 		if (vsw_versions[i].ver_major == vp->ver_major) {
4399 			/*
4400 			 * Matching or lower major version found. Update
4401 			 * minor number if necessary.
4402 			 */
4403 			if (vp->ver_minor > vsw_versions[i].ver_minor) {
4404 				D2(NULL, "%s: adjusting minor value"
4405 					" from %d to %d", __func__,
4406 					vp->ver_minor,
4407 					vsw_versions[i].ver_minor);
4408 				vp->ver_minor = vsw_versions[i].ver_minor;
4409 			}
4410 
4411 			return (0);
4412 		}
4413 
4414 		if (vsw_versions[i].ver_major < vp->ver_major) {
4415 			if (vp->ver_minor > vsw_versions[i].ver_minor) {
4416 				D2(NULL, "%s: adjusting minor value"
4417 					" from %d to %d", __func__,
4418 					vp->ver_minor,
4419 					vsw_versions[i].ver_minor);
4420 				vp->ver_minor = vsw_versions[i].ver_minor;
4421 			}
4422 			return (1);
4423 		}
4424 	}
4425 
4426 	/* No match was possible, zero out fields */
4427 	vp->ver_major = 0;
4428 	vp->ver_minor = 0;
4429 
4430 	D1(NULL, "vsw_supported_version: exit");
4431 
4432 	return (1);
4433 }
4434 
4435 /*
4436  * Main routine for processing messages received over LDC.
4437  */
4438 static void
4439 vsw_process_pkt(void *arg)
4440 {
4441 	vsw_ldc_t	*ldcp = (vsw_ldc_t  *)arg;
4442 	vsw_t 		*vswp = ldcp->ldc_vswp;
4443 	size_t		msglen;
4444 	vio_msg_tag_t	tag;
4445 	def_msg_t	dmsg;
4446 	int 		rv = 0;
4447 
4448 
4449 	D1(vswp, "%s enter: ldcid (%lld)\n", __func__, ldcp->ldc_id);
4450 
4451 	/*
4452 	 * If channel is up read messages until channel is empty.
4453 	 */
4454 	do {
4455 		msglen = sizeof (dmsg);
4456 		rv = ldc_read(ldcp->ldc_handle, (caddr_t)&dmsg, &msglen);
4457 
4458 		if (rv != 0) {
4459 			DERR(vswp, "%s :ldc_read err id(%lld) rv(%d) "
4460 				"len(%d)\n", __func__, ldcp->ldc_id,
4461 							rv, msglen);
4462 		}
4463 
4464 		/* channel has been reset */
4465 		if (rv == ECONNRESET) {
4466 			vsw_handle_reset(ldcp);
4467 			break;
4468 		}
4469 
4470 		if (msglen == 0) {
4471 			D2(vswp, "%s: ldc_read id(%lld) NODATA", __func__,
4472 			ldcp->ldc_id);
4473 			break;
4474 		}
4475 
4476 		D2(vswp, "%s: ldc_read id(%lld): msglen(%d)", __func__,
4477 		    ldcp->ldc_id, msglen);
4478 
4479 		/*
4480 		 * Figure out what sort of packet we have gotten by
4481 		 * examining the msg tag, and then switch it appropriately.
4482 		 */
4483 		bcopy(&dmsg, &tag, sizeof (vio_msg_tag_t));
4484 
4485 		switch (tag.vio_msgtype) {
4486 		case VIO_TYPE_CTRL:
4487 			vsw_dispatch_ctrl_task(ldcp, &dmsg, tag);
4488 			break;
4489 		case VIO_TYPE_DATA:
4490 			vsw_process_data_pkt(ldcp, &dmsg, tag);
4491 			break;
4492 		case VIO_TYPE_ERR:
4493 			vsw_process_err_pkt(ldcp, &dmsg, tag);
4494 			break;
4495 		default:
4496 			DERR(vswp, "%s: Unknown tag(%lx) ", __func__,
4497 				"id(%lx)\n", tag.vio_msgtype, ldcp->ldc_id);
4498 			break;
4499 		}
4500 	} while (msglen);
4501 
4502 	D1(vswp, "%s exit: ldcid (%lld)\n", __func__, ldcp->ldc_id);
4503 }
4504 
4505 /*
4506  * Dispatch a task to process a VIO control message.
4507  */
4508 static void
4509 vsw_dispatch_ctrl_task(vsw_ldc_t *ldcp, void *cpkt, vio_msg_tag_t tag)
4510 {
4511 	vsw_ctrl_task_t		*ctaskp = NULL;
4512 	vsw_port_t		*port = ldcp->ldc_port;
4513 	vsw_t			*vswp = port->p_vswp;
4514 
4515 	D1(vswp, "%s: enter", __func__);
4516 
4517 	/*
4518 	 * We need to handle RDX ACK messages in-band as once they
4519 	 * are exchanged it is possible that we will get an
4520 	 * immediate (legitimate) data packet.
4521 	 */
4522 	if ((tag.vio_subtype_env == VIO_RDX) &&
4523 		(tag.vio_subtype == VIO_SUBTYPE_ACK)) {
4524 
4525 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_RDX_ACK_RECV))
4526 			return;
4527 
4528 		ldcp->lane_out.lstate |= VSW_RDX_ACK_RECV;
4529 		D2(vswp, "%s (%ld) handling RDX_ACK in place "
4530 			"(ostate 0x%llx : hphase %d)", __func__,
4531 			ldcp->ldc_id, ldcp->lane_out.lstate, ldcp->hphase);
4532 		vsw_next_milestone(ldcp);
4533 		return;
4534 	}
4535 
4536 	ctaskp = kmem_alloc(sizeof (vsw_ctrl_task_t), KM_NOSLEEP);
4537 
4538 	if (ctaskp == NULL) {
4539 		DERR(vswp, "%s: unable to alloc space for ctrl"
4540 			" msg", __func__);
4541 		vsw_restart_handshake(ldcp);
4542 		return;
4543 	}
4544 
4545 	ctaskp->ldcp = ldcp;
4546 	bcopy((def_msg_t *)cpkt, &ctaskp->pktp, sizeof (def_msg_t));
4547 	mutex_enter(&ldcp->hss_lock);
4548 	ctaskp->hss_id = ldcp->hss_id;
4549 	mutex_exit(&ldcp->hss_lock);
4550 
4551 	/*
4552 	 * Dispatch task to processing taskq if port is not in
4553 	 * the process of being detached.
4554 	 */
4555 	mutex_enter(&port->state_lock);
4556 	if (port->state == VSW_PORT_INIT) {
4557 		if ((vswp->taskq_p == NULL) ||
4558 			(ddi_taskq_dispatch(vswp->taskq_p,
4559 			vsw_process_ctrl_pkt, ctaskp, DDI_NOSLEEP)
4560 							!= DDI_SUCCESS)) {
4561 			DERR(vswp, "%s: unable to dispatch task to taskq",
4562 				__func__);
4563 			kmem_free(ctaskp, sizeof (vsw_ctrl_task_t));
4564 			mutex_exit(&port->state_lock);
4565 			vsw_restart_handshake(ldcp);
4566 			return;
4567 		}
4568 	} else {
4569 		DWARN(vswp, "%s: port %d detaching, not dispatching "
4570 			"task", __func__, port->p_instance);
4571 	}
4572 
4573 	mutex_exit(&port->state_lock);
4574 
4575 	D2(vswp, "%s: dispatched task to taskq for chan %d", __func__,
4576 			ldcp->ldc_id);
4577 	D1(vswp, "%s: exit", __func__);
4578 }
4579 
4580 /*
4581  * Process a VIO ctrl message. Invoked from taskq.
4582  */
4583 static void
4584 vsw_process_ctrl_pkt(void *arg)
4585 {
4586 	vsw_ctrl_task_t	*ctaskp = (vsw_ctrl_task_t *)arg;
4587 	vsw_ldc_t	*ldcp = ctaskp->ldcp;
4588 	vsw_t 		*vswp = ldcp->ldc_vswp;
4589 	vio_msg_tag_t	tag;
4590 	uint16_t	env;
4591 
4592 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
4593 
4594 	bcopy(&ctaskp->pktp, &tag, sizeof (vio_msg_tag_t));
4595 	env = tag.vio_subtype_env;
4596 
4597 	/* stale pkt check */
4598 	mutex_enter(&ldcp->hss_lock);
4599 	if (ctaskp->hss_id < ldcp->hss_id) {
4600 		DWARN(vswp, "%s: discarding stale packet belonging to"
4601 			" earlier (%ld) handshake session", __func__,
4602 			ctaskp->hss_id);
4603 		mutex_exit(&ldcp->hss_lock);
4604 		return;
4605 	}
4606 	mutex_exit(&ldcp->hss_lock);
4607 
4608 	/* session id check */
4609 	if (ldcp->session_status & VSW_PEER_SESSION) {
4610 		if (ldcp->peer_session != tag.vio_sid) {
4611 			DERR(vswp, "%s (chan %d): invalid session id (%llx)",
4612 				__func__, ldcp->ldc_id, tag.vio_sid);
4613 			kmem_free(ctaskp, sizeof (vsw_ctrl_task_t));
4614 			vsw_restart_handshake(ldcp);
4615 			return;
4616 		}
4617 	}
4618 
4619 	/*
4620 	 * Switch on vio_subtype envelope, then let lower routines
4621 	 * decide if its an INFO, ACK or NACK packet.
4622 	 */
4623 	switch (env) {
4624 	case VIO_VER_INFO:
4625 		vsw_process_ctrl_ver_pkt(ldcp, &ctaskp->pktp);
4626 		break;
4627 	case VIO_DRING_REG:
4628 		vsw_process_ctrl_dring_reg_pkt(ldcp, &ctaskp->pktp);
4629 		break;
4630 	case VIO_DRING_UNREG:
4631 		vsw_process_ctrl_dring_unreg_pkt(ldcp, &ctaskp->pktp);
4632 		break;
4633 	case VIO_ATTR_INFO:
4634 		vsw_process_ctrl_attr_pkt(ldcp, &ctaskp->pktp);
4635 		break;
4636 	case VNET_MCAST_INFO:
4637 		vsw_process_ctrl_mcst_pkt(ldcp, &ctaskp->pktp);
4638 		break;
4639 	case VIO_RDX:
4640 		vsw_process_ctrl_rdx_pkt(ldcp, &ctaskp->pktp);
4641 		break;
4642 	default:
4643 		DERR(vswp, "%s : unknown vio_subtype_env (%x)\n",
4644 							__func__, env);
4645 	}
4646 
4647 	kmem_free(ctaskp, sizeof (vsw_ctrl_task_t));
4648 	D1(vswp, "%s(%lld): exit", __func__, ldcp->ldc_id);
4649 }
4650 
4651 /*
4652  * Version negotiation. We can end up here either because our peer
4653  * has responded to a handshake message we have sent it, or our peer
4654  * has initiated a handshake with us. If its the former then can only
4655  * be ACK or NACK, if its the later can only be INFO.
4656  *
4657  * If its an ACK we move to the next stage of the handshake, namely
4658  * attribute exchange. If its a NACK we see if we can specify another
4659  * version, if we can't we stop.
4660  *
4661  * If it is an INFO we reset all params associated with communication
4662  * in that direction over this channel (remember connection is
4663  * essentially 2 independent simplex channels).
4664  */
4665 void
4666 vsw_process_ctrl_ver_pkt(vsw_ldc_t *ldcp, void *pkt)
4667 {
4668 	vio_ver_msg_t	*ver_pkt;
4669 	vsw_t 		*vswp = ldcp->ldc_vswp;
4670 
4671 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
4672 
4673 	/*
4674 	 * We know this is a ctrl/version packet so
4675 	 * cast it into the correct structure.
4676 	 */
4677 	ver_pkt = (vio_ver_msg_t *)pkt;
4678 
4679 	switch (ver_pkt->tag.vio_subtype) {
4680 	case VIO_SUBTYPE_INFO:
4681 		D2(vswp, "vsw_process_ctrl_ver_pkt: VIO_SUBTYPE_INFO\n");
4682 
4683 		/*
4684 		 * Record the session id, which we will use from now
4685 		 * until we see another VER_INFO msg. Even then the
4686 		 * session id in most cases will be unchanged, execpt
4687 		 * if channel was reset.
4688 		 */
4689 		if ((ldcp->session_status & VSW_PEER_SESSION) &&
4690 			(ldcp->peer_session != ver_pkt->tag.vio_sid)) {
4691 			DERR(vswp, "%s: updating session id for chan %lld "
4692 				"from %llx to %llx", __func__, ldcp->ldc_id,
4693 				ldcp->peer_session, ver_pkt->tag.vio_sid);
4694 		}
4695 
4696 		ldcp->peer_session = ver_pkt->tag.vio_sid;
4697 		ldcp->session_status |= VSW_PEER_SESSION;
4698 
4699 		/* Legal message at this time ? */
4700 		if (vsw_check_flag(ldcp, INBOUND, VSW_VER_INFO_RECV))
4701 			return;
4702 
4703 		/*
4704 		 * First check the device class. Currently only expect
4705 		 * to be talking to a network device. In the future may
4706 		 * also talk to another switch.
4707 		 */
4708 		if (ver_pkt->dev_class != VDEV_NETWORK) {
4709 			DERR(vswp, "%s: illegal device class %d", __func__,
4710 				ver_pkt->dev_class);
4711 
4712 			ver_pkt->tag.vio_sid = ldcp->local_session;
4713 			ver_pkt->tag.vio_subtype = VIO_SUBTYPE_NACK;
4714 
4715 			DUMP_TAG_PTR((vio_msg_tag_t *)ver_pkt);
4716 
4717 			vsw_send_msg(ldcp, (void *)ver_pkt,
4718 					sizeof (vio_ver_msg_t));
4719 
4720 			ldcp->lane_in.lstate |= VSW_VER_NACK_SENT;
4721 			vsw_next_milestone(ldcp);
4722 			return;
4723 		} else {
4724 			ldcp->dev_class = ver_pkt->dev_class;
4725 		}
4726 
4727 		/*
4728 		 * Now check the version.
4729 		 */
4730 		if (vsw_supported_version(ver_pkt) == 0) {
4731 			/*
4732 			 * Support this major version and possibly
4733 			 * adjusted minor version.
4734 			 */
4735 
4736 			D2(vswp, "%s: accepted ver %d:%d", __func__,
4737 				ver_pkt->ver_major, ver_pkt->ver_minor);
4738 
4739 			/* Store accepted values */
4740 			ldcp->lane_in.ver_major = ver_pkt->ver_major;
4741 			ldcp->lane_in.ver_minor = ver_pkt->ver_minor;
4742 
4743 			ver_pkt->tag.vio_subtype = VIO_SUBTYPE_ACK;
4744 
4745 			ldcp->lane_in.lstate |= VSW_VER_ACK_SENT;
4746 		} else {
4747 			/*
4748 			 * NACK back with the next lower major/minor
4749 			 * pairing we support (if don't suuport any more
4750 			 * versions then they will be set to zero.
4751 			 */
4752 
4753 			D2(vswp, "%s: replying with ver %d:%d", __func__,
4754 				ver_pkt->ver_major, ver_pkt->ver_minor);
4755 
4756 			/* Store updated values */
4757 			ldcp->lane_in.ver_major = ver_pkt->ver_major;
4758 			ldcp->lane_in.ver_minor = ver_pkt->ver_minor;
4759 
4760 			ver_pkt->tag.vio_subtype = VIO_SUBTYPE_NACK;
4761 
4762 			ldcp->lane_in.lstate |= VSW_VER_NACK_SENT;
4763 		}
4764 
4765 		DUMP_TAG_PTR((vio_msg_tag_t *)ver_pkt);
4766 		ver_pkt->tag.vio_sid = ldcp->local_session;
4767 		vsw_send_msg(ldcp, (void *)ver_pkt, sizeof (vio_ver_msg_t));
4768 
4769 		vsw_next_milestone(ldcp);
4770 		break;
4771 
4772 	case VIO_SUBTYPE_ACK:
4773 		D2(vswp, "%s: VIO_SUBTYPE_ACK\n", __func__);
4774 
4775 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_VER_ACK_RECV))
4776 			return;
4777 
4778 		/* Store updated values */
4779 		ldcp->lane_in.ver_major = ver_pkt->ver_major;
4780 		ldcp->lane_in.ver_minor = ver_pkt->ver_minor;
4781 
4782 
4783 		ldcp->lane_out.lstate |= VSW_VER_ACK_RECV;
4784 		vsw_next_milestone(ldcp);
4785 
4786 		break;
4787 
4788 	case VIO_SUBTYPE_NACK:
4789 		D2(vswp, "%s: VIO_SUBTYPE_NACK\n", __func__);
4790 
4791 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_VER_NACK_RECV))
4792 			return;
4793 
4794 		/*
4795 		 * If our peer sent us a NACK with the ver fields set to
4796 		 * zero then there is nothing more we can do. Otherwise see
4797 		 * if we support either the version suggested, or a lesser
4798 		 * one.
4799 		 */
4800 		if ((ver_pkt->ver_major == 0) && (ver_pkt->ver_minor == 0)) {
4801 			DERR(vswp, "%s: peer unable to negotiate any "
4802 				"further.", __func__);
4803 			ldcp->lane_out.lstate |= VSW_VER_NACK_RECV;
4804 			vsw_next_milestone(ldcp);
4805 			return;
4806 		}
4807 
4808 		/*
4809 		 * Check to see if we support this major version or
4810 		 * a lower one. If we don't then maj/min will be set
4811 		 * to zero.
4812 		 */
4813 		(void) vsw_supported_version(ver_pkt);
4814 		if ((ver_pkt->ver_major == 0) && (ver_pkt->ver_minor == 0)) {
4815 			/* Nothing more we can do */
4816 			DERR(vswp, "%s: version negotiation failed.\n",
4817 								__func__);
4818 			ldcp->lane_out.lstate |= VSW_VER_NACK_RECV;
4819 			vsw_next_milestone(ldcp);
4820 		} else {
4821 			/* found a supported major version */
4822 			ldcp->lane_out.ver_major = ver_pkt->ver_major;
4823 			ldcp->lane_out.ver_minor = ver_pkt->ver_minor;
4824 
4825 			D2(vswp, "%s: resending with updated values (%x, %x)",
4826 				__func__, ver_pkt->ver_major,
4827 				ver_pkt->ver_minor);
4828 
4829 			ldcp->lane_out.lstate |= VSW_VER_INFO_SENT;
4830 			ver_pkt->tag.vio_sid = ldcp->local_session;
4831 			ver_pkt->tag.vio_subtype = VIO_SUBTYPE_INFO;
4832 
4833 			DUMP_TAG_PTR((vio_msg_tag_t *)ver_pkt);
4834 
4835 			vsw_send_msg(ldcp, (void *)ver_pkt,
4836 					sizeof (vio_ver_msg_t));
4837 
4838 			vsw_next_milestone(ldcp);
4839 
4840 		}
4841 		break;
4842 
4843 	default:
4844 		DERR(vswp, "%s: unknown vio_subtype %x\n", __func__,
4845 			ver_pkt->tag.vio_subtype);
4846 	}
4847 
4848 	D1(vswp, "%s(%lld): exit\n", __func__, ldcp->ldc_id);
4849 }
4850 
4851 /*
4852  * Process an attribute packet. We can end up here either because our peer
4853  * has ACK/NACK'ed back to an earlier ATTR msg we had sent it, or our
4854  * peer has sent us an attribute INFO message
4855  *
4856  * If its an ACK we then move to the next stage of the handshake which
4857  * is to send our descriptor ring info to our peer. If its a NACK then
4858  * there is nothing more we can (currently) do.
4859  *
4860  * If we get a valid/acceptable INFO packet (and we have already negotiated
4861  * a version) we ACK back and set channel state to ATTR_RECV, otherwise we
4862  * NACK back and reset channel state to INACTIV.
4863  *
4864  * FUTURE: in time we will probably negotiate over attributes, but for
4865  * the moment unacceptable attributes are regarded as a fatal error.
4866  *
4867  */
4868 void
4869 vsw_process_ctrl_attr_pkt(vsw_ldc_t *ldcp, void *pkt)
4870 {
4871 	vnet_attr_msg_t		*attr_pkt;
4872 	vsw_t			*vswp = ldcp->ldc_vswp;
4873 	vsw_port_t		*port = ldcp->ldc_port;
4874 	uint64_t		macaddr = 0;
4875 	int			i;
4876 
4877 	D1(vswp, "%s(%lld) enter", __func__, ldcp->ldc_id);
4878 
4879 	/*
4880 	 * We know this is a ctrl/attr packet so
4881 	 * cast it into the correct structure.
4882 	 */
4883 	attr_pkt = (vnet_attr_msg_t *)pkt;
4884 
4885 	switch (attr_pkt->tag.vio_subtype) {
4886 	case VIO_SUBTYPE_INFO:
4887 		D2(vswp, "%s: VIO_SUBTYPE_INFO", __func__);
4888 
4889 		if (vsw_check_flag(ldcp, INBOUND, VSW_ATTR_INFO_RECV))
4890 			return;
4891 
4892 		/*
4893 		 * If the attributes are unacceptable then we NACK back.
4894 		 */
4895 		if (vsw_check_attr(attr_pkt, ldcp->ldc_port)) {
4896 
4897 			DERR(vswp, "%s (chan %d): invalid attributes",
4898 				__func__, ldcp->ldc_id);
4899 
4900 			vsw_free_lane_resources(ldcp, INBOUND);
4901 
4902 			attr_pkt->tag.vio_sid = ldcp->local_session;
4903 			attr_pkt->tag.vio_subtype = VIO_SUBTYPE_NACK;
4904 
4905 			DUMP_TAG_PTR((vio_msg_tag_t *)attr_pkt);
4906 			ldcp->lane_in.lstate |= VSW_ATTR_NACK_SENT;
4907 			vsw_send_msg(ldcp, (void *)attr_pkt,
4908 					sizeof (vnet_attr_msg_t));
4909 
4910 			vsw_next_milestone(ldcp);
4911 			return;
4912 		}
4913 
4914 		/*
4915 		 * Otherwise store attributes for this lane and update
4916 		 * lane state.
4917 		 */
4918 		ldcp->lane_in.mtu = attr_pkt->mtu;
4919 		ldcp->lane_in.addr = attr_pkt->addr;
4920 		ldcp->lane_in.addr_type = attr_pkt->addr_type;
4921 		ldcp->lane_in.xfer_mode = attr_pkt->xfer_mode;
4922 		ldcp->lane_in.ack_freq = attr_pkt->ack_freq;
4923 
4924 		macaddr = ldcp->lane_in.addr;
4925 		for (i = ETHERADDRL - 1; i >= 0; i--) {
4926 			port->p_macaddr.ether_addr_octet[i] = macaddr & 0xFF;
4927 			macaddr >>= 8;
4928 		}
4929 
4930 		/* create the fdb entry for this port/mac address */
4931 		(void) vsw_add_fdb(vswp, port);
4932 
4933 		/* setup device specifc xmit routines */
4934 		mutex_enter(&port->tx_lock);
4935 		if (ldcp->lane_in.xfer_mode == VIO_DRING_MODE) {
4936 			D2(vswp, "%s: mode = VIO_DRING_MODE", __func__);
4937 			port->transmit = vsw_dringsend;
4938 		} else if (ldcp->lane_in.xfer_mode == VIO_DESC_MODE) {
4939 			D2(vswp, "%s: mode = VIO_DESC_MODE", __func__);
4940 			vsw_create_privring(ldcp);
4941 			port->transmit = vsw_descrsend;
4942 		}
4943 		mutex_exit(&port->tx_lock);
4944 
4945 		attr_pkt->tag.vio_sid = ldcp->local_session;
4946 		attr_pkt->tag.vio_subtype = VIO_SUBTYPE_ACK;
4947 
4948 		DUMP_TAG_PTR((vio_msg_tag_t *)attr_pkt);
4949 
4950 		ldcp->lane_in.lstate |= VSW_ATTR_ACK_SENT;
4951 
4952 		vsw_send_msg(ldcp, (void *)attr_pkt,
4953 					sizeof (vnet_attr_msg_t));
4954 
4955 		vsw_next_milestone(ldcp);
4956 		break;
4957 
4958 	case VIO_SUBTYPE_ACK:
4959 		D2(vswp, "%s: VIO_SUBTYPE_ACK", __func__);
4960 
4961 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_ATTR_ACK_RECV))
4962 			return;
4963 
4964 		ldcp->lane_out.lstate |= VSW_ATTR_ACK_RECV;
4965 		vsw_next_milestone(ldcp);
4966 		break;
4967 
4968 	case VIO_SUBTYPE_NACK:
4969 		D2(vswp, "%s: VIO_SUBTYPE_NACK", __func__);
4970 
4971 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_ATTR_NACK_RECV))
4972 			return;
4973 
4974 		ldcp->lane_out.lstate |= VSW_ATTR_NACK_RECV;
4975 		vsw_next_milestone(ldcp);
4976 		break;
4977 
4978 	default:
4979 		DERR(vswp, "%s: unknown vio_subtype %x\n", __func__,
4980 			attr_pkt->tag.vio_subtype);
4981 	}
4982 
4983 	D1(vswp, "%s(%lld) exit", __func__, ldcp->ldc_id);
4984 }
4985 
4986 /*
4987  * Process a dring info packet. We can end up here either because our peer
4988  * has ACK/NACK'ed back to an earlier DRING msg we had sent it, or our
4989  * peer has sent us a dring INFO message.
4990  *
4991  * If we get a valid/acceptable INFO packet (and we have already negotiated
4992  * a version) we ACK back and update the lane state, otherwise we NACK back.
4993  *
4994  * FUTURE: nothing to stop client from sending us info on multiple dring's
4995  * but for the moment we will just use the first one we are given.
4996  *
4997  */
4998 void
4999 vsw_process_ctrl_dring_reg_pkt(vsw_ldc_t *ldcp, void *pkt)
5000 {
5001 	vio_dring_reg_msg_t	*dring_pkt;
5002 	vsw_t			*vswp = ldcp->ldc_vswp;
5003 	ldc_mem_info_t		minfo;
5004 	dring_info_t		*dp, *dbp;
5005 	int			dring_found = 0;
5006 
5007 	/*
5008 	 * We know this is a ctrl/dring packet so
5009 	 * cast it into the correct structure.
5010 	 */
5011 	dring_pkt = (vio_dring_reg_msg_t *)pkt;
5012 
5013 	D1(vswp, "%s(%lld) enter", __func__, ldcp->ldc_id);
5014 
5015 	switch (dring_pkt->tag.vio_subtype) {
5016 	case VIO_SUBTYPE_INFO:
5017 		D2(vswp, "%s: VIO_SUBTYPE_INFO", __func__);
5018 
5019 		if (vsw_check_flag(ldcp, INBOUND, VSW_DRING_INFO_RECV))
5020 			return;
5021 
5022 		/*
5023 		 * If the dring params are unacceptable then we NACK back.
5024 		 */
5025 		if (vsw_check_dring_info(dring_pkt)) {
5026 
5027 			DERR(vswp, "%s (%lld): invalid dring info",
5028 				__func__, ldcp->ldc_id);
5029 
5030 			vsw_free_lane_resources(ldcp, INBOUND);
5031 
5032 			dring_pkt->tag.vio_sid = ldcp->local_session;
5033 			dring_pkt->tag.vio_subtype = VIO_SUBTYPE_NACK;
5034 
5035 			DUMP_TAG_PTR((vio_msg_tag_t *)dring_pkt);
5036 
5037 			ldcp->lane_in.lstate |= VSW_DRING_NACK_SENT;
5038 
5039 			vsw_send_msg(ldcp, (void *)dring_pkt,
5040 					sizeof (vio_dring_reg_msg_t));
5041 
5042 			vsw_next_milestone(ldcp);
5043 			return;
5044 		}
5045 
5046 		/*
5047 		 * Otherwise, attempt to map in the dring using the
5048 		 * cookie. If that succeeds we send back a unique dring
5049 		 * identifier that the sending side will use in future
5050 		 * to refer to this descriptor ring.
5051 		 */
5052 		dp = kmem_zalloc(sizeof (dring_info_t), KM_SLEEP);
5053 
5054 		dp->num_descriptors = dring_pkt->num_descriptors;
5055 		dp->descriptor_size = dring_pkt->descriptor_size;
5056 		dp->options = dring_pkt->options;
5057 		dp->ncookies = dring_pkt->ncookies;
5058 
5059 		/*
5060 		 * Note: should only get one cookie. Enforced in
5061 		 * the ldc layer.
5062 		 */
5063 		bcopy(&dring_pkt->cookie[0], &dp->cookie[0],
5064 			sizeof (ldc_mem_cookie_t));
5065 
5066 		D2(vswp, "%s: num_desc %ld : desc_size %ld", __func__,
5067 			dp->num_descriptors, dp->descriptor_size);
5068 		D2(vswp, "%s: options 0x%lx: ncookies %ld", __func__,
5069 			dp->options, dp->ncookies);
5070 
5071 		if ((ldc_mem_dring_map(ldcp->ldc_handle, &dp->cookie[0],
5072 			dp->ncookies, dp->num_descriptors,
5073 			dp->descriptor_size, LDC_SHADOW_MAP,
5074 			&(dp->handle))) != 0) {
5075 
5076 			DERR(vswp, "%s: dring_map failed\n", __func__);
5077 
5078 			kmem_free(dp, sizeof (dring_info_t));
5079 			vsw_free_lane_resources(ldcp, INBOUND);
5080 
5081 			dring_pkt->tag.vio_sid = ldcp->local_session;
5082 			dring_pkt->tag.vio_subtype = VIO_SUBTYPE_NACK;
5083 
5084 			DUMP_TAG_PTR((vio_msg_tag_t *)dring_pkt);
5085 
5086 			ldcp->lane_in.lstate |= VSW_DRING_NACK_SENT;
5087 			vsw_send_msg(ldcp, (void *)dring_pkt,
5088 				sizeof (vio_dring_reg_msg_t));
5089 
5090 			vsw_next_milestone(ldcp);
5091 			return;
5092 		}
5093 
5094 		if ((ldc_mem_dring_info(dp->handle, &minfo)) != 0) {
5095 
5096 			DERR(vswp, "%s: dring_addr failed\n", __func__);
5097 
5098 			kmem_free(dp, sizeof (dring_info_t));
5099 			vsw_free_lane_resources(ldcp, INBOUND);
5100 
5101 			dring_pkt->tag.vio_sid = ldcp->local_session;
5102 			dring_pkt->tag.vio_subtype = VIO_SUBTYPE_NACK;
5103 
5104 			DUMP_TAG_PTR((vio_msg_tag_t *)dring_pkt);
5105 
5106 			ldcp->lane_in.lstate |= VSW_DRING_NACK_SENT;
5107 			vsw_send_msg(ldcp, (void *)dring_pkt,
5108 				sizeof (vio_dring_reg_msg_t));
5109 
5110 			vsw_next_milestone(ldcp);
5111 			return;
5112 		} else {
5113 			/* store the address of the pub part of ring */
5114 			dp->pub_addr = minfo.vaddr;
5115 		}
5116 
5117 		/* no private section as we are importing */
5118 		dp->priv_addr = NULL;
5119 
5120 		/*
5121 		 * Using simple mono increasing int for ident at
5122 		 * the moment.
5123 		 */
5124 		dp->ident = ldcp->next_ident;
5125 		ldcp->next_ident++;
5126 
5127 		dp->end_idx = 0;
5128 		dp->next = NULL;
5129 
5130 		/*
5131 		 * Link it onto the end of the list of drings
5132 		 * for this lane.
5133 		 */
5134 		if (ldcp->lane_in.dringp == NULL) {
5135 			D2(vswp, "%s: adding first INBOUND dring", __func__);
5136 			ldcp->lane_in.dringp = dp;
5137 		} else {
5138 			dbp = ldcp->lane_in.dringp;
5139 
5140 			while (dbp->next != NULL)
5141 				dbp = dbp->next;
5142 
5143 			dbp->next = dp;
5144 		}
5145 
5146 		/* acknowledge it */
5147 		dring_pkt->tag.vio_sid = ldcp->local_session;
5148 		dring_pkt->tag.vio_subtype = VIO_SUBTYPE_ACK;
5149 		dring_pkt->dring_ident = dp->ident;
5150 
5151 		vsw_send_msg(ldcp, (void *)dring_pkt,
5152 				sizeof (vio_dring_reg_msg_t));
5153 
5154 		ldcp->lane_in.lstate |= VSW_DRING_ACK_SENT;
5155 		vsw_next_milestone(ldcp);
5156 		break;
5157 
5158 	case VIO_SUBTYPE_ACK:
5159 		D2(vswp, "%s: VIO_SUBTYPE_ACK", __func__);
5160 
5161 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_DRING_ACK_RECV))
5162 			return;
5163 
5164 		/*
5165 		 * Peer is acknowledging our dring info and will have
5166 		 * sent us a dring identifier which we will use to
5167 		 * refer to this ring w.r.t. our peer.
5168 		 */
5169 		dp = ldcp->lane_out.dringp;
5170 		if (dp != NULL) {
5171 			/*
5172 			 * Find the ring this ident should be associated
5173 			 * with.
5174 			 */
5175 			if (vsw_dring_match(dp, dring_pkt)) {
5176 				dring_found = 1;
5177 
5178 			} else while (dp != NULL) {
5179 				if (vsw_dring_match(dp, dring_pkt)) {
5180 					dring_found = 1;
5181 					break;
5182 				}
5183 				dp = dp->next;
5184 			}
5185 
5186 			if (dring_found == 0) {
5187 				DERR(NULL, "%s: unrecognised ring cookie",
5188 					__func__);
5189 				vsw_restart_handshake(ldcp);
5190 				return;
5191 			}
5192 
5193 		} else {
5194 			DERR(vswp, "%s: DRING ACK received but no drings "
5195 				"allocated", __func__);
5196 			vsw_restart_handshake(ldcp);
5197 			return;
5198 		}
5199 
5200 		/* store ident */
5201 		dp->ident = dring_pkt->dring_ident;
5202 		ldcp->lane_out.lstate |= VSW_DRING_ACK_RECV;
5203 		vsw_next_milestone(ldcp);
5204 		break;
5205 
5206 	case VIO_SUBTYPE_NACK:
5207 		D2(vswp, "%s: VIO_SUBTYPE_NACK", __func__);
5208 
5209 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_DRING_NACK_RECV))
5210 			return;
5211 
5212 		ldcp->lane_out.lstate |= VSW_DRING_NACK_RECV;
5213 		vsw_next_milestone(ldcp);
5214 		break;
5215 
5216 	default:
5217 		DERR(vswp, "%s: Unknown vio_subtype %x\n", __func__,
5218 			dring_pkt->tag.vio_subtype);
5219 	}
5220 
5221 	D1(vswp, "%s(%lld) exit", __func__, ldcp->ldc_id);
5222 }
5223 
5224 /*
5225  * Process a request from peer to unregister a dring.
5226  *
5227  * For the moment we just restart the handshake if our
5228  * peer endpoint attempts to unregister a dring.
5229  */
5230 void
5231 vsw_process_ctrl_dring_unreg_pkt(vsw_ldc_t *ldcp, void *pkt)
5232 {
5233 	vsw_t			*vswp = ldcp->ldc_vswp;
5234 	vio_dring_unreg_msg_t	*dring_pkt;
5235 
5236 	/*
5237 	 * We know this is a ctrl/dring packet so
5238 	 * cast it into the correct structure.
5239 	 */
5240 	dring_pkt = (vio_dring_unreg_msg_t *)pkt;
5241 
5242 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
5243 
5244 	switch (dring_pkt->tag.vio_subtype) {
5245 	case VIO_SUBTYPE_INFO:
5246 		D2(vswp, "%s: VIO_SUBTYPE_INFO", __func__);
5247 
5248 		DWARN(vswp, "%s: restarting handshake..", __func__);
5249 		vsw_restart_handshake(ldcp);
5250 		break;
5251 
5252 	case VIO_SUBTYPE_ACK:
5253 		D2(vswp, "%s: VIO_SUBTYPE_ACK", __func__);
5254 
5255 		DWARN(vswp, "%s: restarting handshake..", __func__);
5256 		vsw_restart_handshake(ldcp);
5257 		break;
5258 
5259 	case VIO_SUBTYPE_NACK:
5260 		D2(vswp, "%s: VIO_SUBTYPE_NACK", __func__);
5261 
5262 		DWARN(vswp, "%s: restarting handshake..", __func__);
5263 		vsw_restart_handshake(ldcp);
5264 		break;
5265 
5266 	default:
5267 		DERR(vswp, "%s: Unknown vio_subtype %x\n", __func__,
5268 			dring_pkt->tag.vio_subtype);
5269 		vsw_restart_handshake(ldcp);
5270 	}
5271 
5272 	D1(vswp, "%s(%lld): exit", __func__, ldcp->ldc_id);
5273 }
5274 
5275 #define	SND_MCST_NACK(ldcp, pkt) \
5276 	pkt->tag.vio_subtype = VIO_SUBTYPE_NACK; \
5277 	pkt->tag.vio_sid = ldcp->local_session; \
5278 	vsw_send_msg(ldcp, (void *)pkt, sizeof (vnet_mcast_msg_t));
5279 
5280 /*
5281  * Process a multicast request from a vnet.
5282  *
5283  * Vnet's specify a multicast address that they are interested in. This
5284  * address is used as a key into the hash table which forms the multicast
5285  * forwarding database (mFDB).
5286  *
5287  * The table keys are the multicast addresses, while the table entries
5288  * are pointers to lists of ports which wish to receive packets for the
5289  * specified multicast address.
5290  *
5291  * When a multicast packet is being switched we use the address as a key
5292  * into the hash table, and then walk the appropriate port list forwarding
5293  * the pkt to each port in turn.
5294  *
5295  * If a vnet is no longer interested in a particular multicast grouping
5296  * we simply find the correct location in the hash table and then delete
5297  * the relevant port from the port list.
5298  *
5299  * To deal with the case whereby a port is being deleted without first
5300  * removing itself from the lists in the hash table, we maintain a list
5301  * of multicast addresses the port has registered an interest in, within
5302  * the port structure itself. We then simply walk that list of addresses
5303  * using them as keys into the hash table and remove the port from the
5304  * appropriate lists.
5305  */
5306 static void
5307 vsw_process_ctrl_mcst_pkt(vsw_ldc_t *ldcp, void *pkt)
5308 {
5309 	vnet_mcast_msg_t	*mcst_pkt;
5310 	vsw_port_t		*port = ldcp->ldc_port;
5311 	vsw_t			*vswp = ldcp->ldc_vswp;
5312 	int			i;
5313 
5314 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
5315 
5316 	/*
5317 	 * We know this is a ctrl/mcast packet so
5318 	 * cast it into the correct structure.
5319 	 */
5320 	mcst_pkt = (vnet_mcast_msg_t *)pkt;
5321 
5322 	switch (mcst_pkt->tag.vio_subtype) {
5323 	case VIO_SUBTYPE_INFO:
5324 		D2(vswp, "%s: VIO_SUBTYPE_INFO", __func__);
5325 
5326 		/*
5327 		 * Check if in correct state to receive a multicast
5328 		 * message (i.e. handshake complete). If not reset
5329 		 * the handshake.
5330 		 */
5331 		if (vsw_check_flag(ldcp, INBOUND, VSW_MCST_INFO_RECV))
5332 			return;
5333 
5334 		/*
5335 		 * Before attempting to add or remove address check
5336 		 * that they are valid multicast addresses.
5337 		 * If not, then NACK back.
5338 		 */
5339 		for (i = 0; i < mcst_pkt->count; i++) {
5340 			if ((mcst_pkt->mca[i].ether_addr_octet[0] & 01) != 1) {
5341 				DERR(vswp, "%s: invalid multicast address",
5342 								__func__);
5343 				SND_MCST_NACK(ldcp, mcst_pkt);
5344 				return;
5345 			}
5346 		}
5347 
5348 		/*
5349 		 * Now add/remove the addresses. If this fails we
5350 		 * NACK back.
5351 		 */
5352 		if (vsw_add_rem_mcst(mcst_pkt, port) != 0) {
5353 			SND_MCST_NACK(ldcp, mcst_pkt);
5354 			return;
5355 		}
5356 
5357 		mcst_pkt->tag.vio_subtype = VIO_SUBTYPE_ACK;
5358 		mcst_pkt->tag.vio_sid = ldcp->local_session;
5359 
5360 		DUMP_TAG_PTR((vio_msg_tag_t *)mcst_pkt);
5361 
5362 		vsw_send_msg(ldcp, (void *)mcst_pkt,
5363 					sizeof (vnet_mcast_msg_t));
5364 		break;
5365 
5366 	case VIO_SUBTYPE_ACK:
5367 		DWARN(vswp, "%s: VIO_SUBTYPE_ACK", __func__);
5368 
5369 		/*
5370 		 * We shouldn't ever get a multicast ACK message as
5371 		 * at the moment we never request multicast addresses
5372 		 * to be set on some other device. This may change in
5373 		 * the future if we have cascading switches.
5374 		 */
5375 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_MCST_ACK_RECV))
5376 			return;
5377 
5378 				/* Do nothing */
5379 		break;
5380 
5381 	case VIO_SUBTYPE_NACK:
5382 		DWARN(vswp, "%s: VIO_SUBTYPE_NACK", __func__);
5383 
5384 		/*
5385 		 * We shouldn't get a multicast NACK packet for the
5386 		 * same reasons as we shouldn't get a ACK packet.
5387 		 */
5388 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_MCST_NACK_RECV))
5389 			return;
5390 
5391 				/* Do nothing */
5392 		break;
5393 
5394 	default:
5395 		DERR(vswp, "%s: unknown vio_subtype %x\n", __func__,
5396 			mcst_pkt->tag.vio_subtype);
5397 	}
5398 
5399 	D1(vswp, "%s(%lld): exit", __func__, ldcp->ldc_id);
5400 }
5401 
5402 static void
5403 vsw_process_ctrl_rdx_pkt(vsw_ldc_t *ldcp, void *pkt)
5404 {
5405 	vio_rdx_msg_t	*rdx_pkt;
5406 	vsw_t		*vswp = ldcp->ldc_vswp;
5407 
5408 	/*
5409 	 * We know this is a ctrl/rdx packet so
5410 	 * cast it into the correct structure.
5411 	 */
5412 	rdx_pkt = (vio_rdx_msg_t *)pkt;
5413 
5414 	D1(vswp, "%s(%lld) enter", __func__, ldcp->ldc_id);
5415 
5416 	switch (rdx_pkt->tag.vio_subtype) {
5417 	case VIO_SUBTYPE_INFO:
5418 		D2(vswp, "%s: VIO_SUBTYPE_INFO", __func__);
5419 
5420 		if (vsw_check_flag(ldcp, INBOUND, VSW_RDX_INFO_RECV))
5421 			return;
5422 
5423 		rdx_pkt->tag.vio_sid = ldcp->local_session;
5424 		rdx_pkt->tag.vio_subtype = VIO_SUBTYPE_ACK;
5425 
5426 		DUMP_TAG_PTR((vio_msg_tag_t *)rdx_pkt);
5427 
5428 		ldcp->lane_in.lstate |= VSW_RDX_ACK_SENT;
5429 
5430 		vsw_send_msg(ldcp, (void *)rdx_pkt,
5431 				sizeof (vio_rdx_msg_t));
5432 
5433 		vsw_next_milestone(ldcp);
5434 		break;
5435 
5436 	case VIO_SUBTYPE_ACK:
5437 		/*
5438 		 * Should be handled in-band by callback handler.
5439 		 */
5440 		DERR(vswp, "%s: Unexpected VIO_SUBTYPE_ACK", __func__);
5441 		vsw_restart_handshake(ldcp);
5442 		break;
5443 
5444 	case VIO_SUBTYPE_NACK:
5445 		D2(vswp, "%s: VIO_SUBTYPE_NACK", __func__);
5446 
5447 		if (vsw_check_flag(ldcp, OUTBOUND, VSW_RDX_NACK_RECV))
5448 			return;
5449 
5450 		ldcp->lane_out.lstate |= VSW_RDX_NACK_RECV;
5451 		vsw_next_milestone(ldcp);
5452 		break;
5453 
5454 	default:
5455 		DERR(vswp, "%s: Unknown vio_subtype %x\n", __func__,
5456 			rdx_pkt->tag.vio_subtype);
5457 	}
5458 
5459 	D1(vswp, "%s(%lld): exit", __func__, ldcp->ldc_id);
5460 }
5461 
5462 static void
5463 vsw_process_data_pkt(vsw_ldc_t *ldcp, void *dpkt, vio_msg_tag_t tag)
5464 {
5465 	uint16_t	env = tag.vio_subtype_env;
5466 	vsw_t		*vswp = ldcp->ldc_vswp;
5467 
5468 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
5469 
5470 	/* session id check */
5471 	if (ldcp->session_status & VSW_PEER_SESSION) {
5472 		if (ldcp->peer_session != tag.vio_sid) {
5473 			DERR(vswp, "%s (chan %d): invalid session id (%llx)",
5474 				__func__, ldcp->ldc_id, tag.vio_sid);
5475 			vsw_restart_handshake(ldcp);
5476 			return;
5477 		}
5478 	}
5479 
5480 	/*
5481 	 * It is an error for us to be getting data packets
5482 	 * before the handshake has completed.
5483 	 */
5484 	if (ldcp->hphase != VSW_MILESTONE4) {
5485 		DERR(vswp, "%s: got data packet before handshake complete "
5486 			"hphase %d (%x: %x)", __func__, ldcp->hphase,
5487 			ldcp->lane_in.lstate, ldcp->lane_out.lstate);
5488 		DUMP_FLAGS(ldcp->lane_in.lstate);
5489 		DUMP_FLAGS(ldcp->lane_out.lstate);
5490 		vsw_restart_handshake(ldcp);
5491 		return;
5492 	}
5493 
5494 	/*
5495 	 * Switch on vio_subtype envelope, then let lower routines
5496 	 * decide if its an INFO, ACK or NACK packet.
5497 	 */
5498 	if (env == VIO_DRING_DATA) {
5499 		vsw_process_data_dring_pkt(ldcp, dpkt);
5500 	} else if (env == VIO_PKT_DATA) {
5501 		vsw_process_data_raw_pkt(ldcp, dpkt);
5502 	} else if (env == VIO_DESC_DATA) {
5503 		vsw_process_data_ibnd_pkt(ldcp, dpkt);
5504 	} else {
5505 		DERR(vswp, "%s : unknown vio_subtype_env (%x)\n",
5506 							__func__, env);
5507 	}
5508 
5509 	D1(vswp, "%s(%lld): exit", __func__, ldcp->ldc_id);
5510 }
5511 
5512 #define	SND_DRING_NACK(ldcp, pkt) \
5513 	pkt->tag.vio_subtype = VIO_SUBTYPE_NACK; \
5514 	pkt->tag.vio_sid = ldcp->local_session; \
5515 	vsw_send_msg(ldcp, (void *)pkt, sizeof (vio_dring_msg_t));
5516 
5517 static void
5518 vsw_process_data_dring_pkt(vsw_ldc_t *ldcp, void *dpkt)
5519 {
5520 	vio_dring_msg_t		*dring_pkt;
5521 	vnet_public_desc_t	*pub_addr = NULL;
5522 	vsw_private_desc_t	*priv_addr = NULL;
5523 	dring_info_t		*dp = NULL;
5524 	vsw_t			*vswp = ldcp->ldc_vswp;
5525 	mblk_t			*mp = NULL;
5526 	mblk_t			*bp = NULL;
5527 	mblk_t			*bpt = NULL;
5528 	size_t			nbytes = 0;
5529 	size_t			off = 0;
5530 	uint64_t		ncookies = 0;
5531 	uint64_t		chain = 0;
5532 	uint64_t		j, len;
5533 	uint32_t		pos, start, datalen;
5534 	uint32_t		range_start, range_end;
5535 	int32_t			end, num, cnt = 0;
5536 	int			i, rv;
5537 	boolean_t		ack_needed = B_FALSE;
5538 	boolean_t		prev_desc_ack = B_FALSE;
5539 	int			read_attempts = 0;
5540 
5541 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
5542 
5543 	/*
5544 	 * We know this is a data/dring packet so
5545 	 * cast it into the correct structure.
5546 	 */
5547 	dring_pkt = (vio_dring_msg_t *)dpkt;
5548 
5549 	/*
5550 	 * Switch on the vio_subtype. If its INFO then we need to
5551 	 * process the data. If its an ACK we need to make sure
5552 	 * it makes sense (i.e did we send an earlier data/info),
5553 	 * and if its a NACK then we maybe attempt a retry.
5554 	 */
5555 	switch (dring_pkt->tag.vio_subtype) {
5556 	case VIO_SUBTYPE_INFO:
5557 		D2(vswp, "%s(%lld): VIO_SUBTYPE_INFO", __func__, ldcp->ldc_id);
5558 
5559 		READ_ENTER(&ldcp->lane_in.dlistrw);
5560 		if ((dp = vsw_ident2dring(&ldcp->lane_in,
5561 				dring_pkt->dring_ident)) == NULL) {
5562 			RW_EXIT(&ldcp->lane_in.dlistrw);
5563 
5564 			DERR(vswp, "%s(%lld): unable to find dring from "
5565 				"ident 0x%llx", __func__, ldcp->ldc_id,
5566 				dring_pkt->dring_ident);
5567 
5568 			SND_DRING_NACK(ldcp, dring_pkt);
5569 			return;
5570 		}
5571 
5572 		start = pos = dring_pkt->start_idx;
5573 		end = dring_pkt->end_idx;
5574 		len = dp->num_descriptors;
5575 
5576 		range_start = range_end = pos;
5577 
5578 		D2(vswp, "%s(%lld): start index %ld : end %ld\n",
5579 			__func__, ldcp->ldc_id, start, end);
5580 
5581 		if (end == -1) {
5582 			num = -1;
5583 		} else if (end >= 0) {
5584 			num = end >= pos ?
5585 				end - pos + 1: (len - pos + 1) + end;
5586 
5587 			/* basic sanity check */
5588 			if (end > len) {
5589 				RW_EXIT(&ldcp->lane_in.dlistrw);
5590 				DERR(vswp, "%s(%lld): endpoint %lld outside "
5591 					"ring length %lld", __func__,
5592 					ldcp->ldc_id, end, len);
5593 
5594 				SND_DRING_NACK(ldcp, dring_pkt);
5595 				return;
5596 			}
5597 		} else {
5598 			RW_EXIT(&ldcp->lane_in.dlistrw);
5599 			DERR(vswp, "%s(%lld): invalid endpoint %lld",
5600 				__func__, ldcp->ldc_id, end);
5601 			SND_DRING_NACK(ldcp, dring_pkt);
5602 			return;
5603 		}
5604 
5605 		while (cnt != num) {
5606 vsw_recheck_desc:
5607 			if ((rv = ldc_mem_dring_acquire(dp->handle,
5608 							pos, pos)) != 0) {
5609 				RW_EXIT(&ldcp->lane_in.dlistrw);
5610 				DERR(vswp, "%s(%lld): unable to acquire "
5611 					"descriptor at pos %d: err %d",
5612 					__func__, pos, ldcp->ldc_id, rv);
5613 				SND_DRING_NACK(ldcp, dring_pkt);
5614 				return;
5615 			}
5616 
5617 			pub_addr = (vnet_public_desc_t *)dp->pub_addr + pos;
5618 
5619 			/*
5620 			 * When given a bounded range of descriptors
5621 			 * to process, its an error to hit a descriptor
5622 			 * which is not ready. In the non-bounded case
5623 			 * (end_idx == -1) this simply indicates we have
5624 			 * reached the end of the current active range.
5625 			 */
5626 			if (pub_addr->hdr.dstate != VIO_DESC_READY) {
5627 				/* unbound - no error */
5628 				if (end == -1) {
5629 					if (read_attempts == vsw_read_attempts)
5630 						break;
5631 
5632 					delay(drv_usectohz(vsw_desc_delay));
5633 					read_attempts++;
5634 					goto vsw_recheck_desc;
5635 				}
5636 
5637 				/* bounded - error - so NACK back */
5638 				RW_EXIT(&ldcp->lane_in.dlistrw);
5639 				DERR(vswp, "%s(%lld): descriptor not READY "
5640 					"(%d)", __func__, ldcp->ldc_id,
5641 					pub_addr->hdr.dstate);
5642 				SND_DRING_NACK(ldcp, dring_pkt);
5643 				return;
5644 			}
5645 
5646 			DTRACE_PROBE1(read_attempts, int, read_attempts);
5647 
5648 			range_end = pos;
5649 
5650 			/*
5651 			 * If we ACK'd the previous descriptor then now
5652 			 * record the new range start position for later
5653 			 * ACK's.
5654 			 */
5655 			if (prev_desc_ack) {
5656 				range_start = pos;
5657 
5658 				D2(vswp, "%s(%lld): updating range start "
5659 					"to be %d", __func__, ldcp->ldc_id,
5660 					range_start);
5661 
5662 				prev_desc_ack = B_FALSE;
5663 			}
5664 
5665 			/*
5666 			 * Data is padded to align on 8 byte boundary,
5667 			 * datalen is actual data length, i.e. minus that
5668 			 * padding.
5669 			 */
5670 			datalen = pub_addr->nbytes;
5671 
5672 			/*
5673 			 * Does peer wish us to ACK when we have finished
5674 			 * with this descriptor ?
5675 			 */
5676 			if (pub_addr->hdr.ack)
5677 				ack_needed = B_TRUE;
5678 
5679 			D2(vswp, "%s(%lld): processing desc %lld at pos"
5680 				" 0x%llx : dstate 0x%lx : datalen 0x%lx",
5681 				__func__, ldcp->ldc_id, pos, pub_addr,
5682 				pub_addr->hdr.dstate, datalen);
5683 
5684 			/*
5685 			 * Mark that we are starting to process descriptor.
5686 			 */
5687 			pub_addr->hdr.dstate = VIO_DESC_ACCEPTED;
5688 
5689 			mp = vio_allocb(ldcp->rxh);
5690 			if (mp == NULL) {
5691 				/*
5692 				 * No free receive buffers available, so
5693 				 * fallback onto allocb(9F). Make sure that
5694 				 * we get a data buffer which is a multiple
5695 				 * of 8 as this is required by ldc_mem_copy.
5696 				 */
5697 				DTRACE_PROBE(allocb);
5698 				mp = allocb(datalen + VNET_IPALIGN + 8,
5699 								BPRI_MED);
5700 			}
5701 
5702 			/*
5703 			 * Ensure that we ask ldc for an aligned
5704 			 * number of bytes.
5705 			 */
5706 			nbytes = datalen + VNET_IPALIGN;
5707 			if (nbytes & 0x7) {
5708 				off = 8 - (nbytes & 0x7);
5709 				nbytes += off;
5710 			}
5711 
5712 			ncookies = pub_addr->ncookies;
5713 			rv = ldc_mem_copy(ldcp->ldc_handle,
5714 				(caddr_t)mp->b_rptr, 0, &nbytes,
5715 				pub_addr->memcookie, ncookies,
5716 				LDC_COPY_IN);
5717 
5718 			if (rv != 0) {
5719 				DERR(vswp, "%s(%d): unable to copy in "
5720 					"data from %d cookies in desc %d"
5721 					" (rv %d)", __func__, ldcp->ldc_id,
5722 					ncookies, pos, rv);
5723 				freemsg(mp);
5724 
5725 				pub_addr->hdr.dstate = VIO_DESC_DONE;
5726 				(void) ldc_mem_dring_release(dp->handle,
5727 								pos, pos);
5728 				break;
5729 			} else {
5730 				D2(vswp, "%s(%d): copied in %ld bytes"
5731 					" using %d cookies", __func__,
5732 					ldcp->ldc_id, nbytes, ncookies);
5733 			}
5734 
5735 			/* adjust the read pointer to skip over the padding */
5736 			mp->b_rptr += VNET_IPALIGN;
5737 
5738 			/* point to the actual end of data */
5739 			mp->b_wptr = mp->b_rptr + datalen;
5740 
5741 			/* build a chain of received packets */
5742 			if (bp == NULL) {
5743 				/* first pkt */
5744 				bp = mp;
5745 				bp->b_next = bp->b_prev = NULL;
5746 				bpt = bp;
5747 				chain = 1;
5748 			} else {
5749 				mp->b_next = NULL;
5750 				mp->b_prev = bpt;
5751 				bpt->b_next = mp;
5752 				bpt = mp;
5753 				chain++;
5754 			}
5755 
5756 			/* mark we are finished with this descriptor */
5757 			pub_addr->hdr.dstate = VIO_DESC_DONE;
5758 
5759 			(void) ldc_mem_dring_release(dp->handle, pos, pos);
5760 
5761 			/*
5762 			 * Send an ACK back to peer if requested.
5763 			 */
5764 			if (ack_needed) {
5765 				ack_needed = B_FALSE;
5766 
5767 				dring_pkt->start_idx = range_start;
5768 				dring_pkt->end_idx = range_end;
5769 
5770 				DERR(vswp, "%s(%lld): processed %d %d, ACK"
5771 					" requested", __func__, ldcp->ldc_id,
5772 					dring_pkt->start_idx,
5773 					dring_pkt->end_idx);
5774 
5775 				dring_pkt->dring_process_state = VIO_DP_ACTIVE;
5776 				dring_pkt->tag.vio_subtype = VIO_SUBTYPE_ACK;
5777 				dring_pkt->tag.vio_sid = ldcp->local_session;
5778 				vsw_send_msg(ldcp, (void *)dring_pkt,
5779 					sizeof (vio_dring_msg_t));
5780 
5781 				prev_desc_ack = B_TRUE;
5782 				range_start = pos;
5783 			}
5784 
5785 			/* next descriptor */
5786 			pos = (pos + 1) % len;
5787 			cnt++;
5788 
5789 			/*
5790 			 * Break out of loop here and stop processing to
5791 			 * allow some other network device (or disk) to
5792 			 * get access to the cpu.
5793 			 */
5794 			/* send the chain of packets to be switched */
5795 			if (chain > vsw_chain_len) {
5796 				D3(vswp, "%s(%lld): switching chain of %d "
5797 					"msgs", __func__, ldcp->ldc_id, chain);
5798 				vswp->vsw_switch_frame(vswp, bp, VSW_VNETPORT,
5799 							ldcp->ldc_port, NULL);
5800 				bp = NULL;
5801 				break;
5802 			}
5803 		}
5804 		RW_EXIT(&ldcp->lane_in.dlistrw);
5805 
5806 		/* send the chain of packets to be switched */
5807 		if (bp != NULL) {
5808 			D3(vswp, "%s(%lld): switching chain of %d msgs",
5809 					__func__, ldcp->ldc_id, chain);
5810 			vswp->vsw_switch_frame(vswp, bp, VSW_VNETPORT,
5811 							ldcp->ldc_port, NULL);
5812 		}
5813 
5814 		DTRACE_PROBE1(msg_cnt, int, cnt);
5815 
5816 		/*
5817 		 * We are now finished so ACK back with the state
5818 		 * set to STOPPING so our peer knows we are finished
5819 		 */
5820 		dring_pkt->tag.vio_subtype = VIO_SUBTYPE_ACK;
5821 		dring_pkt->tag.vio_sid = ldcp->local_session;
5822 
5823 		dring_pkt->dring_process_state = VIO_DP_STOPPED;
5824 
5825 		DTRACE_PROBE(stop_process_sent);
5826 
5827 		/*
5828 		 * We have not processed any more descriptors beyond
5829 		 * the last one we ACK'd.
5830 		 */
5831 		if (prev_desc_ack)
5832 			range_start = range_end;
5833 
5834 		dring_pkt->start_idx = range_start;
5835 		dring_pkt->end_idx = range_end;
5836 
5837 		D2(vswp, "%s(%lld) processed : %d : %d, now stopping",
5838 			__func__, ldcp->ldc_id, dring_pkt->start_idx,
5839 			dring_pkt->end_idx);
5840 
5841 		vsw_send_msg(ldcp, (void *)dring_pkt,
5842 					sizeof (vio_dring_msg_t));
5843 		break;
5844 
5845 	case VIO_SUBTYPE_ACK:
5846 		D2(vswp, "%s(%lld): VIO_SUBTYPE_ACK", __func__, ldcp->ldc_id);
5847 		/*
5848 		 * Verify that the relevant descriptors are all
5849 		 * marked as DONE
5850 		 */
5851 		READ_ENTER(&ldcp->lane_out.dlistrw);
5852 		if ((dp = vsw_ident2dring(&ldcp->lane_out,
5853 			dring_pkt->dring_ident)) == NULL) {
5854 			RW_EXIT(&ldcp->lane_out.dlistrw);
5855 			DERR(vswp, "%s: unknown ident in ACK", __func__);
5856 			return;
5857 		}
5858 
5859 		pub_addr = (vnet_public_desc_t *)dp->pub_addr;
5860 		priv_addr = (vsw_private_desc_t *)dp->priv_addr;
5861 
5862 		start = end = 0;
5863 		start = dring_pkt->start_idx;
5864 		end = dring_pkt->end_idx;
5865 		len = dp->num_descriptors;
5866 
5867 		j = num = 0;
5868 		/* calculate # descriptors taking into a/c wrap around */
5869 		num = end >= start ? end - start + 1: (len - start + 1) + end;
5870 
5871 		D2(vswp, "%s(%lld): start index %ld : end %ld : num %ld\n",
5872 			__func__, ldcp->ldc_id, start, end, num);
5873 
5874 		mutex_enter(&dp->dlock);
5875 		dp->last_ack_recv = end;
5876 		mutex_exit(&dp->dlock);
5877 
5878 		for (i = start; j < num; i = (i + 1) % len, j++) {
5879 			pub_addr = (vnet_public_desc_t *)dp->pub_addr + i;
5880 			priv_addr = (vsw_private_desc_t *)dp->priv_addr + i;
5881 
5882 			/*
5883 			 * If the last descriptor in a range has the ACK
5884 			 * bit set then we will get two messages from our
5885 			 * peer relating to it. The normal ACK msg and then
5886 			 * a subsequent STOP msg. The first message will have
5887 			 * resulted in the descriptor being reclaimed and
5888 			 * its state set to FREE so when we encounter a non
5889 			 * DONE descriptor we need to check to see if its
5890 			 * because we have just reclaimed it.
5891 			 */
5892 			mutex_enter(&priv_addr->dstate_lock);
5893 			if (pub_addr->hdr.dstate == VIO_DESC_DONE) {
5894 				/* clear all the fields */
5895 				bzero(priv_addr->datap, priv_addr->datalen);
5896 				priv_addr->datalen = 0;
5897 
5898 				pub_addr->hdr.dstate = VIO_DESC_FREE;
5899 				pub_addr->hdr.ack = 0;
5900 
5901 				priv_addr->dstate = VIO_DESC_FREE;
5902 				mutex_exit(&priv_addr->dstate_lock);
5903 
5904 				D3(vswp, "clearing descp %d : pub state "
5905 					"0x%llx : priv state 0x%llx", i,
5906 					pub_addr->hdr.dstate,
5907 					priv_addr->dstate);
5908 
5909 			} else {
5910 				mutex_exit(&priv_addr->dstate_lock);
5911 
5912 				if (dring_pkt->dring_process_state !=
5913 							VIO_DP_STOPPED) {
5914 					DERR(vswp, "%s: descriptor %lld at pos "
5915 						" 0x%llx not DONE (0x%lx)\n",
5916 						__func__, i, pub_addr,
5917 						pub_addr->hdr.dstate);
5918 					RW_EXIT(&ldcp->lane_out.dlistrw);
5919 					return;
5920 				}
5921 			}
5922 		}
5923 
5924 		/*
5925 		 * If our peer is stopping processing descriptors then
5926 		 * we check to make sure it has processed all the descriptors
5927 		 * we have updated. If not then we send it a new message
5928 		 * to prompt it to restart.
5929 		 */
5930 		if (dring_pkt->dring_process_state == VIO_DP_STOPPED) {
5931 			DTRACE_PROBE(stop_process_recv);
5932 			D2(vswp, "%s(%lld): got stopping msg : %d : %d",
5933 				__func__, ldcp->ldc_id, dring_pkt->start_idx,
5934 				dring_pkt->end_idx);
5935 
5936 			/*
5937 			 * Check next descriptor in public section of ring.
5938 			 * If its marked as READY then we need to prompt our
5939 			 * peer to start processing the ring again.
5940 			 */
5941 			i = (end + 1) % len;
5942 			pub_addr = (vnet_public_desc_t *)dp->pub_addr + i;
5943 			priv_addr = (vsw_private_desc_t *)dp->priv_addr + i;
5944 
5945 			/*
5946 			 * Hold the restart lock across all of this to
5947 			 * make sure that its not possible for us to
5948 			 * decide that a msg needs to be sent in the future
5949 			 * but the sending code having already checked is
5950 			 * about to exit.
5951 			 */
5952 			mutex_enter(&dp->restart_lock);
5953 			mutex_enter(&priv_addr->dstate_lock);
5954 			if (pub_addr->hdr.dstate == VIO_DESC_READY) {
5955 
5956 				mutex_exit(&priv_addr->dstate_lock);
5957 
5958 				dring_pkt->tag.vio_subtype = VIO_SUBTYPE_INFO;
5959 				dring_pkt->tag.vio_sid = ldcp->local_session;
5960 
5961 				mutex_enter(&ldcp->lane_out.seq_lock);
5962 				dring_pkt->seq_num = ldcp->lane_out.seq_num++;
5963 				mutex_exit(&ldcp->lane_out.seq_lock);
5964 
5965 				dring_pkt->start_idx = (end + 1) % len;
5966 				dring_pkt->end_idx = -1;
5967 
5968 				D2(vswp, "%s(%lld) : sending restart msg:"
5969 					" %d : %d", __func__, ldcp->ldc_id,
5970 					dring_pkt->start_idx,
5971 					dring_pkt->end_idx);
5972 
5973 				vsw_send_msg(ldcp, (void *)dring_pkt,
5974 						sizeof (vio_dring_msg_t));
5975 			} else {
5976 				mutex_exit(&priv_addr->dstate_lock);
5977 				dp->restart_reqd = B_TRUE;
5978 			}
5979 			mutex_exit(&dp->restart_lock);
5980 		}
5981 		RW_EXIT(&ldcp->lane_out.dlistrw);
5982 		break;
5983 
5984 	case VIO_SUBTYPE_NACK:
5985 		DWARN(vswp, "%s(%lld): VIO_SUBTYPE_NACK",
5986 						__func__, ldcp->ldc_id);
5987 		/*
5988 		 * Something is badly wrong if we are getting NACK's
5989 		 * for our data pkts. So reset the channel.
5990 		 */
5991 		vsw_restart_handshake(ldcp);
5992 
5993 		break;
5994 
5995 	default:
5996 		DERR(vswp, "%s(%lld): Unknown vio_subtype %x\n", __func__,
5997 			ldcp->ldc_id, dring_pkt->tag.vio_subtype);
5998 	}
5999 
6000 	D1(vswp, "%s(%lld) exit", __func__, ldcp->ldc_id);
6001 }
6002 
6003 /*
6004  * VIO_PKT_DATA (a.k.a raw data mode )
6005  *
6006  * Note - currently not supported. Do nothing.
6007  */
6008 static void
6009 vsw_process_data_raw_pkt(vsw_ldc_t *ldcp, void *dpkt)
6010 {
6011 	_NOTE(ARGUNUSED(dpkt))
6012 
6013 	D1(NULL, "%s (%lld): enter\n", __func__, ldcp->ldc_id);
6014 
6015 	DERR(NULL, "%s (%lld): currently  not supported",
6016 						__func__, ldcp->ldc_id);
6017 
6018 	D1(NULL, "%s (%lld): exit\n", __func__, ldcp->ldc_id);
6019 }
6020 
6021 /*
6022  * Process an in-band descriptor message (most likely from
6023  * OBP).
6024  */
6025 static void
6026 vsw_process_data_ibnd_pkt(vsw_ldc_t *ldcp, void *pkt)
6027 {
6028 	vnet_ibnd_desc_t	*ibnd_desc;
6029 	dring_info_t		*dp = NULL;
6030 	vsw_private_desc_t	*priv_addr = NULL;
6031 	vsw_t			*vswp = ldcp->ldc_vswp;
6032 	mblk_t			*mp = NULL;
6033 	size_t			nbytes = 0;
6034 	size_t			off = 0;
6035 	uint64_t		idx = 0;
6036 	uint32_t		num = 1, len, datalen = 0;
6037 	uint64_t		ncookies = 0;
6038 	int			i, rv;
6039 	int			j = 0;
6040 
6041 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
6042 
6043 	ibnd_desc = (vnet_ibnd_desc_t *)pkt;
6044 
6045 	switch (ibnd_desc->hdr.tag.vio_subtype) {
6046 	case VIO_SUBTYPE_INFO:
6047 		D1(vswp, "%s: VIO_SUBTYPE_INFO", __func__);
6048 
6049 		if (vsw_check_flag(ldcp, INBOUND, VSW_DRING_INFO_RECV))
6050 			return;
6051 
6052 		/*
6053 		 * Data is padded to align on a 8 byte boundary,
6054 		 * nbytes is actual data length, i.e. minus that
6055 		 * padding.
6056 		 */
6057 		datalen = ibnd_desc->nbytes;
6058 
6059 		D2(vswp, "%s(%lld): processing inband desc : "
6060 			": datalen 0x%lx", __func__, ldcp->ldc_id, datalen);
6061 
6062 		ncookies = ibnd_desc->ncookies;
6063 
6064 		/*
6065 		 * allocb(9F) returns an aligned data block. We
6066 		 * need to ensure that we ask ldc for an aligned
6067 		 * number of bytes also.
6068 		 */
6069 		nbytes = datalen;
6070 		if (nbytes & 0x7) {
6071 			off = 8 - (nbytes & 0x7);
6072 			nbytes += off;
6073 		}
6074 
6075 		mp = allocb(datalen, BPRI_MED);
6076 		if (mp == NULL) {
6077 			DERR(vswp, "%s(%lld): allocb failed",
6078 					__func__, ldcp->ldc_id);
6079 			return;
6080 		}
6081 
6082 		rv = ldc_mem_copy(ldcp->ldc_handle, (caddr_t)mp->b_rptr,
6083 			0, &nbytes, ibnd_desc->memcookie, (uint64_t)ncookies,
6084 			LDC_COPY_IN);
6085 
6086 		if (rv != 0) {
6087 			DERR(vswp, "%s(%d): unable to copy in data from "
6088 				"%d cookie(s)", __func__,
6089 				ldcp->ldc_id, ncookies);
6090 			freemsg(mp);
6091 			return;
6092 		} else {
6093 			D2(vswp, "%s(%d): copied in %ld bytes using %d "
6094 				"cookies", __func__, ldcp->ldc_id, nbytes,
6095 				ncookies);
6096 		}
6097 
6098 		/* point to the actual end of data */
6099 		mp->b_wptr = mp->b_rptr + datalen;
6100 
6101 		/*
6102 		 * We ACK back every in-band descriptor message we process
6103 		 */
6104 		ibnd_desc->hdr.tag.vio_subtype = VIO_SUBTYPE_ACK;
6105 		ibnd_desc->hdr.tag.vio_sid = ldcp->local_session;
6106 		vsw_send_msg(ldcp, (void *)ibnd_desc,
6107 				sizeof (vnet_ibnd_desc_t));
6108 
6109 		/* send the packet to be switched */
6110 		vswp->vsw_switch_frame(vswp, mp, VSW_VNETPORT,
6111 					ldcp->ldc_port, NULL);
6112 
6113 		break;
6114 
6115 	case VIO_SUBTYPE_ACK:
6116 		D1(vswp, "%s: VIO_SUBTYPE_ACK", __func__);
6117 
6118 		/* Verify the ACK is valid */
6119 		idx = ibnd_desc->hdr.desc_handle;
6120 
6121 		if (idx >= VSW_RING_NUM_EL) {
6122 			cmn_err(CE_WARN, "!vsw%d: corrupted ACK received "
6123 				"(idx %ld)", vswp->instance, idx);
6124 			return;
6125 		}
6126 
6127 		if ((dp = ldcp->lane_out.dringp) == NULL) {
6128 			DERR(vswp, "%s: no dring found", __func__);
6129 			return;
6130 		}
6131 
6132 		len = dp->num_descriptors;
6133 		/*
6134 		 * If the descriptor we are being ACK'ed for is not the
6135 		 * one we expected, then pkts were lost somwhere, either
6136 		 * when we tried to send a msg, or a previous ACK msg from
6137 		 * our peer. In either case we now reclaim the descriptors
6138 		 * in the range from the last ACK we received up to the
6139 		 * current ACK.
6140 		 */
6141 		if (idx != dp->last_ack_recv) {
6142 			DWARN(vswp, "%s: dropped pkts detected, (%ld, %ld)",
6143 				__func__, dp->last_ack_recv, idx);
6144 			num = idx >= dp->last_ack_recv ?
6145 				idx - dp->last_ack_recv + 1:
6146 				(len - dp->last_ack_recv + 1) + idx;
6147 		}
6148 
6149 		/*
6150 		 * When we sent the in-band message to our peer we
6151 		 * marked the copy in our private ring as READY. We now
6152 		 * check that the descriptor we are being ACK'ed for is in
6153 		 * fact READY, i.e. it is one we have shared with our peer.
6154 		 *
6155 		 * If its not we flag an error, but still reset the descr
6156 		 * back to FREE.
6157 		 */
6158 		for (i = dp->last_ack_recv; j < num; i = (i + 1) % len, j++) {
6159 			priv_addr = (vsw_private_desc_t *)dp->priv_addr + i;
6160 			mutex_enter(&priv_addr->dstate_lock);
6161 			if (priv_addr->dstate != VIO_DESC_READY) {
6162 				DERR(vswp, "%s: (%ld) desc at index %ld not "
6163 					"READY (0x%lx)", __func__,
6164 					ldcp->ldc_id, idx, priv_addr->dstate);
6165 				DERR(vswp, "%s: bound %d: ncookies %ld : "
6166 					"datalen %ld", __func__,
6167 					priv_addr->bound, priv_addr->ncookies,
6168 					priv_addr->datalen);
6169 			}
6170 			D2(vswp, "%s: (%lld) freeing descp at %lld", __func__,
6171 				ldcp->ldc_id, idx);
6172 			/* release resources associated with sent msg */
6173 			bzero(priv_addr->datap, priv_addr->datalen);
6174 			priv_addr->datalen = 0;
6175 			priv_addr->dstate = VIO_DESC_FREE;
6176 			mutex_exit(&priv_addr->dstate_lock);
6177 		}
6178 		/* update to next expected value */
6179 		dp->last_ack_recv = (idx + 1) % dp->num_descriptors;
6180 
6181 		break;
6182 
6183 	case VIO_SUBTYPE_NACK:
6184 		DERR(vswp, "%s: VIO_SUBTYPE_NACK", __func__);
6185 
6186 		/*
6187 		 * We should only get a NACK if our peer doesn't like
6188 		 * something about a message we have sent it. If this
6189 		 * happens we just release the resources associated with
6190 		 * the message. (We are relying on higher layers to decide
6191 		 * whether or not to resend.
6192 		 */
6193 
6194 		/* limit check */
6195 		idx = ibnd_desc->hdr.desc_handle;
6196 
6197 		if (idx >= VSW_RING_NUM_EL) {
6198 			DERR(vswp, "%s: corrupted NACK received (idx %lld)",
6199 				__func__, idx);
6200 			return;
6201 		}
6202 
6203 		if ((dp = ldcp->lane_out.dringp) == NULL) {
6204 			DERR(vswp, "%s: no dring found", __func__);
6205 			return;
6206 		}
6207 
6208 		priv_addr = (vsw_private_desc_t *)dp->priv_addr;
6209 
6210 		/* move to correct location in ring */
6211 		priv_addr += idx;
6212 
6213 		/* release resources associated with sent msg */
6214 		mutex_enter(&priv_addr->dstate_lock);
6215 		bzero(priv_addr->datap, priv_addr->datalen);
6216 		priv_addr->datalen = 0;
6217 		priv_addr->dstate = VIO_DESC_FREE;
6218 		mutex_exit(&priv_addr->dstate_lock);
6219 
6220 		break;
6221 
6222 	default:
6223 		DERR(vswp, "%s(%lld): Unknown vio_subtype %x\n", __func__,
6224 			ldcp->ldc_id, ibnd_desc->hdr.tag.vio_subtype);
6225 	}
6226 
6227 	D1(vswp, "%s(%lld) exit", __func__, ldcp->ldc_id);
6228 }
6229 
6230 static void
6231 vsw_process_err_pkt(vsw_ldc_t *ldcp, void *epkt, vio_msg_tag_t tag)
6232 {
6233 	_NOTE(ARGUNUSED(epkt))
6234 
6235 	vsw_t		*vswp = ldcp->ldc_vswp;
6236 	uint16_t	env = tag.vio_subtype_env;
6237 
6238 	D1(vswp, "%s (%lld): enter\n", __func__, ldcp->ldc_id);
6239 
6240 	/*
6241 	 * Error vio_subtypes have yet to be defined. So for
6242 	 * the moment we can't do anything.
6243 	 */
6244 	D2(vswp, "%s: (%x) vio_subtype env", __func__, env);
6245 
6246 	D1(vswp, "%s (%lld): exit\n", __func__, ldcp->ldc_id);
6247 }
6248 
6249 /*
6250  * Switch the given ethernet frame when operating in layer 2 mode.
6251  *
6252  * vswp: pointer to the vsw instance
6253  * mp: pointer to chain of ethernet frame(s) to be switched
6254  * caller: identifies the source of this frame as:
6255  * 		1. VSW_VNETPORT - a vsw port (connected to a vnet).
6256  *		2. VSW_PHYSDEV - the physical ethernet device
6257  *		3. VSW_LOCALDEV - vsw configured as a virtual interface
6258  * arg: argument provided by the caller.
6259  *		1. for VNETPORT - pointer to the corresponding vsw_port_t.
6260  *		2. for PHYSDEV - NULL
6261  *		3. for LOCALDEV - pointer to to this vsw_t(self)
6262  */
6263 void
6264 vsw_switch_l2_frame(vsw_t *vswp, mblk_t *mp, int caller,
6265 			vsw_port_t *arg, mac_resource_handle_t mrh)
6266 {
6267 	struct ether_header	*ehp;
6268 	vsw_port_t		*port = NULL;
6269 	mblk_t			*bp, *ret_m;
6270 	mblk_t			*nmp = NULL;
6271 	vsw_port_list_t		*plist = &vswp->plist;
6272 
6273 	D1(vswp, "%s: enter (caller %d)", __func__, caller);
6274 
6275 	/*
6276 	 * PERF: rather than breaking up the chain here, scan it
6277 	 * to find all mblks heading to same destination and then
6278 	 * pass that sub-chain to the lower transmit functions.
6279 	 */
6280 
6281 	/* process the chain of packets */
6282 	bp = mp;
6283 	while (bp) {
6284 		mp = bp;
6285 		bp = bp->b_next;
6286 		mp->b_next = mp->b_prev = NULL;
6287 		ehp = (struct ether_header *)mp->b_rptr;
6288 
6289 		D2(vswp, "%s: mblk data buffer %lld : actual data size %lld",
6290 			__func__, MBLKSIZE(mp), MBLKL(mp));
6291 
6292 		READ_ENTER(&vswp->if_lockrw);
6293 		if (ether_cmp(&ehp->ether_dhost, &vswp->if_addr) == 0) {
6294 			/*
6295 			 * If destination is VSW_LOCALDEV (vsw as an eth
6296 			 * interface) and if the device is up & running,
6297 			 * send the packet up the stack on this host.
6298 			 * If the virtual interface is down, drop the packet.
6299 			 */
6300 			if (caller != VSW_LOCALDEV) {
6301 				if (vswp->if_state & VSW_IF_UP) {
6302 					RW_EXIT(&vswp->if_lockrw);
6303 					mac_rx(vswp->if_mh, mrh, mp);
6304 				} else {
6305 					RW_EXIT(&vswp->if_lockrw);
6306 					/* Interface down, drop pkt */
6307 					freemsg(mp);
6308 				}
6309 			} else {
6310 				RW_EXIT(&vswp->if_lockrw);
6311 				freemsg(mp);
6312 			}
6313 			continue;
6314 		}
6315 		RW_EXIT(&vswp->if_lockrw);
6316 
6317 		READ_ENTER(&plist->lockrw);
6318 		port = vsw_lookup_fdb(vswp, ehp);
6319 		if (port) {
6320 			/*
6321 			 * Mark the port as in-use.
6322 			 */
6323 			mutex_enter(&port->ref_lock);
6324 			port->ref_cnt++;
6325 			mutex_exit(&port->ref_lock);
6326 			RW_EXIT(&plist->lockrw);
6327 
6328 			/*
6329 			 * If plumbed and in promisc mode then copy msg
6330 			 * and send up the stack.
6331 			 */
6332 			READ_ENTER(&vswp->if_lockrw);
6333 			if (VSW_U_P(vswp->if_state)) {
6334 				RW_EXIT(&vswp->if_lockrw);
6335 				nmp = copymsg(mp);
6336 				if (nmp)
6337 					mac_rx(vswp->if_mh, mrh, nmp);
6338 			} else {
6339 				RW_EXIT(&vswp->if_lockrw);
6340 			}
6341 
6342 			/*
6343 			 * If the destination is in FDB, the packet
6344 			 * should be forwarded to the correponding
6345 			 * vsw_port (connected to a vnet device -
6346 			 * VSW_VNETPORT)
6347 			 */
6348 			(void) vsw_portsend(port, mp);
6349 
6350 			/*
6351 			 * Decrement use count in port and check if
6352 			 * should wake delete thread.
6353 			 */
6354 			mutex_enter(&port->ref_lock);
6355 			port->ref_cnt--;
6356 			if (port->ref_cnt == 0)
6357 				cv_signal(&port->ref_cv);
6358 			mutex_exit(&port->ref_lock);
6359 		} else {
6360 			RW_EXIT(&plist->lockrw);
6361 			/*
6362 			 * Destination not in FDB.
6363 			 *
6364 			 * If the destination is broadcast or
6365 			 * multicast forward the packet to all
6366 			 * (VNETPORTs, PHYSDEV, LOCALDEV),
6367 			 * except the caller.
6368 			 */
6369 			if (IS_BROADCAST(ehp)) {
6370 				D3(vswp, "%s: BROADCAST pkt", __func__);
6371 				(void) vsw_forward_all(vswp, mp,
6372 								caller, arg);
6373 			} else if (IS_MULTICAST(ehp)) {
6374 				D3(vswp, "%s: MULTICAST pkt", __func__);
6375 				(void) vsw_forward_grp(vswp, mp,
6376 							caller, arg);
6377 			} else {
6378 				/*
6379 				 * If the destination is unicast, and came
6380 				 * from either a logical network device or
6381 				 * the switch itself when it is plumbed, then
6382 				 * send it out on the physical device and also
6383 				 * up the stack if the logical interface is
6384 				 * in promiscious mode.
6385 				 *
6386 				 * NOTE:  The assumption here is that if we
6387 				 * cannot find the destination in our fdb, its
6388 				 * a unicast address, and came from either a
6389 				 * vnet or down the stack (when plumbed) it
6390 				 * must be destinded for an ethernet device
6391 				 * outside our ldoms.
6392 				 */
6393 				if (caller == VSW_VNETPORT) {
6394 					READ_ENTER(&vswp->if_lockrw);
6395 					if (VSW_U_P(vswp->if_state)) {
6396 						RW_EXIT(&vswp->if_lockrw);
6397 						nmp = copymsg(mp);
6398 						if (nmp)
6399 							mac_rx(vswp->if_mh,
6400 								mrh, nmp);
6401 					} else {
6402 						RW_EXIT(&vswp->if_lockrw);
6403 					}
6404 					if ((ret_m = vsw_tx_msg(vswp, mp))
6405 								!= NULL) {
6406 						DERR(vswp, "%s: drop mblks to "
6407 							"phys dev", __func__);
6408 						freemsg(ret_m);
6409 					}
6410 
6411 				} else if (caller == VSW_PHYSDEV) {
6412 					/*
6413 					 * Pkt seen because card in promisc
6414 					 * mode. Send up stack if plumbed in
6415 					 * promisc mode, else drop it.
6416 					 */
6417 					READ_ENTER(&vswp->if_lockrw);
6418 					if (VSW_U_P(vswp->if_state)) {
6419 						RW_EXIT(&vswp->if_lockrw);
6420 						mac_rx(vswp->if_mh, mrh, mp);
6421 					} else {
6422 						RW_EXIT(&vswp->if_lockrw);
6423 						freemsg(mp);
6424 					}
6425 
6426 				} else if (caller == VSW_LOCALDEV) {
6427 					/*
6428 					 * Pkt came down the stack, send out
6429 					 * over physical device.
6430 					 */
6431 					if ((ret_m = vsw_tx_msg(vswp, mp))
6432 								!= NULL) {
6433 						DERR(vswp, "%s: drop mblks to "
6434 							"phys dev", __func__);
6435 						freemsg(ret_m);
6436 					}
6437 				}
6438 			}
6439 		}
6440 	}
6441 	D1(vswp, "%s: exit\n", __func__);
6442 }
6443 
6444 /*
6445  * Switch ethernet frame when in layer 3 mode (i.e. using IP
6446  * layer to do the routing).
6447  *
6448  * There is a large amount of overlap between this function and
6449  * vsw_switch_l2_frame. At some stage we need to revisit and refactor
6450  * both these functions.
6451  */
6452 void
6453 vsw_switch_l3_frame(vsw_t *vswp, mblk_t *mp, int caller,
6454 			vsw_port_t *arg, mac_resource_handle_t mrh)
6455 {
6456 	struct ether_header	*ehp;
6457 	vsw_port_t		*port = NULL;
6458 	mblk_t			*bp = NULL;
6459 	vsw_port_list_t		*plist = &vswp->plist;
6460 
6461 	D1(vswp, "%s: enter (caller %d)", __func__, caller);
6462 
6463 	/*
6464 	 * In layer 3 mode should only ever be switching packets
6465 	 * between IP layer and vnet devices. So make sure thats
6466 	 * who is invoking us.
6467 	 */
6468 	if ((caller != VSW_LOCALDEV) && (caller != VSW_VNETPORT)) {
6469 		DERR(vswp, "%s: unexpected caller (%d)", __func__, caller);
6470 		freemsgchain(mp);
6471 		return;
6472 	}
6473 
6474 	/* process the chain of packets */
6475 	bp = mp;
6476 	while (bp) {
6477 		mp = bp;
6478 		bp = bp->b_next;
6479 		mp->b_next = mp->b_prev = NULL;
6480 		ehp = (struct ether_header *)mp->b_rptr;
6481 
6482 		D2(vswp, "%s: mblk data buffer %lld : actual data size %lld",
6483 			__func__, MBLKSIZE(mp), MBLKL(mp));
6484 
6485 		READ_ENTER(&plist->lockrw);
6486 		port = vsw_lookup_fdb(vswp, ehp);
6487 		if (port) {
6488 			/*
6489 			 * Mark port as in-use.
6490 			 */
6491 			mutex_enter(&port->ref_lock);
6492 			port->ref_cnt++;
6493 			mutex_exit(&port->ref_lock);
6494 			RW_EXIT(&plist->lockrw);
6495 
6496 			D2(vswp, "%s: sending to target port", __func__);
6497 			(void) vsw_portsend(port, mp);
6498 
6499 			/*
6500 			 * Finished with port so decrement ref count and
6501 			 * check if should wake delete thread.
6502 			 */
6503 			mutex_enter(&port->ref_lock);
6504 			port->ref_cnt--;
6505 			if (port->ref_cnt == 0)
6506 				cv_signal(&port->ref_cv);
6507 			mutex_exit(&port->ref_lock);
6508 		} else {
6509 			RW_EXIT(&plist->lockrw);
6510 			/*
6511 			 * Destination not in FDB
6512 			 *
6513 			 * If the destination is broadcast or
6514 			 * multicast forward the packet to all
6515 			 * (VNETPORTs, PHYSDEV, LOCALDEV),
6516 			 * except the caller.
6517 			 */
6518 			if (IS_BROADCAST(ehp)) {
6519 				D2(vswp, "%s: BROADCAST pkt", __func__);
6520 				(void) vsw_forward_all(vswp, mp,
6521 								caller, arg);
6522 			} else if (IS_MULTICAST(ehp)) {
6523 				D2(vswp, "%s: MULTICAST pkt", __func__);
6524 				(void) vsw_forward_grp(vswp, mp,
6525 							caller, arg);
6526 			} else {
6527 				/*
6528 				 * Unicast pkt from vnet that we don't have
6529 				 * an FDB entry for, so must be destinded for
6530 				 * the outside world. Attempt to send up to the
6531 				 * IP layer to allow it to deal with it.
6532 				 */
6533 				if (caller == VSW_VNETPORT) {
6534 					READ_ENTER(&vswp->if_lockrw);
6535 					if (vswp->if_state & VSW_IF_UP) {
6536 						RW_EXIT(&vswp->if_lockrw);
6537 						D2(vswp, "%s: sending up",
6538 							__func__);
6539 						mac_rx(vswp->if_mh, mrh, mp);
6540 					} else {
6541 						RW_EXIT(&vswp->if_lockrw);
6542 						/* Interface down, drop pkt */
6543 						D2(vswp, "%s I/F down",
6544 								__func__);
6545 						freemsg(mp);
6546 					}
6547 				}
6548 			}
6549 		}
6550 	}
6551 
6552 	D1(vswp, "%s: exit", __func__);
6553 }
6554 
6555 /*
6556  * Forward the ethernet frame to all ports (VNETPORTs, PHYSDEV, LOCALDEV),
6557  * except the caller (port on which frame arrived).
6558  */
6559 static int
6560 vsw_forward_all(vsw_t *vswp, mblk_t *mp, int caller, vsw_port_t *arg)
6561 {
6562 	vsw_port_list_t	*plist = &vswp->plist;
6563 	vsw_port_t	*portp;
6564 	mblk_t		*nmp = NULL;
6565 	mblk_t		*ret_m = NULL;
6566 	int		skip_port = 0;
6567 
6568 	D1(vswp, "vsw_forward_all: enter\n");
6569 
6570 	/*
6571 	 * Broadcast message from inside ldoms so send to outside
6572 	 * world if in either of layer 2 modes.
6573 	 */
6574 	if (((vswp->smode[vswp->smode_idx] == VSW_LAYER2) ||
6575 		(vswp->smode[vswp->smode_idx] == VSW_LAYER2_PROMISC)) &&
6576 		((caller == VSW_LOCALDEV) || (caller == VSW_VNETPORT))) {
6577 
6578 		nmp = dupmsg(mp);
6579 		if (nmp) {
6580 			if ((ret_m = vsw_tx_msg(vswp, nmp)) != NULL) {
6581 				DERR(vswp, "%s: dropping pkt(s) "
6582 				"consisting of %ld bytes of data for"
6583 				" physical device", __func__, MBLKL(ret_m));
6584 			freemsg(ret_m);
6585 			}
6586 		}
6587 	}
6588 
6589 	if (caller == VSW_VNETPORT)
6590 		skip_port = 1;
6591 
6592 	/*
6593 	 * Broadcast message from other vnet (layer 2 or 3) or outside
6594 	 * world (layer 2 only), send up stack if plumbed.
6595 	 */
6596 	if ((caller == VSW_PHYSDEV) || (caller == VSW_VNETPORT)) {
6597 		READ_ENTER(&vswp->if_lockrw);
6598 		if (vswp->if_state & VSW_IF_UP) {
6599 			RW_EXIT(&vswp->if_lockrw);
6600 			nmp = copymsg(mp);
6601 			if (nmp)
6602 				mac_rx(vswp->if_mh, NULL, nmp);
6603 		} else {
6604 			RW_EXIT(&vswp->if_lockrw);
6605 		}
6606 	}
6607 
6608 	/* send it to all VNETPORTs */
6609 	READ_ENTER(&plist->lockrw);
6610 	for (portp = plist->head; portp != NULL; portp = portp->p_next) {
6611 		D2(vswp, "vsw_forward_all: port %d", portp->p_instance);
6612 		/*
6613 		 * Caution ! - don't reorder these two checks as arg
6614 		 * will be NULL if the caller is PHYSDEV. skip_port is
6615 		 * only set if caller is VNETPORT.
6616 		 */
6617 		if ((skip_port) && (portp == arg))
6618 			continue;
6619 		else {
6620 			nmp = dupmsg(mp);
6621 			if (nmp) {
6622 				(void) vsw_portsend(portp, nmp);
6623 			} else {
6624 				DERR(vswp, "vsw_forward_all: nmp NULL");
6625 			}
6626 		}
6627 	}
6628 	RW_EXIT(&plist->lockrw);
6629 
6630 	freemsg(mp);
6631 
6632 	D1(vswp, "vsw_forward_all: exit\n");
6633 	return (0);
6634 }
6635 
6636 /*
6637  * Forward pkts to any devices or interfaces which have registered
6638  * an interest in them (i.e. multicast groups).
6639  */
6640 static int
6641 vsw_forward_grp(vsw_t *vswp, mblk_t *mp, int caller, vsw_port_t *arg)
6642 {
6643 	struct ether_header	*ehp = (struct ether_header *)mp->b_rptr;
6644 	mfdb_ent_t		*entp = NULL;
6645 	mfdb_ent_t		*tpp = NULL;
6646 	vsw_port_t 		*port;
6647 	uint64_t		key = 0;
6648 	mblk_t			*nmp = NULL;
6649 	mblk_t			*ret_m = NULL;
6650 	boolean_t		check_if = B_TRUE;
6651 
6652 	/*
6653 	 * Convert address to hash table key
6654 	 */
6655 	KEY_HASH(key, ehp->ether_dhost);
6656 
6657 	D1(vswp, "%s: key 0x%llx", __func__, key);
6658 
6659 	/*
6660 	 * If pkt came from either a vnet or down the stack (if we are
6661 	 * plumbed) and we are in layer 2 mode, then we send the pkt out
6662 	 * over the physical adapter, and then check to see if any other
6663 	 * vnets are interested in it.
6664 	 */
6665 	if (((vswp->smode[vswp->smode_idx] == VSW_LAYER2) ||
6666 		(vswp->smode[vswp->smode_idx] == VSW_LAYER2_PROMISC)) &&
6667 		((caller == VSW_VNETPORT) || (caller == VSW_LOCALDEV))) {
6668 		nmp = dupmsg(mp);
6669 		if (nmp) {
6670 			if ((ret_m = vsw_tx_msg(vswp, nmp)) != NULL) {
6671 				DERR(vswp, "%s: dropping pkt(s) "
6672 					"consisting of %ld bytes of "
6673 					"data for physical device",
6674 					__func__, MBLKL(ret_m));
6675 				freemsg(ret_m);
6676 			}
6677 		}
6678 	}
6679 
6680 	READ_ENTER(&vswp->mfdbrw);
6681 	if (mod_hash_find(vswp->mfdb, (mod_hash_key_t)key,
6682 				(mod_hash_val_t *)&entp) != 0) {
6683 		D3(vswp, "%s: no table entry found for addr 0x%llx",
6684 								__func__, key);
6685 	} else {
6686 		/*
6687 		 * Send to list of devices associated with this address...
6688 		 */
6689 		for (tpp = entp; tpp != NULL; tpp = tpp->nextp) {
6690 
6691 			/* dont send to ourselves */
6692 			if ((caller == VSW_VNETPORT) &&
6693 				(tpp->d_addr == (void *)arg)) {
6694 				port = (vsw_port_t *)tpp->d_addr;
6695 				D3(vswp, "%s: not sending to ourselves"
6696 					" : port %d", __func__,
6697 					port->p_instance);
6698 				continue;
6699 
6700 			} else if ((caller == VSW_LOCALDEV) &&
6701 				(tpp->d_type == VSW_LOCALDEV)) {
6702 				D3(vswp, "%s: not sending back up stack",
6703 					__func__);
6704 				continue;
6705 			}
6706 
6707 			if (tpp->d_type == VSW_VNETPORT) {
6708 				port = (vsw_port_t *)tpp->d_addr;
6709 				D3(vswp, "%s: sending to port %ld for "
6710 					" addr 0x%llx", __func__,
6711 					port->p_instance, key);
6712 
6713 				nmp = dupmsg(mp);
6714 				if (nmp)
6715 					(void) vsw_portsend(port, nmp);
6716 			} else {
6717 				if (vswp->if_state & VSW_IF_UP) {
6718 					nmp = copymsg(mp);
6719 					if (nmp)
6720 						mac_rx(vswp->if_mh, NULL, nmp);
6721 					check_if = B_FALSE;
6722 					D3(vswp, "%s: sending up stack"
6723 						" for addr 0x%llx", __func__,
6724 						key);
6725 				}
6726 			}
6727 		}
6728 	}
6729 
6730 	RW_EXIT(&vswp->mfdbrw);
6731 
6732 	/*
6733 	 * If the pkt came from either a vnet or from physical device,
6734 	 * and if we havent already sent the pkt up the stack then we
6735 	 * check now if we can/should (i.e. the interface is plumbed
6736 	 * and in promisc mode).
6737 	 */
6738 	if ((check_if) &&
6739 		((caller == VSW_VNETPORT) || (caller == VSW_PHYSDEV))) {
6740 		READ_ENTER(&vswp->if_lockrw);
6741 		if (VSW_U_P(vswp->if_state)) {
6742 			RW_EXIT(&vswp->if_lockrw);
6743 			D3(vswp, "%s: (caller %d) finally sending up stack"
6744 				" for addr 0x%llx", __func__, caller, key);
6745 			nmp = copymsg(mp);
6746 			if (nmp)
6747 				mac_rx(vswp->if_mh, NULL, nmp);
6748 		} else {
6749 			RW_EXIT(&vswp->if_lockrw);
6750 		}
6751 	}
6752 
6753 	freemsg(mp);
6754 
6755 	D1(vswp, "%s: exit", __func__);
6756 
6757 	return (0);
6758 }
6759 
6760 /* transmit the packet over the given port */
6761 static int
6762 vsw_portsend(vsw_port_t *port, mblk_t *mp)
6763 {
6764 	vsw_ldc_list_t 	*ldcl = &port->p_ldclist;
6765 	vsw_ldc_t 	*ldcp;
6766 	int		status = 0;
6767 
6768 
6769 	READ_ENTER(&ldcl->lockrw);
6770 	/*
6771 	 * Note for now, we have a single channel.
6772 	 */
6773 	ldcp = ldcl->head;
6774 	if (ldcp == NULL) {
6775 		DERR(port->p_vswp, "vsw_portsend: no ldc: dropping packet\n");
6776 		freemsg(mp);
6777 		RW_EXIT(&ldcl->lockrw);
6778 		return (1);
6779 	}
6780 
6781 	/*
6782 	 * Send the message out using the appropriate
6783 	 * transmit function which will free mblock when it
6784 	 * is finished with it.
6785 	 */
6786 	mutex_enter(&port->tx_lock);
6787 	if (port->transmit != NULL)
6788 		status = (*port->transmit)(ldcp, mp);
6789 	else {
6790 		freemsg(mp);
6791 	}
6792 	mutex_exit(&port->tx_lock);
6793 
6794 	RW_EXIT(&ldcl->lockrw);
6795 
6796 	return (status);
6797 }
6798 
6799 /*
6800  * Send packet out via descriptor ring to a logical device.
6801  */
6802 static int
6803 vsw_dringsend(vsw_ldc_t *ldcp, mblk_t *mp)
6804 {
6805 	vio_dring_msg_t		dring_pkt;
6806 	dring_info_t		*dp = NULL;
6807 	vsw_private_desc_t	*priv_desc = NULL;
6808 	vnet_public_desc_t	*pub = NULL;
6809 	vsw_t			*vswp = ldcp->ldc_vswp;
6810 	mblk_t			*bp;
6811 	size_t			n, size;
6812 	caddr_t			bufp;
6813 	int			idx;
6814 	int			status = LDC_TX_SUCCESS;
6815 
6816 	D1(vswp, "%s(%lld): enter\n", __func__, ldcp->ldc_id);
6817 
6818 	/* TODO: make test a macro */
6819 	if ((!(ldcp->lane_out.lstate & VSW_LANE_ACTIVE)) ||
6820 		(ldcp->ldc_status != LDC_UP) || (ldcp->ldc_handle == NULL)) {
6821 		DWARN(vswp, "%s(%lld) status(%d) lstate(0x%llx), dropping "
6822 			"packet\n", __func__, ldcp->ldc_id, ldcp->ldc_status,
6823 			ldcp->lane_out.lstate);
6824 		freemsg(mp);
6825 		return (LDC_TX_FAILURE);
6826 	}
6827 
6828 	/*
6829 	 * Note - using first ring only, this may change
6830 	 * in the future.
6831 	 */
6832 	READ_ENTER(&ldcp->lane_out.dlistrw);
6833 	if ((dp = ldcp->lane_out.dringp) == NULL) {
6834 		RW_EXIT(&ldcp->lane_out.dlistrw);
6835 		DERR(vswp, "%s(%lld): no dring for outbound lane on"
6836 			" channel %d", __func__, ldcp->ldc_id, ldcp->ldc_id);
6837 		freemsg(mp);
6838 		return (LDC_TX_FAILURE);
6839 	}
6840 
6841 	size = msgsize(mp);
6842 	if (size > (size_t)ETHERMAX) {
6843 		RW_EXIT(&ldcp->lane_out.dlistrw);
6844 		DERR(vswp, "%s(%lld) invalid size (%ld)\n", __func__,
6845 		    ldcp->ldc_id, size);
6846 		freemsg(mp);
6847 		return (LDC_TX_FAILURE);
6848 	}
6849 
6850 	/*
6851 	 * Find a free descriptor
6852 	 *
6853 	 * Note: for the moment we are assuming that we will only
6854 	 * have one dring going from the switch to each of its
6855 	 * peers. This may change in the future.
6856 	 */
6857 	if (vsw_dring_find_free_desc(dp, &priv_desc, &idx) != 0) {
6858 		D2(vswp, "%s(%lld): no descriptor available for ring "
6859 			"at 0x%llx", __func__, ldcp->ldc_id, dp);
6860 
6861 		/* nothing more we can do */
6862 		status = LDC_TX_NORESOURCES;
6863 		goto vsw_dringsend_free_exit;
6864 	} else {
6865 		D2(vswp, "%s(%lld): free private descriptor found at pos "
6866 			"%ld addr 0x%llx\n", __func__, ldcp->ldc_id, idx,
6867 			priv_desc);
6868 	}
6869 
6870 	/* copy data into the descriptor */
6871 	bufp = priv_desc->datap;
6872 	bufp += VNET_IPALIGN;
6873 	for (bp = mp, n = 0; bp != NULL; bp = bp->b_cont) {
6874 		n = MBLKL(bp);
6875 		bcopy(bp->b_rptr, bufp, n);
6876 		bufp += n;
6877 	}
6878 
6879 	priv_desc->datalen = (size < (size_t)ETHERMIN) ? ETHERMIN : size;
6880 
6881 	pub = priv_desc->descp;
6882 	pub->nbytes = priv_desc->datalen;
6883 
6884 	mutex_enter(&priv_desc->dstate_lock);
6885 	pub->hdr.dstate = VIO_DESC_READY;
6886 	mutex_exit(&priv_desc->dstate_lock);
6887 
6888 	/*
6889 	 * Determine whether or not we need to send a message to our
6890 	 * peer prompting them to read our newly updated descriptor(s).
6891 	 */
6892 	mutex_enter(&dp->restart_lock);
6893 	if (dp->restart_reqd) {
6894 		dp->restart_reqd = B_FALSE;
6895 		mutex_exit(&dp->restart_lock);
6896 
6897 		/*
6898 		 * Send a vio_dring_msg to peer to prompt them to read
6899 		 * the updated descriptor ring.
6900 		 */
6901 		dring_pkt.tag.vio_msgtype = VIO_TYPE_DATA;
6902 		dring_pkt.tag.vio_subtype = VIO_SUBTYPE_INFO;
6903 		dring_pkt.tag.vio_subtype_env = VIO_DRING_DATA;
6904 		dring_pkt.tag.vio_sid = ldcp->local_session;
6905 
6906 		/* Note - for now using first ring */
6907 		dring_pkt.dring_ident = dp->ident;
6908 
6909 		mutex_enter(&ldcp->lane_out.seq_lock);
6910 		dring_pkt.seq_num = ldcp->lane_out.seq_num++;
6911 		mutex_exit(&ldcp->lane_out.seq_lock);
6912 
6913 		/*
6914 		 * If last_ack_recv is -1 then we know we've not
6915 		 * received any ack's yet, so this must be the first
6916 		 * msg sent, so set the start to the begining of the ring.
6917 		 */
6918 		mutex_enter(&dp->dlock);
6919 		if (dp->last_ack_recv == -1) {
6920 			dring_pkt.start_idx = 0;
6921 		} else {
6922 			dring_pkt.start_idx = (dp->last_ack_recv + 1) %
6923 						dp->num_descriptors;
6924 		}
6925 		dring_pkt.end_idx = -1;
6926 		mutex_exit(&dp->dlock);
6927 
6928 		D3(vswp, "%s(%lld): dring 0x%llx : ident 0x%llx\n", __func__,
6929 			ldcp->ldc_id, dp, dring_pkt.dring_ident);
6930 		D3(vswp, "%s(%lld): start %lld : end %lld : seq %lld\n",
6931 			__func__, ldcp->ldc_id, dring_pkt.start_idx,
6932 			dring_pkt.end_idx, dring_pkt.seq_num);
6933 
6934 		vsw_send_msg(ldcp, (void *)&dring_pkt,
6935 						sizeof (vio_dring_msg_t));
6936 	} else {
6937 		mutex_exit(&dp->restart_lock);
6938 		D2(vswp, "%s(%lld): updating descp %d", __func__,
6939 			ldcp->ldc_id, idx);
6940 	}
6941 
6942 vsw_dringsend_free_exit:
6943 
6944 	RW_EXIT(&ldcp->lane_out.dlistrw);
6945 
6946 	/* free the message block */
6947 	freemsg(mp);
6948 
6949 	D1(vswp, "%s(%lld): exit\n", __func__, ldcp->ldc_id);
6950 	return (status);
6951 }
6952 
6953 /*
6954  * Send an in-band descriptor message over ldc.
6955  */
6956 static int
6957 vsw_descrsend(vsw_ldc_t *ldcp, mblk_t *mp)
6958 {
6959 	vsw_t			*vswp = ldcp->ldc_vswp;
6960 	vnet_ibnd_desc_t	ibnd_msg;
6961 	vsw_private_desc_t	*priv_desc = NULL;
6962 	dring_info_t		*dp = NULL;
6963 	size_t			n, size = 0;
6964 	caddr_t			bufp;
6965 	mblk_t			*bp;
6966 	int			idx, i;
6967 	int			status = LDC_TX_SUCCESS;
6968 	static int		warn_msg = 1;
6969 
6970 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
6971 
6972 	ASSERT(mp != NULL);
6973 
6974 	if ((!(ldcp->lane_out.lstate & VSW_LANE_ACTIVE)) ||
6975 		(ldcp->ldc_status != LDC_UP) || (ldcp->ldc_handle == NULL)) {
6976 		DERR(vswp, "%s(%lld) status(%d) state (0x%llx), dropping pkt",
6977 			__func__, ldcp->ldc_id, ldcp->ldc_status,
6978 			ldcp->lane_out.lstate);
6979 		freemsg(mp);
6980 		return (LDC_TX_FAILURE);
6981 	}
6982 
6983 	/*
6984 	 * only expect single dring to exist, which we use
6985 	 * as an internal buffer, rather than a transfer channel.
6986 	 */
6987 	READ_ENTER(&ldcp->lane_out.dlistrw);
6988 	if ((dp = ldcp->lane_out.dringp) == NULL) {
6989 		DERR(vswp, "%s(%lld): no dring for outbound lane",
6990 			__func__, ldcp->ldc_id);
6991 		DERR(vswp, "%s(%lld) status(%d) state (0x%llx)",
6992 			__func__, ldcp->ldc_id, ldcp->ldc_status,
6993 			ldcp->lane_out.lstate);
6994 		RW_EXIT(&ldcp->lane_out.dlistrw);
6995 		freemsg(mp);
6996 		return (LDC_TX_FAILURE);
6997 	}
6998 
6999 	size = msgsize(mp);
7000 	if (size > (size_t)ETHERMAX) {
7001 		DERR(vswp, "%s(%lld) invalid size (%ld)\n", __func__,
7002 		    ldcp->ldc_id, size);
7003 		freemsg(mp);
7004 		return (LDC_TX_FAILURE);
7005 	}
7006 
7007 	/*
7008 	 * Find a free descriptor in our buffer ring
7009 	 */
7010 	if (vsw_dring_find_free_desc(dp, &priv_desc, &idx) != 0) {
7011 		if (warn_msg) {
7012 			DERR(vswp, "%s(%lld): no descriptor available for ring "
7013 			"at 0x%llx", __func__, ldcp->ldc_id, dp);
7014 			warn_msg = 0;
7015 		}
7016 
7017 		/* nothing more we can do */
7018 		status = LDC_TX_NORESOURCES;
7019 		goto vsw_descrsend_free_exit;
7020 	} else {
7021 		D2(vswp, "%s(%lld): free private descriptor found at pos "
7022 			"%ld addr 0x%x\n", __func__, ldcp->ldc_id, idx,
7023 			priv_desc);
7024 		warn_msg = 1;
7025 	}
7026 
7027 	/* copy data into the descriptor */
7028 	bufp = priv_desc->datap;
7029 	for (bp = mp, n = 0; bp != NULL; bp = bp->b_cont) {
7030 		n = MBLKL(bp);
7031 		bcopy(bp->b_rptr, bufp, n);
7032 		bufp += n;
7033 	}
7034 
7035 	priv_desc->datalen = (size < (size_t)ETHERMIN) ? ETHERMIN : size;
7036 
7037 	/* create and send the in-band descp msg */
7038 	ibnd_msg.hdr.tag.vio_msgtype = VIO_TYPE_DATA;
7039 	ibnd_msg.hdr.tag.vio_subtype = VIO_SUBTYPE_INFO;
7040 	ibnd_msg.hdr.tag.vio_subtype_env = VIO_DESC_DATA;
7041 	ibnd_msg.hdr.tag.vio_sid = ldcp->local_session;
7042 
7043 	mutex_enter(&ldcp->lane_out.seq_lock);
7044 	ibnd_msg.hdr.seq_num = ldcp->lane_out.seq_num++;
7045 	mutex_exit(&ldcp->lane_out.seq_lock);
7046 
7047 	/*
7048 	 * Copy the mem cookies describing the data from the
7049 	 * private region of the descriptor ring into the inband
7050 	 * descriptor.
7051 	 */
7052 	for (i = 0; i < priv_desc->ncookies; i++) {
7053 		bcopy(&priv_desc->memcookie[i], &ibnd_msg.memcookie[i],
7054 			sizeof (ldc_mem_cookie_t));
7055 	}
7056 
7057 	ibnd_msg.hdr.desc_handle = idx;
7058 	ibnd_msg.ncookies = priv_desc->ncookies;
7059 	ibnd_msg.nbytes = size;
7060 
7061 	vsw_send_msg(ldcp, (void *)&ibnd_msg, sizeof (vnet_ibnd_desc_t));
7062 
7063 vsw_descrsend_free_exit:
7064 
7065 	RW_EXIT(&ldcp->lane_out.dlistrw);
7066 
7067 	/* free the allocated message blocks */
7068 	freemsg(mp);
7069 
7070 	D1(vswp, "%s(%lld): exit", __func__, ldcp->ldc_id);
7071 	return (status);
7072 }
7073 
7074 static void
7075 vsw_send_ver(void *arg)
7076 {
7077 	vsw_ldc_t	*ldcp = (vsw_ldc_t *)arg;
7078 	vsw_t		*vswp = ldcp->ldc_vswp;
7079 	lane_t		*lp = &ldcp->lane_out;
7080 	vio_ver_msg_t	ver_msg;
7081 
7082 	D1(vswp, "%s enter", __func__);
7083 
7084 	ver_msg.tag.vio_msgtype = VIO_TYPE_CTRL;
7085 	ver_msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
7086 	ver_msg.tag.vio_subtype_env = VIO_VER_INFO;
7087 	ver_msg.tag.vio_sid = ldcp->local_session;
7088 
7089 	ver_msg.ver_major = vsw_versions[0].ver_major;
7090 	ver_msg.ver_minor = vsw_versions[0].ver_minor;
7091 	ver_msg.dev_class = VDEV_NETWORK_SWITCH;
7092 
7093 	lp->lstate |= VSW_VER_INFO_SENT;
7094 	lp->ver_major = ver_msg.ver_major;
7095 	lp->ver_minor = ver_msg.ver_minor;
7096 
7097 	DUMP_TAG(ver_msg.tag);
7098 
7099 	vsw_send_msg(ldcp, &ver_msg, sizeof (vio_ver_msg_t));
7100 
7101 	D1(vswp, "%s (%d): exit", __func__, ldcp->ldc_id);
7102 }
7103 
7104 static void
7105 vsw_send_attr(vsw_ldc_t *ldcp)
7106 {
7107 	vsw_t			*vswp = ldcp->ldc_vswp;
7108 	lane_t			*lp = &ldcp->lane_out;
7109 	vnet_attr_msg_t		attr_msg;
7110 
7111 	D1(vswp, "%s (%ld) enter", __func__, ldcp->ldc_id);
7112 
7113 	/*
7114 	 * Subtype is set to INFO by default
7115 	 */
7116 	attr_msg.tag.vio_msgtype = VIO_TYPE_CTRL;
7117 	attr_msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
7118 	attr_msg.tag.vio_subtype_env = VIO_ATTR_INFO;
7119 	attr_msg.tag.vio_sid = ldcp->local_session;
7120 
7121 	/* payload copied from default settings for lane */
7122 	attr_msg.mtu = lp->mtu;
7123 	attr_msg.addr_type = lp->addr_type;
7124 	attr_msg.xfer_mode = lp->xfer_mode;
7125 	attr_msg.ack_freq = lp->xfer_mode;
7126 
7127 	READ_ENTER(&vswp->if_lockrw);
7128 	bcopy(&(vswp->if_addr), &(attr_msg.addr), ETHERADDRL);
7129 	RW_EXIT(&vswp->if_lockrw);
7130 
7131 	ldcp->lane_out.lstate |= VSW_ATTR_INFO_SENT;
7132 
7133 	DUMP_TAG(attr_msg.tag);
7134 
7135 	vsw_send_msg(ldcp, &attr_msg, sizeof (vnet_attr_msg_t));
7136 
7137 	D1(vswp, "%s (%ld) enter", __func__, ldcp->ldc_id);
7138 }
7139 
7140 /*
7141  * Create dring info msg (which also results in the creation of
7142  * a dring).
7143  */
7144 static vio_dring_reg_msg_t *
7145 vsw_create_dring_info_pkt(vsw_ldc_t *ldcp)
7146 {
7147 	vio_dring_reg_msg_t	*mp;
7148 	dring_info_t		*dp;
7149 	vsw_t			*vswp = ldcp->ldc_vswp;
7150 
7151 	D1(vswp, "vsw_create_dring_info_pkt enter\n");
7152 
7153 	/*
7154 	 * If we can't create a dring, obviously no point sending
7155 	 * a message.
7156 	 */
7157 	if ((dp = vsw_create_dring(ldcp)) == NULL)
7158 		return (NULL);
7159 
7160 	mp = kmem_zalloc(sizeof (vio_dring_reg_msg_t), KM_SLEEP);
7161 
7162 	mp->tag.vio_msgtype = VIO_TYPE_CTRL;
7163 	mp->tag.vio_subtype = VIO_SUBTYPE_INFO;
7164 	mp->tag.vio_subtype_env = VIO_DRING_REG;
7165 	mp->tag.vio_sid = ldcp->local_session;
7166 
7167 	/* payload */
7168 	mp->num_descriptors = dp->num_descriptors;
7169 	mp->descriptor_size = dp->descriptor_size;
7170 	mp->options = dp->options;
7171 	mp->ncookies = dp->ncookies;
7172 	bcopy(&dp->cookie[0], &mp->cookie[0], sizeof (ldc_mem_cookie_t));
7173 
7174 	mp->dring_ident = 0;
7175 
7176 	D1(vswp, "vsw_create_dring_info_pkt exit\n");
7177 
7178 	return (mp);
7179 }
7180 
7181 static void
7182 vsw_send_dring_info(vsw_ldc_t *ldcp)
7183 {
7184 	vio_dring_reg_msg_t	*dring_msg;
7185 	vsw_t			*vswp = ldcp->ldc_vswp;
7186 
7187 	D1(vswp, "%s: (%ld) enter", __func__, ldcp->ldc_id);
7188 
7189 	dring_msg = vsw_create_dring_info_pkt(ldcp);
7190 	if (dring_msg == NULL) {
7191 		cmn_err(CE_WARN, "!vsw%d: %s: error creating msg",
7192 			vswp->instance, __func__);
7193 		return;
7194 	}
7195 
7196 	ldcp->lane_out.lstate |= VSW_DRING_INFO_SENT;
7197 
7198 	DUMP_TAG_PTR((vio_msg_tag_t *)dring_msg);
7199 
7200 	vsw_send_msg(ldcp, dring_msg,
7201 		sizeof (vio_dring_reg_msg_t));
7202 
7203 	kmem_free(dring_msg, sizeof (vio_dring_reg_msg_t));
7204 
7205 	D1(vswp, "%s: (%ld) exit", __func__, ldcp->ldc_id);
7206 }
7207 
7208 static void
7209 vsw_send_rdx(vsw_ldc_t *ldcp)
7210 {
7211 	vsw_t		*vswp = ldcp->ldc_vswp;
7212 	vio_rdx_msg_t	rdx_msg;
7213 
7214 	D1(vswp, "%s (%ld) enter", __func__, ldcp->ldc_id);
7215 
7216 	rdx_msg.tag.vio_msgtype = VIO_TYPE_CTRL;
7217 	rdx_msg.tag.vio_subtype = VIO_SUBTYPE_INFO;
7218 	rdx_msg.tag.vio_subtype_env = VIO_RDX;
7219 	rdx_msg.tag.vio_sid = ldcp->local_session;
7220 
7221 	ldcp->lane_out.lstate |= VSW_RDX_INFO_SENT;
7222 
7223 	DUMP_TAG(rdx_msg.tag);
7224 
7225 	vsw_send_msg(ldcp, &rdx_msg, sizeof (vio_rdx_msg_t));
7226 
7227 	D1(vswp, "%s (%ld) exit", __func__, ldcp->ldc_id);
7228 }
7229 
7230 /*
7231  * Generic routine to send message out over ldc channel.
7232  */
7233 static void
7234 vsw_send_msg(vsw_ldc_t *ldcp, void *msgp, int size)
7235 {
7236 	int		rv;
7237 	size_t		msglen = size;
7238 	vio_msg_tag_t	*tag = (vio_msg_tag_t *)msgp;
7239 	vsw_t		*vswp = ldcp->ldc_vswp;
7240 
7241 	D1(vswp, "vsw_send_msg (%lld) enter : sending %d bytes",
7242 			ldcp->ldc_id, size);
7243 
7244 	D2(vswp, "send_msg: type 0x%llx", tag->vio_msgtype);
7245 	D2(vswp, "send_msg: stype 0x%llx", tag->vio_subtype);
7246 	D2(vswp, "send_msg: senv 0x%llx", tag->vio_subtype_env);
7247 
7248 	mutex_enter(&ldcp->ldc_txlock);
7249 	do {
7250 		msglen = size;
7251 		rv = ldc_write(ldcp->ldc_handle, (caddr_t)msgp, &msglen);
7252 	} while (rv == EWOULDBLOCK && --vsw_wretries > 0);
7253 
7254 	if ((rv != 0) || (msglen != size)) {
7255 		DERR(vswp, "vsw_send_msg:ldc_write failed: chan(%lld) "
7256 			"rv(%d) size (%d) msglen(%d)\n", ldcp->ldc_id,
7257 			rv, size, msglen);
7258 	}
7259 	mutex_exit(&ldcp->ldc_txlock);
7260 
7261 	/* channel has been reset */
7262 	if (rv == ECONNRESET) {
7263 		vsw_handle_reset(ldcp);
7264 	}
7265 
7266 	D1(vswp, "vsw_send_msg (%lld) exit : sent %d bytes",
7267 			ldcp->ldc_id, msglen);
7268 }
7269 
7270 /*
7271  * Add an entry into FDB, for the given mac address and port_id.
7272  * Returns 0 on success, 1 on failure.
7273  *
7274  * Lock protecting FDB must be held by calling process.
7275  */
7276 static int
7277 vsw_add_fdb(vsw_t *vswp, vsw_port_t *port)
7278 {
7279 	uint64_t	addr = 0;
7280 
7281 	D1(vswp, "%s: enter", __func__);
7282 
7283 	KEY_HASH(addr, port->p_macaddr);
7284 
7285 	D2(vswp, "%s: key = 0x%llx", __func__, addr);
7286 
7287 	/*
7288 	 * Note: duplicate keys will be rejected by mod_hash.
7289 	 */
7290 	if (mod_hash_insert(vswp->fdb, (mod_hash_key_t)addr,
7291 				(mod_hash_val_t)port) != 0) {
7292 		DERR(vswp, "%s: unable to add entry into fdb.", __func__);
7293 		return (1);
7294 	}
7295 
7296 	D1(vswp, "%s: exit", __func__);
7297 	return (0);
7298 }
7299 
7300 /*
7301  * Remove an entry from FDB.
7302  * Returns 0 on success, 1 on failure.
7303  */
7304 static int
7305 vsw_del_fdb(vsw_t *vswp, vsw_port_t *port)
7306 {
7307 	uint64_t	addr = 0;
7308 
7309 	D1(vswp, "%s: enter", __func__);
7310 
7311 	KEY_HASH(addr, port->p_macaddr);
7312 
7313 	D2(vswp, "%s: key = 0x%llx", __func__, addr);
7314 
7315 	(void) mod_hash_destroy(vswp->fdb, (mod_hash_val_t)addr);
7316 
7317 	D1(vswp, "%s: enter", __func__);
7318 
7319 	return (0);
7320 }
7321 
7322 /*
7323  * Search fdb for a given mac address.
7324  * Returns pointer to the entry if found, else returns NULL.
7325  */
7326 static vsw_port_t *
7327 vsw_lookup_fdb(vsw_t *vswp, struct ether_header *ehp)
7328 {
7329 	uint64_t	key = 0;
7330 	vsw_port_t	*port = NULL;
7331 
7332 	D1(vswp, "%s: enter", __func__);
7333 
7334 	KEY_HASH(key, ehp->ether_dhost);
7335 
7336 	D2(vswp, "%s: key = 0x%llx", __func__, key);
7337 
7338 	if (mod_hash_find(vswp->fdb, (mod_hash_key_t)key,
7339 				(mod_hash_val_t *)&port) != 0) {
7340 		D2(vswp, "%s: no port found", __func__);
7341 		return (NULL);
7342 	}
7343 
7344 	D1(vswp, "%s: exit", __func__);
7345 
7346 	return (port);
7347 }
7348 
7349 /*
7350  * Add or remove multicast address(es).
7351  *
7352  * Returns 0 on success, 1 on failure.
7353  */
7354 static int
7355 vsw_add_rem_mcst(vnet_mcast_msg_t *mcst_pkt, vsw_port_t *port)
7356 {
7357 	mcst_addr_t		*mcst_p = NULL;
7358 	vsw_t			*vswp = port->p_vswp;
7359 	uint64_t		addr = 0x0;
7360 	int			i;
7361 
7362 	D1(vswp, "%s: enter", __func__);
7363 
7364 	D2(vswp, "%s: %d addresses", __func__, mcst_pkt->count);
7365 
7366 	mutex_enter(&vswp->mac_lock);
7367 	if (vswp->mh == NULL) {
7368 		mutex_exit(&vswp->mac_lock);
7369 		return (1);
7370 	}
7371 	mutex_exit(&vswp->mac_lock);
7372 
7373 	for (i = 0; i < mcst_pkt->count; i++) {
7374 		/*
7375 		 * Convert address into form that can be used
7376 		 * as hash table key.
7377 		 */
7378 		KEY_HASH(addr, mcst_pkt->mca[i]);
7379 
7380 		/*
7381 		 * Add or delete the specified address/port combination.
7382 		 */
7383 		if (mcst_pkt->set == 0x1) {
7384 			D3(vswp, "%s: adding multicast address 0x%llx for "
7385 				"port %ld", __func__, addr, port->p_instance);
7386 			if (vsw_add_mcst(vswp, VSW_VNETPORT, addr, port) == 0) {
7387 				/*
7388 				 * Update the list of multicast
7389 				 * addresses contained within the
7390 				 * port structure to include this new
7391 				 * one.
7392 				 */
7393 				mcst_p = kmem_alloc(sizeof (mcst_addr_t),
7394 								KM_NOSLEEP);
7395 				if (mcst_p == NULL) {
7396 					DERR(vswp, "%s: unable to alloc mem",
7397 						__func__);
7398 					return (1);
7399 				}
7400 
7401 				mcst_p->nextp = NULL;
7402 				mcst_p->addr = addr;
7403 
7404 				mutex_enter(&port->mca_lock);
7405 				mcst_p->nextp = port->mcap;
7406 				port->mcap = mcst_p;
7407 				mutex_exit(&port->mca_lock);
7408 
7409 				/*
7410 				 * Program the address into HW. If the addr
7411 				 * has already been programmed then the MAC
7412 				 * just increments a ref counter (which is
7413 				 * used when the address is being deleted)
7414 				 */
7415 				mutex_enter(&vswp->mac_lock);
7416 				if ((vswp->mh == NULL) ||
7417 					mac_multicst_add(vswp->mh,
7418 						(uchar_t *)&mcst_pkt->mca[i])) {
7419 					mutex_exit(&vswp->mac_lock);
7420 					cmn_err(CE_WARN, "!vsw%d: unable to "
7421 						"add multicast address",
7422 						vswp->instance);
7423 					(void) vsw_del_mcst(vswp, VSW_VNETPORT,
7424 						addr, port);
7425 					vsw_del_addr(VSW_VNETPORT, port, addr);
7426 					return (1);
7427 				}
7428 				mutex_exit(&vswp->mac_lock);
7429 
7430 			} else {
7431 				DERR(vswp, "%s: error adding multicast "
7432 					"address 0x%llx for port %ld",
7433 					__func__, addr, port->p_instance);
7434 				return (1);
7435 			}
7436 		} else {
7437 			/*
7438 			 * Delete an entry from the multicast hash
7439 			 * table and update the address list
7440 			 * appropriately.
7441 			 */
7442 			if (vsw_del_mcst(vswp, VSW_VNETPORT, addr, port) == 0) {
7443 				D3(vswp, "%s: deleting multicast address "
7444 					"0x%llx for port %ld", __func__, addr,
7445 					port->p_instance);
7446 
7447 				vsw_del_addr(VSW_VNETPORT, port, addr);
7448 
7449 				/*
7450 				 * Remove the address from HW. The address
7451 				 * will actually only be removed once the ref
7452 				 * count within the MAC layer has dropped to
7453 				 * zero. I.e. we can safely call this fn even
7454 				 * if other ports are interested in this
7455 				 * address.
7456 				 */
7457 				mutex_enter(&vswp->mac_lock);
7458 				if ((vswp->mh == NULL) ||
7459 					mac_multicst_remove(vswp->mh,
7460 						(uchar_t *)&mcst_pkt->mca[i])) {
7461 					mutex_exit(&vswp->mac_lock);
7462 					cmn_err(CE_WARN, "!vsw%d: unable to "
7463 						"remove multicast address",
7464 						vswp->instance);
7465 					return (1);
7466 				}
7467 				mutex_exit(&vswp->mac_lock);
7468 
7469 			} else {
7470 				DERR(vswp, "%s: error deleting multicast "
7471 					"addr 0x%llx for port %ld",
7472 					__func__, addr, port->p_instance);
7473 				return (1);
7474 			}
7475 		}
7476 	}
7477 	D1(vswp, "%s: exit", __func__);
7478 	return (0);
7479 }
7480 
7481 /*
7482  * Add a new multicast entry.
7483  *
7484  * Search hash table based on address. If match found then
7485  * update associated val (which is chain of ports), otherwise
7486  * create new key/val (addr/port) pair and insert into table.
7487  */
7488 static int
7489 vsw_add_mcst(vsw_t *vswp, uint8_t devtype, uint64_t addr, void *arg)
7490 {
7491 	int		dup = 0;
7492 	int		rv = 0;
7493 	mfdb_ent_t	*ment = NULL;
7494 	mfdb_ent_t	*tmp_ent = NULL;
7495 	mfdb_ent_t	*new_ent = NULL;
7496 	void		*tgt = NULL;
7497 
7498 	if (devtype == VSW_VNETPORT) {
7499 		/*
7500 		 * Being invoked from a vnet.
7501 		 */
7502 		ASSERT(arg != NULL);
7503 		tgt = arg;
7504 		D2(NULL, "%s: port %d : address 0x%llx", __func__,
7505 			((vsw_port_t *)arg)->p_instance, addr);
7506 	} else {
7507 		/*
7508 		 * We are being invoked via the m_multicst mac entry
7509 		 * point.
7510 		 */
7511 		D2(NULL, "%s: address 0x%llx", __func__, addr);
7512 		tgt = (void *)vswp;
7513 	}
7514 
7515 	WRITE_ENTER(&vswp->mfdbrw);
7516 	if (mod_hash_find(vswp->mfdb, (mod_hash_key_t)addr,
7517 				(mod_hash_val_t *)&ment) != 0) {
7518 
7519 		/* address not currently in table */
7520 		ment = kmem_alloc(sizeof (mfdb_ent_t), KM_SLEEP);
7521 		ment->d_addr = (void *)tgt;
7522 		ment->d_type = devtype;
7523 		ment->nextp = NULL;
7524 
7525 		if (mod_hash_insert(vswp->mfdb, (mod_hash_key_t)addr,
7526 			(mod_hash_val_t)ment) != 0) {
7527 			DERR(vswp, "%s: hash table insertion failed", __func__);
7528 			kmem_free(ment, sizeof (mfdb_ent_t));
7529 			rv = 1;
7530 		} else {
7531 			D2(vswp, "%s: added initial entry for 0x%llx to "
7532 				"table", __func__, addr);
7533 		}
7534 	} else {
7535 		/*
7536 		 * Address in table. Check to see if specified port
7537 		 * is already associated with the address. If not add
7538 		 * it now.
7539 		 */
7540 		tmp_ent = ment;
7541 		while (tmp_ent != NULL) {
7542 			if (tmp_ent->d_addr == (void *)tgt) {
7543 				if (devtype == VSW_VNETPORT) {
7544 					DERR(vswp, "%s: duplicate port entry "
7545 						"found for portid %ld and key "
7546 						"0x%llx", __func__,
7547 						((vsw_port_t *)arg)->p_instance,
7548 						addr);
7549 				} else {
7550 					DERR(vswp, "%s: duplicate entry found"
7551 						"for key 0x%llx",
7552 						__func__, addr);
7553 				}
7554 				rv = 1;
7555 				dup = 1;
7556 				break;
7557 			}
7558 			tmp_ent = tmp_ent->nextp;
7559 		}
7560 
7561 		/*
7562 		 * Port not on list so add it to end now.
7563 		 */
7564 		if (0 == dup) {
7565 			D2(vswp, "%s: added entry for 0x%llx to table",
7566 				__func__, addr);
7567 			new_ent = kmem_alloc(sizeof (mfdb_ent_t), KM_SLEEP);
7568 			new_ent->d_addr = (void *)tgt;
7569 			new_ent->d_type = devtype;
7570 			new_ent->nextp = NULL;
7571 
7572 			tmp_ent = ment;
7573 			while (tmp_ent->nextp != NULL)
7574 				tmp_ent = tmp_ent->nextp;
7575 
7576 			tmp_ent->nextp = new_ent;
7577 		}
7578 	}
7579 
7580 	RW_EXIT(&vswp->mfdbrw);
7581 	return (rv);
7582 }
7583 
7584 /*
7585  * Remove a multicast entry from the hashtable.
7586  *
7587  * Search hash table based on address. If match found, scan
7588  * list of ports associated with address. If specified port
7589  * found remove it from list.
7590  */
7591 static int
7592 vsw_del_mcst(vsw_t *vswp, uint8_t devtype, uint64_t addr, void *arg)
7593 {
7594 	mfdb_ent_t	*ment = NULL;
7595 	mfdb_ent_t	*curr_p, *prev_p;
7596 	void		*tgt = NULL;
7597 
7598 	D1(vswp, "%s: enter", __func__);
7599 
7600 	if (devtype == VSW_VNETPORT) {
7601 		tgt = (vsw_port_t *)arg;
7602 		D2(vswp, "%s: removing port %d from mFDB for address"
7603 			" 0x%llx", __func__, ((vsw_port_t *)tgt)->p_instance,
7604 			addr);
7605 	} else {
7606 		D2(vswp, "%s: removing entry", __func__);
7607 		tgt = (void *)vswp;
7608 	}
7609 
7610 	WRITE_ENTER(&vswp->mfdbrw);
7611 	if (mod_hash_find(vswp->mfdb, (mod_hash_key_t)addr,
7612 				(mod_hash_val_t *)&ment) != 0) {
7613 		D2(vswp, "%s: address 0x%llx not in table", __func__, addr);
7614 		RW_EXIT(&vswp->mfdbrw);
7615 		return (1);
7616 	}
7617 
7618 	prev_p = curr_p = ment;
7619 
7620 	while (curr_p != NULL) {
7621 		if (curr_p->d_addr == (void *)tgt) {
7622 			if (devtype == VSW_VNETPORT) {
7623 				D2(vswp, "%s: port %d found", __func__,
7624 					((vsw_port_t *)tgt)->p_instance);
7625 			} else {
7626 				D2(vswp, "%s: instance found", __func__);
7627 			}
7628 
7629 			if (prev_p == curr_p) {
7630 				/*
7631 				 * head of list, if no other element is in
7632 				 * list then destroy this entry, otherwise
7633 				 * just replace it with updated value.
7634 				 */
7635 				ment = curr_p->nextp;
7636 				kmem_free(curr_p, sizeof (mfdb_ent_t));
7637 				if (ment == NULL) {
7638 					(void) mod_hash_destroy(vswp->mfdb,
7639 							(mod_hash_val_t)addr);
7640 				} else {
7641 					(void) mod_hash_replace(vswp->mfdb,
7642 							(mod_hash_key_t)addr,
7643 							(mod_hash_val_t)ment);
7644 				}
7645 			} else {
7646 				/*
7647 				 * Not head of list, no need to do
7648 				 * replacement, just adjust list pointers.
7649 				 */
7650 				prev_p->nextp = curr_p->nextp;
7651 				kmem_free(curr_p, sizeof (mfdb_ent_t));
7652 			}
7653 			break;
7654 		}
7655 
7656 		prev_p = curr_p;
7657 		curr_p = curr_p->nextp;
7658 	}
7659 
7660 	RW_EXIT(&vswp->mfdbrw);
7661 
7662 	D1(vswp, "%s: exit", __func__);
7663 
7664 	return (0);
7665 }
7666 
7667 /*
7668  * Port is being deleted, but has registered an interest in one
7669  * or more multicast groups. Using the list of addresses maintained
7670  * within the port structure find the appropriate entry in the hash
7671  * table and remove this port from the list of interested ports.
7672  */
7673 static void
7674 vsw_del_mcst_port(vsw_port_t *port)
7675 {
7676 	mcst_addr_t	*mcst_p = NULL;
7677 	vsw_t		*vswp = port->p_vswp;
7678 
7679 	D1(vswp, "%s: enter", __func__);
7680 
7681 	mutex_enter(&port->mca_lock);
7682 	while (port->mcap != NULL) {
7683 		(void) vsw_del_mcst(vswp, VSW_VNETPORT,
7684 					port->mcap->addr, port);
7685 
7686 		mcst_p = port->mcap->nextp;
7687 		kmem_free(port->mcap, sizeof (mcst_addr_t));
7688 		port->mcap = mcst_p;
7689 	}
7690 	mutex_exit(&port->mca_lock);
7691 
7692 	D1(vswp, "%s: exit", __func__);
7693 }
7694 
7695 /*
7696  * This vsw instance is detaching, but has registered an interest in one
7697  * or more multicast groups. Using the list of addresses maintained
7698  * within the vsw structure find the appropriate entry in the hash
7699  * table and remove this instance from the list of interested ports.
7700  */
7701 static void
7702 vsw_del_mcst_vsw(vsw_t *vswp)
7703 {
7704 	mcst_addr_t	*next_p = NULL;
7705 
7706 	D1(vswp, "%s: enter", __func__);
7707 
7708 	mutex_enter(&vswp->mca_lock);
7709 
7710 	while (vswp->mcap != NULL) {
7711 		DERR(vswp, "%s: deleting addr 0x%llx",
7712 			__func__, vswp->mcap->addr);
7713 		(void) vsw_del_mcst(vswp, VSW_LOCALDEV,
7714 				vswp->mcap->addr, NULL);
7715 
7716 		next_p = vswp->mcap->nextp;
7717 		kmem_free(vswp->mcap, sizeof (mcst_addr_t));
7718 		vswp->mcap = next_p;
7719 	}
7720 
7721 	vswp->mcap = NULL;
7722 	mutex_exit(&vswp->mca_lock);
7723 
7724 	D1(vswp, "%s: exit", __func__);
7725 }
7726 
7727 
7728 /*
7729  * Remove the specified address from the list of address maintained
7730  * in this port node.
7731  */
7732 static void
7733 vsw_del_addr(uint8_t devtype, void *arg, uint64_t addr)
7734 {
7735 	vsw_t		*vswp = NULL;
7736 	vsw_port_t	*port = NULL;
7737 	mcst_addr_t	*prev_p = NULL;
7738 	mcst_addr_t	*curr_p = NULL;
7739 
7740 	D1(NULL, "%s: enter : devtype %d : addr 0x%llx",
7741 		__func__, devtype, addr);
7742 
7743 	if (devtype == VSW_VNETPORT) {
7744 		port = (vsw_port_t *)arg;
7745 		mutex_enter(&port->mca_lock);
7746 		prev_p = curr_p = port->mcap;
7747 	} else {
7748 		vswp = (vsw_t *)arg;
7749 		mutex_enter(&vswp->mca_lock);
7750 		prev_p = curr_p = vswp->mcap;
7751 	}
7752 
7753 	while (curr_p != NULL) {
7754 		if (curr_p->addr == addr) {
7755 			D2(NULL, "%s: address found", __func__);
7756 			/* match found */
7757 			if (prev_p == curr_p) {
7758 				/* list head */
7759 				if (devtype == VSW_VNETPORT)
7760 					port->mcap = curr_p->nextp;
7761 				else
7762 					vswp->mcap = curr_p->nextp;
7763 			} else {
7764 				prev_p->nextp = curr_p->nextp;
7765 			}
7766 			kmem_free(curr_p, sizeof (mcst_addr_t));
7767 			break;
7768 		} else {
7769 			prev_p = curr_p;
7770 			curr_p = curr_p->nextp;
7771 		}
7772 	}
7773 
7774 	if (devtype == VSW_VNETPORT)
7775 		mutex_exit(&port->mca_lock);
7776 	else
7777 		mutex_exit(&vswp->mca_lock);
7778 
7779 	D1(NULL, "%s: exit", __func__);
7780 }
7781 
7782 /*
7783  * Creates a descriptor ring (dring) and links it into the
7784  * link of outbound drings for this channel.
7785  *
7786  * Returns NULL if creation failed.
7787  */
7788 static dring_info_t *
7789 vsw_create_dring(vsw_ldc_t *ldcp)
7790 {
7791 	vsw_private_desc_t	*priv_addr = NULL;
7792 	vsw_t			*vswp = ldcp->ldc_vswp;
7793 	ldc_mem_info_t		minfo;
7794 	dring_info_t		*dp, *tp;
7795 	int			i;
7796 
7797 	dp = (dring_info_t *)kmem_zalloc(sizeof (dring_info_t), KM_SLEEP);
7798 
7799 	mutex_init(&dp->dlock, NULL, MUTEX_DRIVER, NULL);
7800 
7801 	/* create public section of ring */
7802 	if ((ldc_mem_dring_create(VSW_RING_NUM_EL,
7803 			VSW_PUB_SIZE, &dp->handle)) != 0) {
7804 
7805 		DERR(vswp, "vsw_create_dring(%lld): ldc dring create "
7806 			"failed", ldcp->ldc_id);
7807 		goto create_fail_exit;
7808 	}
7809 
7810 	ASSERT(dp->handle != NULL);
7811 
7812 	/*
7813 	 * Get the base address of the public section of the ring.
7814 	 */
7815 	if ((ldc_mem_dring_info(dp->handle, &minfo)) != 0) {
7816 		DERR(vswp, "vsw_create_dring(%lld): dring info failed\n",
7817 			ldcp->ldc_id);
7818 		goto dring_fail_exit;
7819 	} else {
7820 		ASSERT(minfo.vaddr != 0);
7821 		dp->pub_addr = minfo.vaddr;
7822 	}
7823 
7824 	dp->num_descriptors = VSW_RING_NUM_EL;
7825 	dp->descriptor_size = VSW_PUB_SIZE;
7826 	dp->options = VIO_TX_DRING;
7827 	dp->ncookies = 1;	/* guaranteed by ldc */
7828 
7829 	/*
7830 	 * create private portion of ring
7831 	 */
7832 	dp->priv_addr = (vsw_private_desc_t *)kmem_zalloc(
7833 		(sizeof (vsw_private_desc_t) * VSW_RING_NUM_EL), KM_SLEEP);
7834 
7835 	if (vsw_setup_ring(ldcp, dp)) {
7836 		DERR(vswp, "%s: unable to setup ring", __func__);
7837 		goto dring_fail_exit;
7838 	}
7839 
7840 	/* haven't used any descriptors yet */
7841 	dp->end_idx = 0;
7842 	dp->last_ack_recv = -1;
7843 
7844 	/* bind dring to the channel */
7845 	if ((ldc_mem_dring_bind(ldcp->ldc_handle, dp->handle,
7846 		LDC_SHADOW_MAP, LDC_MEM_RW,
7847 		&dp->cookie[0], &dp->ncookies)) != 0) {
7848 		DERR(vswp, "vsw_create_dring: unable to bind to channel "
7849 			"%lld", ldcp->ldc_id);
7850 		goto dring_fail_exit;
7851 	}
7852 
7853 	mutex_init(&dp->restart_lock, NULL, MUTEX_DRIVER, NULL);
7854 	dp->restart_reqd = B_TRUE;
7855 
7856 	/*
7857 	 * Only ever create rings for outgoing lane. Link it onto
7858 	 * end of list.
7859 	 */
7860 	WRITE_ENTER(&ldcp->lane_out.dlistrw);
7861 	if (ldcp->lane_out.dringp == NULL) {
7862 		D2(vswp, "vsw_create_dring: adding first outbound ring");
7863 		ldcp->lane_out.dringp = dp;
7864 	} else {
7865 		tp = ldcp->lane_out.dringp;
7866 		while (tp->next != NULL)
7867 			tp = tp->next;
7868 
7869 		tp->next = dp;
7870 	}
7871 	RW_EXIT(&ldcp->lane_out.dlistrw);
7872 
7873 	return (dp);
7874 
7875 dring_fail_exit:
7876 	(void) ldc_mem_dring_destroy(dp->handle);
7877 
7878 create_fail_exit:
7879 	if (dp->priv_addr != NULL) {
7880 		priv_addr = dp->priv_addr;
7881 		for (i = 0; i < VSW_RING_NUM_EL; i++) {
7882 			if (priv_addr->memhandle != NULL)
7883 				(void) ldc_mem_free_handle(
7884 						priv_addr->memhandle);
7885 			priv_addr++;
7886 		}
7887 		kmem_free(dp->priv_addr,
7888 			(sizeof (vsw_private_desc_t) * VSW_RING_NUM_EL));
7889 	}
7890 	mutex_destroy(&dp->dlock);
7891 
7892 	kmem_free(dp, sizeof (dring_info_t));
7893 	return (NULL);
7894 }
7895 
7896 /*
7897  * Create a ring consisting of just a private portion and link
7898  * it into the list of rings for the outbound lane.
7899  *
7900  * These type of rings are used primarily for temporary data
7901  * storage (i.e. as data buffers).
7902  */
7903 void
7904 vsw_create_privring(vsw_ldc_t *ldcp)
7905 {
7906 	dring_info_t		*dp, *tp;
7907 	vsw_t			*vswp = ldcp->ldc_vswp;
7908 
7909 	D1(vswp, "%s(%lld): enter", __func__, ldcp->ldc_id);
7910 
7911 	dp = kmem_zalloc(sizeof (dring_info_t), KM_SLEEP);
7912 
7913 	mutex_init(&dp->dlock, NULL, MUTEX_DRIVER, NULL);
7914 
7915 	/* no public section */
7916 	dp->pub_addr = NULL;
7917 
7918 	dp->priv_addr = kmem_zalloc((sizeof (vsw_private_desc_t) *
7919 					VSW_RING_NUM_EL), KM_SLEEP);
7920 
7921 	dp->num_descriptors = VSW_RING_NUM_EL;
7922 
7923 	if (vsw_setup_ring(ldcp, dp)) {
7924 		DERR(vswp, "%s: setup of ring failed", __func__);
7925 		kmem_free(dp->priv_addr,
7926 			(sizeof (vsw_private_desc_t) * VSW_RING_NUM_EL));
7927 		mutex_destroy(&dp->dlock);
7928 		kmem_free(dp, sizeof (dring_info_t));
7929 		return;
7930 	}
7931 
7932 	/* haven't used any descriptors yet */
7933 	dp->end_idx = 0;
7934 
7935 	mutex_init(&dp->restart_lock, NULL, MUTEX_DRIVER, NULL);
7936 	dp->restart_reqd = B_TRUE;
7937 
7938 	/*
7939 	 * Only ever create rings for outgoing lane. Link it onto
7940 	 * end of list.
7941 	 */
7942 	WRITE_ENTER(&ldcp->lane_out.dlistrw);
7943 	if (ldcp->lane_out.dringp == NULL) {
7944 		D2(vswp, "%s: adding first outbound privring", __func__);
7945 		ldcp->lane_out.dringp = dp;
7946 	} else {
7947 		tp = ldcp->lane_out.dringp;
7948 		while (tp->next != NULL)
7949 			tp = tp->next;
7950 
7951 		tp->next = dp;
7952 	}
7953 	RW_EXIT(&ldcp->lane_out.dlistrw);
7954 
7955 	D1(vswp, "%s(%lld): exit", __func__, ldcp->ldc_id);
7956 }
7957 
7958 /*
7959  * Setup the descriptors in the dring. Returns 0 on success, 1 on
7960  * failure.
7961  */
7962 int
7963 vsw_setup_ring(vsw_ldc_t *ldcp, dring_info_t *dp)
7964 {
7965 	vnet_public_desc_t	*pub_addr = NULL;
7966 	vsw_private_desc_t	*priv_addr = NULL;
7967 	vsw_t			*vswp = ldcp->ldc_vswp;
7968 	uint64_t		*tmpp;
7969 	uint64_t		offset = 0;
7970 	uint32_t		ncookies = 0;
7971 	static char		*name = "vsw_setup_ring";
7972 	int			i, j, nc, rv;
7973 
7974 	priv_addr = dp->priv_addr;
7975 	pub_addr = dp->pub_addr;
7976 
7977 	/* public section may be null but private should never be */
7978 	ASSERT(priv_addr != NULL);
7979 
7980 	/*
7981 	 * Allocate the region of memory which will be used to hold
7982 	 * the data the descriptors will refer to.
7983 	 */
7984 	dp->data_sz = (VSW_RING_NUM_EL * VSW_RING_EL_DATA_SZ);
7985 	dp->data_addr = kmem_alloc(dp->data_sz, KM_SLEEP);
7986 
7987 	D2(vswp, "%s: allocated %lld bytes at 0x%llx\n", name,
7988 		dp->data_sz, dp->data_addr);
7989 
7990 	tmpp = (uint64_t *)dp->data_addr;
7991 	offset = VSW_RING_EL_DATA_SZ / sizeof (tmpp);
7992 
7993 	/*
7994 	 * Initialise some of the private and public (if they exist)
7995 	 * descriptor fields.
7996 	 */
7997 	for (i = 0; i < VSW_RING_NUM_EL; i++) {
7998 		mutex_init(&priv_addr->dstate_lock, NULL, MUTEX_DRIVER, NULL);
7999 
8000 		if ((ldc_mem_alloc_handle(ldcp->ldc_handle,
8001 			&priv_addr->memhandle)) != 0) {
8002 			DERR(vswp, "%s: alloc mem handle failed", name);
8003 			goto setup_ring_cleanup;
8004 		}
8005 
8006 		priv_addr->datap = (void *)tmpp;
8007 
8008 		rv = ldc_mem_bind_handle(priv_addr->memhandle,
8009 			(caddr_t)priv_addr->datap, VSW_RING_EL_DATA_SZ,
8010 			LDC_SHADOW_MAP, LDC_MEM_R|LDC_MEM_W,
8011 			&(priv_addr->memcookie[0]), &ncookies);
8012 		if (rv != 0) {
8013 			DERR(vswp, "%s(%lld): ldc_mem_bind_handle failed "
8014 				"(rv %d)", name, ldcp->ldc_id, rv);
8015 			goto setup_ring_cleanup;
8016 		}
8017 		priv_addr->bound = 1;
8018 
8019 		D2(vswp, "%s: %d: memcookie 0 : addr 0x%llx : size 0x%llx",
8020 			name, i, priv_addr->memcookie[0].addr,
8021 			priv_addr->memcookie[0].size);
8022 
8023 		if (ncookies >= (uint32_t)(VSW_MAX_COOKIES + 1)) {
8024 			DERR(vswp, "%s(%lld) ldc_mem_bind_handle returned "
8025 				"invalid num of cookies (%d) for size 0x%llx",
8026 				name, ldcp->ldc_id, ncookies,
8027 				VSW_RING_EL_DATA_SZ);
8028 
8029 			goto setup_ring_cleanup;
8030 		} else {
8031 			for (j = 1; j < ncookies; j++) {
8032 				rv = ldc_mem_nextcookie(priv_addr->memhandle,
8033 					&(priv_addr->memcookie[j]));
8034 				if (rv != 0) {
8035 					DERR(vswp, "%s: ldc_mem_nextcookie "
8036 						"failed rv (%d)", name, rv);
8037 					goto setup_ring_cleanup;
8038 				}
8039 				D3(vswp, "%s: memcookie %d : addr 0x%llx : "
8040 					"size 0x%llx", name, j,
8041 					priv_addr->memcookie[j].addr,
8042 					priv_addr->memcookie[j].size);
8043 			}
8044 
8045 		}
8046 		priv_addr->ncookies = ncookies;
8047 		priv_addr->dstate = VIO_DESC_FREE;
8048 
8049 		if (pub_addr != NULL) {
8050 
8051 			/* link pub and private sides */
8052 			priv_addr->descp = pub_addr;
8053 
8054 			pub_addr->ncookies = priv_addr->ncookies;
8055 
8056 			for (nc = 0; nc < pub_addr->ncookies; nc++) {
8057 				bcopy(&priv_addr->memcookie[nc],
8058 					&pub_addr->memcookie[nc],
8059 					sizeof (ldc_mem_cookie_t));
8060 			}
8061 
8062 			pub_addr->hdr.dstate = VIO_DESC_FREE;
8063 			pub_addr++;
8064 		}
8065 
8066 		/*
8067 		 * move to next element in the dring and the next
8068 		 * position in the data buffer.
8069 		 */
8070 		priv_addr++;
8071 		tmpp += offset;
8072 	}
8073 
8074 	return (0);
8075 
8076 setup_ring_cleanup:
8077 	priv_addr = dp->priv_addr;
8078 
8079 	for (j = 0; j < i; j++) {
8080 		(void) ldc_mem_unbind_handle(priv_addr->memhandle);
8081 		(void) ldc_mem_free_handle(priv_addr->memhandle);
8082 
8083 		mutex_destroy(&priv_addr->dstate_lock);
8084 
8085 		priv_addr++;
8086 	}
8087 	kmem_free(dp->data_addr, dp->data_sz);
8088 
8089 	return (1);
8090 }
8091 
8092 /*
8093  * Searches the private section of a ring for a free descriptor,
8094  * starting at the location of the last free descriptor found
8095  * previously.
8096  *
8097  * Returns 0 if free descriptor is available, and updates state
8098  * of private descriptor to VIO_DESC_READY,  otherwise returns 1.
8099  *
8100  * FUTURE: might need to return contiguous range of descriptors
8101  * as dring info msg assumes all will be contiguous.
8102  */
8103 static int
8104 vsw_dring_find_free_desc(dring_info_t *dringp,
8105 		vsw_private_desc_t **priv_p, int *idx)
8106 {
8107 	vsw_private_desc_t	*addr = NULL;
8108 	int			num = VSW_RING_NUM_EL;
8109 	int			ret = 1;
8110 
8111 	D1(NULL, "%s enter\n", __func__);
8112 
8113 	ASSERT(dringp->priv_addr != NULL);
8114 
8115 	D2(NULL, "%s: searching ring, dringp 0x%llx : start pos %lld",
8116 			__func__, dringp, dringp->end_idx);
8117 
8118 	addr = (vsw_private_desc_t *)dringp->priv_addr + dringp->end_idx;
8119 
8120 	mutex_enter(&addr->dstate_lock);
8121 	if (addr->dstate == VIO_DESC_FREE) {
8122 		addr->dstate = VIO_DESC_READY;
8123 		*priv_p = addr;
8124 		*idx = dringp->end_idx;
8125 		dringp->end_idx = (dringp->end_idx + 1) % num;
8126 		ret = 0;
8127 
8128 	}
8129 	mutex_exit(&addr->dstate_lock);
8130 
8131 	/* ring full */
8132 	if (ret == 1) {
8133 		D2(NULL, "%s: no desp free: started at %d", __func__,
8134 			dringp->end_idx);
8135 	}
8136 
8137 	D1(NULL, "%s: exit\n", __func__);
8138 
8139 	return (ret);
8140 }
8141 
8142 /*
8143  * Map from a dring identifier to the ring itself. Returns
8144  * pointer to ring or NULL if no match found.
8145  *
8146  * Should be called with dlistrw rwlock held as reader.
8147  */
8148 static dring_info_t *
8149 vsw_ident2dring(lane_t *lane, uint64_t ident)
8150 {
8151 	dring_info_t	*dp = NULL;
8152 
8153 	if ((dp = lane->dringp) == NULL) {
8154 		return (NULL);
8155 	} else {
8156 		if (dp->ident == ident)
8157 			return (dp);
8158 
8159 		while (dp != NULL) {
8160 			if (dp->ident == ident)
8161 				break;
8162 			dp = dp->next;
8163 		}
8164 	}
8165 
8166 	return (dp);
8167 }
8168 
8169 /*
8170  * Set the default lane attributes. These are copied into
8171  * the attr msg we send to our peer. If they are not acceptable
8172  * then (currently) the handshake ends.
8173  */
8174 static void
8175 vsw_set_lane_attr(vsw_t *vswp, lane_t *lp)
8176 {
8177 	bzero(lp, sizeof (lane_t));
8178 
8179 	READ_ENTER(&vswp->if_lockrw);
8180 	ether_copy(&(vswp->if_addr), &(lp->addr));
8181 	RW_EXIT(&vswp->if_lockrw);
8182 
8183 	lp->mtu = VSW_MTU;
8184 	lp->addr_type = ADDR_TYPE_MAC;
8185 	lp->xfer_mode = VIO_DRING_MODE;
8186 	lp->ack_freq = 0;	/* for shared mode */
8187 
8188 	mutex_enter(&lp->seq_lock);
8189 	lp->seq_num = VNET_ISS;
8190 	mutex_exit(&lp->seq_lock);
8191 }
8192 
8193 /*
8194  * Verify that the attributes are acceptable.
8195  *
8196  * FUTURE: If some attributes are not acceptable, change them
8197  * our desired values.
8198  */
8199 static int
8200 vsw_check_attr(vnet_attr_msg_t *pkt, vsw_port_t *port)
8201 {
8202 	int	ret = 0;
8203 
8204 	D1(NULL, "vsw_check_attr enter\n");
8205 
8206 	/*
8207 	 * Note we currently only support in-band descriptors
8208 	 * and descriptor rings, not packet based transfer (VIO_PKT_MODE)
8209 	 */
8210 	if ((pkt->xfer_mode != VIO_DESC_MODE) &&
8211 			(pkt->xfer_mode != VIO_DRING_MODE)) {
8212 		D2(NULL, "vsw_check_attr: unknown mode %x\n",
8213 			pkt->xfer_mode);
8214 		ret = 1;
8215 	}
8216 
8217 	/* Only support MAC addresses at moment. */
8218 	if ((pkt->addr_type != ADDR_TYPE_MAC) || (pkt->addr == 0)) {
8219 		D2(NULL, "vsw_check_attr: invalid addr_type %x, "
8220 			"or address 0x%llx\n", pkt->addr_type,
8221 			pkt->addr);
8222 		ret = 1;
8223 	}
8224 
8225 	/*
8226 	 * MAC address supplied by device should match that stored
8227 	 * in the vsw-port OBP node. Need to decide what to do if they
8228 	 * don't match, for the moment just warn but don't fail.
8229 	 */
8230 	if (bcmp(&pkt->addr, &port->p_macaddr, ETHERADDRL) != 0) {
8231 		DERR(NULL, "vsw_check_attr: device supplied address "
8232 			"0x%llx doesn't match node address 0x%llx\n",
8233 			pkt->addr, port->p_macaddr);
8234 	}
8235 
8236 	/*
8237 	 * Ack freq only makes sense in pkt mode, in shared
8238 	 * mode the ring descriptors say whether or not to
8239 	 * send back an ACK.
8240 	 */
8241 	if ((pkt->xfer_mode == VIO_DRING_MODE) &&
8242 				(pkt->ack_freq > 0)) {
8243 		D2(NULL, "vsw_check_attr: non zero ack freq "
8244 			" in SHM mode\n");
8245 		ret = 1;
8246 	}
8247 
8248 	/*
8249 	 * Note: for the moment we only support ETHER
8250 	 * frames. This may change in the future.
8251 	 */
8252 	if ((pkt->mtu > VSW_MTU) || (pkt->mtu <= 0)) {
8253 		D2(NULL, "vsw_check_attr: invalid MTU (0x%llx)\n",
8254 			pkt->mtu);
8255 		ret = 1;
8256 	}
8257 
8258 	D1(NULL, "vsw_check_attr exit\n");
8259 
8260 	return (ret);
8261 }
8262 
8263 /*
8264  * Returns 1 if there is a problem, 0 otherwise.
8265  */
8266 static int
8267 vsw_check_dring_info(vio_dring_reg_msg_t *pkt)
8268 {
8269 	_NOTE(ARGUNUSED(pkt))
8270 
8271 	int	ret = 0;
8272 
8273 	D1(NULL, "vsw_check_dring_info enter\n");
8274 
8275 	if ((pkt->num_descriptors == 0) ||
8276 		(pkt->descriptor_size == 0) ||
8277 		(pkt->ncookies != 1)) {
8278 		DERR(NULL, "vsw_check_dring_info: invalid dring msg");
8279 		ret = 1;
8280 	}
8281 
8282 	D1(NULL, "vsw_check_dring_info exit\n");
8283 
8284 	return (ret);
8285 }
8286 
8287 /*
8288  * Returns 1 if two memory cookies match. Otherwise returns 0.
8289  */
8290 static int
8291 vsw_mem_cookie_match(ldc_mem_cookie_t *m1, ldc_mem_cookie_t *m2)
8292 {
8293 	if ((m1->addr != m2->addr) ||
8294 		(m2->size != m2->size)) {
8295 		return (0);
8296 	} else {
8297 		return (1);
8298 	}
8299 }
8300 
8301 /*
8302  * Returns 1 if ring described in reg message matches that
8303  * described by dring_info structure. Otherwise returns 0.
8304  */
8305 static int
8306 vsw_dring_match(dring_info_t *dp, vio_dring_reg_msg_t *msg)
8307 {
8308 	if ((msg->descriptor_size != dp->descriptor_size) ||
8309 		(msg->num_descriptors != dp->num_descriptors) ||
8310 		(msg->ncookies != dp->ncookies) ||
8311 		!(vsw_mem_cookie_match(&msg->cookie[0], &dp->cookie[0]))) {
8312 		return (0);
8313 	} else {
8314 		return (1);
8315 	}
8316 
8317 }
8318 
8319 static caddr_t
8320 vsw_print_ethaddr(uint8_t *a, char *ebuf)
8321 {
8322 	(void) sprintf(ebuf, "%x:%x:%x:%x:%x:%x",
8323 	    a[0], a[1], a[2], a[3], a[4], a[5]);
8324 	return (ebuf);
8325 }
8326 
8327 /*
8328  * Reset and free all the resources associated with
8329  * the channel.
8330  */
8331 static void
8332 vsw_free_lane_resources(vsw_ldc_t *ldcp, uint64_t dir)
8333 {
8334 	dring_info_t		*dp, *dpp;
8335 	lane_t			*lp = NULL;
8336 	int			rv = 0;
8337 
8338 	ASSERT(ldcp != NULL);
8339 
8340 	D1(ldcp->ldc_vswp, "%s (%lld): enter", __func__, ldcp->ldc_id);
8341 
8342 	if (dir == INBOUND) {
8343 		D2(ldcp->ldc_vswp, "%s: freeing INBOUND lane"
8344 			" of channel %lld", __func__, ldcp->ldc_id);
8345 		lp = &ldcp->lane_in;
8346 	} else {
8347 		D2(ldcp->ldc_vswp, "%s: freeing OUTBOUND lane"
8348 			" of channel %lld", __func__, ldcp->ldc_id);
8349 		lp = &ldcp->lane_out;
8350 	}
8351 
8352 	lp->lstate = VSW_LANE_INACTIV;
8353 	mutex_enter(&lp->seq_lock);
8354 	lp->seq_num = VNET_ISS;
8355 	mutex_exit(&lp->seq_lock);
8356 	if (lp->dringp) {
8357 		if (dir == INBOUND) {
8358 			WRITE_ENTER(&lp->dlistrw);
8359 			dp = lp->dringp;
8360 			while (dp != NULL) {
8361 				dpp = dp->next;
8362 				if (dp->handle != NULL)
8363 					(void) ldc_mem_dring_unmap(dp->handle);
8364 				kmem_free(dp, sizeof (dring_info_t));
8365 				dp = dpp;
8366 			}
8367 			RW_EXIT(&lp->dlistrw);
8368 		} else {
8369 			/*
8370 			 * unbind, destroy exported dring, free dring struct
8371 			 */
8372 			WRITE_ENTER(&lp->dlistrw);
8373 			dp = lp->dringp;
8374 			rv = vsw_free_ring(dp);
8375 			RW_EXIT(&lp->dlistrw);
8376 		}
8377 		if (rv == 0) {
8378 			lp->dringp = NULL;
8379 		}
8380 	}
8381 
8382 	D1(ldcp->ldc_vswp, "%s (%lld): exit", __func__, ldcp->ldc_id);
8383 }
8384 
8385 /*
8386  * Free ring and all associated resources.
8387  *
8388  * Should be called with dlistrw rwlock held as writer.
8389  */
8390 static int
8391 vsw_free_ring(dring_info_t *dp)
8392 {
8393 	vsw_private_desc_t	*paddr = NULL;
8394 	dring_info_t		*dpp;
8395 	int			i, rv = 1;
8396 
8397 	while (dp != NULL) {
8398 		mutex_enter(&dp->dlock);
8399 		dpp = dp->next;
8400 		if (dp->priv_addr != NULL) {
8401 			/*
8402 			 * First unbind and free the memory handles
8403 			 * stored in each descriptor within the ring.
8404 			 */
8405 			for (i = 0; i < VSW_RING_NUM_EL; i++) {
8406 				paddr = (vsw_private_desc_t *)
8407 						dp->priv_addr + i;
8408 				if (paddr->memhandle != NULL) {
8409 					if (paddr->bound == 1) {
8410 						rv = ldc_mem_unbind_handle(
8411 							paddr->memhandle);
8412 
8413 						if (rv != 0) {
8414 							DERR(NULL, "error "
8415 							"unbinding handle for "
8416 							"ring 0x%llx at pos %d",
8417 							dp, i);
8418 							mutex_exit(&dp->dlock);
8419 							return (rv);
8420 						}
8421 						paddr->bound = 0;
8422 					}
8423 
8424 					rv = ldc_mem_free_handle(
8425 							paddr->memhandle);
8426 					if (rv != 0) {
8427 						DERR(NULL, "error freeing "
8428 							"handle for ring "
8429 							"0x%llx at pos %d",
8430 							dp, i);
8431 						mutex_exit(&dp->dlock);
8432 						return (rv);
8433 					}
8434 					paddr->memhandle = NULL;
8435 				}
8436 				mutex_destroy(&paddr->dstate_lock);
8437 			}
8438 			kmem_free(dp->priv_addr, (sizeof (vsw_private_desc_t)
8439 					* VSW_RING_NUM_EL));
8440 		}
8441 
8442 		/*
8443 		 * Now unbind and destroy the ring itself.
8444 		 */
8445 		if (dp->handle != NULL) {
8446 			(void) ldc_mem_dring_unbind(dp->handle);
8447 			(void) ldc_mem_dring_destroy(dp->handle);
8448 		}
8449 
8450 		if (dp->data_addr != NULL) {
8451 			kmem_free(dp->data_addr, dp->data_sz);
8452 		}
8453 
8454 		mutex_exit(&dp->dlock);
8455 		mutex_destroy(&dp->dlock);
8456 		mutex_destroy(&dp->restart_lock);
8457 		kmem_free(dp, sizeof (dring_info_t));
8458 
8459 		dp = dpp;
8460 	}
8461 	return (0);
8462 }
8463 
8464 /*
8465  * Debugging routines
8466  */
8467 static void
8468 display_state(void)
8469 {
8470 	vsw_t		*vswp;
8471 	vsw_port_list_t	*plist;
8472 	vsw_port_t 	*port;
8473 	vsw_ldc_list_t	*ldcl;
8474 	vsw_ldc_t 	*ldcp;
8475 
8476 	cmn_err(CE_NOTE, "***** system state *****");
8477 
8478 	for (vswp = vsw_head; vswp; vswp = vswp->next) {
8479 		plist = &vswp->plist;
8480 		READ_ENTER(&plist->lockrw);
8481 		cmn_err(CE_CONT, "vsw instance %d has %d ports attached\n",
8482 			vswp->instance, plist->num_ports);
8483 
8484 		for (port = plist->head; port != NULL; port = port->p_next) {
8485 			ldcl = &port->p_ldclist;
8486 			cmn_err(CE_CONT, "port %d : %d ldcs attached\n",
8487 				port->p_instance, ldcl->num_ldcs);
8488 			READ_ENTER(&ldcl->lockrw);
8489 			ldcp = ldcl->head;
8490 			for (; ldcp != NULL; ldcp = ldcp->ldc_next) {
8491 				cmn_err(CE_CONT, "chan %lu : dev %d : "
8492 					"status %d : phase %u\n",
8493 					ldcp->ldc_id, ldcp->dev_class,
8494 					ldcp->ldc_status, ldcp->hphase);
8495 				cmn_err(CE_CONT, "chan %lu : lsession %lu : "
8496 					"psession %lu\n",
8497 					ldcp->ldc_id,
8498 					ldcp->local_session,
8499 					ldcp->peer_session);
8500 
8501 				cmn_err(CE_CONT, "Inbound lane:\n");
8502 				display_lane(&ldcp->lane_in);
8503 				cmn_err(CE_CONT, "Outbound lane:\n");
8504 				display_lane(&ldcp->lane_out);
8505 			}
8506 			RW_EXIT(&ldcl->lockrw);
8507 		}
8508 		RW_EXIT(&plist->lockrw);
8509 	}
8510 	cmn_err(CE_NOTE, "***** system state *****");
8511 }
8512 
8513 static void
8514 display_lane(lane_t *lp)
8515 {
8516 	dring_info_t	*drp;
8517 
8518 	cmn_err(CE_CONT, "ver 0x%x:0x%x : state %lx : mtu 0x%lx\n",
8519 		lp->ver_major, lp->ver_minor, lp->lstate, lp->mtu);
8520 	cmn_err(CE_CONT, "addr_type %d : addr 0x%lx : xmode %d\n",
8521 		lp->addr_type, lp->addr, lp->xfer_mode);
8522 	cmn_err(CE_CONT, "dringp 0x%lx\n", (uint64_t)lp->dringp);
8523 
8524 	cmn_err(CE_CONT, "Dring info:\n");
8525 	for (drp = lp->dringp; drp != NULL; drp = drp->next) {
8526 		cmn_err(CE_CONT, "\tnum_desc %u : dsize %u\n",
8527 			drp->num_descriptors, drp->descriptor_size);
8528 		cmn_err(CE_CONT, "\thandle 0x%lx\n", drp->handle);
8529 		cmn_err(CE_CONT, "\tpub_addr 0x%lx : priv_addr 0x%lx\n",
8530 			(uint64_t)drp->pub_addr, (uint64_t)drp->priv_addr);
8531 		cmn_err(CE_CONT, "\tident 0x%lx : end_idx %lu\n",
8532 			drp->ident, drp->end_idx);
8533 		display_ring(drp);
8534 	}
8535 }
8536 
8537 static void
8538 display_ring(dring_info_t *dringp)
8539 {
8540 	uint64_t		i;
8541 	uint64_t		priv_count = 0;
8542 	uint64_t		pub_count = 0;
8543 	vnet_public_desc_t	*pub_addr = NULL;
8544 	vsw_private_desc_t	*priv_addr = NULL;
8545 
8546 	for (i = 0; i < VSW_RING_NUM_EL; i++) {
8547 		if (dringp->pub_addr != NULL) {
8548 			pub_addr = (vnet_public_desc_t *)dringp->pub_addr + i;
8549 
8550 			if (pub_addr->hdr.dstate == VIO_DESC_FREE)
8551 				pub_count++;
8552 		}
8553 
8554 		if (dringp->priv_addr != NULL) {
8555 			priv_addr =
8556 				(vsw_private_desc_t *)dringp->priv_addr + i;
8557 
8558 			if (priv_addr->dstate == VIO_DESC_FREE)
8559 				priv_count++;
8560 		}
8561 	}
8562 	cmn_err(CE_CONT, "\t%lu elements: %lu priv free: %lu pub free\n",
8563 			i, priv_count, pub_count);
8564 }
8565 
8566 static void
8567 dump_flags(uint64_t state)
8568 {
8569 	int	i;
8570 
8571 	typedef struct flag_name {
8572 		int	flag_val;
8573 		char	*flag_name;
8574 	} flag_name_t;
8575 
8576 	flag_name_t	flags[] = {
8577 		VSW_VER_INFO_SENT, "VSW_VER_INFO_SENT",
8578 		VSW_VER_INFO_RECV, "VSW_VER_INFO_RECV",
8579 		VSW_VER_ACK_RECV, "VSW_VER_ACK_RECV",
8580 		VSW_VER_ACK_SENT, "VSW_VER_ACK_SENT",
8581 		VSW_VER_NACK_RECV, "VSW_VER_NACK_RECV",
8582 		VSW_VER_NACK_SENT, "VSW_VER_NACK_SENT",
8583 		VSW_ATTR_INFO_SENT, "VSW_ATTR_INFO_SENT",
8584 		VSW_ATTR_INFO_RECV, "VSW_ATTR_INFO_RECV",
8585 		VSW_ATTR_ACK_SENT, "VSW_ATTR_ACK_SENT",
8586 		VSW_ATTR_ACK_RECV, "VSW_ATTR_ACK_RECV",
8587 		VSW_ATTR_NACK_SENT, "VSW_ATTR_NACK_SENT",
8588 		VSW_ATTR_NACK_RECV, "VSW_ATTR_NACK_RECV",
8589 		VSW_DRING_INFO_SENT, "VSW_DRING_INFO_SENT",
8590 		VSW_DRING_INFO_RECV, "VSW_DRING_INFO_RECV",
8591 		VSW_DRING_ACK_SENT, "VSW_DRING_ACK_SENT",
8592 		VSW_DRING_ACK_RECV, "VSW_DRING_ACK_RECV",
8593 		VSW_DRING_NACK_SENT, "VSW_DRING_NACK_SENT",
8594 		VSW_DRING_NACK_RECV, "VSW_DRING_NACK_RECV",
8595 		VSW_RDX_INFO_SENT, "VSW_RDX_INFO_SENT",
8596 		VSW_RDX_INFO_RECV, "VSW_RDX_INFO_RECV",
8597 		VSW_RDX_ACK_SENT, "VSW_RDX_ACK_SENT",
8598 		VSW_RDX_ACK_RECV, "VSW_RDX_ACK_RECV",
8599 		VSW_RDX_NACK_SENT, "VSW_RDX_NACK_SENT",
8600 		VSW_RDX_NACK_RECV, "VSW_RDX_NACK_RECV",
8601 		VSW_MCST_INFO_SENT, "VSW_MCST_INFO_SENT",
8602 		VSW_MCST_INFO_RECV, "VSW_MCST_INFO_RECV",
8603 		VSW_MCST_ACK_SENT, "VSW_MCST_ACK_SENT",
8604 		VSW_MCST_ACK_RECV, "VSW_MCST_ACK_RECV",
8605 		VSW_MCST_NACK_SENT, "VSW_MCST_NACK_SENT",
8606 		VSW_MCST_NACK_RECV, "VSW_MCST_NACK_RECV",
8607 		VSW_LANE_ACTIVE, "VSW_LANE_ACTIVE"};
8608 
8609 	DERR(NULL, "DUMP_FLAGS: %llx\n", state);
8610 	for (i = 0; i < sizeof (flags)/sizeof (flag_name_t); i++) {
8611 		if (state & flags[i].flag_val)
8612 			DERR(NULL, "DUMP_FLAGS %s", flags[i].flag_name);
8613 	}
8614 }
8615