xref: /illumos-gate/usr/src/uts/sun4v/io/ldc.c (revision f67ca41a3fe371a8ac34045eb45b3c5449ee601c)
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 /*
30  * sun4v LDC Link Layer
31  */
32 #include <sys/types.h>
33 #include <sys/file.h>
34 #include <sys/errno.h>
35 #include <sys/open.h>
36 #include <sys/cred.h>
37 #include <sys/kmem.h>
38 #include <sys/conf.h>
39 #include <sys/cmn_err.h>
40 #include <sys/ksynch.h>
41 #include <sys/modctl.h>
42 #include <sys/stat.h> /* needed for S_IFBLK and S_IFCHR */
43 #include <sys/debug.h>
44 #include <sys/types.h>
45 #include <sys/cred.h>
46 #include <sys/promif.h>
47 #include <sys/ddi.h>
48 #include <sys/sunddi.h>
49 #include <sys/cyclic.h>
50 #include <sys/machsystm.h>
51 #include <sys/vm.h>
52 #include <sys/cpu.h>
53 #include <sys/intreg.h>
54 #include <sys/machcpuvar.h>
55 #include <sys/mmu.h>
56 #include <sys/pte.h>
57 #include <vm/hat.h>
58 #include <vm/as.h>
59 #include <vm/hat_sfmmu.h>
60 #include <sys/vm_machparam.h>
61 #include <vm/seg_kmem.h>
62 #include <vm/seg_kpm.h>
63 #include <sys/note.h>
64 #include <sys/ivintr.h>
65 #include <sys/hypervisor_api.h>
66 #include <sys/ldc.h>
67 #include <sys/ldc_impl.h>
68 #include <sys/cnex.h>
69 #include <sys/hsvc.h>
70 
71 /* Core internal functions */
72 static int i_ldc_h2v_error(int h_error);
73 static int i_ldc_txq_reconf(ldc_chan_t *ldcp);
74 static int i_ldc_rxq_reconf(ldc_chan_t *ldcp, boolean_t force_reset);
75 static int i_ldc_rxq_drain(ldc_chan_t *ldcp);
76 static void i_ldc_reset_state(ldc_chan_t *ldcp);
77 static void i_ldc_reset(ldc_chan_t *ldcp, boolean_t force_reset);
78 
79 static int i_ldc_get_tx_tail(ldc_chan_t *ldcp, uint64_t *tail);
80 static int i_ldc_set_tx_tail(ldc_chan_t *ldcp, uint64_t tail);
81 static int i_ldc_set_rx_head(ldc_chan_t *ldcp, uint64_t head);
82 static int i_ldc_send_pkt(ldc_chan_t *ldcp, uint8_t pkttype, uint8_t subtype,
83     uint8_t ctrlmsg);
84 
85 /* Interrupt handling functions */
86 static uint_t i_ldc_tx_hdlr(caddr_t arg1, caddr_t arg2);
87 static uint_t i_ldc_rx_hdlr(caddr_t arg1, caddr_t arg2);
88 static void i_ldc_clear_intr(ldc_chan_t *ldcp, cnex_intrtype_t itype);
89 
90 /* Read method functions */
91 static int i_ldc_read_raw(ldc_chan_t *ldcp, caddr_t target_bufp, size_t *sizep);
92 static int i_ldc_read_packet(ldc_chan_t *ldcp, caddr_t target_bufp,
93 	size_t *sizep);
94 static int i_ldc_read_stream(ldc_chan_t *ldcp, caddr_t target_bufp,
95 	size_t *sizep);
96 
97 /* Write method functions */
98 static int i_ldc_write_raw(ldc_chan_t *ldcp, caddr_t target_bufp,
99 	size_t *sizep);
100 static int i_ldc_write_packet(ldc_chan_t *ldcp, caddr_t target_bufp,
101 	size_t *sizep);
102 static int i_ldc_write_stream(ldc_chan_t *ldcp, caddr_t target_bufp,
103 	size_t *sizep);
104 
105 /* Pkt processing internal functions */
106 static int i_ldc_check_seqid(ldc_chan_t *ldcp, ldc_msg_t *ldcmsg);
107 static int i_ldc_ctrlmsg(ldc_chan_t *ldcp, ldc_msg_t *ldcmsg);
108 static int i_ldc_process_VER(ldc_chan_t *ldcp, ldc_msg_t *msg);
109 static int i_ldc_process_RTS(ldc_chan_t *ldcp, ldc_msg_t *msg);
110 static int i_ldc_process_RTR(ldc_chan_t *ldcp, ldc_msg_t *msg);
111 static int i_ldc_process_RDX(ldc_chan_t *ldcp, ldc_msg_t *msg);
112 static int i_ldc_process_data_ACK(ldc_chan_t *ldcp, ldc_msg_t *msg);
113 
114 /* Memory synchronization internal functions */
115 static int i_ldc_mem_acquire_release(ldc_mem_handle_t mhandle,
116     uint8_t direction, uint64_t offset, size_t size);
117 static int i_ldc_dring_acquire_release(ldc_dring_handle_t dhandle,
118     uint8_t direction, uint64_t start, uint64_t end);
119 
120 /* LDC Version */
121 static ldc_ver_t ldc_versions[] = { {1, 0} };
122 
123 /* number of supported versions */
124 #define	LDC_NUM_VERS	(sizeof (ldc_versions) / sizeof (ldc_versions[0]))
125 
126 /* Module State Pointer */
127 static ldc_soft_state_t *ldcssp;
128 
129 static struct modldrv md = {
130 	&mod_miscops,			/* This is a misc module */
131 	"sun4v LDC module v%I%",	/* Name of the module */
132 };
133 
134 static struct modlinkage ml = {
135 	MODREV_1,
136 	&md,
137 	NULL
138 };
139 
140 static uint64_t ldc_sup_minor;		/* Supported minor number */
141 static hsvc_info_t ldc_hsvc = {
142 	HSVC_REV_1, NULL, HSVC_GROUP_LDC, 1, 0, "ldc"
143 };
144 
145 static uint64_t intr_sup_minor;		/* Supported minor number */
146 static hsvc_info_t intr_hsvc = {
147 	HSVC_REV_1, NULL, HSVC_GROUP_INTR, 1, 0, "ldc"
148 };
149 
150 /*
151  * LDC framework supports mapping remote domain's memory
152  * either directly or via shadow memory pages. Default
153  * support is currently implemented via shadow copy.
154  * Direct map can be enabled by setting 'ldc_shmem_enabled'
155  */
156 int ldc_shmem_enabled = 0;
157 
158 /*
159  * The no. of MTU size messages that can be stored in
160  * the LDC Tx queue. The number of Tx queue entries is
161  * then computed as (mtu * mtu_msgs)/sizeof(queue_entry)
162  */
163 uint64_t ldc_mtu_msgs = LDC_MTU_MSGS;
164 
165 /*
166  * The minimum queue length. This is the size of the smallest
167  * LDC queue. If the computed value is less than this default,
168  * the queue length is rounded up to 'ldc_queue_entries'.
169  */
170 uint64_t ldc_queue_entries = LDC_QUEUE_ENTRIES;
171 
172 /*
173  * Pages exported for remote access over each channel is
174  * maintained in a table registered with the Hypervisor.
175  * The default number of entries in the table is set to
176  * 'ldc_mtbl_entries'.
177  */
178 uint64_t ldc_maptable_entries = LDC_MTBL_ENTRIES;
179 
180 /*
181  * LDC retry count and delay - when the HV returns EWOULDBLOCK
182  * the operation is retried 'ldc_max_retries' times with a
183  * wait of 'ldc_delay' usecs between each retry.
184  */
185 int ldc_max_retries = LDC_MAX_RETRIES;
186 clock_t ldc_delay = LDC_DELAY;
187 
188 #ifdef DEBUG
189 
190 /*
191  * Print debug messages
192  *
193  * set ldcdbg to 0x7 for enabling all msgs
194  * 0x4 - Warnings
195  * 0x2 - All debug messages
196  * 0x1 - Minimal debug messages
197  *
198  * set ldcdbgchan to the channel number you want to debug
199  * setting it to -1 prints debug messages for all channels
200  * NOTE: ldcdbgchan has no effect on error messages
201  */
202 
203 #define	DBG_ALL_LDCS -1
204 
205 int ldcdbg = 0x0;
206 int64_t ldcdbgchan = DBG_ALL_LDCS;
207 boolean_t ldc_inject_reset_flag = B_FALSE;
208 
209 static void
210 ldcdebug(int64_t id, const char *fmt, ...)
211 {
212 	char buf[512];
213 	va_list ap;
214 
215 	/*
216 	 * Do not return if,
217 	 * caller wants to print it anyway - (id == DBG_ALL_LDCS)
218 	 * debug channel is set to all LDCs - (ldcdbgchan == DBG_ALL_LDCS)
219 	 * debug channel = caller specified channel
220 	 */
221 	if ((id != DBG_ALL_LDCS) &&
222 	    (ldcdbgchan != DBG_ALL_LDCS) &&
223 	    (ldcdbgchan != id)) {
224 		return;
225 	}
226 
227 	va_start(ap, fmt);
228 	(void) vsprintf(buf, fmt, ap);
229 	va_end(ap);
230 
231 	cmn_err(CE_CONT, "?%s", buf);
232 }
233 
234 static boolean_t
235 ldc_inject_reset(ldc_chan_t *ldcp)
236 {
237 	if ((ldcdbgchan != DBG_ALL_LDCS) && (ldcdbgchan != ldcp->id))
238 		return (B_FALSE);
239 
240 	if (!ldc_inject_reset_flag)
241 		return (B_FALSE);
242 
243 	/* clear the injection state */
244 	ldc_inject_reset_flag = 0;
245 
246 	return (B_TRUE);
247 }
248 
249 #define	D1		\
250 if (ldcdbg & 0x01)	\
251 	ldcdebug
252 
253 #define	D2		\
254 if (ldcdbg & 0x02)	\
255 	ldcdebug
256 
257 #define	DWARN		\
258 if (ldcdbg & 0x04)	\
259 	ldcdebug
260 
261 #define	DUMP_PAYLOAD(id, addr)						\
262 {									\
263 	char buf[65*3];							\
264 	int i;								\
265 	uint8_t *src = (uint8_t *)addr;					\
266 	for (i = 0; i < 64; i++, src++)					\
267 		(void) sprintf(&buf[i * 3], "|%02x", *src);		\
268 	(void) sprintf(&buf[i * 3], "|\n");				\
269 	D2((id), "payload: %s", buf);					\
270 }
271 
272 #define	DUMP_LDC_PKT(c, s, addr)					\
273 {									\
274 	ldc_msg_t *msg = (ldc_msg_t *)(addr);				\
275 	uint32_t mid = ((c)->mode != LDC_MODE_RAW) ? msg->seqid : 0;	\
276 	if (msg->type == LDC_DATA) {                                    \
277 	    D2((c)->id, "%s: msg%d (/%x/%x/%x/,env[%c%c,sz=%d])",	\
278 	    (s), mid, msg->type, msg->stype, msg->ctrl,			\
279 	    (msg->env & LDC_FRAG_START) ? 'B' : ' ',                    \
280 	    (msg->env & LDC_FRAG_STOP) ? 'E' : ' ',                     \
281 	    (msg->env & LDC_LEN_MASK));					\
282 	} else { 							\
283 	    D2((c)->id, "%s: msg%d (/%x/%x/%x/,env=%x)", (s),		\
284 	    mid, msg->type, msg->stype, msg->ctrl, msg->env);		\
285 	} 								\
286 }
287 
288 #define	LDC_INJECT_RESET(_ldcp)	ldc_inject_reset(_ldcp)
289 
290 #else
291 
292 #define	DBG_ALL_LDCS -1
293 
294 #define	D1
295 #define	D2
296 #define	DWARN
297 
298 #define	DUMP_PAYLOAD(id, addr)
299 #define	DUMP_LDC_PKT(c, s, addr)
300 
301 #define	LDC_INJECT_RESET(_ldcp)	(B_FALSE)
302 
303 #endif
304 
305 #define	ZERO_PKT(p)			\
306 	bzero((p), sizeof (ldc_msg_t));
307 
308 #define	IDX2COOKIE(idx, pg_szc, pg_shift)				\
309 	(((pg_szc) << LDC_COOKIE_PGSZC_SHIFT) | ((idx) << (pg_shift)))
310 
311 
312 int
313 _init(void)
314 {
315 	int status;
316 
317 	status = hsvc_register(&ldc_hsvc, &ldc_sup_minor);
318 	if (status != 0) {
319 		cmn_err(CE_WARN, "%s: cannot negotiate hypervisor LDC services"
320 		    " group: 0x%lx major: %ld minor: %ld errno: %d",
321 		    ldc_hsvc.hsvc_modname, ldc_hsvc.hsvc_group,
322 		    ldc_hsvc.hsvc_major, ldc_hsvc.hsvc_minor, status);
323 		return (-1);
324 	}
325 
326 	status = hsvc_register(&intr_hsvc, &intr_sup_minor);
327 	if (status != 0) {
328 		cmn_err(CE_WARN, "%s: cannot negotiate hypervisor interrupt "
329 		    "services group: 0x%lx major: %ld minor: %ld errno: %d",
330 		    intr_hsvc.hsvc_modname, intr_hsvc.hsvc_group,
331 		    intr_hsvc.hsvc_major, intr_hsvc.hsvc_minor, status);
332 		(void) hsvc_unregister(&ldc_hsvc);
333 		return (-1);
334 	}
335 
336 	/* allocate soft state structure */
337 	ldcssp = kmem_zalloc(sizeof (ldc_soft_state_t), KM_SLEEP);
338 
339 	/* Link the module into the system */
340 	status = mod_install(&ml);
341 	if (status != 0) {
342 		kmem_free(ldcssp, sizeof (ldc_soft_state_t));
343 		return (status);
344 	}
345 
346 	/* Initialize the LDC state structure */
347 	mutex_init(&ldcssp->lock, NULL, MUTEX_DRIVER, NULL);
348 
349 	mutex_enter(&ldcssp->lock);
350 
351 	/* Create a cache for memory handles */
352 	ldcssp->memhdl_cache = kmem_cache_create("ldc_memhdl_cache",
353 	    sizeof (ldc_mhdl_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
354 	if (ldcssp->memhdl_cache == NULL) {
355 		DWARN(DBG_ALL_LDCS, "_init: ldc_memhdl cache create failed\n");
356 		mutex_exit(&ldcssp->lock);
357 		return (-1);
358 	}
359 
360 	/* Create cache for memory segment structures */
361 	ldcssp->memseg_cache = kmem_cache_create("ldc_memseg_cache",
362 	    sizeof (ldc_memseg_t), 0, NULL, NULL, NULL, NULL, NULL, 0);
363 	if (ldcssp->memseg_cache == NULL) {
364 		DWARN(DBG_ALL_LDCS, "_init: ldc_memseg cache create failed\n");
365 		mutex_exit(&ldcssp->lock);
366 		return (-1);
367 	}
368 
369 
370 	ldcssp->channel_count = 0;
371 	ldcssp->channels_open = 0;
372 	ldcssp->chan_list = NULL;
373 	ldcssp->dring_list = NULL;
374 
375 	mutex_exit(&ldcssp->lock);
376 
377 	return (0);
378 }
379 
380 int
381 _info(struct modinfo *modinfop)
382 {
383 	/* Report status of the dynamically loadable driver module */
384 	return (mod_info(&ml, modinfop));
385 }
386 
387 int
388 _fini(void)
389 {
390 	int 		rv, status;
391 	ldc_chan_t 	*ldcp;
392 	ldc_dring_t 	*dringp;
393 	ldc_mem_info_t 	minfo;
394 
395 	/* Unlink the driver module from the system */
396 	status = mod_remove(&ml);
397 	if (status) {
398 		DWARN(DBG_ALL_LDCS, "_fini: mod_remove failed\n");
399 		return (EIO);
400 	}
401 
402 	/* close and finalize channels */
403 	ldcp = ldcssp->chan_list;
404 	while (ldcp != NULL) {
405 		(void) ldc_close((ldc_handle_t)ldcp);
406 		(void) ldc_fini((ldc_handle_t)ldcp);
407 
408 		ldcp = ldcp->next;
409 	}
410 
411 	/* Free descriptor rings */
412 	dringp = ldcssp->dring_list;
413 	while (dringp != NULL) {
414 		dringp = dringp->next;
415 
416 		rv = ldc_mem_dring_info((ldc_dring_handle_t)dringp, &minfo);
417 		if (rv == 0 && minfo.status != LDC_UNBOUND) {
418 			if (minfo.status == LDC_BOUND) {
419 				(void) ldc_mem_dring_unbind(
420 						(ldc_dring_handle_t)dringp);
421 			}
422 			if (minfo.status == LDC_MAPPED) {
423 				(void) ldc_mem_dring_unmap(
424 						(ldc_dring_handle_t)dringp);
425 			}
426 		}
427 
428 		(void) ldc_mem_dring_destroy((ldc_dring_handle_t)dringp);
429 	}
430 	ldcssp->dring_list = NULL;
431 
432 	/* Destroy kmem caches */
433 	kmem_cache_destroy(ldcssp->memhdl_cache);
434 	kmem_cache_destroy(ldcssp->memseg_cache);
435 
436 	/*
437 	 * We have successfully "removed" the driver.
438 	 * Destroying soft states
439 	 */
440 	mutex_destroy(&ldcssp->lock);
441 	kmem_free(ldcssp, sizeof (ldc_soft_state_t));
442 
443 	(void) hsvc_unregister(&ldc_hsvc);
444 	(void) hsvc_unregister(&intr_hsvc);
445 
446 	return (status);
447 }
448 
449 /* -------------------------------------------------------------------------- */
450 
451 /*
452  * LDC Link Layer Internal Functions
453  */
454 
455 /*
456  * Translate HV Errors to sun4v error codes
457  */
458 static int
459 i_ldc_h2v_error(int h_error)
460 {
461 	switch (h_error) {
462 
463 	case	H_EOK:
464 		return (0);
465 
466 	case	H_ENORADDR:
467 		return (EFAULT);
468 
469 	case	H_EBADPGSZ:
470 	case	H_EINVAL:
471 		return (EINVAL);
472 
473 	case	H_EWOULDBLOCK:
474 		return (EWOULDBLOCK);
475 
476 	case	H_ENOACCESS:
477 	case	H_ENOMAP:
478 		return (EACCES);
479 
480 	case	H_EIO:
481 	case	H_ECPUERROR:
482 		return (EIO);
483 
484 	case	H_ENOTSUPPORTED:
485 		return (ENOTSUP);
486 
487 	case 	H_ETOOMANY:
488 		return (ENOSPC);
489 
490 	case	H_ECHANNEL:
491 		return (ECHRNG);
492 	default:
493 		break;
494 	}
495 
496 	return (EIO);
497 }
498 
499 /*
500  * Reconfigure the transmit queue
501  */
502 static int
503 i_ldc_txq_reconf(ldc_chan_t *ldcp)
504 {
505 	int rv;
506 
507 	ASSERT(MUTEX_HELD(&ldcp->lock));
508 	ASSERT(MUTEX_HELD(&ldcp->tx_lock));
509 
510 	rv = hv_ldc_tx_qconf(ldcp->id, ldcp->tx_q_ra, ldcp->tx_q_entries);
511 	if (rv) {
512 		cmn_err(CE_WARN,
513 		    "i_ldc_txq_reconf: (0x%lx) cannot set qconf", ldcp->id);
514 		return (EIO);
515 	}
516 	rv = hv_ldc_tx_get_state(ldcp->id, &(ldcp->tx_head),
517 	    &(ldcp->tx_tail), &(ldcp->link_state));
518 	if (rv) {
519 		cmn_err(CE_WARN,
520 		    "i_ldc_txq_reconf: (0x%lx) cannot get qptrs", ldcp->id);
521 		return (EIO);
522 	}
523 	D1(ldcp->id, "i_ldc_txq_reconf: (0x%llx) h=0x%llx,t=0x%llx,"
524 	    "s=0x%llx\n", ldcp->id, ldcp->tx_head, ldcp->tx_tail,
525 	    ldcp->link_state);
526 
527 	return (0);
528 }
529 
530 /*
531  * Reconfigure the receive queue
532  */
533 static int
534 i_ldc_rxq_reconf(ldc_chan_t *ldcp, boolean_t force_reset)
535 {
536 	int rv;
537 	uint64_t rx_head, rx_tail;
538 
539 	ASSERT(MUTEX_HELD(&ldcp->lock));
540 	rv = hv_ldc_rx_get_state(ldcp->id, &rx_head, &rx_tail,
541 	    &(ldcp->link_state));
542 	if (rv) {
543 		cmn_err(CE_WARN,
544 		    "i_ldc_rxq_reconf: (0x%lx) cannot get state",
545 		    ldcp->id);
546 		return (EIO);
547 	}
548 
549 	if (force_reset || (ldcp->tstate & ~TS_IN_RESET) == TS_UP) {
550 		rv = hv_ldc_rx_qconf(ldcp->id, ldcp->rx_q_ra,
551 			ldcp->rx_q_entries);
552 		if (rv) {
553 			cmn_err(CE_WARN,
554 			    "i_ldc_rxq_reconf: (0x%lx) cannot set qconf",
555 			    ldcp->id);
556 			return (EIO);
557 		}
558 		D1(ldcp->id, "i_ldc_rxq_reconf: (0x%llx) completed q reconf",
559 		    ldcp->id);
560 	}
561 
562 	return (0);
563 }
564 
565 
566 /*
567  * Drain the contents of the receive queue
568  */
569 static int
570 i_ldc_rxq_drain(ldc_chan_t *ldcp)
571 {
572 	int rv;
573 	uint64_t rx_head, rx_tail;
574 
575 	ASSERT(MUTEX_HELD(&ldcp->lock));
576 	rv = hv_ldc_rx_get_state(ldcp->id, &rx_head, &rx_tail,
577 	    &(ldcp->link_state));
578 	if (rv) {
579 		cmn_err(CE_WARN, "i_ldc_rxq_drain: (0x%lx) cannot get state",
580 		    ldcp->id);
581 		return (EIO);
582 	}
583 
584 	/* flush contents by setting the head = tail */
585 	return (i_ldc_set_rx_head(ldcp, rx_tail));
586 }
587 
588 
589 /*
590  * Reset LDC state structure and its contents
591  */
592 static void
593 i_ldc_reset_state(ldc_chan_t *ldcp)
594 {
595 	ASSERT(MUTEX_HELD(&ldcp->lock));
596 	ldcp->last_msg_snt = LDC_INIT_SEQID;
597 	ldcp->last_ack_rcd = 0;
598 	ldcp->last_msg_rcd = 0;
599 	ldcp->tx_ackd_head = ldcp->tx_head;
600 	ldcp->next_vidx = 0;
601 	ldcp->hstate = 0;
602 	ldcp->tstate = TS_OPEN;
603 	ldcp->status = LDC_OPEN;
604 
605 	if (ldcp->link_state == LDC_CHANNEL_UP ||
606 	    ldcp->link_state == LDC_CHANNEL_RESET) {
607 
608 		if (ldcp->mode == LDC_MODE_RAW) {
609 			ldcp->status = LDC_UP;
610 			ldcp->tstate = TS_UP;
611 		} else {
612 			ldcp->status = LDC_READY;
613 			ldcp->tstate |= TS_LINK_READY;
614 		}
615 	}
616 }
617 
618 /*
619  * Reset a LDC channel
620  */
621 static void
622 i_ldc_reset(ldc_chan_t *ldcp, boolean_t force_reset)
623 {
624 	DWARN(ldcp->id, "i_ldc_reset: (0x%llx) channel reset\n", ldcp->id);
625 
626 	ASSERT(MUTEX_HELD(&ldcp->lock));
627 	ASSERT(MUTEX_HELD(&ldcp->tx_lock));
628 
629 	/* reconfig Tx and Rx queues */
630 	(void) i_ldc_txq_reconf(ldcp);
631 	(void) i_ldc_rxq_reconf(ldcp, force_reset);
632 
633 	/* Clear Tx and Rx interrupts */
634 	(void) i_ldc_clear_intr(ldcp, CNEX_TX_INTR);
635 	(void) i_ldc_clear_intr(ldcp, CNEX_RX_INTR);
636 
637 	/* Reset channel state */
638 	i_ldc_reset_state(ldcp);
639 
640 	/* Mark channel in reset */
641 	ldcp->tstate |= TS_IN_RESET;
642 }
643 
644 
645 /*
646  * Clear pending interrupts
647  */
648 static void
649 i_ldc_clear_intr(ldc_chan_t *ldcp, cnex_intrtype_t itype)
650 {
651 	ldc_cnex_t *cinfo = &ldcssp->cinfo;
652 
653 	ASSERT(MUTEX_HELD(&ldcp->lock));
654 	ASSERT(cinfo->dip != NULL);
655 
656 	switch (itype) {
657 	case CNEX_TX_INTR:
658 		/* check Tx interrupt */
659 		if (ldcp->tx_intr_state)
660 			ldcp->tx_intr_state = LDC_INTR_NONE;
661 		else
662 			return;
663 		break;
664 
665 	case CNEX_RX_INTR:
666 		/* check Rx interrupt */
667 		if (ldcp->rx_intr_state)
668 			ldcp->rx_intr_state = LDC_INTR_NONE;
669 		else
670 			return;
671 		break;
672 	}
673 
674 	(void) cinfo->clr_intr(cinfo->dip, ldcp->id, itype);
675 	D2(ldcp->id,
676 	    "i_ldc_clear_intr: (0x%llx) cleared 0x%x intr\n",
677 	    ldcp->id, itype);
678 }
679 
680 /*
681  * Set the receive queue head
682  * Resets connection and returns an error if it fails.
683  */
684 static int
685 i_ldc_set_rx_head(ldc_chan_t *ldcp, uint64_t head)
686 {
687 	int 	rv;
688 	int 	retries;
689 
690 	ASSERT(MUTEX_HELD(&ldcp->lock));
691 	for (retries = 0; retries < ldc_max_retries; retries++) {
692 
693 		if ((rv = hv_ldc_rx_set_qhead(ldcp->id, head)) == 0)
694 			return (0);
695 
696 		if (rv != H_EWOULDBLOCK)
697 			break;
698 
699 		/* wait for ldc_delay usecs */
700 		drv_usecwait(ldc_delay);
701 	}
702 
703 	cmn_err(CE_WARN, "ldc_rx_set_qhead: (0x%lx) cannot set qhead 0x%lx",
704 		ldcp->id, head);
705 	mutex_enter(&ldcp->tx_lock);
706 	i_ldc_reset(ldcp, B_TRUE);
707 	mutex_exit(&ldcp->tx_lock);
708 
709 	return (ECONNRESET);
710 }
711 
712 
713 /*
714  * Returns the tx_tail to be used for transfer
715  * Re-reads the TX queue ptrs if and only if the
716  * the cached head and tail are equal (queue is full)
717  */
718 static int
719 i_ldc_get_tx_tail(ldc_chan_t *ldcp, uint64_t *tail)
720 {
721 	int 		rv;
722 	uint64_t 	current_head, new_tail;
723 
724 	ASSERT(MUTEX_HELD(&ldcp->tx_lock));
725 	/* Read the head and tail ptrs from HV */
726 	rv = hv_ldc_tx_get_state(ldcp->id,
727 	    &ldcp->tx_head, &ldcp->tx_tail, &ldcp->link_state);
728 	if (rv) {
729 		cmn_err(CE_WARN,
730 		    "i_ldc_get_tx_tail: (0x%lx) cannot read qptrs\n",
731 		    ldcp->id);
732 		return (EIO);
733 	}
734 	if (ldcp->link_state == LDC_CHANNEL_DOWN) {
735 		DWARN(DBG_ALL_LDCS,
736 		    "i_ldc_get_tx_tail: (0x%llx) channel not ready\n",
737 		    ldcp->id);
738 		return (ECONNRESET);
739 	}
740 
741 	/* In reliable mode, check against last ACKd msg */
742 	current_head = (ldcp->mode == LDC_MODE_RELIABLE ||
743 		ldcp->mode == LDC_MODE_STREAM)
744 		? ldcp->tx_ackd_head : ldcp->tx_head;
745 
746 	/* increment the tail */
747 	new_tail = (ldcp->tx_tail + LDC_PACKET_SIZE) %
748 		(ldcp->tx_q_entries << LDC_PACKET_SHIFT);
749 
750 	if (new_tail == current_head) {
751 		DWARN(ldcp->id,
752 		    "i_ldc_get_tx_tail: (0x%llx) TX queue is full\n",
753 		    ldcp->id);
754 		return (EWOULDBLOCK);
755 	}
756 
757 	D2(ldcp->id, "i_ldc_get_tx_tail: (0x%llx) head=0x%llx, tail=0x%llx\n",
758 	    ldcp->id, ldcp->tx_head, ldcp->tx_tail);
759 
760 	*tail = ldcp->tx_tail;
761 	return (0);
762 }
763 
764 /*
765  * Set the tail pointer. If HV returns EWOULDBLOCK, it will back off
766  * and retry ldc_max_retries times before returning an error.
767  * Returns 0, EWOULDBLOCK or EIO
768  */
769 static int
770 i_ldc_set_tx_tail(ldc_chan_t *ldcp, uint64_t tail)
771 {
772 	int		rv, retval = EWOULDBLOCK;
773 	int 		retries;
774 
775 	ASSERT(MUTEX_HELD(&ldcp->tx_lock));
776 	for (retries = 0; retries < ldc_max_retries; retries++) {
777 
778 		if ((rv = hv_ldc_tx_set_qtail(ldcp->id, tail)) == 0) {
779 			retval = 0;
780 			break;
781 		}
782 		if (rv != H_EWOULDBLOCK) {
783 			DWARN(ldcp->id, "i_ldc_set_tx_tail: (0x%llx) set "
784 			    "qtail=0x%llx failed, rv=%d\n", ldcp->id, tail, rv);
785 			retval = EIO;
786 			break;
787 		}
788 
789 		/* wait for ldc_delay usecs */
790 		drv_usecwait(ldc_delay);
791 	}
792 	return (retval);
793 }
794 
795 /*
796  * Send a LDC message
797  */
798 static int
799 i_ldc_send_pkt(ldc_chan_t *ldcp, uint8_t pkttype, uint8_t subtype,
800     uint8_t ctrlmsg)
801 {
802 	int		rv;
803 	ldc_msg_t 	*pkt;
804 	uint64_t	tx_tail;
805 	uint32_t	curr_seqid = ldcp->last_msg_snt;
806 
807 	/* Obtain Tx lock */
808 	mutex_enter(&ldcp->tx_lock);
809 
810 	/* get the current tail for the message */
811 	rv = i_ldc_get_tx_tail(ldcp, &tx_tail);
812 	if (rv) {
813 		DWARN(ldcp->id,
814 		    "i_ldc_send_pkt: (0x%llx) error sending pkt, "
815 		    "type=0x%x,subtype=0x%x,ctrl=0x%x\n",
816 		    ldcp->id, pkttype, subtype, ctrlmsg);
817 		mutex_exit(&ldcp->tx_lock);
818 		return (rv);
819 	}
820 
821 	pkt = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail);
822 	ZERO_PKT(pkt);
823 
824 	/* Initialize the packet */
825 	pkt->type = pkttype;
826 	pkt->stype = subtype;
827 	pkt->ctrl = ctrlmsg;
828 
829 	/* Store ackid/seqid iff it is RELIABLE mode & not a RTS/RTR message */
830 	if (((ctrlmsg & LDC_CTRL_MASK) != LDC_RTS) &&
831 	    ((ctrlmsg & LDC_CTRL_MASK) != LDC_RTR)) {
832 		curr_seqid++;
833 		if (ldcp->mode != LDC_MODE_RAW) {
834 			pkt->seqid = curr_seqid;
835 			pkt->ackid = ldcp->last_msg_rcd;
836 		}
837 	}
838 	DUMP_LDC_PKT(ldcp, "i_ldc_send_pkt", (uint64_t)pkt);
839 
840 	/* initiate the send by calling into HV and set the new tail */
841 	tx_tail = (tx_tail + LDC_PACKET_SIZE) %
842 		(ldcp->tx_q_entries << LDC_PACKET_SHIFT);
843 
844 	rv = i_ldc_set_tx_tail(ldcp, tx_tail);
845 	if (rv) {
846 		DWARN(ldcp->id,
847 		    "i_ldc_send_pkt:(0x%llx) error sending pkt, "
848 		    "type=0x%x,stype=0x%x,ctrl=0x%x\n",
849 		    ldcp->id, pkttype, subtype, ctrlmsg);
850 		mutex_exit(&ldcp->tx_lock);
851 		return (EIO);
852 	}
853 
854 	ldcp->last_msg_snt = curr_seqid;
855 	ldcp->tx_tail = tx_tail;
856 
857 	mutex_exit(&ldcp->tx_lock);
858 	return (0);
859 }
860 
861 /*
862  * Checks if packet was received in right order
863  * in the case of a reliable link.
864  * Returns 0 if in order, else EIO
865  */
866 static int
867 i_ldc_check_seqid(ldc_chan_t *ldcp, ldc_msg_t *msg)
868 {
869 	/* No seqid checking for RAW mode */
870 	if (ldcp->mode == LDC_MODE_RAW)
871 		return (0);
872 
873 	/* No seqid checking for version, RTS, RTR message */
874 	if (msg->ctrl == LDC_VER ||
875 	    msg->ctrl == LDC_RTS ||
876 	    msg->ctrl == LDC_RTR)
877 		return (0);
878 
879 	/* Initial seqid to use is sent in RTS/RTR and saved in last_msg_rcd */
880 	if (msg->seqid != (ldcp->last_msg_rcd + 1)) {
881 		DWARN(ldcp->id,
882 		    "i_ldc_check_seqid: (0x%llx) out-of-order pkt, got 0x%x, "
883 		    "expecting 0x%x\n", ldcp->id, msg->seqid,
884 		    (ldcp->last_msg_rcd + 1));
885 		return (EIO);
886 	}
887 
888 	return (0);
889 }
890 
891 
892 /*
893  * Process an incoming version ctrl message
894  */
895 static int
896 i_ldc_process_VER(ldc_chan_t *ldcp, ldc_msg_t *msg)
897 {
898 	int 		rv = 0, idx = ldcp->next_vidx;
899 	ldc_msg_t 	*pkt;
900 	uint64_t	tx_tail;
901 	ldc_ver_t	*rcvd_ver;
902 
903 	/* get the received version */
904 	rcvd_ver = (ldc_ver_t *)((uint64_t)msg + LDC_PAYLOAD_VER_OFF);
905 
906 	D2(ldcp->id, "i_ldc_process_VER: (0x%llx) received VER v%u.%u\n",
907 	    ldcp->id, rcvd_ver->major, rcvd_ver->minor);
908 
909 	/* Obtain Tx lock */
910 	mutex_enter(&ldcp->tx_lock);
911 
912 	switch (msg->stype) {
913 	case LDC_INFO:
914 
915 		if ((ldcp->tstate & ~TS_IN_RESET) == TS_VREADY) {
916 			(void) i_ldc_txq_reconf(ldcp);
917 			i_ldc_reset_state(ldcp);
918 			mutex_exit(&ldcp->tx_lock);
919 			return (EAGAIN);
920 		}
921 
922 		/* get the current tail and pkt for the response */
923 		rv = i_ldc_get_tx_tail(ldcp, &tx_tail);
924 		if (rv != 0) {
925 			DWARN(ldcp->id,
926 			    "i_ldc_process_VER: (0x%llx) err sending "
927 			    "version ACK/NACK\n", ldcp->id);
928 			i_ldc_reset(ldcp, B_TRUE);
929 			mutex_exit(&ldcp->tx_lock);
930 			return (ECONNRESET);
931 		}
932 
933 		pkt = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail);
934 		ZERO_PKT(pkt);
935 
936 		/* initialize the packet */
937 		pkt->type = LDC_CTRL;
938 		pkt->ctrl = LDC_VER;
939 
940 		for (;;) {
941 
942 			D1(ldcp->id, "i_ldc_process_VER: got %u.%u chk %u.%u\n",
943 			    rcvd_ver->major, rcvd_ver->minor,
944 			    ldc_versions[idx].major, ldc_versions[idx].minor);
945 
946 			if (rcvd_ver->major == ldc_versions[idx].major) {
947 				/* major version match - ACK version */
948 				pkt->stype = LDC_ACK;
949 
950 				/*
951 				 * lower minor version to the one this endpt
952 				 * supports, if necessary
953 				 */
954 				if (rcvd_ver->minor > ldc_versions[idx].minor)
955 					rcvd_ver->minor =
956 						ldc_versions[idx].minor;
957 				bcopy(rcvd_ver, pkt->udata, sizeof (*rcvd_ver));
958 
959 				break;
960 			}
961 
962 			if (rcvd_ver->major > ldc_versions[idx].major) {
963 
964 				D1(ldcp->id, "i_ldc_process_VER: using next"
965 				    " lower idx=%d, v%u.%u\n", idx,
966 				    ldc_versions[idx].major,
967 				    ldc_versions[idx].minor);
968 
969 				/* nack with next lower version */
970 				pkt->stype = LDC_NACK;
971 				bcopy(&ldc_versions[idx], pkt->udata,
972 				    sizeof (ldc_versions[idx]));
973 				ldcp->next_vidx = idx;
974 				break;
975 			}
976 
977 			/* next major version */
978 			idx++;
979 
980 			D1(ldcp->id, "i_ldc_process_VER: inc idx %x\n", idx);
981 
982 			if (idx == LDC_NUM_VERS) {
983 				/* no version match - send NACK */
984 				pkt->stype = LDC_NACK;
985 				bzero(pkt->udata, sizeof (ldc_ver_t));
986 				ldcp->next_vidx = 0;
987 				break;
988 			}
989 		}
990 
991 		/* initiate the send by calling into HV and set the new tail */
992 		tx_tail = (tx_tail + LDC_PACKET_SIZE) %
993 			(ldcp->tx_q_entries << LDC_PACKET_SHIFT);
994 
995 		rv = i_ldc_set_tx_tail(ldcp, tx_tail);
996 		if (rv == 0) {
997 			ldcp->tx_tail = tx_tail;
998 			if (pkt->stype == LDC_ACK) {
999 				D2(ldcp->id, "i_ldc_process_VER: (0x%llx) sent"
1000 				    " version ACK\n", ldcp->id);
1001 				/* Save the ACK'd version */
1002 				ldcp->version.major = rcvd_ver->major;
1003 				ldcp->version.minor = rcvd_ver->minor;
1004 				ldcp->hstate |= TS_RCVD_VER;
1005 				ldcp->tstate |= TS_VER_DONE;
1006 				DWARN(DBG_ALL_LDCS,
1007 				    "(0x%llx) Sent ACK, "
1008 				    "Agreed on version v%u.%u\n",
1009 				    ldcp->id, rcvd_ver->major, rcvd_ver->minor);
1010 			}
1011 		} else {
1012 			DWARN(ldcp->id,
1013 			    "i_ldc_process_VER: (0x%llx) error sending "
1014 			    "ACK/NACK\n", ldcp->id);
1015 			i_ldc_reset(ldcp, B_TRUE);
1016 			mutex_exit(&ldcp->tx_lock);
1017 			return (ECONNRESET);
1018 		}
1019 
1020 		break;
1021 
1022 	case LDC_ACK:
1023 		if ((ldcp->tstate & ~TS_IN_RESET) == TS_VREADY) {
1024 			if (ldcp->version.major != rcvd_ver->major ||
1025 				ldcp->version.minor != rcvd_ver->minor) {
1026 
1027 				/* mismatched version - reset connection */
1028 				DWARN(ldcp->id,
1029 					"i_ldc_process_VER: (0x%llx) recvd"
1030 					" ACK ver != sent ACK ver\n", ldcp->id);
1031 				i_ldc_reset(ldcp, B_TRUE);
1032 				mutex_exit(&ldcp->tx_lock);
1033 				return (ECONNRESET);
1034 			}
1035 		} else {
1036 			/* SUCCESS - we have agreed on a version */
1037 			ldcp->version.major = rcvd_ver->major;
1038 			ldcp->version.minor = rcvd_ver->minor;
1039 			ldcp->tstate |= TS_VER_DONE;
1040 		}
1041 
1042 		DWARN(DBG_ALL_LDCS,
1043 		    "(0x%llx) Got ACK, Agreed on version v%u.%u\n",
1044 		    ldcp->id, rcvd_ver->major, rcvd_ver->minor);
1045 
1046 		/* initiate RTS-RTR-RDX handshake */
1047 		rv = i_ldc_get_tx_tail(ldcp, &tx_tail);
1048 		if (rv) {
1049 			DWARN(ldcp->id,
1050 		    "i_ldc_process_VER: (0x%llx) cannot send RTS\n",
1051 			    ldcp->id);
1052 			i_ldc_reset(ldcp, B_TRUE);
1053 			mutex_exit(&ldcp->tx_lock);
1054 			return (ECONNRESET);
1055 		}
1056 
1057 		pkt = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail);
1058 		ZERO_PKT(pkt);
1059 
1060 		pkt->type = LDC_CTRL;
1061 		pkt->stype = LDC_INFO;
1062 		pkt->ctrl = LDC_RTS;
1063 		pkt->env = ldcp->mode;
1064 		if (ldcp->mode != LDC_MODE_RAW)
1065 			pkt->seqid = LDC_INIT_SEQID;
1066 
1067 		ldcp->last_msg_rcd = LDC_INIT_SEQID;
1068 
1069 		DUMP_LDC_PKT(ldcp, "i_ldc_process_VER snd rts", (uint64_t)pkt);
1070 
1071 		/* initiate the send by calling into HV and set the new tail */
1072 		tx_tail = (tx_tail + LDC_PACKET_SIZE) %
1073 			(ldcp->tx_q_entries << LDC_PACKET_SHIFT);
1074 
1075 		rv = i_ldc_set_tx_tail(ldcp, tx_tail);
1076 		if (rv) {
1077 			D2(ldcp->id,
1078 			    "i_ldc_process_VER: (0x%llx) no listener\n",
1079 			    ldcp->id);
1080 			i_ldc_reset(ldcp, B_TRUE);
1081 			mutex_exit(&ldcp->tx_lock);
1082 			return (ECONNRESET);
1083 		}
1084 
1085 		ldcp->tx_tail = tx_tail;
1086 		ldcp->hstate |= TS_SENT_RTS;
1087 
1088 		break;
1089 
1090 	case LDC_NACK:
1091 		/* check if version in NACK is zero */
1092 		if (rcvd_ver->major == 0 && rcvd_ver->minor == 0) {
1093 			/* version handshake failure */
1094 			DWARN(DBG_ALL_LDCS,
1095 			    "i_ldc_process_VER: (0x%llx) no version match\n",
1096 			    ldcp->id);
1097 			i_ldc_reset(ldcp, B_TRUE);
1098 			mutex_exit(&ldcp->tx_lock);
1099 			return (ECONNRESET);
1100 		}
1101 
1102 		/* get the current tail and pkt for the response */
1103 		rv = i_ldc_get_tx_tail(ldcp, &tx_tail);
1104 		if (rv != 0) {
1105 			cmn_err(CE_NOTE,
1106 			    "i_ldc_process_VER: (0x%lx) err sending "
1107 			    "version ACK/NACK\n", ldcp->id);
1108 			i_ldc_reset(ldcp, B_TRUE);
1109 			mutex_exit(&ldcp->tx_lock);
1110 			return (ECONNRESET);
1111 		}
1112 
1113 		pkt = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail);
1114 		ZERO_PKT(pkt);
1115 
1116 		/* initialize the packet */
1117 		pkt->type = LDC_CTRL;
1118 		pkt->ctrl = LDC_VER;
1119 		pkt->stype = LDC_INFO;
1120 
1121 		/* check ver in NACK msg has a match */
1122 		for (;;) {
1123 			if (rcvd_ver->major == ldc_versions[idx].major) {
1124 				/*
1125 				 * major version match - resubmit request
1126 				 * if lower minor version to the one this endpt
1127 				 * supports, if necessary
1128 				 */
1129 				if (rcvd_ver->minor > ldc_versions[idx].minor)
1130 					rcvd_ver->minor =
1131 						ldc_versions[idx].minor;
1132 				bcopy(rcvd_ver, pkt->udata, sizeof (*rcvd_ver));
1133 				break;
1134 
1135 			}
1136 
1137 			if (rcvd_ver->major > ldc_versions[idx].major) {
1138 
1139 				D1(ldcp->id, "i_ldc_process_VER: using next"
1140 				    " lower idx=%d, v%u.%u\n", idx,
1141 				    ldc_versions[idx].major,
1142 				    ldc_versions[idx].minor);
1143 
1144 				/* send next lower version */
1145 				bcopy(&ldc_versions[idx], pkt->udata,
1146 				    sizeof (ldc_versions[idx]));
1147 				ldcp->next_vidx = idx;
1148 				break;
1149 			}
1150 
1151 			/* next version */
1152 			idx++;
1153 
1154 			D1(ldcp->id, "i_ldc_process_VER: inc idx %x\n", idx);
1155 
1156 			if (idx == LDC_NUM_VERS) {
1157 				/* no version match - terminate */
1158 				ldcp->next_vidx = 0;
1159 				mutex_exit(&ldcp->tx_lock);
1160 				return (ECONNRESET);
1161 			}
1162 		}
1163 
1164 		/* initiate the send by calling into HV and set the new tail */
1165 		tx_tail = (tx_tail + LDC_PACKET_SIZE) %
1166 			(ldcp->tx_q_entries << LDC_PACKET_SHIFT);
1167 
1168 		rv = i_ldc_set_tx_tail(ldcp, tx_tail);
1169 		if (rv == 0) {
1170 			D2(ldcp->id, "i_ldc_process_VER: (0x%llx) sent version"
1171 			    "INFO v%u.%u\n", ldcp->id, ldc_versions[idx].major,
1172 			    ldc_versions[idx].minor);
1173 			ldcp->tx_tail = tx_tail;
1174 		} else {
1175 			cmn_err(CE_NOTE,
1176 			    "i_ldc_process_VER: (0x%lx) error sending version"
1177 			    "INFO\n", ldcp->id);
1178 			i_ldc_reset(ldcp, B_TRUE);
1179 			mutex_exit(&ldcp->tx_lock);
1180 			return (ECONNRESET);
1181 		}
1182 
1183 		break;
1184 	}
1185 
1186 	mutex_exit(&ldcp->tx_lock);
1187 	return (rv);
1188 }
1189 
1190 
1191 /*
1192  * Process an incoming RTS ctrl message
1193  */
1194 static int
1195 i_ldc_process_RTS(ldc_chan_t *ldcp, ldc_msg_t *msg)
1196 {
1197 	int 		rv = 0;
1198 	ldc_msg_t 	*pkt;
1199 	uint64_t	tx_tail;
1200 	boolean_t	sent_NACK = B_FALSE;
1201 
1202 	D2(ldcp->id, "i_ldc_process_RTS: (0x%llx) received RTS\n", ldcp->id);
1203 
1204 	switch (msg->stype) {
1205 	case LDC_NACK:
1206 		DWARN(ldcp->id,
1207 		    "i_ldc_process_RTS: (0x%llx) RTS NACK received\n",
1208 		    ldcp->id);
1209 
1210 		/* Reset the channel -- as we cannot continue */
1211 		mutex_enter(&ldcp->tx_lock);
1212 		i_ldc_reset(ldcp, B_TRUE);
1213 		mutex_exit(&ldcp->tx_lock);
1214 		rv = ECONNRESET;
1215 		break;
1216 
1217 	case LDC_INFO:
1218 
1219 		/* check mode */
1220 		if (ldcp->mode != (ldc_mode_t)msg->env) {
1221 			cmn_err(CE_NOTE,
1222 			    "i_ldc_process_RTS: (0x%lx) mode mismatch\n",
1223 			    ldcp->id);
1224 			/*
1225 			 * send NACK in response to MODE message
1226 			 * get the current tail for the response
1227 			 */
1228 			rv = i_ldc_send_pkt(ldcp, LDC_CTRL, LDC_NACK, LDC_RTS);
1229 			if (rv) {
1230 				/* if cannot send NACK - reset channel */
1231 				mutex_enter(&ldcp->tx_lock);
1232 				i_ldc_reset(ldcp, B_TRUE);
1233 				mutex_exit(&ldcp->tx_lock);
1234 				rv = ECONNRESET;
1235 				break;
1236 			}
1237 			sent_NACK = B_TRUE;
1238 		}
1239 		break;
1240 	default:
1241 		DWARN(ldcp->id, "i_ldc_process_RTS: (0x%llx) unexp ACK\n",
1242 		    ldcp->id);
1243 		mutex_enter(&ldcp->tx_lock);
1244 		i_ldc_reset(ldcp, B_TRUE);
1245 		mutex_exit(&ldcp->tx_lock);
1246 		rv = ECONNRESET;
1247 		break;
1248 	}
1249 
1250 	/*
1251 	 * If either the connection was reset (when rv != 0) or
1252 	 * a NACK was sent, we return. In the case of a NACK
1253 	 * we dont want to consume the packet that came in but
1254 	 * not record that we received the RTS
1255 	 */
1256 	if (rv || sent_NACK)
1257 		return (rv);
1258 
1259 	/* record RTS received */
1260 	ldcp->hstate |= TS_RCVD_RTS;
1261 
1262 	/* store initial SEQID info */
1263 	ldcp->last_msg_snt = msg->seqid;
1264 
1265 	/* Obtain Tx lock */
1266 	mutex_enter(&ldcp->tx_lock);
1267 
1268 	/* get the current tail for the response */
1269 	rv = i_ldc_get_tx_tail(ldcp, &tx_tail);
1270 	if (rv != 0) {
1271 		cmn_err(CE_NOTE,
1272 		    "i_ldc_process_RTS: (0x%lx) err sending RTR\n",
1273 		    ldcp->id);
1274 		i_ldc_reset(ldcp, B_TRUE);
1275 		mutex_exit(&ldcp->tx_lock);
1276 		return (ECONNRESET);
1277 	}
1278 
1279 	pkt = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail);
1280 	ZERO_PKT(pkt);
1281 
1282 	/* initialize the packet */
1283 	pkt->type = LDC_CTRL;
1284 	pkt->stype = LDC_INFO;
1285 	pkt->ctrl = LDC_RTR;
1286 	pkt->env = ldcp->mode;
1287 	if (ldcp->mode != LDC_MODE_RAW)
1288 		pkt->seqid = LDC_INIT_SEQID;
1289 
1290 	ldcp->last_msg_rcd = msg->seqid;
1291 
1292 	/* initiate the send by calling into HV and set the new tail */
1293 	tx_tail = (tx_tail + LDC_PACKET_SIZE) %
1294 		(ldcp->tx_q_entries << LDC_PACKET_SHIFT);
1295 
1296 	rv = i_ldc_set_tx_tail(ldcp, tx_tail);
1297 	if (rv == 0) {
1298 		D2(ldcp->id,
1299 		    "i_ldc_process_RTS: (0x%llx) sent RTR\n", ldcp->id);
1300 		DUMP_LDC_PKT(ldcp, "i_ldc_process_RTS sent rtr", (uint64_t)pkt);
1301 
1302 		ldcp->tx_tail = tx_tail;
1303 		ldcp->hstate |= TS_SENT_RTR;
1304 
1305 	} else {
1306 		cmn_err(CE_NOTE,
1307 		    "i_ldc_process_RTS: (0x%lx) error sending RTR\n",
1308 		    ldcp->id);
1309 		i_ldc_reset(ldcp, B_TRUE);
1310 		mutex_exit(&ldcp->tx_lock);
1311 		return (ECONNRESET);
1312 	}
1313 
1314 	mutex_exit(&ldcp->tx_lock);
1315 	return (0);
1316 }
1317 
1318 /*
1319  * Process an incoming RTR ctrl message
1320  */
1321 static int
1322 i_ldc_process_RTR(ldc_chan_t *ldcp, ldc_msg_t *msg)
1323 {
1324 	int 		rv = 0;
1325 	boolean_t	sent_NACK = B_FALSE;
1326 
1327 	D2(ldcp->id, "i_ldc_process_RTR: (0x%llx) received RTR\n", ldcp->id);
1328 
1329 	switch (msg->stype) {
1330 	case LDC_NACK:
1331 		/* RTR NACK received */
1332 		DWARN(ldcp->id,
1333 		    "i_ldc_process_RTR: (0x%llx) RTR NACK received\n",
1334 		    ldcp->id);
1335 
1336 		/* Reset the channel -- as we cannot continue */
1337 		mutex_enter(&ldcp->tx_lock);
1338 		i_ldc_reset(ldcp, B_TRUE);
1339 		mutex_exit(&ldcp->tx_lock);
1340 		rv = ECONNRESET;
1341 
1342 		break;
1343 
1344 	case LDC_INFO:
1345 
1346 		/* check mode */
1347 		if (ldcp->mode != (ldc_mode_t)msg->env) {
1348 			DWARN(ldcp->id,
1349 			    "i_ldc_process_RTR: (0x%llx) mode mismatch\n",
1350 			    ldcp->id);
1351 			/*
1352 			 * send NACK in response to MODE message
1353 			 * get the current tail for the response
1354 			 */
1355 			rv = i_ldc_send_pkt(ldcp, LDC_CTRL, LDC_NACK, LDC_RTR);
1356 			if (rv) {
1357 				/* if cannot send NACK - reset channel */
1358 				mutex_enter(&ldcp->tx_lock);
1359 				i_ldc_reset(ldcp, B_TRUE);
1360 				mutex_exit(&ldcp->tx_lock);
1361 				rv = ECONNRESET;
1362 				break;
1363 			}
1364 			sent_NACK = B_TRUE;
1365 		}
1366 		break;
1367 
1368 	default:
1369 		DWARN(ldcp->id, "i_ldc_process_RTR: (0x%llx) unexp ACK\n",
1370 		    ldcp->id);
1371 
1372 		/* Reset the channel -- as we cannot continue */
1373 		mutex_enter(&ldcp->tx_lock);
1374 		i_ldc_reset(ldcp, B_TRUE);
1375 		mutex_exit(&ldcp->tx_lock);
1376 		rv = ECONNRESET;
1377 		break;
1378 	}
1379 
1380 	/*
1381 	 * If either the connection was reset (when rv != 0) or
1382 	 * a NACK was sent, we return. In the case of a NACK
1383 	 * we dont want to consume the packet that came in but
1384 	 * not record that we received the RTR
1385 	 */
1386 	if (rv || sent_NACK)
1387 		return (rv);
1388 
1389 	ldcp->last_msg_snt = msg->seqid;
1390 	ldcp->hstate |= TS_RCVD_RTR;
1391 
1392 	rv = i_ldc_send_pkt(ldcp, LDC_CTRL, LDC_INFO, LDC_RDX);
1393 	if (rv) {
1394 		cmn_err(CE_NOTE,
1395 		    "i_ldc_process_RTR: (0x%lx) cannot send RDX\n",
1396 		    ldcp->id);
1397 		mutex_enter(&ldcp->tx_lock);
1398 		i_ldc_reset(ldcp, B_TRUE);
1399 		mutex_exit(&ldcp->tx_lock);
1400 		return (ECONNRESET);
1401 	}
1402 	D2(ldcp->id,
1403 	    "i_ldc_process_RTR: (0x%llx) sent RDX\n", ldcp->id);
1404 
1405 	ldcp->hstate |= TS_SENT_RDX;
1406 	ldcp->tstate |= TS_HSHAKE_DONE;
1407 	if ((ldcp->tstate & TS_IN_RESET) == 0)
1408 		ldcp->status = LDC_UP;
1409 
1410 	DWARN(DBG_ALL_LDCS, "(0x%llx) Handshake Complete\n", ldcp->id);
1411 
1412 	return (0);
1413 }
1414 
1415 
1416 /*
1417  * Process an incoming RDX ctrl message
1418  */
1419 static int
1420 i_ldc_process_RDX(ldc_chan_t *ldcp, ldc_msg_t *msg)
1421 {
1422 	int	rv = 0;
1423 
1424 	D2(ldcp->id, "i_ldc_process_RDX: (0x%llx) received RDX\n", ldcp->id);
1425 
1426 	switch (msg->stype) {
1427 	case LDC_NACK:
1428 		/* RDX NACK received */
1429 		DWARN(ldcp->id,
1430 		    "i_ldc_process_RDX: (0x%llx) RDX NACK received\n",
1431 		    ldcp->id);
1432 
1433 		/* Reset the channel -- as we cannot continue */
1434 		mutex_enter(&ldcp->tx_lock);
1435 		i_ldc_reset(ldcp, B_TRUE);
1436 		mutex_exit(&ldcp->tx_lock);
1437 		rv = ECONNRESET;
1438 
1439 		break;
1440 
1441 	case LDC_INFO:
1442 
1443 		/*
1444 		 * if channel is UP and a RDX received after data transmission
1445 		 * has commenced it is an error
1446 		 */
1447 		if ((ldcp->tstate == TS_UP) && (ldcp->hstate & TS_RCVD_RDX)) {
1448 			DWARN(DBG_ALL_LDCS,
1449 			    "i_ldc_process_RDX: (0x%llx) unexpected RDX"
1450 			    " - LDC reset\n", ldcp->id);
1451 			mutex_enter(&ldcp->tx_lock);
1452 			i_ldc_reset(ldcp, B_TRUE);
1453 			mutex_exit(&ldcp->tx_lock);
1454 			return (ECONNRESET);
1455 		}
1456 
1457 		ldcp->hstate |= TS_RCVD_RDX;
1458 		ldcp->tstate |= TS_HSHAKE_DONE;
1459 		if ((ldcp->tstate & TS_IN_RESET) == 0)
1460 			ldcp->status = LDC_UP;
1461 
1462 		D1(DBG_ALL_LDCS, "(0x%llx) Handshake Complete\n", ldcp->id);
1463 		break;
1464 
1465 	default:
1466 		DWARN(ldcp->id, "i_ldc_process_RDX: (0x%llx) unexp ACK\n",
1467 		    ldcp->id);
1468 
1469 		/* Reset the channel -- as we cannot continue */
1470 		mutex_enter(&ldcp->tx_lock);
1471 		i_ldc_reset(ldcp, B_TRUE);
1472 		mutex_exit(&ldcp->tx_lock);
1473 		rv = ECONNRESET;
1474 		break;
1475 	}
1476 
1477 	return (rv);
1478 }
1479 
1480 /*
1481  * Process an incoming ACK for a data packet
1482  */
1483 static int
1484 i_ldc_process_data_ACK(ldc_chan_t *ldcp, ldc_msg_t *msg)
1485 {
1486 	int		rv;
1487 	uint64_t 	tx_head;
1488 	ldc_msg_t	*pkt;
1489 
1490 	/* Obtain Tx lock */
1491 	mutex_enter(&ldcp->tx_lock);
1492 
1493 	/*
1494 	 * Read the current Tx head and tail
1495 	 */
1496 	rv = hv_ldc_tx_get_state(ldcp->id,
1497 	    &ldcp->tx_head, &ldcp->tx_tail, &ldcp->link_state);
1498 	if (rv != 0) {
1499 		cmn_err(CE_WARN,
1500 		    "i_ldc_process_data_ACK: (0x%lx) cannot read qptrs\n",
1501 		    ldcp->id);
1502 
1503 		/* Reset the channel -- as we cannot continue */
1504 		i_ldc_reset(ldcp, B_TRUE);
1505 		mutex_exit(&ldcp->tx_lock);
1506 		return (ECONNRESET);
1507 	}
1508 
1509 	/*
1510 	 * loop from where the previous ACK location was to the
1511 	 * current head location. This is how far the HV has
1512 	 * actually send pkts. Pkts between head and tail are
1513 	 * yet to be sent by HV.
1514 	 */
1515 	tx_head = ldcp->tx_ackd_head;
1516 	for (;;) {
1517 		pkt = (ldc_msg_t *)(ldcp->tx_q_va + tx_head);
1518 		tx_head = (tx_head + LDC_PACKET_SIZE) %
1519 			(ldcp->tx_q_entries << LDC_PACKET_SHIFT);
1520 
1521 		if (pkt->seqid == msg->ackid) {
1522 			D2(ldcp->id,
1523 			    "i_ldc_process_data_ACK: (0x%llx) found packet\n",
1524 			    ldcp->id);
1525 			ldcp->last_ack_rcd = msg->ackid;
1526 			ldcp->tx_ackd_head = tx_head;
1527 			break;
1528 		}
1529 		if (tx_head == ldcp->tx_head) {
1530 			/* could not find packet */
1531 			DWARN(ldcp->id,
1532 			    "i_ldc_process_data_ACK: (0x%llx) invalid ACKid\n",
1533 			    ldcp->id);
1534 
1535 			/* Reset the channel -- as we cannot continue */
1536 			i_ldc_reset(ldcp, B_TRUE);
1537 			mutex_exit(&ldcp->tx_lock);
1538 			return (ECONNRESET);
1539 		}
1540 	}
1541 
1542 	mutex_exit(&ldcp->tx_lock);
1543 	return (0);
1544 }
1545 
1546 /*
1547  * Process incoming control message
1548  * Return 0 - session can continue
1549  *        EAGAIN - reprocess packet - state was changed
1550  *	  ECONNRESET - channel was reset
1551  */
1552 static int
1553 i_ldc_ctrlmsg(ldc_chan_t *ldcp, ldc_msg_t *msg)
1554 {
1555 	int 		rv = 0;
1556 
1557 	D1(ldcp->id, "i_ldc_ctrlmsg: (%llx) tstate = %lx, hstate = %lx\n",
1558 	    ldcp->id, ldcp->tstate, ldcp->hstate);
1559 
1560 	switch (ldcp->tstate & ~TS_IN_RESET) {
1561 
1562 	case TS_OPEN:
1563 	case TS_READY:
1564 
1565 		switch (msg->ctrl & LDC_CTRL_MASK) {
1566 		case LDC_VER:
1567 			/* process version message */
1568 			rv = i_ldc_process_VER(ldcp, msg);
1569 			break;
1570 		default:
1571 			DWARN(ldcp->id,
1572 			    "i_ldc_ctrlmsg: (0x%llx) unexp ctrl 0x%x "
1573 			    "tstate=0x%x\n", ldcp->id,
1574 			    (msg->ctrl & LDC_CTRL_MASK), ldcp->tstate);
1575 			break;
1576 		}
1577 
1578 		break;
1579 
1580 	case TS_VREADY:
1581 
1582 		switch (msg->ctrl & LDC_CTRL_MASK) {
1583 		case LDC_VER:
1584 			/* process version message */
1585 			rv = i_ldc_process_VER(ldcp, msg);
1586 			break;
1587 		case LDC_RTS:
1588 			/* process RTS message */
1589 			rv = i_ldc_process_RTS(ldcp, msg);
1590 			break;
1591 		case LDC_RTR:
1592 			/* process RTR message */
1593 			rv = i_ldc_process_RTR(ldcp, msg);
1594 			break;
1595 		case LDC_RDX:
1596 			/* process RDX message */
1597 			rv = i_ldc_process_RDX(ldcp, msg);
1598 			break;
1599 		default:
1600 			DWARN(ldcp->id,
1601 			    "i_ldc_ctrlmsg: (0x%llx) unexp ctrl 0x%x "
1602 			    "tstate=0x%x\n", ldcp->id,
1603 			    (msg->ctrl & LDC_CTRL_MASK), ldcp->tstate);
1604 			break;
1605 		}
1606 
1607 		break;
1608 
1609 	case TS_UP:
1610 
1611 		switch (msg->ctrl & LDC_CTRL_MASK) {
1612 		case LDC_VER:
1613 			DWARN(ldcp->id,
1614 			    "i_ldc_ctrlmsg: (0x%llx) unexpected VER "
1615 			    "- LDC reset\n", ldcp->id);
1616 			/* peer is redoing version negotiation */
1617 			mutex_enter(&ldcp->tx_lock);
1618 			(void) i_ldc_txq_reconf(ldcp);
1619 			i_ldc_reset_state(ldcp);
1620 			mutex_exit(&ldcp->tx_lock);
1621 			rv = EAGAIN;
1622 			break;
1623 
1624 		case LDC_RDX:
1625 			/* process RDX message */
1626 			rv = i_ldc_process_RDX(ldcp, msg);
1627 			break;
1628 
1629 		default:
1630 			DWARN(ldcp->id,
1631 			    "i_ldc_ctrlmsg: (0x%llx) unexp ctrl 0x%x "
1632 			    "tstate=0x%x\n", ldcp->id,
1633 			    (msg->ctrl & LDC_CTRL_MASK), ldcp->tstate);
1634 			break;
1635 		}
1636 	}
1637 
1638 	return (rv);
1639 }
1640 
1641 /*
1642  * Register channel with the channel nexus
1643  */
1644 static int
1645 i_ldc_register_channel(ldc_chan_t *ldcp)
1646 {
1647 	int		rv = 0;
1648 	ldc_cnex_t	*cinfo = &ldcssp->cinfo;
1649 
1650 	if (cinfo->dip == NULL) {
1651 		DWARN(ldcp->id,
1652 		    "i_ldc_register_channel: cnex has not registered\n");
1653 		return (EAGAIN);
1654 	}
1655 
1656 	rv = cinfo->reg_chan(cinfo->dip, ldcp->id, ldcp->devclass);
1657 	if (rv) {
1658 		DWARN(ldcp->id,
1659 		    "i_ldc_register_channel: cannot register channel\n");
1660 		return (rv);
1661 	}
1662 
1663 	rv = cinfo->add_intr(cinfo->dip, ldcp->id, CNEX_TX_INTR,
1664 	    i_ldc_tx_hdlr, ldcp, NULL);
1665 	if (rv) {
1666 		DWARN(ldcp->id,
1667 		    "i_ldc_register_channel: cannot add Tx interrupt\n");
1668 		(void) cinfo->unreg_chan(cinfo->dip, ldcp->id);
1669 		return (rv);
1670 	}
1671 
1672 	rv = cinfo->add_intr(cinfo->dip, ldcp->id, CNEX_RX_INTR,
1673 	    i_ldc_rx_hdlr, ldcp, NULL);
1674 	if (rv) {
1675 		DWARN(ldcp->id,
1676 		    "i_ldc_register_channel: cannot add Rx interrupt\n");
1677 		(void) cinfo->rem_intr(cinfo->dip, ldcp->id, CNEX_TX_INTR);
1678 		(void) cinfo->unreg_chan(cinfo->dip, ldcp->id);
1679 		return (rv);
1680 	}
1681 
1682 	ldcp->tstate |= TS_CNEX_RDY;
1683 
1684 	return (0);
1685 }
1686 
1687 /*
1688  * Unregister a channel with the channel nexus
1689  */
1690 static int
1691 i_ldc_unregister_channel(ldc_chan_t *ldcp)
1692 {
1693 	int		rv = 0;
1694 	ldc_cnex_t	*cinfo = &ldcssp->cinfo;
1695 
1696 	if (cinfo->dip == NULL) {
1697 		DWARN(ldcp->id,
1698 		    "i_ldc_unregister_channel: cnex has not registered\n");
1699 		return (EAGAIN);
1700 	}
1701 
1702 	if (ldcp->tstate & TS_CNEX_RDY) {
1703 
1704 		/* Remove the Rx interrupt */
1705 		rv = cinfo->rem_intr(cinfo->dip, ldcp->id, CNEX_RX_INTR);
1706 		if (rv) {
1707 			if (rv != EAGAIN) {
1708 				DWARN(ldcp->id,
1709 				    "i_ldc_unregister_channel: err removing "
1710 				    "Rx intr\n");
1711 				return (rv);
1712 			}
1713 
1714 			/*
1715 			 * If interrupts are pending and handler has
1716 			 * finished running, clear interrupt and try
1717 			 * again
1718 			 */
1719 			if (ldcp->rx_intr_state != LDC_INTR_PEND)
1720 				return (rv);
1721 
1722 			(void) i_ldc_clear_intr(ldcp, CNEX_RX_INTR);
1723 			rv = cinfo->rem_intr(cinfo->dip, ldcp->id,
1724 			    CNEX_RX_INTR);
1725 			if (rv) {
1726 				DWARN(ldcp->id, "i_ldc_unregister_channel: "
1727 				    "err removing Rx interrupt\n");
1728 				return (rv);
1729 			}
1730 		}
1731 
1732 		/* Remove the Tx interrupt */
1733 		rv = cinfo->rem_intr(cinfo->dip, ldcp->id, CNEX_TX_INTR);
1734 		if (rv) {
1735 			DWARN(ldcp->id,
1736 			    "i_ldc_unregister_channel: err removing Tx intr\n");
1737 			return (rv);
1738 		}
1739 
1740 		/* Unregister the channel */
1741 		rv = cinfo->unreg_chan(ldcssp->cinfo.dip, ldcp->id);
1742 		if (rv) {
1743 			DWARN(ldcp->id,
1744 			    "i_ldc_unregister_channel: cannot unreg channel\n");
1745 			return (rv);
1746 		}
1747 
1748 		ldcp->tstate &= ~TS_CNEX_RDY;
1749 	}
1750 
1751 	return (0);
1752 }
1753 
1754 
1755 /*
1756  * LDC transmit interrupt handler
1757  *    triggered for chanel up/down/reset events
1758  *    and Tx queue content changes
1759  */
1760 static uint_t
1761 i_ldc_tx_hdlr(caddr_t arg1, caddr_t arg2)
1762 {
1763 	_NOTE(ARGUNUSED(arg2))
1764 
1765 	int 		rv;
1766 	ldc_chan_t 	*ldcp;
1767 	boolean_t 	notify_client = B_FALSE;
1768 	uint64_t	notify_event = 0, link_state;
1769 
1770 	/* Get the channel for which interrupt was received */
1771 	ASSERT(arg1 != NULL);
1772 	ldcp = (ldc_chan_t *)arg1;
1773 
1774 	D1(ldcp->id, "i_ldc_tx_hdlr: (0x%llx) Received intr, ldcp=0x%p\n",
1775 	    ldcp->id, ldcp);
1776 
1777 	/* Lock channel */
1778 	mutex_enter(&ldcp->lock);
1779 
1780 	/* Obtain Tx lock */
1781 	mutex_enter(&ldcp->tx_lock);
1782 
1783 	/* mark interrupt as pending */
1784 	ldcp->tx_intr_state = LDC_INTR_ACTIVE;
1785 
1786 	/* save current link state */
1787 	link_state = ldcp->link_state;
1788 
1789 	rv = hv_ldc_tx_get_state(ldcp->id, &ldcp->tx_head, &ldcp->tx_tail,
1790 	    &ldcp->link_state);
1791 	if (rv) {
1792 		cmn_err(CE_WARN,
1793 		    "i_ldc_tx_hdlr: (0x%lx) cannot read queue ptrs rv=0x%d\n",
1794 		    ldcp->id, rv);
1795 		i_ldc_clear_intr(ldcp, CNEX_TX_INTR);
1796 		mutex_exit(&ldcp->tx_lock);
1797 		mutex_exit(&ldcp->lock);
1798 		return (DDI_INTR_CLAIMED);
1799 	}
1800 
1801 	/*
1802 	 * reset the channel state if the channel went down
1803 	 * (other side unconfigured queue) or channel was reset
1804 	 * (other side reconfigured its queue)
1805 	 */
1806 	if (link_state != ldcp->link_state &&
1807 	    ldcp->link_state == LDC_CHANNEL_DOWN) {
1808 		D1(ldcp->id, "i_ldc_tx_hdlr: channel link down\n", ldcp->id);
1809 		i_ldc_reset(ldcp, B_FALSE);
1810 		notify_client = B_TRUE;
1811 		notify_event = LDC_EVT_DOWN;
1812 	}
1813 
1814 	if (link_state != ldcp->link_state &&
1815 	    ldcp->link_state == LDC_CHANNEL_RESET) {
1816 		D1(ldcp->id, "i_ldc_tx_hdlr: channel link reset\n", ldcp->id);
1817 		i_ldc_reset(ldcp, B_FALSE);
1818 		notify_client = B_TRUE;
1819 		notify_event = LDC_EVT_RESET;
1820 	}
1821 
1822 	if (link_state != ldcp->link_state &&
1823 	    (ldcp->tstate & ~TS_IN_RESET) == TS_OPEN &&
1824 	    ldcp->link_state == LDC_CHANNEL_UP) {
1825 		D1(ldcp->id, "i_ldc_tx_hdlr: channel link up\n", ldcp->id);
1826 		notify_client = B_TRUE;
1827 		notify_event = LDC_EVT_RESET;
1828 		ldcp->tstate |= TS_LINK_READY;
1829 		ldcp->status = LDC_READY;
1830 	}
1831 
1832 	/* if callbacks are disabled, do not notify */
1833 	if (!ldcp->cb_enabled)
1834 		notify_client = B_FALSE;
1835 
1836 	/* Unlock channel */
1837 
1838 	if (notify_client) {
1839 		ldcp->cb_inprogress = B_TRUE;
1840 		mutex_exit(&ldcp->tx_lock);
1841 		mutex_exit(&ldcp->lock);
1842 		rv = ldcp->cb(notify_event, ldcp->cb_arg);
1843 		if (rv) {
1844 			DWARN(ldcp->id, "i_ldc_tx_hdlr: (0x%llx) callback "
1845 			    "failure", ldcp->id);
1846 		}
1847 		mutex_enter(&ldcp->lock);
1848 		ldcp->cb_inprogress = B_FALSE;
1849 	}
1850 
1851 	i_ldc_clear_intr(ldcp, CNEX_TX_INTR);
1852 	mutex_exit(&ldcp->lock);
1853 
1854 	D1(ldcp->id, "i_ldc_tx_hdlr: (0x%llx) exiting handler", ldcp->id);
1855 
1856 	return (DDI_INTR_CLAIMED);
1857 }
1858 
1859 /*
1860  * LDC receive interrupt handler
1861  *    triggered for channel with data pending to read
1862  *    i.e. Rx queue content changes
1863  */
1864 static uint_t
1865 i_ldc_rx_hdlr(caddr_t arg1, caddr_t arg2)
1866 {
1867 	_NOTE(ARGUNUSED(arg2))
1868 
1869 	int		rv;
1870 	uint64_t 	rx_head, rx_tail;
1871 	ldc_msg_t 	*msg;
1872 	ldc_chan_t 	*ldcp;
1873 	boolean_t 	notify_client = B_FALSE;
1874 	uint64_t	notify_event = 0;
1875 	uint64_t	link_state, first_fragment = 0;
1876 
1877 
1878 	/* Get the channel for which interrupt was received */
1879 	if (arg1 == NULL) {
1880 		cmn_err(CE_WARN, "i_ldc_rx_hdlr: invalid arg\n");
1881 		return (DDI_INTR_UNCLAIMED);
1882 	}
1883 
1884 	ldcp = (ldc_chan_t *)arg1;
1885 
1886 	D1(ldcp->id, "i_ldc_rx_hdlr: (0x%llx) Received intr, ldcp=0x%p\n",
1887 	    ldcp->id, ldcp);
1888 	D1(ldcp->id, "i_ldc_rx_hdlr: (%llx) USR%lx/TS%lx/HS%lx, LSTATE=%lx\n",
1889 	    ldcp->id, ldcp->status, ldcp->tstate, ldcp->hstate,
1890 	    ldcp->link_state);
1891 
1892 	/* Lock channel */
1893 	mutex_enter(&ldcp->lock);
1894 
1895 	/* mark interrupt as pending */
1896 	ldcp->rx_intr_state = LDC_INTR_ACTIVE;
1897 
1898 	/*
1899 	 * Read packet(s) from the queue
1900 	 */
1901 	for (;;) {
1902 
1903 		link_state = ldcp->link_state;
1904 		rv = hv_ldc_rx_get_state(ldcp->id, &rx_head, &rx_tail,
1905 		    &ldcp->link_state);
1906 		if (rv) {
1907 			cmn_err(CE_WARN,
1908 			    "i_ldc_rx_hdlr: (0x%lx) cannot read "
1909 			    "queue ptrs, rv=0x%d\n", ldcp->id, rv);
1910 			i_ldc_clear_intr(ldcp, CNEX_RX_INTR);
1911 			mutex_exit(&ldcp->lock);
1912 			return (DDI_INTR_CLAIMED);
1913 		}
1914 
1915 		/*
1916 		 * reset the channel state if the channel went down
1917 		 * (other side unconfigured queue) or channel was reset
1918 		 * (other side reconfigured its queue)
1919 		 */
1920 
1921 		if (link_state != ldcp->link_state) {
1922 			switch (ldcp->link_state) {
1923 			case LDC_CHANNEL_DOWN:
1924 				D1(ldcp->id, "i_ldc_rx_hdlr: channel "
1925 				    "link down\n", ldcp->id);
1926 				mutex_enter(&ldcp->tx_lock);
1927 				i_ldc_reset(ldcp, B_FALSE);
1928 				mutex_exit(&ldcp->tx_lock);
1929 				notify_client = B_TRUE;
1930 				notify_event = LDC_EVT_DOWN;
1931 				goto loop_exit;
1932 
1933 			case LDC_CHANNEL_UP:
1934 				D1(ldcp->id, "i_ldc_rx_hdlr: "
1935 				    "channel link up\n", ldcp->id);
1936 
1937 				if ((ldcp->tstate & ~TS_IN_RESET) == TS_OPEN) {
1938 					notify_client = B_TRUE;
1939 					notify_event = LDC_EVT_RESET;
1940 					ldcp->tstate |= TS_LINK_READY;
1941 					ldcp->status = LDC_READY;
1942 				}
1943 				break;
1944 
1945 			case LDC_CHANNEL_RESET:
1946 			default:
1947 #ifdef DEBUG
1948 force_reset:
1949 #endif
1950 				D1(ldcp->id, "i_ldc_rx_hdlr: channel "
1951 				    "link reset\n", ldcp->id);
1952 				mutex_enter(&ldcp->tx_lock);
1953 				i_ldc_reset(ldcp, B_FALSE);
1954 				mutex_exit(&ldcp->tx_lock);
1955 				notify_client = B_TRUE;
1956 				notify_event = LDC_EVT_RESET;
1957 				break;
1958 			}
1959 		}
1960 
1961 #ifdef DEBUG
1962 		if (LDC_INJECT_RESET(ldcp))
1963 			goto force_reset;
1964 #endif
1965 
1966 		if (rx_head == rx_tail) {
1967 			D2(ldcp->id, "i_ldc_rx_hdlr: (0x%llx) No packets\n",
1968 			    ldcp->id);
1969 			break;
1970 		}
1971 
1972 		D2(ldcp->id, "i_ldc_rx_hdlr: head=0x%llx, tail=0x%llx\n",
1973 		    rx_head, rx_tail);
1974 		DUMP_LDC_PKT(ldcp, "i_ldc_rx_hdlr rcd",
1975 		    ldcp->rx_q_va + rx_head);
1976 
1977 		/* get the message */
1978 		msg = (ldc_msg_t *)(ldcp->rx_q_va + rx_head);
1979 
1980 		/* if channel is in RAW mode or data pkt, notify and return */
1981 		if (ldcp->mode == LDC_MODE_RAW) {
1982 			notify_client = B_TRUE;
1983 			notify_event |= LDC_EVT_READ;
1984 			break;
1985 		}
1986 
1987 		if ((msg->type & LDC_DATA) && (msg->stype & LDC_INFO)) {
1988 
1989 			/* discard packet if channel is not up */
1990 			if ((ldcp->tstate & ~TS_IN_RESET) != TS_UP) {
1991 
1992 				/* move the head one position */
1993 				rx_head = (rx_head + LDC_PACKET_SIZE) %
1994 				(ldcp->rx_q_entries << LDC_PACKET_SHIFT);
1995 
1996 				if (rv = i_ldc_set_rx_head(ldcp, rx_head))
1997 					break;
1998 
1999 				continue;
2000 			} else {
2001 				if ((ldcp->tstate & TS_IN_RESET) == 0)
2002 					notify_client = B_TRUE;
2003 				notify_event |= LDC_EVT_READ;
2004 				break;
2005 			}
2006 		}
2007 
2008 		/* Check the sequence ID for the message received */
2009 		rv = i_ldc_check_seqid(ldcp, msg);
2010 		if (rv != 0) {
2011 
2012 			DWARN(ldcp->id, "i_ldc_rx_hdlr: (0x%llx) seqid error, "
2013 			    "q_ptrs=0x%lx,0x%lx", ldcp->id, rx_head, rx_tail);
2014 
2015 			/* Reset last_msg_rcd to start of message */
2016 			if (first_fragment != 0) {
2017 				ldcp->last_msg_rcd = first_fragment - 1;
2018 				first_fragment = 0;
2019 			}
2020 
2021 			/*
2022 			 * Send a NACK due to seqid mismatch
2023 			 */
2024 			rv = i_ldc_send_pkt(ldcp, LDC_CTRL, LDC_NACK,
2025 			    (msg->ctrl & LDC_CTRL_MASK));
2026 
2027 			if (rv) {
2028 				cmn_err(CE_NOTE,
2029 				    "i_ldc_rx_hdlr: (0x%lx) err sending "
2030 				    "CTRL/NACK msg\n", ldcp->id);
2031 
2032 				/* if cannot send NACK - reset channel */
2033 				mutex_enter(&ldcp->tx_lock);
2034 				i_ldc_reset(ldcp, B_TRUE);
2035 				mutex_exit(&ldcp->tx_lock);
2036 				rv = ECONNRESET;
2037 				break;
2038 			}
2039 
2040 			/* purge receive queue */
2041 			(void) i_ldc_set_rx_head(ldcp, rx_tail);
2042 			break;
2043 		}
2044 
2045 		/* record the message ID */
2046 		ldcp->last_msg_rcd = msg->seqid;
2047 
2048 		/* process control messages */
2049 		if (msg->type & LDC_CTRL) {
2050 			/* save current internal state */
2051 			uint64_t tstate = ldcp->tstate;
2052 
2053 			rv = i_ldc_ctrlmsg(ldcp, msg);
2054 			if (rv == EAGAIN) {
2055 				/* re-process pkt - state was adjusted */
2056 				continue;
2057 			}
2058 			if (rv == ECONNRESET) {
2059 				notify_client = B_TRUE;
2060 				notify_event = LDC_EVT_RESET;
2061 				break;
2062 			}
2063 
2064 			/*
2065 			 * control message processing was successful
2066 			 * channel transitioned to ready for communication
2067 			 */
2068 			if (rv == 0 && ldcp->tstate == TS_UP &&
2069 			    (tstate & ~TS_IN_RESET) !=
2070 			    (ldcp->tstate & ~TS_IN_RESET)) {
2071 				notify_client = B_TRUE;
2072 				notify_event = LDC_EVT_UP;
2073 			}
2074 		}
2075 
2076 		/* process data ACKs */
2077 		if ((msg->type & LDC_DATA) && (msg->stype & LDC_ACK)) {
2078 			if (rv = i_ldc_process_data_ACK(ldcp, msg)) {
2079 				notify_client = B_TRUE;
2080 				notify_event = LDC_EVT_RESET;
2081 				break;
2082 			}
2083 		}
2084 
2085 		/* move the head one position */
2086 		rx_head = (rx_head + LDC_PACKET_SIZE) %
2087 			(ldcp->rx_q_entries << LDC_PACKET_SHIFT);
2088 		if (rv = i_ldc_set_rx_head(ldcp, rx_head)) {
2089 			notify_client = B_TRUE;
2090 			notify_event = LDC_EVT_RESET;
2091 			break;
2092 		}
2093 
2094 	} /* for */
2095 
2096 loop_exit:
2097 
2098 	/* if callbacks are disabled, do not notify */
2099 	if (!ldcp->cb_enabled)
2100 		notify_client = B_FALSE;
2101 
2102 	/*
2103 	 * If there are data packets in the queue, the ldc_read will
2104 	 * clear interrupts after draining the queue, else clear interrupts
2105 	 */
2106 	if ((notify_event & LDC_EVT_READ) == 0) {
2107 		i_ldc_clear_intr(ldcp, CNEX_RX_INTR);
2108 	} else
2109 		ldcp->rx_intr_state = LDC_INTR_PEND;
2110 
2111 	mutex_exit(&ldcp->lock);
2112 
2113 	if (notify_client) {
2114 		rv = ldcp->cb(notify_event, ldcp->cb_arg);
2115 		if (rv) {
2116 			DWARN(ldcp->id,
2117 			    "i_ldc_rx_hdlr: (0x%llx) callback failure",
2118 			    ldcp->id);
2119 		}
2120 	}
2121 
2122 	D1(ldcp->id, "i_ldc_rx_hdlr: (0x%llx) exiting handler", ldcp->id);
2123 	return (DDI_INTR_CLAIMED);
2124 }
2125 
2126 
2127 /* -------------------------------------------------------------------------- */
2128 
2129 /*
2130  * LDC API functions
2131  */
2132 
2133 /*
2134  * Initialize the channel. Allocate internal structure and memory for
2135  * TX/RX queues, and initialize locks.
2136  */
2137 int
2138 ldc_init(uint64_t id, ldc_attr_t *attr, ldc_handle_t *handle)
2139 {
2140 	ldc_chan_t 	*ldcp;
2141 	int		rv, exit_val;
2142 	uint64_t	ra_base, nentries;
2143 	uint64_t	qlen;
2144 
2145 	exit_val = EINVAL;	/* guarantee an error if exit on failure */
2146 
2147 	if (attr == NULL) {
2148 		DWARN(id, "ldc_init: (0x%llx) invalid attr\n", id);
2149 		return (EINVAL);
2150 	}
2151 	if (handle == NULL) {
2152 		DWARN(id, "ldc_init: (0x%llx) invalid handle\n", id);
2153 		return (EINVAL);
2154 	}
2155 
2156 	/* check if channel is valid */
2157 	rv = hv_ldc_tx_qinfo(id, &ra_base, &nentries);
2158 	if (rv == H_ECHANNEL) {
2159 		DWARN(id, "ldc_init: (0x%llx) invalid channel id\n", id);
2160 		return (EINVAL);
2161 	}
2162 
2163 	/* check if the channel has already been initialized */
2164 	mutex_enter(&ldcssp->lock);
2165 	ldcp = ldcssp->chan_list;
2166 	while (ldcp != NULL) {
2167 		if (ldcp->id == id) {
2168 			DWARN(id, "ldc_init: (0x%llx) already initialized\n",
2169 			    id);
2170 			mutex_exit(&ldcssp->lock);
2171 			return (EADDRINUSE);
2172 		}
2173 		ldcp = ldcp->next;
2174 	}
2175 	mutex_exit(&ldcssp->lock);
2176 
2177 	ASSERT(ldcp == NULL);
2178 
2179 	*handle = 0;
2180 
2181 	/* Allocate an ldcp structure */
2182 	ldcp = kmem_zalloc(sizeof (ldc_chan_t), KM_SLEEP);
2183 
2184 	/*
2185 	 * Initialize the channel and Tx lock
2186 	 *
2187 	 * The channel 'lock' protects the entire channel and
2188 	 * should be acquired before initializing, resetting,
2189 	 * destroying or reading from a channel.
2190 	 *
2191 	 * The 'tx_lock' should be acquired prior to transmitting
2192 	 * data over the channel. The lock should also be acquired
2193 	 * prior to channel reconfiguration (in order to prevent
2194 	 * concurrent writes).
2195 	 *
2196 	 * ORDERING: When both locks are being acquired, to prevent
2197 	 * deadlocks, the channel lock should be always acquired prior
2198 	 * to the tx_lock.
2199 	 */
2200 	mutex_init(&ldcp->lock, NULL, MUTEX_DRIVER, NULL);
2201 	mutex_init(&ldcp->tx_lock, NULL, MUTEX_DRIVER, NULL);
2202 
2203 	/* Initialize the channel */
2204 	ldcp->id = id;
2205 	ldcp->cb = NULL;
2206 	ldcp->cb_arg = NULL;
2207 	ldcp->cb_inprogress = B_FALSE;
2208 	ldcp->cb_enabled = B_FALSE;
2209 	ldcp->next = NULL;
2210 
2211 	/* Read attributes */
2212 	ldcp->mode = attr->mode;
2213 	ldcp->devclass = attr->devclass;
2214 	ldcp->devinst = attr->instance;
2215 	ldcp->mtu = (attr->mtu > 0) ? attr->mtu : LDC_DEFAULT_MTU;
2216 
2217 	D1(ldcp->id,
2218 	    "ldc_init: (0x%llx) channel attributes, class=0x%x, "
2219 	    "instance=0x%llx, mode=%d, mtu=%d\n",
2220 	    ldcp->id, ldcp->devclass, ldcp->devinst, ldcp->mode, ldcp->mtu);
2221 
2222 	ldcp->next_vidx = 0;
2223 	ldcp->tstate = TS_IN_RESET;
2224 	ldcp->hstate = 0;
2225 	ldcp->last_msg_snt = LDC_INIT_SEQID;
2226 	ldcp->last_ack_rcd = 0;
2227 	ldcp->last_msg_rcd = 0;
2228 
2229 	ldcp->stream_bufferp = NULL;
2230 	ldcp->exp_dring_list = NULL;
2231 	ldcp->imp_dring_list = NULL;
2232 	ldcp->mhdl_list = NULL;
2233 
2234 	ldcp->tx_intr_state = LDC_INTR_NONE;
2235 	ldcp->rx_intr_state = LDC_INTR_NONE;
2236 
2237 	/* Initialize payload size depending on whether channel is reliable */
2238 	switch (ldcp->mode) {
2239 	case LDC_MODE_RAW:
2240 		ldcp->pkt_payload = LDC_PAYLOAD_SIZE_RAW;
2241 		ldcp->read_p = i_ldc_read_raw;
2242 		ldcp->write_p = i_ldc_write_raw;
2243 		break;
2244 	case LDC_MODE_UNRELIABLE:
2245 		ldcp->pkt_payload = LDC_PAYLOAD_SIZE_UNRELIABLE;
2246 		ldcp->read_p = i_ldc_read_packet;
2247 		ldcp->write_p = i_ldc_write_packet;
2248 		break;
2249 	case LDC_MODE_RELIABLE:
2250 		ldcp->pkt_payload = LDC_PAYLOAD_SIZE_RELIABLE;
2251 		ldcp->read_p = i_ldc_read_packet;
2252 		ldcp->write_p = i_ldc_write_packet;
2253 		break;
2254 	case LDC_MODE_STREAM:
2255 		ldcp->pkt_payload = LDC_PAYLOAD_SIZE_RELIABLE;
2256 
2257 		ldcp->stream_remains = 0;
2258 		ldcp->stream_offset = 0;
2259 		ldcp->stream_bufferp = kmem_alloc(ldcp->mtu, KM_SLEEP);
2260 		ldcp->read_p = i_ldc_read_stream;
2261 		ldcp->write_p = i_ldc_write_stream;
2262 		break;
2263 	default:
2264 		exit_val = EINVAL;
2265 		goto cleanup_on_exit;
2266 	}
2267 
2268 	/*
2269 	 * qlen is (mtu * ldc_mtu_msgs) / pkt_payload. If this
2270 	 * value is smaller than default length of ldc_queue_entries,
2271 	 * qlen is set to ldc_queue_entries..
2272 	 */
2273 	qlen = (ldcp->mtu * ldc_mtu_msgs) / ldcp->pkt_payload;
2274 	ldcp->rx_q_entries =
2275 		(qlen < ldc_queue_entries) ? ldc_queue_entries : qlen;
2276 	ldcp->tx_q_entries = ldcp->rx_q_entries;
2277 
2278 	D1(ldcp->id, "ldc_init: queue length = 0x%llx\n", qlen);
2279 
2280 	/* Create a transmit queue */
2281 	ldcp->tx_q_va = (uint64_t)
2282 		contig_mem_alloc(ldcp->tx_q_entries << LDC_PACKET_SHIFT);
2283 	if (ldcp->tx_q_va == NULL) {
2284 		cmn_err(CE_WARN,
2285 		    "ldc_init: (0x%lx) TX queue allocation failed\n",
2286 		    ldcp->id);
2287 		exit_val = ENOMEM;
2288 		goto cleanup_on_exit;
2289 	}
2290 	ldcp->tx_q_ra = va_to_pa((caddr_t)ldcp->tx_q_va);
2291 
2292 	D2(ldcp->id, "ldc_init: txq_va=0x%llx, txq_ra=0x%llx, entries=0x%llx\n",
2293 	    ldcp->tx_q_va, ldcp->tx_q_ra, ldcp->tx_q_entries);
2294 
2295 	ldcp->tstate |= TS_TXQ_RDY;
2296 
2297 	/* Create a receive queue */
2298 	ldcp->rx_q_va = (uint64_t)
2299 		contig_mem_alloc(ldcp->rx_q_entries << LDC_PACKET_SHIFT);
2300 	if (ldcp->rx_q_va == NULL) {
2301 		cmn_err(CE_WARN,
2302 		    "ldc_init: (0x%lx) RX queue allocation failed\n",
2303 		    ldcp->id);
2304 		exit_val = ENOMEM;
2305 		goto cleanup_on_exit;
2306 	}
2307 	ldcp->rx_q_ra = va_to_pa((caddr_t)ldcp->rx_q_va);
2308 
2309 	D2(ldcp->id, "ldc_init: rxq_va=0x%llx, rxq_ra=0x%llx, entries=0x%llx\n",
2310 	    ldcp->rx_q_va, ldcp->rx_q_ra, ldcp->rx_q_entries);
2311 
2312 	ldcp->tstate |= TS_RXQ_RDY;
2313 
2314 	/* Init descriptor ring and memory handle list lock */
2315 	mutex_init(&ldcp->exp_dlist_lock, NULL, MUTEX_DRIVER, NULL);
2316 	mutex_init(&ldcp->imp_dlist_lock, NULL, MUTEX_DRIVER, NULL);
2317 	mutex_init(&ldcp->mlist_lock, NULL, MUTEX_DRIVER, NULL);
2318 
2319 	/* mark status as INITialized */
2320 	ldcp->status = LDC_INIT;
2321 
2322 	/* Add to channel list */
2323 	mutex_enter(&ldcssp->lock);
2324 	ldcp->next = ldcssp->chan_list;
2325 	ldcssp->chan_list = ldcp;
2326 	ldcssp->channel_count++;
2327 	mutex_exit(&ldcssp->lock);
2328 
2329 	/* set the handle */
2330 	*handle = (ldc_handle_t)ldcp;
2331 
2332 	D1(ldcp->id, "ldc_init: (0x%llx) channel initialized\n", ldcp->id);
2333 
2334 	return (0);
2335 
2336 cleanup_on_exit:
2337 
2338 	if (ldcp->mode == LDC_MODE_STREAM && ldcp->stream_bufferp)
2339 		kmem_free(ldcp->stream_bufferp, ldcp->mtu);
2340 
2341 	if (ldcp->tstate & TS_TXQ_RDY)
2342 		contig_mem_free((caddr_t)ldcp->tx_q_va,
2343 		    (ldcp->tx_q_entries << LDC_PACKET_SHIFT));
2344 
2345 	if (ldcp->tstate & TS_RXQ_RDY)
2346 		contig_mem_free((caddr_t)ldcp->rx_q_va,
2347 		    (ldcp->rx_q_entries << LDC_PACKET_SHIFT));
2348 
2349 	mutex_destroy(&ldcp->tx_lock);
2350 	mutex_destroy(&ldcp->lock);
2351 
2352 	if (ldcp)
2353 		kmem_free(ldcp, sizeof (ldc_chan_t));
2354 
2355 	return (exit_val);
2356 }
2357 
2358 /*
2359  * Finalizes the LDC connection. It will return EBUSY if the
2360  * channel is open. A ldc_close() has to be done prior to
2361  * a ldc_fini operation. It frees TX/RX queues, associated
2362  * with the channel
2363  */
2364 int
2365 ldc_fini(ldc_handle_t handle)
2366 {
2367 	ldc_chan_t 	*ldcp;
2368 	ldc_chan_t 	*tmp_ldcp;
2369 	uint64_t 	id;
2370 
2371 	if (handle == NULL) {
2372 		DWARN(DBG_ALL_LDCS, "ldc_fini: invalid channel handle\n");
2373 		return (EINVAL);
2374 	}
2375 	ldcp = (ldc_chan_t *)handle;
2376 	id = ldcp->id;
2377 
2378 	mutex_enter(&ldcp->lock);
2379 
2380 	if ((ldcp->tstate & ~TS_IN_RESET) > TS_INIT) {
2381 		DWARN(ldcp->id, "ldc_fini: (0x%llx) channel is open\n",
2382 		    ldcp->id);
2383 		mutex_exit(&ldcp->lock);
2384 		return (EBUSY);
2385 	}
2386 
2387 	/* Remove from the channel list */
2388 	mutex_enter(&ldcssp->lock);
2389 	tmp_ldcp = ldcssp->chan_list;
2390 	if (tmp_ldcp == ldcp) {
2391 		ldcssp->chan_list = ldcp->next;
2392 		ldcp->next = NULL;
2393 	} else {
2394 		while (tmp_ldcp != NULL) {
2395 			if (tmp_ldcp->next == ldcp) {
2396 				tmp_ldcp->next = ldcp->next;
2397 				ldcp->next = NULL;
2398 				break;
2399 			}
2400 			tmp_ldcp = tmp_ldcp->next;
2401 		}
2402 		if (tmp_ldcp == NULL) {
2403 			DWARN(DBG_ALL_LDCS, "ldc_fini: invalid channel hdl\n");
2404 			mutex_exit(&ldcssp->lock);
2405 			mutex_exit(&ldcp->lock);
2406 			return (EINVAL);
2407 		}
2408 	}
2409 
2410 	ldcssp->channel_count--;
2411 
2412 	mutex_exit(&ldcssp->lock);
2413 
2414 	/* Free the map table for this channel */
2415 	if (ldcp->mtbl) {
2416 		(void) hv_ldc_set_map_table(ldcp->id, NULL, NULL);
2417 		if (ldcp->mtbl->contigmem)
2418 			contig_mem_free(ldcp->mtbl->table, ldcp->mtbl->size);
2419 		else
2420 			kmem_free(ldcp->mtbl->table, ldcp->mtbl->size);
2421 		mutex_destroy(&ldcp->mtbl->lock);
2422 		kmem_free(ldcp->mtbl, sizeof (ldc_mtbl_t));
2423 	}
2424 
2425 	/* Destroy descriptor ring and memory handle list lock */
2426 	mutex_destroy(&ldcp->exp_dlist_lock);
2427 	mutex_destroy(&ldcp->imp_dlist_lock);
2428 	mutex_destroy(&ldcp->mlist_lock);
2429 
2430 	/* Free the stream buffer for STREAM_MODE */
2431 	if (ldcp->mode == LDC_MODE_STREAM && ldcp->stream_bufferp)
2432 		kmem_free(ldcp->stream_bufferp, ldcp->mtu);
2433 
2434 	/* Free the RX queue */
2435 	contig_mem_free((caddr_t)ldcp->rx_q_va,
2436 	    (ldcp->rx_q_entries << LDC_PACKET_SHIFT));
2437 	ldcp->tstate &= ~TS_RXQ_RDY;
2438 
2439 	/* Free the TX queue */
2440 	contig_mem_free((caddr_t)ldcp->tx_q_va,
2441 	    (ldcp->tx_q_entries << LDC_PACKET_SHIFT));
2442 	ldcp->tstate &= ~TS_TXQ_RDY;
2443 
2444 	mutex_exit(&ldcp->lock);
2445 
2446 	/* Destroy mutex */
2447 	mutex_destroy(&ldcp->tx_lock);
2448 	mutex_destroy(&ldcp->lock);
2449 
2450 	/* free channel structure */
2451 	kmem_free(ldcp, sizeof (ldc_chan_t));
2452 
2453 	D1(id, "ldc_fini: (0x%llx) channel finalized\n", id);
2454 
2455 	return (0);
2456 }
2457 
2458 /*
2459  * Open the LDC channel for use. It registers the TX/RX queues
2460  * with the Hypervisor. It also specifies the interrupt number
2461  * and target CPU for this channel
2462  */
2463 int
2464 ldc_open(ldc_handle_t handle)
2465 {
2466 	ldc_chan_t 	*ldcp;
2467 	int 		rv;
2468 
2469 	if (handle == NULL) {
2470 		DWARN(DBG_ALL_LDCS, "ldc_open: invalid channel handle\n");
2471 		return (EINVAL);
2472 	}
2473 
2474 	ldcp = (ldc_chan_t *)handle;
2475 
2476 	mutex_enter(&ldcp->lock);
2477 
2478 	if (ldcp->tstate < TS_INIT) {
2479 		DWARN(ldcp->id,
2480 		    "ldc_open: (0x%llx) channel not initialized\n", ldcp->id);
2481 		mutex_exit(&ldcp->lock);
2482 		return (EFAULT);
2483 	}
2484 	if ((ldcp->tstate & ~TS_IN_RESET) >= TS_OPEN) {
2485 		DWARN(ldcp->id,
2486 		    "ldc_open: (0x%llx) channel is already open\n", ldcp->id);
2487 		mutex_exit(&ldcp->lock);
2488 		return (EFAULT);
2489 	}
2490 
2491 	/*
2492 	 * Unregister/Register the tx queue with the hypervisor
2493 	 */
2494 	rv = hv_ldc_tx_qconf(ldcp->id, NULL, NULL);
2495 	if (rv) {
2496 		cmn_err(CE_WARN,
2497 		    "ldc_open: (0x%lx) channel tx queue unconf failed\n",
2498 		    ldcp->id);
2499 		mutex_exit(&ldcp->lock);
2500 		return (EIO);
2501 	}
2502 
2503 	rv = hv_ldc_tx_qconf(ldcp->id, ldcp->tx_q_ra, ldcp->tx_q_entries);
2504 	if (rv) {
2505 		cmn_err(CE_WARN,
2506 		    "ldc_open: (0x%lx) channel tx queue conf failed\n",
2507 		    ldcp->id);
2508 		mutex_exit(&ldcp->lock);
2509 		return (EIO);
2510 	}
2511 
2512 	D2(ldcp->id, "ldc_open: (0x%llx) registered tx queue with LDC\n",
2513 	    ldcp->id);
2514 
2515 	/*
2516 	 * Unregister/Register the rx queue with the hypervisor
2517 	 */
2518 	rv = hv_ldc_rx_qconf(ldcp->id, NULL, NULL);
2519 	if (rv) {
2520 		cmn_err(CE_WARN,
2521 		    "ldc_open: (0x%lx) channel rx queue unconf failed\n",
2522 		    ldcp->id);
2523 		mutex_exit(&ldcp->lock);
2524 		return (EIO);
2525 	}
2526 
2527 	rv = hv_ldc_rx_qconf(ldcp->id, ldcp->rx_q_ra, ldcp->rx_q_entries);
2528 	if (rv) {
2529 		cmn_err(CE_WARN,
2530 		    "ldc_open: (0x%lx) channel rx queue conf failed\n",
2531 		    ldcp->id);
2532 		mutex_exit(&ldcp->lock);
2533 		return (EIO);
2534 	}
2535 
2536 	D2(ldcp->id, "ldc_open: (0x%llx) registered rx queue with LDC\n",
2537 	    ldcp->id);
2538 
2539 	ldcp->tstate |= TS_QCONF_RDY;
2540 
2541 	/* Register the channel with the channel nexus */
2542 	rv = i_ldc_register_channel(ldcp);
2543 	if (rv && rv != EAGAIN) {
2544 		cmn_err(CE_WARN,
2545 		    "ldc_open: (0x%lx) channel register failed\n", ldcp->id);
2546 		(void) hv_ldc_tx_qconf(ldcp->id, NULL, NULL);
2547 		(void) hv_ldc_rx_qconf(ldcp->id, NULL, NULL);
2548 		mutex_exit(&ldcp->lock);
2549 		return (EIO);
2550 	}
2551 
2552 	/* mark channel in OPEN state */
2553 	ldcp->status = LDC_OPEN;
2554 
2555 	/* Read channel state */
2556 	rv = hv_ldc_tx_get_state(ldcp->id,
2557 	    &ldcp->tx_head, &ldcp->tx_tail, &ldcp->link_state);
2558 	if (rv) {
2559 		cmn_err(CE_WARN,
2560 		    "ldc_open: (0x%lx) cannot read channel state\n",
2561 		    ldcp->id);
2562 		(void) i_ldc_unregister_channel(ldcp);
2563 		(void) hv_ldc_tx_qconf(ldcp->id, NULL, NULL);
2564 		(void) hv_ldc_rx_qconf(ldcp->id, NULL, NULL);
2565 		mutex_exit(&ldcp->lock);
2566 		return (EIO);
2567 	}
2568 
2569 	/*
2570 	 * set the ACKd head to current head location for reliable &
2571 	 * streaming mode
2572 	 */
2573 	ldcp->tx_ackd_head = ldcp->tx_head;
2574 
2575 	/* mark channel ready if HV report link is UP (peer alloc'd Rx queue) */
2576 	if (ldcp->link_state == LDC_CHANNEL_UP ||
2577 	    ldcp->link_state == LDC_CHANNEL_RESET) {
2578 		ldcp->tstate |= TS_LINK_READY;
2579 		ldcp->status = LDC_READY;
2580 	}
2581 
2582 	/*
2583 	 * if channel is being opened in RAW mode - no handshake is needed
2584 	 * switch the channel READY and UP state
2585 	 */
2586 	if (ldcp->mode == LDC_MODE_RAW) {
2587 		ldcp->tstate = TS_UP;	/* set bits associated with LDC UP */
2588 		ldcp->status = LDC_UP;
2589 	}
2590 
2591 	mutex_exit(&ldcp->lock);
2592 
2593 	/*
2594 	 * Increment number of open channels
2595 	 */
2596 	mutex_enter(&ldcssp->lock);
2597 	ldcssp->channels_open++;
2598 	mutex_exit(&ldcssp->lock);
2599 
2600 	DWARN(ldcp->id,
2601 	    "ldc_open: (0x%llx) channel (0x%p) open for use "
2602 	    "(tstate=0x%x, status=0x%x)\n",
2603 	    ldcp->id, ldcp, ldcp->tstate, ldcp->status);
2604 
2605 	return (0);
2606 }
2607 
2608 /*
2609  * Close the LDC connection. It will return EBUSY if there
2610  * are memory segments or descriptor rings either bound to or
2611  * mapped over the channel
2612  */
2613 int
2614 ldc_close(ldc_handle_t handle)
2615 {
2616 	ldc_chan_t 	*ldcp;
2617 	int		rv = 0, retries = 0;
2618 	boolean_t	chk_done = B_FALSE;
2619 
2620 	if (handle == NULL) {
2621 		DWARN(DBG_ALL_LDCS, "ldc_close: invalid channel handle\n");
2622 		return (EINVAL);
2623 	}
2624 	ldcp = (ldc_chan_t *)handle;
2625 
2626 	mutex_enter(&ldcp->lock);
2627 
2628 	/* return error if channel is not open */
2629 	if ((ldcp->tstate & ~TS_IN_RESET) < TS_OPEN) {
2630 		DWARN(ldcp->id,
2631 		    "ldc_close: (0x%llx) channel is not open\n", ldcp->id);
2632 		mutex_exit(&ldcp->lock);
2633 		return (EFAULT);
2634 	}
2635 
2636 	/* if any memory handles, drings, are bound or mapped cannot close */
2637 	if (ldcp->mhdl_list != NULL) {
2638 		DWARN(ldcp->id,
2639 		    "ldc_close: (0x%llx) channel has bound memory handles\n",
2640 		    ldcp->id);
2641 		mutex_exit(&ldcp->lock);
2642 		return (EBUSY);
2643 	}
2644 	if (ldcp->exp_dring_list != NULL) {
2645 		DWARN(ldcp->id,
2646 		    "ldc_close: (0x%llx) channel has bound descriptor rings\n",
2647 		    ldcp->id);
2648 		mutex_exit(&ldcp->lock);
2649 		return (EBUSY);
2650 	}
2651 	if (ldcp->imp_dring_list != NULL) {
2652 		DWARN(ldcp->id,
2653 		    "ldc_close: (0x%llx) channel has mapped descriptor rings\n",
2654 		    ldcp->id);
2655 		mutex_exit(&ldcp->lock);
2656 		return (EBUSY);
2657 	}
2658 
2659 	/* Obtain Tx lock */
2660 	mutex_enter(&ldcp->tx_lock);
2661 
2662 	/*
2663 	 * Wait for pending transmits to complete i.e Tx queue to drain
2664 	 * if there are pending pkts - wait 1 ms and retry again
2665 	 */
2666 	for (;;) {
2667 
2668 		rv = hv_ldc_tx_get_state(ldcp->id,
2669 		    &ldcp->tx_head, &ldcp->tx_tail, &ldcp->link_state);
2670 		if (rv) {
2671 			cmn_err(CE_WARN,
2672 			    "ldc_close: (0x%lx) cannot read qptrs\n", ldcp->id);
2673 			mutex_exit(&ldcp->tx_lock);
2674 			mutex_exit(&ldcp->lock);
2675 			return (EIO);
2676 		}
2677 
2678 		if (ldcp->tx_head == ldcp->tx_tail ||
2679 		    ldcp->link_state != LDC_CHANNEL_UP) {
2680 			break;
2681 		}
2682 
2683 		if (chk_done) {
2684 			DWARN(ldcp->id,
2685 			    "ldc_close: (0x%llx) Tx queue drain timeout\n",
2686 			    ldcp->id);
2687 			break;
2688 		}
2689 
2690 		/* wait for one ms and try again */
2691 		delay(drv_usectohz(1000));
2692 		chk_done = B_TRUE;
2693 	}
2694 
2695 	/*
2696 	 * Drain the Tx and Rx queues as we are closing the
2697 	 * channel. We dont care about any pending packets.
2698 	 * We have to also drain the queue prior to clearing
2699 	 * pending interrupts, otherwise the HV will trigger
2700 	 * an interrupt the moment the interrupt state is
2701 	 * cleared.
2702 	 */
2703 	(void) i_ldc_txq_reconf(ldcp);
2704 	(void) i_ldc_rxq_drain(ldcp);
2705 
2706 	/*
2707 	 * Unregister the channel with the nexus
2708 	 */
2709 	while ((rv = i_ldc_unregister_channel(ldcp)) != 0) {
2710 
2711 		mutex_exit(&ldcp->tx_lock);
2712 		mutex_exit(&ldcp->lock);
2713 
2714 		/* if any error other than EAGAIN return back */
2715 		if (rv != EAGAIN || retries >= ldc_max_retries) {
2716 			cmn_err(CE_WARN,
2717 			    "ldc_close: (0x%lx) unregister failed, %d\n",
2718 			    ldcp->id, rv);
2719 			return (rv);
2720 		}
2721 
2722 		/*
2723 		 * As there could be pending interrupts we need
2724 		 * to wait and try again
2725 		 */
2726 		drv_usecwait(ldc_delay);
2727 		mutex_enter(&ldcp->lock);
2728 		mutex_enter(&ldcp->tx_lock);
2729 		retries++;
2730 	}
2731 
2732 	/*
2733 	 * Unregister queues
2734 	 */
2735 	rv = hv_ldc_tx_qconf(ldcp->id, NULL, NULL);
2736 	if (rv) {
2737 		cmn_err(CE_WARN,
2738 		    "ldc_close: (0x%lx) channel TX queue unconf failed\n",
2739 		    ldcp->id);
2740 		mutex_exit(&ldcp->tx_lock);
2741 		mutex_exit(&ldcp->lock);
2742 		return (EIO);
2743 	}
2744 	rv = hv_ldc_rx_qconf(ldcp->id, NULL, NULL);
2745 	if (rv) {
2746 		cmn_err(CE_WARN,
2747 		    "ldc_close: (0x%lx) channel RX queue unconf failed\n",
2748 		    ldcp->id);
2749 		mutex_exit(&ldcp->tx_lock);
2750 		mutex_exit(&ldcp->lock);
2751 		return (EIO);
2752 	}
2753 
2754 	ldcp->tstate &= ~TS_QCONF_RDY;
2755 
2756 	/* Reset channel state information */
2757 	i_ldc_reset_state(ldcp);
2758 
2759 	/* Mark channel as down and in initialized state */
2760 	ldcp->tx_ackd_head = 0;
2761 	ldcp->tx_head = 0;
2762 	ldcp->tstate = TS_IN_RESET|TS_INIT;
2763 	ldcp->status = LDC_INIT;
2764 
2765 	mutex_exit(&ldcp->tx_lock);
2766 	mutex_exit(&ldcp->lock);
2767 
2768 	/* Decrement number of open channels */
2769 	mutex_enter(&ldcssp->lock);
2770 	ldcssp->channels_open--;
2771 	mutex_exit(&ldcssp->lock);
2772 
2773 	D1(ldcp->id, "ldc_close: (0x%llx) channel closed\n", ldcp->id);
2774 
2775 	return (0);
2776 }
2777 
2778 /*
2779  * Register channel callback
2780  */
2781 int
2782 ldc_reg_callback(ldc_handle_t handle,
2783     uint_t(*cb)(uint64_t event, caddr_t arg), caddr_t arg)
2784 {
2785 	ldc_chan_t *ldcp;
2786 
2787 	if (handle == NULL) {
2788 		DWARN(DBG_ALL_LDCS,
2789 		    "ldc_reg_callback: invalid channel handle\n");
2790 		return (EINVAL);
2791 	}
2792 	if (((uint64_t)cb) < KERNELBASE) {
2793 		DWARN(DBG_ALL_LDCS, "ldc_reg_callback: invalid callback\n");
2794 		return (EINVAL);
2795 	}
2796 	ldcp = (ldc_chan_t *)handle;
2797 
2798 	mutex_enter(&ldcp->lock);
2799 
2800 	if (ldcp->cb) {
2801 		DWARN(ldcp->id, "ldc_reg_callback: (0x%llx) callback exists\n",
2802 		    ldcp->id);
2803 		mutex_exit(&ldcp->lock);
2804 		return (EIO);
2805 	}
2806 	if (ldcp->cb_inprogress) {
2807 		DWARN(ldcp->id, "ldc_reg_callback: (0x%llx) callback active\n",
2808 		    ldcp->id);
2809 		mutex_exit(&ldcp->lock);
2810 		return (EWOULDBLOCK);
2811 	}
2812 
2813 	ldcp->cb = cb;
2814 	ldcp->cb_arg = arg;
2815 	ldcp->cb_enabled = B_TRUE;
2816 
2817 	D1(ldcp->id,
2818 	    "ldc_reg_callback: (0x%llx) registered callback for channel\n",
2819 	    ldcp->id);
2820 
2821 	mutex_exit(&ldcp->lock);
2822 
2823 	return (0);
2824 }
2825 
2826 /*
2827  * Unregister channel callback
2828  */
2829 int
2830 ldc_unreg_callback(ldc_handle_t handle)
2831 {
2832 	ldc_chan_t *ldcp;
2833 
2834 	if (handle == NULL) {
2835 		DWARN(DBG_ALL_LDCS,
2836 		    "ldc_unreg_callback: invalid channel handle\n");
2837 		return (EINVAL);
2838 	}
2839 	ldcp = (ldc_chan_t *)handle;
2840 
2841 	mutex_enter(&ldcp->lock);
2842 
2843 	if (ldcp->cb == NULL) {
2844 		DWARN(ldcp->id,
2845 		    "ldc_unreg_callback: (0x%llx) no callback exists\n",
2846 		    ldcp->id);
2847 		mutex_exit(&ldcp->lock);
2848 		return (EIO);
2849 	}
2850 	if (ldcp->cb_inprogress) {
2851 		DWARN(ldcp->id,
2852 		    "ldc_unreg_callback: (0x%llx) callback active\n",
2853 		    ldcp->id);
2854 		mutex_exit(&ldcp->lock);
2855 		return (EWOULDBLOCK);
2856 	}
2857 
2858 	ldcp->cb = NULL;
2859 	ldcp->cb_arg = NULL;
2860 	ldcp->cb_enabled = B_FALSE;
2861 
2862 	D1(ldcp->id,
2863 	    "ldc_unreg_callback: (0x%llx) unregistered callback for channel\n",
2864 	    ldcp->id);
2865 
2866 	mutex_exit(&ldcp->lock);
2867 
2868 	return (0);
2869 }
2870 
2871 
2872 /*
2873  * Bring a channel up by initiating a handshake with the peer
2874  * This call is asynchronous. It will complete at a later point
2875  * in time when the peer responds back with an RTR.
2876  */
2877 int
2878 ldc_up(ldc_handle_t handle)
2879 {
2880 	int 		rv;
2881 	ldc_chan_t 	*ldcp;
2882 	ldc_msg_t 	*ldcmsg;
2883 	uint64_t 	tx_tail, tstate;
2884 
2885 	if (handle == NULL) {
2886 		DWARN(DBG_ALL_LDCS, "ldc_up: invalid channel handle\n");
2887 		return (EINVAL);
2888 	}
2889 	ldcp = (ldc_chan_t *)handle;
2890 
2891 	mutex_enter(&ldcp->lock);
2892 
2893 	D1(ldcp->id, "ldc_up: (0x%llx) doing channel UP\n", ldcp->id);
2894 
2895 	/* clear the reset state */
2896 	tstate = ldcp->tstate;
2897 	ldcp->tstate &= ~TS_IN_RESET;
2898 
2899 	if (ldcp->tstate == TS_UP) {
2900 		DWARN(ldcp->id,
2901 		    "ldc_up: (0x%llx) channel is already in UP state\n",
2902 		    ldcp->id);
2903 
2904 		/* mark channel as up */
2905 		ldcp->status = LDC_UP;
2906 
2907 		/*
2908 		 * if channel was in reset state and there was
2909 		 * pending data clear interrupt state. this will
2910 		 * trigger an interrupt, causing the RX handler to
2911 		 * to invoke the client's callback
2912 		 */
2913 		if ((tstate & TS_IN_RESET) &&
2914 		    ldcp->rx_intr_state == LDC_INTR_PEND) {
2915 			DWARN(ldcp->id,
2916 			    "ldc_up: (0x%llx) channel has pending data, "
2917 			    "clearing interrupt\n", ldcp->id);
2918 			i_ldc_clear_intr(ldcp, CNEX_RX_INTR);
2919 		}
2920 
2921 		mutex_exit(&ldcp->lock);
2922 		return (0);
2923 	}
2924 
2925 	/* if the channel is in RAW mode - mark it as UP, if READY */
2926 	if (ldcp->mode == LDC_MODE_RAW && ldcp->tstate >= TS_READY) {
2927 		ldcp->tstate = TS_UP;
2928 		mutex_exit(&ldcp->lock);
2929 		return (0);
2930 	}
2931 
2932 	/* Don't start another handshake if there is one in progress */
2933 	if (ldcp->hstate) {
2934 		D1(ldcp->id,
2935 		    "ldc_up: (0x%llx) channel handshake in progress\n",
2936 		    ldcp->id);
2937 		mutex_exit(&ldcp->lock);
2938 		return (0);
2939 	}
2940 
2941 	mutex_enter(&ldcp->tx_lock);
2942 
2943 	/* get the current tail for the LDC msg */
2944 	rv = i_ldc_get_tx_tail(ldcp, &tx_tail);
2945 	if (rv) {
2946 		DWARN(ldcp->id, "ldc_up: (0x%llx) cannot initiate handshake\n",
2947 		    ldcp->id);
2948 		mutex_exit(&ldcp->tx_lock);
2949 		mutex_exit(&ldcp->lock);
2950 		return (ECONNREFUSED);
2951 	}
2952 
2953 	ldcmsg = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail);
2954 	ZERO_PKT(ldcmsg);
2955 
2956 	ldcmsg->type = LDC_CTRL;
2957 	ldcmsg->stype = LDC_INFO;
2958 	ldcmsg->ctrl = LDC_VER;
2959 	ldcp->next_vidx = 0;
2960 	bcopy(&ldc_versions[0], ldcmsg->udata, sizeof (ldc_versions[0]));
2961 
2962 	DUMP_LDC_PKT(ldcp, "ldc_up snd ver", (uint64_t)ldcmsg);
2963 
2964 	/* initiate the send by calling into HV and set the new tail */
2965 	tx_tail = (tx_tail + LDC_PACKET_SIZE) %
2966 		(ldcp->tx_q_entries << LDC_PACKET_SHIFT);
2967 
2968 	rv = i_ldc_set_tx_tail(ldcp, tx_tail);
2969 	if (rv) {
2970 		DWARN(ldcp->id,
2971 		    "ldc_up: (0x%llx) cannot initiate handshake rv=%d\n",
2972 		    ldcp->id, rv);
2973 		mutex_exit(&ldcp->tx_lock);
2974 		mutex_exit(&ldcp->lock);
2975 		return (rv);
2976 	}
2977 
2978 	ldcp->hstate |= TS_SENT_VER;
2979 	ldcp->tx_tail = tx_tail;
2980 	D1(ldcp->id, "ldc_up: (0x%llx) channel up initiated\n", ldcp->id);
2981 
2982 	mutex_exit(&ldcp->tx_lock);
2983 	mutex_exit(&ldcp->lock);
2984 
2985 	return (rv);
2986 }
2987 
2988 
2989 /*
2990  * Bring a channel down by resetting its state and queues
2991  */
2992 int
2993 ldc_down(ldc_handle_t handle)
2994 {
2995 	ldc_chan_t 	*ldcp;
2996 
2997 	if (handle == NULL) {
2998 		DWARN(DBG_ALL_LDCS, "ldc_down: invalid channel handle\n");
2999 		return (EINVAL);
3000 	}
3001 	ldcp = (ldc_chan_t *)handle;
3002 	mutex_enter(&ldcp->lock);
3003 	mutex_enter(&ldcp->tx_lock);
3004 	i_ldc_reset(ldcp, B_TRUE);
3005 	mutex_exit(&ldcp->tx_lock);
3006 	mutex_exit(&ldcp->lock);
3007 
3008 	return (0);
3009 }
3010 
3011 /*
3012  * Get the current channel status
3013  */
3014 int
3015 ldc_status(ldc_handle_t handle, ldc_status_t *status)
3016 {
3017 	ldc_chan_t *ldcp;
3018 
3019 	if (handle == NULL || status == NULL) {
3020 		DWARN(DBG_ALL_LDCS, "ldc_status: invalid argument\n");
3021 		return (EINVAL);
3022 	}
3023 	ldcp = (ldc_chan_t *)handle;
3024 
3025 	*status = ((ldc_chan_t *)handle)->status;
3026 
3027 	DWARN(ldcp->id,
3028 	    "ldc_status: (0x%llx) returned status %d\n", ldcp->id, *status);
3029 	return (0);
3030 }
3031 
3032 
3033 /*
3034  * Set the channel's callback mode - enable/disable callbacks
3035  */
3036 int
3037 ldc_set_cb_mode(ldc_handle_t handle, ldc_cb_mode_t cmode)
3038 {
3039 	ldc_chan_t 	*ldcp;
3040 
3041 	if (handle == NULL) {
3042 		DWARN(DBG_ALL_LDCS,
3043 		    "ldc_set_intr_mode: invalid channel handle\n");
3044 		return (EINVAL);
3045 	}
3046 	ldcp = (ldc_chan_t *)handle;
3047 
3048 	/*
3049 	 * Record no callbacks should be invoked
3050 	 */
3051 	mutex_enter(&ldcp->lock);
3052 
3053 	switch (cmode) {
3054 	case LDC_CB_DISABLE:
3055 		if (!ldcp->cb_enabled) {
3056 			DWARN(ldcp->id,
3057 			    "ldc_set_cb_mode: (0x%llx) callbacks disabled\n",
3058 			    ldcp->id);
3059 			break;
3060 		}
3061 		ldcp->cb_enabled = B_FALSE;
3062 
3063 		D1(ldcp->id, "ldc_set_cb_mode: (0x%llx) disabled callbacks\n",
3064 		    ldcp->id);
3065 		break;
3066 
3067 	case LDC_CB_ENABLE:
3068 		if (ldcp->cb_enabled) {
3069 			DWARN(ldcp->id,
3070 			    "ldc_set_cb_mode: (0x%llx) callbacks enabled\n",
3071 			    ldcp->id);
3072 			break;
3073 		}
3074 		ldcp->cb_enabled = B_TRUE;
3075 
3076 		D1(ldcp->id, "ldc_set_cb_mode: (0x%llx) enabled callbacks\n",
3077 		    ldcp->id);
3078 		break;
3079 	}
3080 
3081 	mutex_exit(&ldcp->lock);
3082 
3083 	return (0);
3084 }
3085 
3086 /*
3087  * Check to see if there are packets on the incoming queue
3088  * Will return hasdata = B_FALSE if there are no packets
3089  */
3090 int
3091 ldc_chkq(ldc_handle_t handle, boolean_t *hasdata)
3092 {
3093 	int 		rv;
3094 	uint64_t 	rx_head, rx_tail;
3095 	ldc_chan_t 	*ldcp;
3096 
3097 	if (handle == NULL) {
3098 		DWARN(DBG_ALL_LDCS, "ldc_chkq: invalid channel handle\n");
3099 		return (EINVAL);
3100 	}
3101 	ldcp = (ldc_chan_t *)handle;
3102 
3103 	*hasdata = B_FALSE;
3104 
3105 	mutex_enter(&ldcp->lock);
3106 
3107 	if (ldcp->tstate != TS_UP) {
3108 		D1(ldcp->id,
3109 		    "ldc_chkq: (0x%llx) channel is not up\n", ldcp->id);
3110 		mutex_exit(&ldcp->lock);
3111 		return (ECONNRESET);
3112 	}
3113 
3114 	/* Read packet(s) from the queue */
3115 	rv = hv_ldc_rx_get_state(ldcp->id, &rx_head, &rx_tail,
3116 	    &ldcp->link_state);
3117 	if (rv != 0) {
3118 		cmn_err(CE_WARN,
3119 		    "ldc_chkq: (0x%lx) unable to read queue ptrs", ldcp->id);
3120 		mutex_exit(&ldcp->lock);
3121 		return (EIO);
3122 	}
3123 	/* reset the channel state if the channel went down */
3124 	if (ldcp->link_state == LDC_CHANNEL_DOWN ||
3125 	    ldcp->link_state == LDC_CHANNEL_RESET) {
3126 		mutex_enter(&ldcp->tx_lock);
3127 		i_ldc_reset(ldcp, B_FALSE);
3128 		mutex_exit(&ldcp->tx_lock);
3129 		mutex_exit(&ldcp->lock);
3130 		return (ECONNRESET);
3131 	}
3132 
3133 	if ((rx_head != rx_tail) ||
3134 	    (ldcp->mode == LDC_MODE_STREAM && ldcp->stream_remains > 0)) {
3135 		D1(ldcp->id,
3136 		    "ldc_chkq: (0x%llx) queue has pkt(s) or buffered data\n",
3137 		    ldcp->id);
3138 		*hasdata = B_TRUE;
3139 	}
3140 
3141 	mutex_exit(&ldcp->lock);
3142 
3143 	return (0);
3144 }
3145 
3146 
3147 /*
3148  * Read 'size' amount of bytes or less. If incoming buffer
3149  * is more than 'size', ENOBUFS is returned.
3150  *
3151  * On return, size contains the number of bytes read.
3152  */
3153 int
3154 ldc_read(ldc_handle_t handle, caddr_t bufp, size_t *sizep)
3155 {
3156 	ldc_chan_t 	*ldcp;
3157 	uint64_t 	rx_head = 0, rx_tail = 0;
3158 	int		rv = 0, exit_val;
3159 
3160 	if (handle == NULL) {
3161 		DWARN(DBG_ALL_LDCS, "ldc_read: invalid channel handle\n");
3162 		return (EINVAL);
3163 	}
3164 
3165 	ldcp = (ldc_chan_t *)handle;
3166 
3167 	/* channel lock */
3168 	mutex_enter(&ldcp->lock);
3169 
3170 	if (ldcp->tstate != TS_UP) {
3171 		DWARN(ldcp->id,
3172 		    "ldc_read: (0x%llx) channel is not in UP state\n",
3173 		    ldcp->id);
3174 		exit_val = ECONNRESET;
3175 	} else {
3176 		exit_val = ldcp->read_p(ldcp, bufp, sizep);
3177 	}
3178 
3179 	/*
3180 	 * if queue has been drained - clear interrupt
3181 	 */
3182 	rv = hv_ldc_rx_get_state(ldcp->id, &rx_head, &rx_tail,
3183 	    &ldcp->link_state);
3184 
3185 	ASSERT(rv == 0);
3186 
3187 	if (exit_val == 0) {
3188 		if (ldcp->link_state == LDC_CHANNEL_DOWN ||
3189 		    ldcp->link_state == LDC_CHANNEL_RESET) {
3190 			mutex_enter(&ldcp->tx_lock);
3191 			i_ldc_reset(ldcp, B_FALSE);
3192 			exit_val = ECONNRESET;
3193 			mutex_exit(&ldcp->tx_lock);
3194 		}
3195 		if ((rv == 0) &&
3196 		    (ldcp->rx_intr_state == LDC_INTR_PEND) &&
3197 		    (rx_head == rx_tail)) {
3198 			i_ldc_clear_intr(ldcp, CNEX_RX_INTR);
3199 		}
3200 	}
3201 
3202 	mutex_exit(&ldcp->lock);
3203 	return (exit_val);
3204 }
3205 
3206 /*
3207  * Basic raw mondo read -
3208  * no interpretation of mondo contents at all.
3209  *
3210  * Enter and exit with ldcp->lock held by caller
3211  */
3212 static int
3213 i_ldc_read_raw(ldc_chan_t *ldcp, caddr_t target_bufp, size_t *sizep)
3214 {
3215 	uint64_t 	q_size_mask;
3216 	ldc_msg_t 	*msgp;
3217 	uint8_t		*msgbufp;
3218 	int		rv = 0, space;
3219 	uint64_t 	rx_head, rx_tail;
3220 
3221 	space = *sizep;
3222 
3223 	if (space < LDC_PAYLOAD_SIZE_RAW)
3224 		return (ENOBUFS);
3225 
3226 	ASSERT(mutex_owned(&ldcp->lock));
3227 
3228 	/* compute mask for increment */
3229 	q_size_mask = (ldcp->rx_q_entries-1)<<LDC_PACKET_SHIFT;
3230 
3231 	/*
3232 	 * Read packet(s) from the queue
3233 	 */
3234 	rv = hv_ldc_rx_get_state(ldcp->id, &rx_head, &rx_tail,
3235 	    &ldcp->link_state);
3236 	if (rv != 0) {
3237 		cmn_err(CE_WARN,
3238 		    "ldc_read_raw: (0x%lx) unable to read queue ptrs",
3239 		    ldcp->id);
3240 		return (EIO);
3241 	}
3242 	D1(ldcp->id, "ldc_read_raw: (0x%llx) rxh=0x%llx,"
3243 		" rxt=0x%llx, st=0x%llx\n",
3244 		ldcp->id, rx_head, rx_tail, ldcp->link_state);
3245 
3246 	/* reset the channel state if the channel went down */
3247 	if (ldcp->link_state == LDC_CHANNEL_DOWN ||
3248 	    ldcp->link_state == LDC_CHANNEL_RESET) {
3249 		mutex_enter(&ldcp->tx_lock);
3250 		i_ldc_reset(ldcp, B_FALSE);
3251 		mutex_exit(&ldcp->tx_lock);
3252 		return (ECONNRESET);
3253 	}
3254 
3255 	/*
3256 	 * Check for empty queue
3257 	 */
3258 	if (rx_head == rx_tail) {
3259 		*sizep = 0;
3260 		return (0);
3261 	}
3262 
3263 	/* get the message */
3264 	msgp = (ldc_msg_t *)(ldcp->rx_q_va + rx_head);
3265 
3266 	/* if channel is in RAW mode, copy data and return */
3267 	msgbufp = (uint8_t *)&(msgp->raw[0]);
3268 
3269 	bcopy(msgbufp, target_bufp, LDC_PAYLOAD_SIZE_RAW);
3270 
3271 	DUMP_PAYLOAD(ldcp->id, msgbufp);
3272 
3273 	*sizep = LDC_PAYLOAD_SIZE_RAW;
3274 
3275 	rx_head = (rx_head + LDC_PACKET_SIZE) & q_size_mask;
3276 	rv = i_ldc_set_rx_head(ldcp, rx_head);
3277 
3278 	return (rv);
3279 }
3280 
3281 /*
3282  * Process LDC mondos to build larger packets
3283  * with either un-reliable or reliable delivery.
3284  *
3285  * Enter and exit with ldcp->lock held by caller
3286  */
3287 static int
3288 i_ldc_read_packet(ldc_chan_t *ldcp, caddr_t target_bufp, size_t *sizep)
3289 {
3290 	int		rv = 0;
3291 	uint64_t 	rx_head = 0, rx_tail = 0;
3292 	uint64_t 	curr_head = 0;
3293 	ldc_msg_t 	*msg;
3294 	caddr_t 	target;
3295 	size_t 		len = 0, bytes_read = 0;
3296 	int 		retries = 0;
3297 	uint64_t 	q_size_mask;
3298 	uint64_t	first_fragment = 0;
3299 
3300 	target = target_bufp;
3301 
3302 	ASSERT(mutex_owned(&ldcp->lock));
3303 
3304 	/* check if the buffer and size are valid */
3305 	if (target_bufp == NULL || *sizep == 0) {
3306 		DWARN(ldcp->id, "ldc_read: (0x%llx) invalid buffer/size\n",
3307 		    ldcp->id);
3308 		return (EINVAL);
3309 	}
3310 
3311 	/* compute mask for increment */
3312 	q_size_mask = (ldcp->rx_q_entries-1)<<LDC_PACKET_SHIFT;
3313 
3314 	/*
3315 	 * Read packet(s) from the queue
3316 	 */
3317 	rv = hv_ldc_rx_get_state(ldcp->id, &curr_head, &rx_tail,
3318 	    &ldcp->link_state);
3319 	if (rv != 0) {
3320 		cmn_err(CE_WARN, "ldc_read: (0x%lx) unable to read queue ptrs",
3321 		    ldcp->id);
3322 		mutex_enter(&ldcp->tx_lock);
3323 		i_ldc_reset(ldcp, B_TRUE);
3324 		mutex_exit(&ldcp->tx_lock);
3325 		return (ECONNRESET);
3326 	}
3327 	D1(ldcp->id, "ldc_read: (0x%llx) chd=0x%llx, tl=0x%llx, st=0x%llx\n",
3328 	    ldcp->id, curr_head, rx_tail, ldcp->link_state);
3329 
3330 	/* reset the channel state if the channel went down */
3331 	if (ldcp->link_state != LDC_CHANNEL_UP)
3332 		goto channel_is_reset;
3333 
3334 	for (;;) {
3335 
3336 		if (curr_head == rx_tail) {
3337 			rv = hv_ldc_rx_get_state(ldcp->id,
3338 			    &rx_head, &rx_tail, &ldcp->link_state);
3339 			if (rv != 0) {
3340 				cmn_err(CE_WARN,
3341 				    "ldc_read: (0x%lx) cannot read queue ptrs",
3342 				    ldcp->id);
3343 				mutex_enter(&ldcp->tx_lock);
3344 				i_ldc_reset(ldcp, B_TRUE);
3345 				mutex_exit(&ldcp->tx_lock);
3346 				return (ECONNRESET);
3347 			}
3348 			if (ldcp->link_state != LDC_CHANNEL_UP)
3349 				goto channel_is_reset;
3350 
3351 			if (curr_head == rx_tail) {
3352 
3353 				/* If in the middle of a fragmented xfer */
3354 				if (first_fragment != 0) {
3355 
3356 					/* wait for ldc_delay usecs */
3357 					drv_usecwait(ldc_delay);
3358 
3359 					if (++retries < ldc_max_retries)
3360 						continue;
3361 
3362 					*sizep = 0;
3363 					ldcp->last_msg_rcd = first_fragment - 1;
3364 					DWARN(DBG_ALL_LDCS, "ldc_read: "
3365 						"(0x%llx) read timeout",
3366 						ldcp->id);
3367 					return (EAGAIN);
3368 				}
3369 				*sizep = 0;
3370 				break;
3371 			}
3372 		}
3373 		retries = 0;
3374 
3375 		D2(ldcp->id,
3376 		    "ldc_read: (0x%llx) chd=0x%llx, rxhd=0x%llx, rxtl=0x%llx\n",
3377 		    ldcp->id, curr_head, rx_head, rx_tail);
3378 
3379 		/* get the message */
3380 		msg = (ldc_msg_t *)(ldcp->rx_q_va + curr_head);
3381 
3382 		DUMP_LDC_PKT(ldcp, "ldc_read received pkt",
3383 		    ldcp->rx_q_va + curr_head);
3384 
3385 		/* Check the message ID for the message received */
3386 		if ((rv = i_ldc_check_seqid(ldcp, msg)) != 0) {
3387 
3388 			DWARN(ldcp->id, "ldc_read: (0x%llx) seqid error, "
3389 			    "q_ptrs=0x%lx,0x%lx", ldcp->id, rx_head, rx_tail);
3390 
3391 			/* throw away data */
3392 			bytes_read = 0;
3393 
3394 			/* Reset last_msg_rcd to start of message */
3395 			if (first_fragment != 0) {
3396 				ldcp->last_msg_rcd = first_fragment - 1;
3397 				first_fragment = 0;
3398 			}
3399 			/*
3400 			 * Send a NACK -- invalid seqid
3401 			 * get the current tail for the response
3402 			 */
3403 			rv = i_ldc_send_pkt(ldcp, msg->type, LDC_NACK,
3404 			    (msg->ctrl & LDC_CTRL_MASK));
3405 			if (rv) {
3406 				cmn_err(CE_NOTE,
3407 				    "ldc_read: (0x%lx) err sending "
3408 				    "NACK msg\n", ldcp->id);
3409 
3410 				/* if cannot send NACK - reset channel */
3411 				mutex_enter(&ldcp->tx_lock);
3412 				i_ldc_reset(ldcp, B_FALSE);
3413 				mutex_exit(&ldcp->tx_lock);
3414 				rv = ECONNRESET;
3415 				break;
3416 			}
3417 
3418 			/* purge receive queue */
3419 			rv = i_ldc_set_rx_head(ldcp, rx_tail);
3420 
3421 			break;
3422 		}
3423 
3424 		/*
3425 		 * Process any messages of type CTRL messages
3426 		 * Future implementations should try to pass these
3427 		 * to LDC link by resetting the intr state.
3428 		 *
3429 		 * NOTE: not done as a switch() as type can be both ctrl+data
3430 		 */
3431 		if (msg->type & LDC_CTRL) {
3432 			if (rv = i_ldc_ctrlmsg(ldcp, msg)) {
3433 				if (rv == EAGAIN)
3434 					continue;
3435 				rv = i_ldc_set_rx_head(ldcp, rx_tail);
3436 				*sizep = 0;
3437 				bytes_read = 0;
3438 				break;
3439 			}
3440 		}
3441 
3442 		/* process data ACKs */
3443 		if ((msg->type & LDC_DATA) && (msg->stype & LDC_ACK)) {
3444 			if (rv = i_ldc_process_data_ACK(ldcp, msg)) {
3445 				*sizep = 0;
3446 				bytes_read = 0;
3447 				break;
3448 			}
3449 		}
3450 
3451 		/* process data messages */
3452 		if ((msg->type & LDC_DATA) && (msg->stype & LDC_INFO)) {
3453 
3454 			uint8_t *msgbuf = (uint8_t *)(
3455 				(ldcp->mode == LDC_MODE_RELIABLE ||
3456 				ldcp->mode == LDC_MODE_STREAM)
3457 				? msg->rdata : msg->udata);
3458 
3459 			D2(ldcp->id,
3460 			    "ldc_read: (0x%llx) received data msg\n", ldcp->id);
3461 
3462 			/* get the packet length */
3463 			len = (msg->env & LDC_LEN_MASK);
3464 
3465 				/*
3466 				 * FUTURE OPTIMIZATION:
3467 				 * dont need to set q head for every
3468 				 * packet we read just need to do this when
3469 				 * we are done or need to wait for more
3470 				 * mondos to make a full packet - this is
3471 				 * currently expensive.
3472 				 */
3473 
3474 			if (first_fragment == 0) {
3475 
3476 				/*
3477 				 * first packets should always have the start
3478 				 * bit set (even for a single packet). If not
3479 				 * throw away the packet
3480 				 */
3481 				if (!(msg->env & LDC_FRAG_START)) {
3482 
3483 					DWARN(DBG_ALL_LDCS,
3484 					    "ldc_read: (0x%llx) not start - "
3485 					    "frag=%x\n", ldcp->id,
3486 					    (msg->env) & LDC_FRAG_MASK);
3487 
3488 					/* toss pkt, inc head, cont reading */
3489 					bytes_read = 0;
3490 					target = target_bufp;
3491 					curr_head =
3492 						(curr_head + LDC_PACKET_SIZE)
3493 						& q_size_mask;
3494 					if (rv = i_ldc_set_rx_head(ldcp,
3495 						curr_head))
3496 						break;
3497 
3498 					continue;
3499 				}
3500 
3501 				first_fragment = msg->seqid;
3502 			} else {
3503 				/* check to see if this is a pkt w/ START bit */
3504 				if (msg->env & LDC_FRAG_START) {
3505 					DWARN(DBG_ALL_LDCS,
3506 					    "ldc_read:(0x%llx) unexpected pkt"
3507 					    " env=0x%x discarding %d bytes,"
3508 					    " lastmsg=%d, currentmsg=%d\n",
3509 					    ldcp->id, msg->env&LDC_FRAG_MASK,
3510 					    bytes_read, ldcp->last_msg_rcd,
3511 					    msg->seqid);
3512 
3513 					/* throw data we have read so far */
3514 					bytes_read = 0;
3515 					target = target_bufp;
3516 					first_fragment = msg->seqid;
3517 
3518 					if (rv = i_ldc_set_rx_head(ldcp,
3519 						curr_head))
3520 						break;
3521 				}
3522 			}
3523 
3524 			/* copy (next) pkt into buffer */
3525 			if (len <= (*sizep - bytes_read)) {
3526 				bcopy(msgbuf, target, len);
3527 				target += len;
3528 				bytes_read += len;
3529 			} else {
3530 				/*
3531 				 * there is not enough space in the buffer to
3532 				 * read this pkt. throw message away & continue
3533 				 * reading data from queue
3534 				 */
3535 				DWARN(DBG_ALL_LDCS,
3536 				    "ldc_read: (0x%llx) buffer too small, "
3537 				    "head=0x%lx, expect=%d, got=%d\n", ldcp->id,
3538 				    curr_head, *sizep, bytes_read+len);
3539 
3540 				first_fragment = 0;
3541 				target = target_bufp;
3542 				bytes_read = 0;
3543 
3544 				/* throw away everything received so far */
3545 				if (rv = i_ldc_set_rx_head(ldcp, curr_head))
3546 					break;
3547 
3548 				/* continue reading remaining pkts */
3549 				continue;
3550 			}
3551 		}
3552 
3553 		/* set the message id */
3554 		ldcp->last_msg_rcd = msg->seqid;
3555 
3556 		/* move the head one position */
3557 		curr_head = (curr_head + LDC_PACKET_SIZE) & q_size_mask;
3558 
3559 		if (msg->env & LDC_FRAG_STOP) {
3560 
3561 			/*
3562 			 * All pkts that are part of this fragmented transfer
3563 			 * have been read or this was a single pkt read
3564 			 * or there was an error
3565 			 */
3566 
3567 			/* set the queue head */
3568 			if (rv = i_ldc_set_rx_head(ldcp, curr_head))
3569 				bytes_read = 0;
3570 
3571 			*sizep = bytes_read;
3572 
3573 			break;
3574 		}
3575 
3576 		/* advance head if it is a DATA ACK */
3577 		if ((msg->type & LDC_DATA) && (msg->stype & LDC_ACK)) {
3578 
3579 			/* set the queue head */
3580 			if (rv = i_ldc_set_rx_head(ldcp, curr_head)) {
3581 				bytes_read = 0;
3582 				break;
3583 			}
3584 
3585 			D2(ldcp->id, "ldc_read: (0x%llx) set ACK qhead 0x%llx",
3586 			    ldcp->id, curr_head);
3587 		}
3588 
3589 	} /* for (;;) */
3590 
3591 
3592 	/*
3593 	 * If useful data was read - Send msg ACK
3594 	 * OPTIMIZE: do not send ACK for all msgs - use some frequency
3595 	 */
3596 	if ((bytes_read > 0) && (ldcp->mode == LDC_MODE_RELIABLE ||
3597 		ldcp->mode == LDC_MODE_STREAM)) {
3598 
3599 		rv = i_ldc_send_pkt(ldcp, LDC_DATA, LDC_ACK, 0);
3600 		if (rv && rv != EWOULDBLOCK) {
3601 			cmn_err(CE_NOTE,
3602 			    "ldc_read: (0x%lx) cannot send ACK\n", ldcp->id);
3603 
3604 			/* if cannot send ACK - reset channel */
3605 			goto channel_is_reset;
3606 		}
3607 	}
3608 
3609 	D2(ldcp->id, "ldc_read: (0x%llx) end size=%d", ldcp->id, *sizep);
3610 
3611 	return (rv);
3612 
3613 channel_is_reset:
3614 	mutex_enter(&ldcp->tx_lock);
3615 	i_ldc_reset(ldcp, B_FALSE);
3616 	mutex_exit(&ldcp->tx_lock);
3617 	return (ECONNRESET);
3618 }
3619 
3620 /*
3621  * Use underlying reliable packet mechanism to fetch
3622  * and buffer incoming packets so we can hand them back as
3623  * a basic byte stream.
3624  *
3625  * Enter and exit with ldcp->lock held by caller
3626  */
3627 static int
3628 i_ldc_read_stream(ldc_chan_t *ldcp, caddr_t target_bufp, size_t *sizep)
3629 {
3630 	int	rv;
3631 	size_t	size;
3632 
3633 	ASSERT(mutex_owned(&ldcp->lock));
3634 
3635 	D2(ldcp->id, "i_ldc_read_stream: (0x%llx) buffer size=%d",
3636 		ldcp->id, *sizep);
3637 
3638 	if (ldcp->stream_remains == 0) {
3639 		size = ldcp->mtu;
3640 		rv = i_ldc_read_packet(ldcp,
3641 			(caddr_t)ldcp->stream_bufferp, &size);
3642 		D2(ldcp->id, "i_ldc_read_stream: read packet (0x%llx) size=%d",
3643 			ldcp->id, size);
3644 
3645 		if (rv != 0)
3646 			return (rv);
3647 
3648 		ldcp->stream_remains = size;
3649 		ldcp->stream_offset = 0;
3650 	}
3651 
3652 	size = MIN(ldcp->stream_remains, *sizep);
3653 
3654 	bcopy(ldcp->stream_bufferp + ldcp->stream_offset, target_bufp, size);
3655 	ldcp->stream_offset += size;
3656 	ldcp->stream_remains -= size;
3657 
3658 	D2(ldcp->id, "i_ldc_read_stream: (0x%llx) fill from buffer size=%d",
3659 		ldcp->id, size);
3660 
3661 	*sizep = size;
3662 	return (0);
3663 }
3664 
3665 /*
3666  * Write specified amount of bytes to the channel
3667  * in multiple pkts of pkt_payload size. Each
3668  * packet is tagged with an unique packet ID in
3669  * the case of a reliable link.
3670  *
3671  * On return, size contains the number of bytes written.
3672  */
3673 int
3674 ldc_write(ldc_handle_t handle, caddr_t buf, size_t *sizep)
3675 {
3676 	ldc_chan_t	*ldcp;
3677 	int		rv = 0;
3678 
3679 	if (handle == NULL) {
3680 		DWARN(DBG_ALL_LDCS, "ldc_write: invalid channel handle\n");
3681 		return (EINVAL);
3682 	}
3683 	ldcp = (ldc_chan_t *)handle;
3684 
3685 	/* check if writes can occur */
3686 	if (!mutex_tryenter(&ldcp->tx_lock)) {
3687 		/*
3688 		 * Could not get the lock - channel could
3689 		 * be in the process of being unconfigured
3690 		 * or reader has encountered an error
3691 		 */
3692 		return (EAGAIN);
3693 	}
3694 
3695 	/* check if non-zero data to write */
3696 	if (buf == NULL || sizep == NULL) {
3697 		DWARN(ldcp->id, "ldc_write: (0x%llx) invalid data write\n",
3698 		    ldcp->id);
3699 		mutex_exit(&ldcp->tx_lock);
3700 		return (EINVAL);
3701 	}
3702 
3703 	if (*sizep == 0) {
3704 		DWARN(ldcp->id, "ldc_write: (0x%llx) write size of zero\n",
3705 		    ldcp->id);
3706 		mutex_exit(&ldcp->tx_lock);
3707 		return (0);
3708 	}
3709 
3710 	/* Check if channel is UP for data exchange */
3711 	if (ldcp->tstate != TS_UP) {
3712 		DWARN(ldcp->id,
3713 		    "ldc_write: (0x%llx) channel is not in UP state\n",
3714 		    ldcp->id);
3715 		*sizep = 0;
3716 		rv = ECONNRESET;
3717 	} else {
3718 		rv = ldcp->write_p(ldcp, buf, sizep);
3719 	}
3720 
3721 	mutex_exit(&ldcp->tx_lock);
3722 
3723 	return (rv);
3724 }
3725 
3726 /*
3727  * Write a raw packet to the channel
3728  * On return, size contains the number of bytes written.
3729  */
3730 static int
3731 i_ldc_write_raw(ldc_chan_t *ldcp, caddr_t buf, size_t *sizep)
3732 {
3733 	ldc_msg_t 	*ldcmsg;
3734 	uint64_t 	tx_head, tx_tail, new_tail;
3735 	int		rv = 0;
3736 	size_t		size;
3737 
3738 	ASSERT(MUTEX_HELD(&ldcp->tx_lock));
3739 	ASSERT(ldcp->mode == LDC_MODE_RAW);
3740 
3741 	size = *sizep;
3742 
3743 	/*
3744 	 * Check to see if the packet size is less than or
3745 	 * equal to packet size support in raw mode
3746 	 */
3747 	if (size > ldcp->pkt_payload) {
3748 		DWARN(ldcp->id,
3749 		    "ldc_write: (0x%llx) invalid size (0x%llx) for RAW mode\n",
3750 		    ldcp->id, *sizep);
3751 		*sizep = 0;
3752 		return (EMSGSIZE);
3753 	}
3754 
3755 	/* get the qptrs for the tx queue */
3756 	rv = hv_ldc_tx_get_state(ldcp->id,
3757 	    &ldcp->tx_head, &ldcp->tx_tail, &ldcp->link_state);
3758 	if (rv != 0) {
3759 		cmn_err(CE_WARN,
3760 		    "ldc_write: (0x%lx) cannot read queue ptrs\n", ldcp->id);
3761 		*sizep = 0;
3762 		return (EIO);
3763 	}
3764 
3765 	if (ldcp->link_state == LDC_CHANNEL_DOWN ||
3766 	    ldcp->link_state == LDC_CHANNEL_RESET) {
3767 		DWARN(ldcp->id,
3768 		    "ldc_write: (0x%llx) channel down/reset\n", ldcp->id);
3769 
3770 		*sizep = 0;
3771 		if (mutex_tryenter(&ldcp->lock)) {
3772 			i_ldc_reset(ldcp, B_FALSE);
3773 			mutex_exit(&ldcp->lock);
3774 		} else {
3775 			/*
3776 			 * Release Tx lock, and then reacquire channel
3777 			 * and Tx lock in correct order
3778 			 */
3779 			mutex_exit(&ldcp->tx_lock);
3780 			mutex_enter(&ldcp->lock);
3781 			mutex_enter(&ldcp->tx_lock);
3782 			i_ldc_reset(ldcp, B_FALSE);
3783 			mutex_exit(&ldcp->lock);
3784 		}
3785 		return (ECONNRESET);
3786 	}
3787 
3788 	tx_tail = ldcp->tx_tail;
3789 	tx_head = ldcp->tx_head;
3790 	new_tail = (tx_tail + LDC_PACKET_SIZE) &
3791 		((ldcp->tx_q_entries-1) << LDC_PACKET_SHIFT);
3792 
3793 	if (new_tail == tx_head) {
3794 		DWARN(DBG_ALL_LDCS,
3795 		    "ldc_write: (0x%llx) TX queue is full\n", ldcp->id);
3796 		*sizep = 0;
3797 		return (EWOULDBLOCK);
3798 	}
3799 
3800 	D2(ldcp->id, "ldc_write: (0x%llx) start xfer size=%d",
3801 	    ldcp->id, size);
3802 
3803 	/* Send the data now */
3804 	ldcmsg = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail);
3805 
3806 	/* copy the data into pkt */
3807 	bcopy((uint8_t *)buf, ldcmsg, size);
3808 
3809 	/* increment tail */
3810 	tx_tail = new_tail;
3811 
3812 	/*
3813 	 * All packets have been copied into the TX queue
3814 	 * update the tail ptr in the HV
3815 	 */
3816 	rv = i_ldc_set_tx_tail(ldcp, tx_tail);
3817 	if (rv) {
3818 		if (rv == EWOULDBLOCK) {
3819 			DWARN(ldcp->id, "ldc_write: (0x%llx) write timed out\n",
3820 			    ldcp->id);
3821 			*sizep = 0;
3822 			return (EWOULDBLOCK);
3823 		}
3824 
3825 		*sizep = 0;
3826 		if (mutex_tryenter(&ldcp->lock)) {
3827 			i_ldc_reset(ldcp, B_FALSE);
3828 			mutex_exit(&ldcp->lock);
3829 		} else {
3830 			/*
3831 			 * Release Tx lock, and then reacquire channel
3832 			 * and Tx lock in correct order
3833 			 */
3834 			mutex_exit(&ldcp->tx_lock);
3835 			mutex_enter(&ldcp->lock);
3836 			mutex_enter(&ldcp->tx_lock);
3837 			i_ldc_reset(ldcp, B_FALSE);
3838 			mutex_exit(&ldcp->lock);
3839 		}
3840 		return (ECONNRESET);
3841 	}
3842 
3843 	ldcp->tx_tail = tx_tail;
3844 	*sizep = size;
3845 
3846 	D2(ldcp->id, "ldc_write: (0x%llx) end xfer size=%d", ldcp->id, size);
3847 
3848 	return (rv);
3849 }
3850 
3851 
3852 /*
3853  * Write specified amount of bytes to the channel
3854  * in multiple pkts of pkt_payload size. Each
3855  * packet is tagged with an unique packet ID in
3856  * the case of a reliable link.
3857  *
3858  * On return, size contains the number of bytes written.
3859  * This function needs to ensure that the write size is < MTU size
3860  */
3861 static int
3862 i_ldc_write_packet(ldc_chan_t *ldcp, caddr_t buf, size_t *size)
3863 {
3864 	ldc_msg_t 	*ldcmsg;
3865 	uint64_t 	tx_head, tx_tail, new_tail, start;
3866 	uint64_t	txq_size_mask, numavail;
3867 	uint8_t 	*msgbuf, *source = (uint8_t *)buf;
3868 	size_t 		len, bytes_written = 0, remaining;
3869 	int		rv;
3870 	uint32_t	curr_seqid;
3871 
3872 	ASSERT(MUTEX_HELD(&ldcp->tx_lock));
3873 
3874 	ASSERT(ldcp->mode == LDC_MODE_RELIABLE ||
3875 		ldcp->mode == LDC_MODE_UNRELIABLE ||
3876 		ldcp->mode == LDC_MODE_STREAM);
3877 
3878 	/* compute mask for increment */
3879 	txq_size_mask = (ldcp->tx_q_entries - 1) << LDC_PACKET_SHIFT;
3880 
3881 	/* get the qptrs for the tx queue */
3882 	rv = hv_ldc_tx_get_state(ldcp->id,
3883 	    &ldcp->tx_head, &ldcp->tx_tail, &ldcp->link_state);
3884 	if (rv != 0) {
3885 		cmn_err(CE_WARN,
3886 		    "ldc_write: (0x%lx) cannot read queue ptrs\n", ldcp->id);
3887 		*size = 0;
3888 		return (EIO);
3889 	}
3890 
3891 	if (ldcp->link_state == LDC_CHANNEL_DOWN ||
3892 	    ldcp->link_state == LDC_CHANNEL_RESET) {
3893 		DWARN(ldcp->id,
3894 		    "ldc_write: (0x%llx) channel down/reset\n", ldcp->id);
3895 		*size = 0;
3896 		if (mutex_tryenter(&ldcp->lock)) {
3897 			i_ldc_reset(ldcp, B_FALSE);
3898 			mutex_exit(&ldcp->lock);
3899 		} else {
3900 			/*
3901 			 * Release Tx lock, and then reacquire channel
3902 			 * and Tx lock in correct order
3903 			 */
3904 			mutex_exit(&ldcp->tx_lock);
3905 			mutex_enter(&ldcp->lock);
3906 			mutex_enter(&ldcp->tx_lock);
3907 			i_ldc_reset(ldcp, B_FALSE);
3908 			mutex_exit(&ldcp->lock);
3909 		}
3910 		return (ECONNRESET);
3911 	}
3912 
3913 	tx_tail = ldcp->tx_tail;
3914 	new_tail = (tx_tail + LDC_PACKET_SIZE) %
3915 		(ldcp->tx_q_entries << LDC_PACKET_SHIFT);
3916 
3917 	/*
3918 	 * Link mode determines whether we use HV Tx head or the
3919 	 * private protocol head (corresponding to last ACKd pkt) for
3920 	 * determining how much we can write
3921 	 */
3922 	tx_head = (ldcp->mode == LDC_MODE_RELIABLE ||
3923 		ldcp->mode == LDC_MODE_STREAM)
3924 		? ldcp->tx_ackd_head : ldcp->tx_head;
3925 	if (new_tail == tx_head) {
3926 		DWARN(DBG_ALL_LDCS,
3927 		    "ldc_write: (0x%llx) TX queue is full\n", ldcp->id);
3928 		*size = 0;
3929 		return (EWOULDBLOCK);
3930 	}
3931 
3932 	/*
3933 	 * Make sure that the LDC Tx queue has enough space
3934 	 */
3935 	numavail = (tx_head >> LDC_PACKET_SHIFT) - (tx_tail >> LDC_PACKET_SHIFT)
3936 		+ ldcp->tx_q_entries - 1;
3937 	numavail %= ldcp->tx_q_entries;
3938 
3939 	if (*size > (numavail * ldcp->pkt_payload)) {
3940 		DWARN(DBG_ALL_LDCS,
3941 		    "ldc_write: (0x%llx) TX queue has no space\n", ldcp->id);
3942 		return (EWOULDBLOCK);
3943 	}
3944 
3945 	D2(ldcp->id, "ldc_write: (0x%llx) start xfer size=%d",
3946 	    ldcp->id, *size);
3947 
3948 	/* Send the data now */
3949 	bytes_written = 0;
3950 	curr_seqid = ldcp->last_msg_snt;
3951 	start = tx_tail;
3952 
3953 	while (*size > bytes_written) {
3954 
3955 		ldcmsg = (ldc_msg_t *)(ldcp->tx_q_va + tx_tail);
3956 
3957 		msgbuf = (uint8_t *)((ldcp->mode == LDC_MODE_RELIABLE ||
3958 			ldcp->mode == LDC_MODE_STREAM)
3959 			? ldcmsg->rdata : ldcmsg->udata);
3960 
3961 		ldcmsg->type = LDC_DATA;
3962 		ldcmsg->stype = LDC_INFO;
3963 		ldcmsg->ctrl = 0;
3964 
3965 		remaining = *size - bytes_written;
3966 		len = min(ldcp->pkt_payload, remaining);
3967 		ldcmsg->env = (uint8_t)len;
3968 
3969 		curr_seqid++;
3970 		ldcmsg->seqid = curr_seqid;
3971 
3972 		/* copy the data into pkt */
3973 		bcopy(source, msgbuf, len);
3974 
3975 		source += len;
3976 		bytes_written += len;
3977 
3978 		/* increment tail */
3979 		tx_tail = (tx_tail + LDC_PACKET_SIZE) & txq_size_mask;
3980 
3981 		ASSERT(tx_tail != tx_head);
3982 	}
3983 
3984 	/* Set the start and stop bits */
3985 	ldcmsg->env |= LDC_FRAG_STOP;
3986 	ldcmsg = (ldc_msg_t *)(ldcp->tx_q_va + start);
3987 	ldcmsg->env |= LDC_FRAG_START;
3988 
3989 	/*
3990 	 * All packets have been copied into the TX queue
3991 	 * update the tail ptr in the HV
3992 	 */
3993 	rv = i_ldc_set_tx_tail(ldcp, tx_tail);
3994 	if (rv == 0) {
3995 		ldcp->tx_tail = tx_tail;
3996 		ldcp->last_msg_snt = curr_seqid;
3997 		*size = bytes_written;
3998 	} else {
3999 		int rv2;
4000 
4001 		if (rv != EWOULDBLOCK) {
4002 			*size = 0;
4003 			if (mutex_tryenter(&ldcp->lock)) {
4004 				i_ldc_reset(ldcp, B_FALSE);
4005 				mutex_exit(&ldcp->lock);
4006 			} else {
4007 				/*
4008 				 * Release Tx lock, and then reacquire channel
4009 				 * and Tx lock in correct order
4010 				 */
4011 				mutex_exit(&ldcp->tx_lock);
4012 				mutex_enter(&ldcp->lock);
4013 				mutex_enter(&ldcp->tx_lock);
4014 				i_ldc_reset(ldcp, B_FALSE);
4015 				mutex_exit(&ldcp->lock);
4016 			}
4017 			return (ECONNRESET);
4018 		}
4019 
4020 		DWARN(ldcp->id, "hv_tx_set_tail returns 0x%x (head 0x%x, "
4021 			"old tail 0x%x, new tail 0x%x, qsize=0x%x)\n",
4022 			rv, ldcp->tx_head, ldcp->tx_tail, tx_tail,
4023 			(ldcp->tx_q_entries << LDC_PACKET_SHIFT));
4024 
4025 		rv2 = hv_ldc_tx_get_state(ldcp->id,
4026 		    &tx_head, &tx_tail, &ldcp->link_state);
4027 
4028 		DWARN(ldcp->id, "hv_ldc_tx_get_state returns 0x%x "
4029 			"(head 0x%x, tail 0x%x state 0x%x)\n",
4030 			rv2, tx_head, tx_tail, ldcp->link_state);
4031 
4032 		*size = 0;
4033 	}
4034 
4035 	D2(ldcp->id, "ldc_write: (0x%llx) end xfer size=%d", ldcp->id, *size);
4036 
4037 	return (rv);
4038 }
4039 
4040 /*
4041  * Write specified amount of bytes to the channel
4042  * in multiple pkts of pkt_payload size. Each
4043  * packet is tagged with an unique packet ID in
4044  * the case of a reliable link.
4045  *
4046  * On return, size contains the number of bytes written.
4047  * This function needs to ensure that the write size is < MTU size
4048  */
4049 static int
4050 i_ldc_write_stream(ldc_chan_t *ldcp, caddr_t buf, size_t *sizep)
4051 {
4052 	ASSERT(MUTEX_HELD(&ldcp->tx_lock));
4053 	ASSERT(ldcp->mode == LDC_MODE_STREAM);
4054 
4055 	/* Truncate packet to max of MTU size */
4056 	if (*sizep > ldcp->mtu) *sizep = ldcp->mtu;
4057 	return (i_ldc_write_packet(ldcp, buf, sizep));
4058 }
4059 
4060 
4061 /*
4062  * Interfaces for channel nexus to register/unregister with LDC module
4063  * The nexus will register functions to be used to register individual
4064  * channels with the nexus and enable interrupts for the channels
4065  */
4066 int
4067 ldc_register(ldc_cnex_t *cinfo)
4068 {
4069 	ldc_chan_t	*ldcp;
4070 
4071 	if (cinfo == NULL || cinfo->dip == NULL ||
4072 	    cinfo->reg_chan == NULL || cinfo->unreg_chan == NULL ||
4073 	    cinfo->add_intr == NULL || cinfo->rem_intr == NULL ||
4074 	    cinfo->clr_intr == NULL) {
4075 
4076 		DWARN(DBG_ALL_LDCS, "ldc_register: invalid nexus info\n");
4077 		return (EINVAL);
4078 	}
4079 
4080 	mutex_enter(&ldcssp->lock);
4081 
4082 	/* nexus registration */
4083 	ldcssp->cinfo.dip = cinfo->dip;
4084 	ldcssp->cinfo.reg_chan = cinfo->reg_chan;
4085 	ldcssp->cinfo.unreg_chan = cinfo->unreg_chan;
4086 	ldcssp->cinfo.add_intr = cinfo->add_intr;
4087 	ldcssp->cinfo.rem_intr = cinfo->rem_intr;
4088 	ldcssp->cinfo.clr_intr = cinfo->clr_intr;
4089 
4090 	/* register any channels that might have been previously initialized */
4091 	ldcp = ldcssp->chan_list;
4092 	while (ldcp) {
4093 		if ((ldcp->tstate & TS_QCONF_RDY) &&
4094 		    (ldcp->tstate & TS_CNEX_RDY) == 0)
4095 			(void) i_ldc_register_channel(ldcp);
4096 
4097 		ldcp = ldcp->next;
4098 	}
4099 
4100 	mutex_exit(&ldcssp->lock);
4101 
4102 	return (0);
4103 }
4104 
4105 int
4106 ldc_unregister(ldc_cnex_t *cinfo)
4107 {
4108 	if (cinfo == NULL || cinfo->dip == NULL) {
4109 		DWARN(DBG_ALL_LDCS, "ldc_unregister: invalid nexus info\n");
4110 		return (EINVAL);
4111 	}
4112 
4113 	mutex_enter(&ldcssp->lock);
4114 
4115 	if (cinfo->dip != ldcssp->cinfo.dip) {
4116 		DWARN(DBG_ALL_LDCS, "ldc_unregister: invalid dip\n");
4117 		mutex_exit(&ldcssp->lock);
4118 		return (EINVAL);
4119 	}
4120 
4121 	/* nexus unregister */
4122 	ldcssp->cinfo.dip = NULL;
4123 	ldcssp->cinfo.reg_chan = NULL;
4124 	ldcssp->cinfo.unreg_chan = NULL;
4125 	ldcssp->cinfo.add_intr = NULL;
4126 	ldcssp->cinfo.rem_intr = NULL;
4127 	ldcssp->cinfo.clr_intr = NULL;
4128 
4129 	mutex_exit(&ldcssp->lock);
4130 
4131 	return (0);
4132 }
4133 
4134 
4135 /* ------------------------------------------------------------------------- */
4136 
4137 /*
4138  * Allocate a memory handle for the channel and link it into the list
4139  * Also choose which memory table to use if this is the first handle
4140  * being assigned to this channel
4141  */
4142 int
4143 ldc_mem_alloc_handle(ldc_handle_t handle, ldc_mem_handle_t *mhandle)
4144 {
4145 	ldc_chan_t 	*ldcp;
4146 	ldc_mhdl_t	*mhdl;
4147 
4148 	if (handle == NULL) {
4149 		DWARN(DBG_ALL_LDCS,
4150 		    "ldc_mem_alloc_handle: invalid channel handle\n");
4151 		return (EINVAL);
4152 	}
4153 	ldcp = (ldc_chan_t *)handle;
4154 
4155 	mutex_enter(&ldcp->lock);
4156 
4157 	/* check to see if channel is initalized */
4158 	if ((ldcp->tstate & ~TS_IN_RESET) < TS_INIT) {
4159 		DWARN(ldcp->id,
4160 		    "ldc_mem_alloc_handle: (0x%llx) channel not initialized\n",
4161 		    ldcp->id);
4162 		mutex_exit(&ldcp->lock);
4163 		return (EINVAL);
4164 	}
4165 
4166 	/* allocate handle for channel */
4167 	mhdl = kmem_cache_alloc(ldcssp->memhdl_cache, KM_SLEEP);
4168 
4169 	/* initialize the lock */
4170 	mutex_init(&mhdl->lock, NULL, MUTEX_DRIVER, NULL);
4171 
4172 	mhdl->myshadow = B_FALSE;
4173 	mhdl->memseg = NULL;
4174 	mhdl->ldcp = ldcp;
4175 	mhdl->status = LDC_UNBOUND;
4176 
4177 	/* insert memory handle (@ head) into list */
4178 	if (ldcp->mhdl_list == NULL) {
4179 		ldcp->mhdl_list = mhdl;
4180 		mhdl->next = NULL;
4181 	} else {
4182 		/* insert @ head */
4183 		mhdl->next = ldcp->mhdl_list;
4184 		ldcp->mhdl_list = mhdl;
4185 	}
4186 
4187 	/* return the handle */
4188 	*mhandle = (ldc_mem_handle_t)mhdl;
4189 
4190 	mutex_exit(&ldcp->lock);
4191 
4192 	D1(ldcp->id, "ldc_mem_alloc_handle: (0x%llx) allocated handle 0x%llx\n",
4193 	    ldcp->id, mhdl);
4194 
4195 	return (0);
4196 }
4197 
4198 /*
4199  * Free memory handle for the channel and unlink it from the list
4200  */
4201 int
4202 ldc_mem_free_handle(ldc_mem_handle_t mhandle)
4203 {
4204 	ldc_mhdl_t 	*mhdl, *phdl;
4205 	ldc_chan_t 	*ldcp;
4206 
4207 	if (mhandle == NULL) {
4208 		DWARN(DBG_ALL_LDCS,
4209 		    "ldc_mem_free_handle: invalid memory handle\n");
4210 		return (EINVAL);
4211 	}
4212 	mhdl = (ldc_mhdl_t *)mhandle;
4213 
4214 	mutex_enter(&mhdl->lock);
4215 
4216 	ldcp = mhdl->ldcp;
4217 
4218 	if (mhdl->status == LDC_BOUND || mhdl->status == LDC_MAPPED) {
4219 		DWARN(ldcp->id,
4220 		    "ldc_mem_free_handle: cannot free, 0x%llx hdl bound\n",
4221 		    mhdl);
4222 		mutex_exit(&mhdl->lock);
4223 		return (EINVAL);
4224 	}
4225 	mutex_exit(&mhdl->lock);
4226 
4227 	mutex_enter(&ldcp->mlist_lock);
4228 
4229 	phdl = ldcp->mhdl_list;
4230 
4231 	/* first handle */
4232 	if (phdl == mhdl) {
4233 		ldcp->mhdl_list = mhdl->next;
4234 		mutex_destroy(&mhdl->lock);
4235 		kmem_cache_free(ldcssp->memhdl_cache, mhdl);
4236 
4237 		D1(ldcp->id,
4238 		    "ldc_mem_free_handle: (0x%llx) freed handle 0x%llx\n",
4239 		    ldcp->id, mhdl);
4240 	} else {
4241 		/* walk the list - unlink and free */
4242 		while (phdl != NULL) {
4243 			if (phdl->next == mhdl) {
4244 				phdl->next = mhdl->next;
4245 				mutex_destroy(&mhdl->lock);
4246 				kmem_cache_free(ldcssp->memhdl_cache, mhdl);
4247 				D1(ldcp->id,
4248 				    "ldc_mem_free_handle: (0x%llx) freed "
4249 				    "handle 0x%llx\n", ldcp->id, mhdl);
4250 				break;
4251 			}
4252 			phdl = phdl->next;
4253 		}
4254 	}
4255 
4256 	if (phdl == NULL) {
4257 		DWARN(ldcp->id,
4258 		    "ldc_mem_free_handle: invalid handle 0x%llx\n", mhdl);
4259 		mutex_exit(&ldcp->mlist_lock);
4260 		return (EINVAL);
4261 	}
4262 
4263 	mutex_exit(&ldcp->mlist_lock);
4264 
4265 	return (0);
4266 }
4267 
4268 /*
4269  * Bind a memory handle to a virtual address.
4270  * The virtual address is converted to the corresponding real addresses.
4271  * Returns pointer to the first ldc_mem_cookie and the total number
4272  * of cookies for this virtual address. Other cookies can be obtained
4273  * using the ldc_mem_nextcookie() call. If the pages are stored in
4274  * consecutive locations in the table, a single cookie corresponding to
4275  * the first location is returned. The cookie size spans all the entries.
4276  *
4277  * If the VA corresponds to a page that is already being exported, reuse
4278  * the page and do not export it again. Bump the page's use count.
4279  */
4280 int
4281 ldc_mem_bind_handle(ldc_mem_handle_t mhandle, caddr_t vaddr, size_t len,
4282     uint8_t mtype, uint8_t perm, ldc_mem_cookie_t *cookie, uint32_t *ccount)
4283 {
4284 	ldc_mhdl_t	*mhdl;
4285 	ldc_chan_t 	*ldcp;
4286 	ldc_mtbl_t	*mtbl;
4287 	ldc_memseg_t	*memseg;
4288 	ldc_mte_t	tmp_mte;
4289 	uint64_t	index, prev_index = 0;
4290 	int64_t		cookie_idx;
4291 	uintptr_t	raddr, ra_aligned;
4292 	uint64_t	psize, poffset, v_offset;
4293 	uint64_t	pg_shift, pg_size, pg_size_code, pg_mask;
4294 	pgcnt_t		npages;
4295 	caddr_t		v_align, addr;
4296 	int 		i, rv;
4297 
4298 	if (mhandle == NULL) {
4299 		DWARN(DBG_ALL_LDCS,
4300 		    "ldc_mem_bind_handle: invalid memory handle\n");
4301 		return (EINVAL);
4302 	}
4303 	mhdl = (ldc_mhdl_t *)mhandle;
4304 	ldcp = mhdl->ldcp;
4305 
4306 	/* clear count */
4307 	*ccount = 0;
4308 
4309 	mutex_enter(&mhdl->lock);
4310 
4311 	if (mhdl->status == LDC_BOUND || mhdl->memseg != NULL) {
4312 		DWARN(ldcp->id,
4313 		    "ldc_mem_bind_handle: (0x%x) handle already bound\n",
4314 		    mhandle);
4315 		mutex_exit(&mhdl->lock);
4316 		return (EINVAL);
4317 	}
4318 
4319 	/* Force address and size to be 8-byte aligned */
4320 	if ((((uintptr_t)vaddr | len) & 0x7) != 0) {
4321 		DWARN(ldcp->id,
4322 		    "ldc_mem_bind_handle: addr/size is not 8-byte aligned\n");
4323 		mutex_exit(&mhdl->lock);
4324 		return (EINVAL);
4325 	}
4326 
4327 	/*
4328 	 * If this channel is binding a memory handle for the
4329 	 * first time allocate it a memory map table and initialize it
4330 	 */
4331 	if ((mtbl = ldcp->mtbl) == NULL) {
4332 
4333 		mutex_enter(&ldcp->lock);
4334 
4335 		/* Allocate and initialize the map table structure */
4336 		mtbl = kmem_zalloc(sizeof (ldc_mtbl_t), KM_SLEEP);
4337 		mtbl->num_entries = mtbl->num_avail = ldc_maptable_entries;
4338 		mtbl->size = ldc_maptable_entries * sizeof (ldc_mte_slot_t);
4339 		mtbl->next_entry = NULL;
4340 		mtbl->contigmem = B_TRUE;
4341 
4342 		/* Allocate the table itself */
4343 		mtbl->table = (ldc_mte_slot_t *)
4344 			contig_mem_alloc_align(mtbl->size, MMU_PAGESIZE);
4345 		if (mtbl->table == NULL) {
4346 
4347 			/* allocate a page of memory using kmem_alloc */
4348 			mtbl->table = kmem_alloc(MMU_PAGESIZE, KM_SLEEP);
4349 			mtbl->size = MMU_PAGESIZE;
4350 			mtbl->contigmem = B_FALSE;
4351 			mtbl->num_entries = mtbl->num_avail =
4352 				mtbl->size / sizeof (ldc_mte_slot_t);
4353 			DWARN(ldcp->id,
4354 			    "ldc_mem_bind_handle: (0x%llx) reduced tbl size "
4355 			    "to %lx entries\n", ldcp->id, mtbl->num_entries);
4356 		}
4357 
4358 		/* zero out the memory */
4359 		bzero(mtbl->table, mtbl->size);
4360 
4361 		/* initialize the lock */
4362 		mutex_init(&mtbl->lock, NULL, MUTEX_DRIVER, NULL);
4363 
4364 		/* register table for this channel */
4365 		rv = hv_ldc_set_map_table(ldcp->id,
4366 		    va_to_pa(mtbl->table), mtbl->num_entries);
4367 		if (rv != 0) {
4368 			cmn_err(CE_WARN,
4369 			    "ldc_mem_bind_handle: (0x%lx) err %d mapping tbl",
4370 			    ldcp->id, rv);
4371 			if (mtbl->contigmem)
4372 				contig_mem_free(mtbl->table, mtbl->size);
4373 			else
4374 				kmem_free(mtbl->table, mtbl->size);
4375 			mutex_destroy(&mtbl->lock);
4376 			kmem_free(mtbl, sizeof (ldc_mtbl_t));
4377 			mutex_exit(&ldcp->lock);
4378 			mutex_exit(&mhdl->lock);
4379 			return (EIO);
4380 		}
4381 
4382 		ldcp->mtbl = mtbl;
4383 		mutex_exit(&ldcp->lock);
4384 
4385 		D1(ldcp->id,
4386 		    "ldc_mem_bind_handle: (0x%llx) alloc'd map table 0x%llx\n",
4387 		    ldcp->id, ldcp->mtbl->table);
4388 	}
4389 
4390 	/* FUTURE: get the page size, pgsz code, and shift */
4391 	pg_size = MMU_PAGESIZE;
4392 	pg_size_code = page_szc(pg_size);
4393 	pg_shift = page_get_shift(pg_size_code);
4394 	pg_mask = ~(pg_size - 1);
4395 
4396 	D1(ldcp->id, "ldc_mem_bind_handle: (0x%llx) binding "
4397 	    "va 0x%llx pgsz=0x%llx, pgszc=0x%llx, pg_shift=0x%llx\n",
4398 	    ldcp->id, vaddr, pg_size, pg_size_code, pg_shift);
4399 
4400 	/* aligned VA and its offset */
4401 	v_align = (caddr_t)(((uintptr_t)vaddr) & ~(pg_size - 1));
4402 	v_offset = ((uintptr_t)vaddr) & (pg_size - 1);
4403 
4404 	npages = (len+v_offset)/pg_size;
4405 	npages = ((len+v_offset)%pg_size == 0) ? npages : npages+1;
4406 
4407 	D1(ldcp->id, "ldc_mem_bind_handle: binding "
4408 	    "(0x%llx) v=0x%llx,val=0x%llx,off=0x%x,pgs=0x%x\n",
4409 	    ldcp->id, vaddr, v_align, v_offset, npages);
4410 
4411 	/* lock the memory table - exclusive access to channel */
4412 	mutex_enter(&mtbl->lock);
4413 
4414 	if (npages > mtbl->num_avail) {
4415 		D1(ldcp->id, "ldc_mem_bind_handle: (0x%llx) no table entries\n",
4416 		    ldcp->id);
4417 		mutex_exit(&mtbl->lock);
4418 		mutex_exit(&mhdl->lock);
4419 		return (ENOMEM);
4420 	}
4421 
4422 	/* Allocate a memseg structure */
4423 	memseg = mhdl->memseg =
4424 		kmem_cache_alloc(ldcssp->memseg_cache, KM_SLEEP);
4425 
4426 	/* Allocate memory to store all pages and cookies */
4427 	memseg->pages = kmem_zalloc((sizeof (ldc_page_t) * npages), KM_SLEEP);
4428 	memseg->cookies =
4429 		kmem_zalloc((sizeof (ldc_mem_cookie_t) * npages), KM_SLEEP);
4430 
4431 	D2(ldcp->id, "ldc_mem_bind_handle: (0x%llx) processing 0x%llx pages\n",
4432 	    ldcp->id, npages);
4433 
4434 	addr = v_align;
4435 
4436 	/*
4437 	 * Check if direct shared memory map is enabled, if not change
4438 	 * the mapping type to include SHADOW_MAP.
4439 	 */
4440 	if (ldc_shmem_enabled == 0)
4441 		mtype = LDC_SHADOW_MAP;
4442 
4443 	/*
4444 	 * Table slots are used in a round-robin manner. The algorithm permits
4445 	 * inserting duplicate entries. Slots allocated earlier will typically
4446 	 * get freed before we get back to reusing the slot.Inserting duplicate
4447 	 * entries should be OK as we only lookup entries using the cookie addr
4448 	 * i.e. tbl index, during export, unexport and copy operation.
4449 	 *
4450 	 * One implementation what was tried was to search for a duplicate
4451 	 * page entry first and reuse it. The search overhead is very high and
4452 	 * in the vnet case dropped the perf by almost half, 50 to 24 mbps.
4453 	 * So it does make sense to avoid searching for duplicates.
4454 	 *
4455 	 * But during the process of searching for a free slot, if we find a
4456 	 * duplicate entry we will go ahead and use it, and bump its use count.
4457 	 */
4458 
4459 	/* index to start searching from */
4460 	index = mtbl->next_entry;
4461 	cookie_idx = -1;
4462 
4463 	tmp_mte.ll = 0;	/* initialise fields to 0 */
4464 
4465 	if (mtype & LDC_DIRECT_MAP) {
4466 		tmp_mte.mte_r = (perm & LDC_MEM_R) ? 1 : 0;
4467 		tmp_mte.mte_w = (perm & LDC_MEM_W) ? 1 : 0;
4468 		tmp_mte.mte_x = (perm & LDC_MEM_X) ? 1 : 0;
4469 	}
4470 
4471 	if (mtype & LDC_SHADOW_MAP) {
4472 		tmp_mte.mte_cr = (perm & LDC_MEM_R) ? 1 : 0;
4473 		tmp_mte.mte_cw = (perm & LDC_MEM_W) ? 1 : 0;
4474 	}
4475 
4476 	if (mtype & LDC_IO_MAP) {
4477 		tmp_mte.mte_ir = (perm & LDC_MEM_R) ? 1 : 0;
4478 		tmp_mte.mte_iw = (perm & LDC_MEM_W) ? 1 : 0;
4479 	}
4480 
4481 	D1(ldcp->id, "ldc_mem_bind_handle mte=0x%llx\n", tmp_mte.ll);
4482 
4483 	tmp_mte.mte_pgszc = pg_size_code;
4484 
4485 	/* initialize each mem table entry */
4486 	for (i = 0; i < npages; i++) {
4487 
4488 		/* check if slot is available in the table */
4489 		while (mtbl->table[index].entry.ll != 0) {
4490 
4491 			index = (index + 1) % mtbl->num_entries;
4492 
4493 			if (index == mtbl->next_entry) {
4494 				/* we have looped around */
4495 				DWARN(DBG_ALL_LDCS,
4496 				    "ldc_mem_bind_handle: (0x%llx) cannot find "
4497 				    "entry\n", ldcp->id);
4498 				*ccount = 0;
4499 
4500 				/* NOTE: free memory, remove previous entries */
4501 				/* this shouldnt happen as num_avail was ok */
4502 
4503 				mutex_exit(&mtbl->lock);
4504 				mutex_exit(&mhdl->lock);
4505 				return (ENOMEM);
4506 			}
4507 		}
4508 
4509 		/* get the real address */
4510 		raddr = va_to_pa((void *)addr);
4511 		ra_aligned = ((uintptr_t)raddr & pg_mask);
4512 
4513 		/* build the mte */
4514 		tmp_mte.mte_rpfn = ra_aligned >> pg_shift;
4515 
4516 		D1(ldcp->id, "ldc_mem_bind_handle mte=0x%llx\n", tmp_mte.ll);
4517 
4518 		/* update entry in table */
4519 		mtbl->table[index].entry = tmp_mte;
4520 
4521 		D2(ldcp->id, "ldc_mem_bind_handle: (0x%llx) stored MTE 0x%llx"
4522 		    " into loc 0x%llx\n", ldcp->id, tmp_mte.ll, index);
4523 
4524 		/* calculate the size and offset for this export range */
4525 		if (i == 0) {
4526 			/* first page */
4527 			psize = min((pg_size - v_offset), len);
4528 			poffset = v_offset;
4529 
4530 		} else if (i == (npages - 1)) {
4531 			/* last page */
4532 			psize =	(((uintptr_t)(vaddr + len)) &
4533 				    ((uint64_t)(pg_size-1)));
4534 			if (psize == 0)
4535 				psize = pg_size;
4536 			poffset = 0;
4537 
4538 		} else {
4539 			/* middle pages */
4540 			psize = pg_size;
4541 			poffset = 0;
4542 		}
4543 
4544 		/* store entry for this page */
4545 		memseg->pages[i].index = index;
4546 		memseg->pages[i].raddr = raddr;
4547 		memseg->pages[i].offset = poffset;
4548 		memseg->pages[i].size = psize;
4549 		memseg->pages[i].mte = &(mtbl->table[index]);
4550 
4551 		/* create the cookie */
4552 		if (i == 0 || (index != prev_index + 1)) {
4553 			cookie_idx++;
4554 			memseg->cookies[cookie_idx].addr =
4555 				IDX2COOKIE(index, pg_size_code, pg_shift);
4556 			memseg->cookies[cookie_idx].addr |= poffset;
4557 			memseg->cookies[cookie_idx].size = psize;
4558 
4559 		} else {
4560 			memseg->cookies[cookie_idx].size += psize;
4561 		}
4562 
4563 		D1(ldcp->id, "ldc_mem_bind_handle: bound "
4564 		    "(0x%llx) va=0x%llx, idx=0x%llx, "
4565 		    "ra=0x%llx(sz=0x%x,off=0x%x)\n",
4566 		    ldcp->id, addr, index, raddr, psize, poffset);
4567 
4568 		/* decrement number of available entries */
4569 		mtbl->num_avail--;
4570 
4571 		/* increment va by page size */
4572 		addr += pg_size;
4573 
4574 		/* increment index */
4575 		prev_index = index;
4576 		index = (index + 1) % mtbl->num_entries;
4577 
4578 		/* save the next slot */
4579 		mtbl->next_entry = index;
4580 	}
4581 
4582 	mutex_exit(&mtbl->lock);
4583 
4584 	/* memory handle = bound */
4585 	mhdl->mtype = mtype;
4586 	mhdl->perm = perm;
4587 	mhdl->status = LDC_BOUND;
4588 
4589 	/* update memseg_t */
4590 	memseg->vaddr = vaddr;
4591 	memseg->raddr = memseg->pages[0].raddr;
4592 	memseg->size = len;
4593 	memseg->npages = npages;
4594 	memseg->ncookies = cookie_idx + 1;
4595 	memseg->next_cookie = (memseg->ncookies > 1) ? 1 : 0;
4596 
4597 	/* return count and first cookie */
4598 	*ccount = memseg->ncookies;
4599 	cookie->addr = memseg->cookies[0].addr;
4600 	cookie->size = memseg->cookies[0].size;
4601 
4602 	D1(ldcp->id,
4603 	    "ldc_mem_bind_handle: (0x%llx) bound 0x%llx, va=0x%llx, "
4604 	    "pgs=0x%llx cookies=0x%llx\n",
4605 	    ldcp->id, mhdl, vaddr, npages, memseg->ncookies);
4606 
4607 	mutex_exit(&mhdl->lock);
4608 	return (0);
4609 }
4610 
4611 /*
4612  * Return the next cookie associated with the specified memory handle
4613  */
4614 int
4615 ldc_mem_nextcookie(ldc_mem_handle_t mhandle, ldc_mem_cookie_t *cookie)
4616 {
4617 	ldc_mhdl_t	*mhdl;
4618 	ldc_chan_t 	*ldcp;
4619 	ldc_memseg_t	*memseg;
4620 
4621 	if (mhandle == NULL) {
4622 		DWARN(DBG_ALL_LDCS,
4623 		    "ldc_mem_nextcookie: invalid memory handle\n");
4624 		return (EINVAL);
4625 	}
4626 	mhdl = (ldc_mhdl_t *)mhandle;
4627 
4628 	mutex_enter(&mhdl->lock);
4629 
4630 	ldcp = mhdl->ldcp;
4631 	memseg = mhdl->memseg;
4632 
4633 	if (cookie == 0) {
4634 		DWARN(ldcp->id,
4635 		    "ldc_mem_nextcookie:(0x%llx) invalid cookie arg\n",
4636 		    ldcp->id);
4637 		mutex_exit(&mhdl->lock);
4638 		return (EINVAL);
4639 	}
4640 
4641 	if (memseg->next_cookie != 0) {
4642 		cookie->addr = memseg->cookies[memseg->next_cookie].addr;
4643 		cookie->size = memseg->cookies[memseg->next_cookie].size;
4644 		memseg->next_cookie++;
4645 		if (memseg->next_cookie == memseg->ncookies)
4646 			memseg->next_cookie = 0;
4647 
4648 	} else {
4649 		DWARN(ldcp->id,
4650 		    "ldc_mem_nextcookie:(0x%llx) no more cookies\n", ldcp->id);
4651 		cookie->addr = 0;
4652 		cookie->size = 0;
4653 		mutex_exit(&mhdl->lock);
4654 		return (EINVAL);
4655 	}
4656 
4657 	D1(ldcp->id,
4658 	    "ldc_mem_nextcookie: (0x%llx) cookie addr=0x%llx,sz=0x%llx\n",
4659 	    ldcp->id, cookie->addr, cookie->size);
4660 
4661 	mutex_exit(&mhdl->lock);
4662 	return (0);
4663 }
4664 
4665 /*
4666  * Unbind the virtual memory region associated with the specified
4667  * memory handle. Allassociated cookies are freed and the corresponding
4668  * RA space is no longer exported.
4669  */
4670 int
4671 ldc_mem_unbind_handle(ldc_mem_handle_t mhandle)
4672 {
4673 	ldc_mhdl_t	*mhdl;
4674 	ldc_chan_t 	*ldcp;
4675 	ldc_mtbl_t	*mtbl;
4676 	ldc_memseg_t	*memseg;
4677 	uint64_t	cookie_addr;
4678 	uint64_t	pg_shift, pg_size_code;
4679 	int		i, rv;
4680 
4681 	if (mhandle == NULL) {
4682 		DWARN(DBG_ALL_LDCS,
4683 		    "ldc_mem_unbind_handle: invalid memory handle\n");
4684 		return (EINVAL);
4685 	}
4686 	mhdl = (ldc_mhdl_t *)mhandle;
4687 
4688 	mutex_enter(&mhdl->lock);
4689 
4690 	if (mhdl->status == LDC_UNBOUND) {
4691 		DWARN(DBG_ALL_LDCS,
4692 		    "ldc_mem_unbind_handle: (0x%x) handle is not bound\n",
4693 		    mhandle);
4694 		mutex_exit(&mhdl->lock);
4695 		return (EINVAL);
4696 	}
4697 
4698 	ldcp = mhdl->ldcp;
4699 	mtbl = ldcp->mtbl;
4700 
4701 	memseg = mhdl->memseg;
4702 
4703 	/* lock the memory table - exclusive access to channel */
4704 	mutex_enter(&mtbl->lock);
4705 
4706 	/* undo the pages exported */
4707 	for (i = 0; i < memseg->npages; i++) {
4708 
4709 		/* check for mapped pages, revocation cookie != 0 */
4710 		if (memseg->pages[i].mte->cookie) {
4711 
4712 			pg_size_code = page_szc(memseg->pages[i].size);
4713 			pg_shift = page_get_shift(memseg->pages[i].size);
4714 			cookie_addr = IDX2COOKIE(memseg->pages[i].index,
4715 			    pg_size_code, pg_shift);
4716 
4717 			D1(ldcp->id, "ldc_mem_unbind_handle: (0x%llx) revoke "
4718 			    "cookie 0x%llx, rcookie 0x%llx\n", ldcp->id,
4719 			    cookie_addr, memseg->pages[i].mte->cookie);
4720 			rv = hv_ldc_revoke(ldcp->id, cookie_addr,
4721 			    memseg->pages[i].mte->cookie);
4722 			if (rv) {
4723 				DWARN(ldcp->id,
4724 				    "ldc_mem_unbind_handle: (0x%llx) cannot "
4725 				    "revoke mapping, cookie %llx\n", ldcp->id,
4726 				    cookie_addr);
4727 			}
4728 		}
4729 
4730 		/* clear the entry from the table */
4731 		memseg->pages[i].mte->entry.ll = 0;
4732 		mtbl->num_avail++;
4733 	}
4734 	mutex_exit(&mtbl->lock);
4735 
4736 	/* free the allocated memseg and page structures */
4737 	kmem_free(memseg->pages, (sizeof (ldc_page_t) * memseg->npages));
4738 	kmem_free(memseg->cookies,
4739 	    (sizeof (ldc_mem_cookie_t) * memseg->npages));
4740 	kmem_cache_free(ldcssp->memseg_cache, memseg);
4741 
4742 	/* uninitialize the memory handle */
4743 	mhdl->memseg = NULL;
4744 	mhdl->status = LDC_UNBOUND;
4745 
4746 	D1(ldcp->id, "ldc_mem_unbind_handle: (0x%llx) unbound handle 0x%llx\n",
4747 	    ldcp->id, mhdl);
4748 
4749 	mutex_exit(&mhdl->lock);
4750 	return (0);
4751 }
4752 
4753 /*
4754  * Get information about the dring. The base address of the descriptor
4755  * ring along with the type and permission are returned back.
4756  */
4757 int
4758 ldc_mem_info(ldc_mem_handle_t mhandle, ldc_mem_info_t *minfo)
4759 {
4760 	ldc_mhdl_t	*mhdl;
4761 
4762 	if (mhandle == NULL) {
4763 		DWARN(DBG_ALL_LDCS, "ldc_mem_info: invalid memory handle\n");
4764 		return (EINVAL);
4765 	}
4766 	mhdl = (ldc_mhdl_t *)mhandle;
4767 
4768 	if (minfo == NULL) {
4769 		DWARN(DBG_ALL_LDCS, "ldc_mem_info: invalid args\n");
4770 		return (EINVAL);
4771 	}
4772 
4773 	mutex_enter(&mhdl->lock);
4774 
4775 	minfo->status = mhdl->status;
4776 	if (mhdl->status == LDC_BOUND || mhdl->status == LDC_MAPPED) {
4777 		minfo->vaddr = mhdl->memseg->vaddr;
4778 		minfo->raddr = mhdl->memseg->raddr;
4779 		minfo->mtype = mhdl->mtype;
4780 		minfo->perm = mhdl->perm;
4781 	}
4782 	mutex_exit(&mhdl->lock);
4783 
4784 	return (0);
4785 }
4786 
4787 /*
4788  * Copy data either from or to the client specified virtual address
4789  * space to or from the exported memory associated with the cookies.
4790  * The direction argument determines whether the data is read from or
4791  * written to exported memory.
4792  */
4793 int
4794 ldc_mem_copy(ldc_handle_t handle, caddr_t vaddr, uint64_t off, size_t *size,
4795     ldc_mem_cookie_t *cookies, uint32_t ccount, uint8_t direction)
4796 {
4797 	ldc_chan_t 	*ldcp;
4798 	uint64_t	local_voff, local_valign;
4799 	uint64_t	cookie_addr, cookie_size;
4800 	uint64_t	pg_shift, pg_size, pg_size_code;
4801 	uint64_t 	export_caddr, export_poff, export_psize, export_size;
4802 	uint64_t	local_ra, local_poff, local_psize;
4803 	uint64_t	copy_size, copied_len = 0, total_bal = 0, idx = 0;
4804 	pgcnt_t		npages;
4805 	size_t		len = *size;
4806 	int 		i, rv = 0;
4807 
4808 	uint64_t	chid;
4809 
4810 	if (handle == NULL) {
4811 		DWARN(DBG_ALL_LDCS, "ldc_mem_copy: invalid channel handle\n");
4812 		return (EINVAL);
4813 	}
4814 	ldcp = (ldc_chan_t *)handle;
4815 	chid = ldcp->id;
4816 
4817 	/* check to see if channel is UP */
4818 	if (ldcp->tstate != TS_UP) {
4819 		DWARN(chid, "ldc_mem_copy: (0x%llx) channel is not UP\n",
4820 		    chid);
4821 		return (ECONNRESET);
4822 	}
4823 
4824 	/* Force address and size to be 8-byte aligned */
4825 	if ((((uintptr_t)vaddr | len) & 0x7) != 0) {
4826 		DWARN(chid,
4827 		    "ldc_mem_copy: addr/sz is not 8-byte aligned\n");
4828 		return (EINVAL);
4829 	}
4830 
4831 	/* Find the size of the exported memory */
4832 	export_size = 0;
4833 	for (i = 0; i < ccount; i++)
4834 		export_size += cookies[i].size;
4835 
4836 	/* check to see if offset is valid */
4837 	if (off > export_size) {
4838 		DWARN(chid,
4839 		    "ldc_mem_copy: (0x%llx) start offset > export mem size\n",
4840 		    chid);
4841 		return (EINVAL);
4842 	}
4843 
4844 	/*
4845 	 * Check to see if the export size is smaller than the size we
4846 	 * are requesting to copy - if so flag an error
4847 	 */
4848 	if ((export_size - off) < *size) {
4849 		DWARN(chid,
4850 		    "ldc_mem_copy: (0x%llx) copy size > export mem size\n",
4851 		    chid);
4852 		return (EINVAL);
4853 	}
4854 
4855 	total_bal = min(export_size, *size);
4856 
4857 	/* FUTURE: get the page size, pgsz code, and shift */
4858 	pg_size = MMU_PAGESIZE;
4859 	pg_size_code = page_szc(pg_size);
4860 	pg_shift = page_get_shift(pg_size_code);
4861 
4862 	D1(chid, "ldc_mem_copy: copying data "
4863 	    "(0x%llx) va 0x%llx pgsz=0x%llx, pgszc=0x%llx, pg_shift=0x%llx\n",
4864 	    chid, vaddr, pg_size, pg_size_code, pg_shift);
4865 
4866 	/* aligned VA and its offset */
4867 	local_valign = (((uintptr_t)vaddr) & ~(pg_size - 1));
4868 	local_voff = ((uintptr_t)vaddr) & (pg_size - 1);
4869 
4870 	npages = (len+local_voff)/pg_size;
4871 	npages = ((len+local_voff)%pg_size == 0) ? npages : npages+1;
4872 
4873 	D1(chid,
4874 	    "ldc_mem_copy: (0x%llx) v=0x%llx,val=0x%llx,off=0x%x,pgs=0x%x\n",
4875 	    chid, vaddr, local_valign, local_voff, npages);
4876 
4877 	local_ra = va_to_pa((void *)local_valign);
4878 	local_poff = local_voff;
4879 	local_psize = min(len, (pg_size - local_voff));
4880 
4881 	len -= local_psize;
4882 
4883 	/*
4884 	 * find the first cookie in the list of cookies
4885 	 * if the offset passed in is not zero
4886 	 */
4887 	for (idx = 0; idx < ccount; idx++) {
4888 		cookie_size = cookies[idx].size;
4889 		if (off < cookie_size)
4890 			break;
4891 		off -= cookie_size;
4892 	}
4893 
4894 	cookie_addr = cookies[idx].addr + off;
4895 	cookie_size = cookies[idx].size - off;
4896 
4897 	export_caddr = cookie_addr & ~(pg_size - 1);
4898 	export_poff = cookie_addr & (pg_size - 1);
4899 	export_psize = min(cookie_size, (pg_size - export_poff));
4900 
4901 	for (;;) {
4902 
4903 		copy_size = min(export_psize, local_psize);
4904 
4905 		D1(chid,
4906 		    "ldc_mem_copy:(0x%llx) dir=0x%x, caddr=0x%llx,"
4907 		    " loc_ra=0x%llx, exp_poff=0x%llx, loc_poff=0x%llx,"
4908 		    " exp_psz=0x%llx, loc_psz=0x%llx, copy_sz=0x%llx,"
4909 		    " total_bal=0x%llx\n",
4910 		    chid, direction, export_caddr, local_ra, export_poff,
4911 		    local_poff, export_psize, local_psize, copy_size,
4912 		    total_bal);
4913 
4914 		rv = hv_ldc_copy(chid, direction,
4915 		    (export_caddr + export_poff), (local_ra + local_poff),
4916 		    copy_size, &copied_len);
4917 
4918 		if (rv != 0) {
4919 			int 		error = EIO;
4920 			uint64_t	rx_hd, rx_tl;
4921 
4922 			DWARN(chid,
4923 			    "ldc_mem_copy: (0x%llx) err %d during copy\n",
4924 			    (unsigned long long)chid, rv);
4925 			DWARN(chid,
4926 			    "ldc_mem_copy: (0x%llx) dir=0x%x, caddr=0x%lx, "
4927 			    "loc_ra=0x%lx, exp_poff=0x%lx, loc_poff=0x%lx,"
4928 			    " exp_psz=0x%lx, loc_psz=0x%lx, copy_sz=0x%lx,"
4929 			    " copied_len=0x%lx, total_bal=0x%lx\n",
4930 			    chid, direction, export_caddr, local_ra,
4931 			    export_poff, local_poff, export_psize, local_psize,
4932 			    copy_size, copied_len, total_bal);
4933 
4934 			*size = *size - total_bal;
4935 
4936 			/*
4937 			 * check if reason for copy error was due to
4938 			 * a channel reset. we need to grab the lock
4939 			 * just in case we have to do a reset.
4940 			 */
4941 			mutex_enter(&ldcp->lock);
4942 			mutex_enter(&ldcp->tx_lock);
4943 
4944 			rv = hv_ldc_rx_get_state(ldcp->id,
4945 			    &rx_hd, &rx_tl, &(ldcp->link_state));
4946 			if (ldcp->link_state == LDC_CHANNEL_DOWN ||
4947 			    ldcp->link_state == LDC_CHANNEL_RESET) {
4948 				i_ldc_reset(ldcp, B_FALSE);
4949 				error = ECONNRESET;
4950 			}
4951 
4952 			mutex_exit(&ldcp->tx_lock);
4953 			mutex_exit(&ldcp->lock);
4954 
4955 			return (error);
4956 		}
4957 
4958 		ASSERT(copied_len <= copy_size);
4959 
4960 		D2(chid, "ldc_mem_copy: copied=0x%llx\n", copied_len);
4961 		export_poff += copied_len;
4962 		local_poff += copied_len;
4963 		export_psize -= copied_len;
4964 		local_psize -= copied_len;
4965 		cookie_size -= copied_len;
4966 
4967 		total_bal -= copied_len;
4968 
4969 		if (copy_size != copied_len)
4970 			continue;
4971 
4972 		if (export_psize == 0 && total_bal != 0) {
4973 
4974 			if (cookie_size == 0) {
4975 				idx++;
4976 				cookie_addr = cookies[idx].addr;
4977 				cookie_size = cookies[idx].size;
4978 
4979 				export_caddr = cookie_addr & ~(pg_size - 1);
4980 				export_poff = cookie_addr & (pg_size - 1);
4981 				export_psize =
4982 					min(cookie_size, (pg_size-export_poff));
4983 			} else {
4984 				export_caddr += pg_size;
4985 				export_poff = 0;
4986 				export_psize = min(cookie_size, pg_size);
4987 			}
4988 		}
4989 
4990 		if (local_psize == 0 && total_bal != 0) {
4991 			local_valign += pg_size;
4992 			local_ra = va_to_pa((void *)local_valign);
4993 			local_poff = 0;
4994 			local_psize = min(pg_size, len);
4995 			len -= local_psize;
4996 		}
4997 
4998 		/* check if we are all done */
4999 		if (total_bal == 0)
5000 			break;
5001 	}
5002 
5003 
5004 	D1(chid,
5005 	    "ldc_mem_copy: (0x%llx) done copying sz=0x%llx\n",
5006 	    chid, *size);
5007 
5008 	return (0);
5009 }
5010 
5011 /*
5012  * Copy data either from or to the client specified virtual address
5013  * space to or from HV physical memory.
5014  *
5015  * The direction argument determines whether the data is read from or
5016  * written to HV memory. direction values are LDC_COPY_IN/OUT similar
5017  * to the ldc_mem_copy interface
5018  */
5019 int
5020 ldc_mem_rdwr_cookie(ldc_handle_t handle, caddr_t vaddr, size_t *size,
5021     caddr_t paddr, uint8_t direction)
5022 {
5023 	ldc_chan_t 	*ldcp;
5024 	uint64_t	local_voff, local_valign;
5025 	uint64_t	pg_shift, pg_size, pg_size_code;
5026 	uint64_t 	target_pa, target_poff, target_psize, target_size;
5027 	uint64_t	local_ra, local_poff, local_psize;
5028 	uint64_t	copy_size, copied_len = 0;
5029 	pgcnt_t		npages;
5030 	size_t		len = *size;
5031 	int 		rv = 0;
5032 
5033 	if (handle == NULL) {
5034 		DWARN(DBG_ALL_LDCS,
5035 		    "ldc_mem_rdwr_cookie: invalid channel handle\n");
5036 		return (EINVAL);
5037 	}
5038 	ldcp = (ldc_chan_t *)handle;
5039 
5040 	mutex_enter(&ldcp->lock);
5041 
5042 	/* check to see if channel is UP */
5043 	if (ldcp->tstate != TS_UP) {
5044 		DWARN(ldcp->id,
5045 		    "ldc_mem_rdwr_cookie: (0x%llx) channel is not UP\n",
5046 		    ldcp->id);
5047 		mutex_exit(&ldcp->lock);
5048 		return (ECONNRESET);
5049 	}
5050 
5051 	/* Force address and size to be 8-byte aligned */
5052 	if ((((uintptr_t)vaddr | len) & 0x7) != 0) {
5053 		DWARN(ldcp->id,
5054 		    "ldc_mem_rdwr_cookie: addr/size is not 8-byte aligned\n");
5055 		mutex_exit(&ldcp->lock);
5056 		return (EINVAL);
5057 	}
5058 
5059 	target_size = *size;
5060 
5061 	/* FUTURE: get the page size, pgsz code, and shift */
5062 	pg_size = MMU_PAGESIZE;
5063 	pg_size_code = page_szc(pg_size);
5064 	pg_shift = page_get_shift(pg_size_code);
5065 
5066 	D1(ldcp->id, "ldc_mem_rdwr_cookie: copying data "
5067 	    "(0x%llx) va 0x%llx pgsz=0x%llx, pgszc=0x%llx, pg_shift=0x%llx\n",
5068 	    ldcp->id, vaddr, pg_size, pg_size_code, pg_shift);
5069 
5070 	/* aligned VA and its offset */
5071 	local_valign = ((uintptr_t)vaddr) & ~(pg_size - 1);
5072 	local_voff = ((uintptr_t)vaddr) & (pg_size - 1);
5073 
5074 	npages = (len + local_voff) / pg_size;
5075 	npages = ((len + local_voff) % pg_size == 0) ? npages : npages+1;
5076 
5077 	D1(ldcp->id, "ldc_mem_rdwr_cookie: (0x%llx) v=0x%llx, "
5078 	    "val=0x%llx,off=0x%x,pgs=0x%x\n",
5079 	    ldcp->id, vaddr, local_valign, local_voff, npages);
5080 
5081 	local_ra = va_to_pa((void *)local_valign);
5082 	local_poff = local_voff;
5083 	local_psize = min(len, (pg_size - local_voff));
5084 
5085 	len -= local_psize;
5086 
5087 	target_pa = ((uintptr_t)paddr) & ~(pg_size - 1);
5088 	target_poff = ((uintptr_t)paddr) & (pg_size - 1);
5089 	target_psize = pg_size - target_poff;
5090 
5091 	for (;;) {
5092 
5093 		copy_size = min(target_psize, local_psize);
5094 
5095 		D1(ldcp->id,
5096 		    "ldc_mem_rdwr_cookie: (0x%llx) dir=0x%x, tar_pa=0x%llx,"
5097 		    " loc_ra=0x%llx, tar_poff=0x%llx, loc_poff=0x%llx,"
5098 		    " tar_psz=0x%llx, loc_psz=0x%llx, copy_sz=0x%llx,"
5099 		    " total_bal=0x%llx\n",
5100 		    ldcp->id, direction, target_pa, local_ra, target_poff,
5101 		    local_poff, target_psize, local_psize, copy_size,
5102 		    target_size);
5103 
5104 		rv = hv_ldc_copy(ldcp->id, direction,
5105 		    (target_pa + target_poff), (local_ra + local_poff),
5106 		    copy_size, &copied_len);
5107 
5108 		if (rv != 0) {
5109 			DWARN(DBG_ALL_LDCS,
5110 			    "ldc_mem_rdwr_cookie: (0x%lx) err %d during copy\n",
5111 			    ldcp->id, rv);
5112 			DWARN(DBG_ALL_LDCS,
5113 			    "ldc_mem_rdwr_cookie: (0x%llx) dir=%lld, "
5114 			    "tar_pa=0x%llx, loc_ra=0x%llx, tar_poff=0x%llx, "
5115 			    "loc_poff=0x%llx, tar_psz=0x%llx, loc_psz=0x%llx, "
5116 			    "copy_sz=0x%llx, total_bal=0x%llx\n",
5117 			    ldcp->id, direction, target_pa, local_ra,
5118 			    target_poff, local_poff, target_psize, local_psize,
5119 			    copy_size, target_size);
5120 
5121 			*size = *size - target_size;
5122 			mutex_exit(&ldcp->lock);
5123 			return (i_ldc_h2v_error(rv));
5124 		}
5125 
5126 		D2(ldcp->id, "ldc_mem_rdwr_cookie: copied=0x%llx\n",
5127 		    copied_len);
5128 		target_poff += copied_len;
5129 		local_poff += copied_len;
5130 		target_psize -= copied_len;
5131 		local_psize -= copied_len;
5132 
5133 		target_size -= copied_len;
5134 
5135 		if (copy_size != copied_len)
5136 			continue;
5137 
5138 		if (target_psize == 0 && target_size != 0) {
5139 			target_pa += pg_size;
5140 			target_poff = 0;
5141 			target_psize = min(pg_size, target_size);
5142 		}
5143 
5144 		if (local_psize == 0 && target_size != 0) {
5145 			local_valign += pg_size;
5146 			local_ra = va_to_pa((void *)local_valign);
5147 			local_poff = 0;
5148 			local_psize = min(pg_size, len);
5149 			len -= local_psize;
5150 		}
5151 
5152 		/* check if we are all done */
5153 		if (target_size == 0)
5154 			break;
5155 	}
5156 
5157 	mutex_exit(&ldcp->lock);
5158 
5159 	D1(ldcp->id, "ldc_mem_rdwr_cookie: (0x%llx) done copying sz=0x%llx\n",
5160 	    ldcp->id, *size);
5161 
5162 	return (0);
5163 }
5164 
5165 /*
5166  * Map an exported memory segment into the local address space. If the
5167  * memory range was exported for direct map access, a HV call is made
5168  * to allocate a RA range. If the map is done via a shadow copy, local
5169  * shadow memory is allocated and the base VA is returned in 'vaddr'. If
5170  * the mapping is a direct map then the RA is returned in 'raddr'.
5171  */
5172 int
5173 ldc_mem_map(ldc_mem_handle_t mhandle, ldc_mem_cookie_t *cookie, uint32_t ccount,
5174     uint8_t mtype, uint8_t perm, caddr_t *vaddr, caddr_t *raddr)
5175 {
5176 	int		i, j, idx, rv, retries;
5177 	ldc_chan_t 	*ldcp;
5178 	ldc_mhdl_t	*mhdl;
5179 	ldc_memseg_t	*memseg;
5180 	caddr_t		tmpaddr;
5181 	uint64_t	map_perm = perm;
5182 	uint64_t	pg_size, pg_shift, pg_size_code, pg_mask;
5183 	uint64_t	exp_size = 0, base_off, map_size, npages;
5184 	uint64_t	cookie_addr, cookie_off, cookie_size;
5185 	tte_t		ldc_tte;
5186 
5187 	if (mhandle == NULL) {
5188 		DWARN(DBG_ALL_LDCS, "ldc_mem_map: invalid memory handle\n");
5189 		return (EINVAL);
5190 	}
5191 	mhdl = (ldc_mhdl_t *)mhandle;
5192 
5193 	mutex_enter(&mhdl->lock);
5194 
5195 	if (mhdl->status == LDC_BOUND || mhdl->status == LDC_MAPPED ||
5196 	    mhdl->memseg != NULL) {
5197 		DWARN(DBG_ALL_LDCS,
5198 		    "ldc_mem_map: (0x%llx) handle bound/mapped\n", mhandle);
5199 		mutex_exit(&mhdl->lock);
5200 		return (EINVAL);
5201 	}
5202 
5203 	ldcp = mhdl->ldcp;
5204 
5205 	mutex_enter(&ldcp->lock);
5206 
5207 	if (ldcp->tstate != TS_UP) {
5208 		DWARN(ldcp->id,
5209 		    "ldc_mem_dring_map: (0x%llx) channel is not UP\n",
5210 		    ldcp->id);
5211 		mutex_exit(&ldcp->lock);
5212 		mutex_exit(&mhdl->lock);
5213 		return (ECONNRESET);
5214 	}
5215 
5216 	if ((mtype & (LDC_SHADOW_MAP|LDC_DIRECT_MAP|LDC_IO_MAP)) == 0) {
5217 		DWARN(ldcp->id, "ldc_mem_map: invalid map type\n");
5218 		mutex_exit(&ldcp->lock);
5219 		mutex_exit(&mhdl->lock);
5220 		return (EINVAL);
5221 	}
5222 
5223 	D1(ldcp->id, "ldc_mem_map: (0x%llx) cookie = 0x%llx,0x%llx\n",
5224 	    ldcp->id, cookie->addr, cookie->size);
5225 
5226 	/* FUTURE: get the page size, pgsz code, and shift */
5227 	pg_size = MMU_PAGESIZE;
5228 	pg_size_code = page_szc(pg_size);
5229 	pg_shift = page_get_shift(pg_size_code);
5230 	pg_mask = ~(pg_size - 1);
5231 
5232 	/* calculate the number of pages in the exported cookie */
5233 	base_off = cookie[0].addr & (pg_size - 1);
5234 	for (idx = 0; idx < ccount; idx++)
5235 		exp_size += cookie[idx].size;
5236 	map_size = P2ROUNDUP((exp_size + base_off), pg_size);
5237 	npages = (map_size >> pg_shift);
5238 
5239 	/* Allocate memseg structure */
5240 	memseg = mhdl->memseg =
5241 		kmem_cache_alloc(ldcssp->memseg_cache, KM_SLEEP);
5242 
5243 	/* Allocate memory to store all pages and cookies */
5244 	memseg->pages =	kmem_zalloc((sizeof (ldc_page_t) * npages), KM_SLEEP);
5245 	memseg->cookies =
5246 		kmem_zalloc((sizeof (ldc_mem_cookie_t) * ccount), KM_SLEEP);
5247 
5248 	D2(ldcp->id, "ldc_mem_map: (0x%llx) exp_size=0x%llx, map_size=0x%llx,"
5249 	    "pages=0x%llx\n", ldcp->id, exp_size, map_size, npages);
5250 
5251 	/*
5252 	 * Check if direct map over shared memory is enabled, if not change
5253 	 * the mapping type to SHADOW_MAP.
5254 	 */
5255 	if (ldc_shmem_enabled == 0)
5256 		mtype = LDC_SHADOW_MAP;
5257 
5258 	/*
5259 	 * Check to see if the client is requesting direct or shadow map
5260 	 * If direct map is requested, try to map remote memory first,
5261 	 * and if that fails, revert to shadow map
5262 	 */
5263 	if (mtype == LDC_DIRECT_MAP) {
5264 
5265 		/* Allocate kernel virtual space for mapping */
5266 		memseg->vaddr = vmem_xalloc(heap_arena, map_size,
5267 		    pg_size, 0, 0, NULL, NULL, VM_NOSLEEP);
5268 		if (memseg->vaddr == NULL) {
5269 			cmn_err(CE_WARN,
5270 			    "ldc_mem_map: (0x%lx) memory map failed\n",
5271 			    ldcp->id);
5272 			kmem_free(memseg->cookies,
5273 			    (sizeof (ldc_mem_cookie_t) * ccount));
5274 			kmem_free(memseg->pages,
5275 			    (sizeof (ldc_page_t) * npages));
5276 			kmem_cache_free(ldcssp->memseg_cache, memseg);
5277 
5278 			mutex_exit(&ldcp->lock);
5279 			mutex_exit(&mhdl->lock);
5280 			return (ENOMEM);
5281 		}
5282 
5283 		/* Unload previous mapping */
5284 		hat_unload(kas.a_hat, memseg->vaddr, map_size,
5285 		    HAT_UNLOAD_NOSYNC | HAT_UNLOAD_UNLOCK);
5286 
5287 		/* for each cookie passed in - map into address space */
5288 		idx = 0;
5289 		cookie_size = 0;
5290 		tmpaddr = memseg->vaddr;
5291 
5292 		for (i = 0; i < npages; i++) {
5293 
5294 			if (cookie_size == 0) {
5295 				ASSERT(idx < ccount);
5296 				cookie_addr = cookie[idx].addr & pg_mask;
5297 				cookie_off = cookie[idx].addr & (pg_size - 1);
5298 				cookie_size =
5299 				    P2ROUNDUP((cookie_off + cookie[idx].size),
5300 					pg_size);
5301 				idx++;
5302 			}
5303 
5304 			D1(ldcp->id, "ldc_mem_map: (0x%llx) mapping "
5305 			    "cookie 0x%llx, bal=0x%llx\n", ldcp->id,
5306 			    cookie_addr, cookie_size);
5307 
5308 			/* map the cookie into address space */
5309 			for (retries = 0; retries < ldc_max_retries;
5310 			    retries++) {
5311 
5312 				rv = hv_ldc_mapin(ldcp->id, cookie_addr,
5313 				    &memseg->pages[i].raddr, &map_perm);
5314 				if (rv != H_EWOULDBLOCK && rv != H_ETOOMANY)
5315 					break;
5316 
5317 				drv_usecwait(ldc_delay);
5318 			}
5319 
5320 			if (rv || memseg->pages[i].raddr == 0) {
5321 				DWARN(ldcp->id,
5322 				    "ldc_mem_map: (0x%llx) hv mapin err %d\n",
5323 				    ldcp->id, rv);
5324 
5325 				/* remove previous mapins */
5326 				hat_unload(kas.a_hat, memseg->vaddr, map_size,
5327 				    HAT_UNLOAD_NOSYNC | HAT_UNLOAD_UNLOCK);
5328 				for (j = 0; j < i; j++) {
5329 					rv = hv_ldc_unmap(
5330 							memseg->pages[j].raddr);
5331 					if (rv) {
5332 						DWARN(ldcp->id,
5333 						    "ldc_mem_map: (0x%llx) "
5334 						    "cannot unmap ra=0x%llx\n",
5335 					    ldcp->id,
5336 						    memseg->pages[j].raddr);
5337 					}
5338 				}
5339 
5340 				/* free kernel virtual space */
5341 				vmem_free(heap_arena, (void *)memseg->vaddr,
5342 				    map_size);
5343 
5344 				/* direct map failed - revert to shadow map */
5345 				mtype = LDC_SHADOW_MAP;
5346 				break;
5347 
5348 			} else {
5349 
5350 				D1(ldcp->id,
5351 				    "ldc_mem_map: (0x%llx) vtop map 0x%llx -> "
5352 				    "0x%llx, cookie=0x%llx, perm=0x%llx\n",
5353 				    ldcp->id, tmpaddr, memseg->pages[i].raddr,
5354 				    cookie_addr, perm);
5355 
5356 				/*
5357 				 * NOTE: Calling hat_devload directly, causes it
5358 				 * to look for page_t using the pfn. Since this
5359 				 * addr is greater than the memlist, it treates
5360 				 * it as non-memory
5361 				 */
5362 				sfmmu_memtte(&ldc_tte,
5363 				    (pfn_t)(memseg->pages[i].raddr >> pg_shift),
5364 				    PROT_READ | PROT_WRITE | HAT_NOSYNC, TTE8K);
5365 
5366 				D1(ldcp->id,
5367 				    "ldc_mem_map: (0x%llx) ra 0x%llx -> "
5368 				    "tte 0x%llx\n", ldcp->id,
5369 				    memseg->pages[i].raddr, ldc_tte);
5370 
5371 				sfmmu_tteload(kas.a_hat, &ldc_tte, tmpaddr,
5372 				    NULL, HAT_LOAD_LOCK);
5373 
5374 				cookie_size -= pg_size;
5375 				cookie_addr += pg_size;
5376 				tmpaddr += pg_size;
5377 			}
5378 		}
5379 	}
5380 
5381 	if (mtype == LDC_SHADOW_MAP) {
5382 		if (*vaddr == NULL) {
5383 			memseg->vaddr = kmem_zalloc(exp_size, KM_SLEEP);
5384 			mhdl->myshadow = B_TRUE;
5385 
5386 			D1(ldcp->id, "ldc_mem_map: (0x%llx) allocated "
5387 			    "shadow page va=0x%llx\n", ldcp->id, memseg->vaddr);
5388 		} else {
5389 			/*
5390 			 * Use client supplied memory for memseg->vaddr
5391 			 * WARNING: assuming that client mem is >= exp_size
5392 			 */
5393 			memseg->vaddr = *vaddr;
5394 		}
5395 
5396 		/* Save all page and cookie information */
5397 		for (i = 0, tmpaddr = memseg->vaddr; i < npages; i++) {
5398 			memseg->pages[i].raddr = va_to_pa(tmpaddr);
5399 			memseg->pages[i].size = pg_size;
5400 			tmpaddr += pg_size;
5401 		}
5402 
5403 	}
5404 
5405 	/* save all cookies */
5406 	bcopy(cookie, memseg->cookies, ccount * sizeof (ldc_mem_cookie_t));
5407 
5408 	/* update memseg_t */
5409 	memseg->raddr = memseg->pages[0].raddr;
5410 	memseg->size = (mtype == LDC_SHADOW_MAP) ? exp_size : map_size;
5411 	memseg->npages = npages;
5412 	memseg->ncookies = ccount;
5413 	memseg->next_cookie = 0;
5414 
5415 	/* memory handle = mapped */
5416 	mhdl->mtype = mtype;
5417 	mhdl->perm = perm;
5418 	mhdl->status = LDC_MAPPED;
5419 
5420 	D1(ldcp->id, "ldc_mem_map: (0x%llx) mapped 0x%llx, ra=0x%llx, "
5421 	    "va=0x%llx, pgs=0x%llx cookies=0x%llx\n",
5422 	    ldcp->id, mhdl, memseg->raddr, memseg->vaddr,
5423 	    memseg->npages, memseg->ncookies);
5424 
5425 	if (mtype == LDC_SHADOW_MAP)
5426 		base_off = 0;
5427 	if (raddr)
5428 		*raddr = (caddr_t)(memseg->raddr | base_off);
5429 	if (vaddr)
5430 		*vaddr = (caddr_t)((uintptr_t)memseg->vaddr | base_off);
5431 
5432 	mutex_exit(&ldcp->lock);
5433 	mutex_exit(&mhdl->lock);
5434 	return (0);
5435 }
5436 
5437 /*
5438  * Unmap a memory segment. Free shadow memory (if any).
5439  */
5440 int
5441 ldc_mem_unmap(ldc_mem_handle_t mhandle)
5442 {
5443 	int		i, rv;
5444 	ldc_mhdl_t	*mhdl = (ldc_mhdl_t *)mhandle;
5445 	ldc_chan_t 	*ldcp;
5446 	ldc_memseg_t	*memseg;
5447 
5448 	if (mhdl == 0 || mhdl->status != LDC_MAPPED) {
5449 		DWARN(DBG_ALL_LDCS,
5450 		    "ldc_mem_unmap: (0x%llx) handle is not mapped\n",
5451 		    mhandle);
5452 		return (EINVAL);
5453 	}
5454 
5455 	mutex_enter(&mhdl->lock);
5456 
5457 	ldcp = mhdl->ldcp;
5458 	memseg = mhdl->memseg;
5459 
5460 	D1(ldcp->id, "ldc_mem_unmap: (0x%llx) unmapping handle 0x%llx\n",
5461 	    ldcp->id, mhdl);
5462 
5463 	/* if we allocated shadow memory - free it */
5464 	if (mhdl->mtype == LDC_SHADOW_MAP && mhdl->myshadow) {
5465 		kmem_free(memseg->vaddr, memseg->size);
5466 	} else if (mhdl->mtype == LDC_DIRECT_MAP) {
5467 
5468 		/* unmap in the case of DIRECT_MAP */
5469 		hat_unload(kas.a_hat, memseg->vaddr, memseg->size,
5470 		    HAT_UNLOAD_UNLOCK);
5471 
5472 		for (i = 0; i < memseg->npages; i++) {
5473 			rv = hv_ldc_unmap(memseg->pages[i].raddr);
5474 			if (rv) {
5475 				cmn_err(CE_WARN,
5476 				    "ldc_mem_map: (0x%lx) hv unmap err %d\n",
5477 				    ldcp->id, rv);
5478 			}
5479 		}
5480 
5481 		vmem_free(heap_arena, (void *)memseg->vaddr, memseg->size);
5482 	}
5483 
5484 	/* free the allocated memseg and page structures */
5485 	kmem_free(memseg->pages, (sizeof (ldc_page_t) * memseg->npages));
5486 	kmem_free(memseg->cookies,
5487 	    (sizeof (ldc_mem_cookie_t) * memseg->ncookies));
5488 	kmem_cache_free(ldcssp->memseg_cache, memseg);
5489 
5490 	/* uninitialize the memory handle */
5491 	mhdl->memseg = NULL;
5492 	mhdl->status = LDC_UNBOUND;
5493 
5494 	D1(ldcp->id, "ldc_mem_unmap: (0x%llx) unmapped handle 0x%llx\n",
5495 	    ldcp->id, mhdl);
5496 
5497 	mutex_exit(&mhdl->lock);
5498 	return (0);
5499 }
5500 
5501 /*
5502  * Internal entry point for LDC mapped memory entry consistency
5503  * semantics. Acquire copies the contents of the remote memory
5504  * into the local shadow copy. The release operation copies the local
5505  * contents into the remote memory. The offset and size specify the
5506  * bounds for the memory range being synchronized.
5507  */
5508 static int
5509 i_ldc_mem_acquire_release(ldc_mem_handle_t mhandle, uint8_t direction,
5510     uint64_t offset, size_t size)
5511 {
5512 	int 		err;
5513 	ldc_mhdl_t	*mhdl;
5514 	ldc_chan_t	*ldcp;
5515 	ldc_memseg_t	*memseg;
5516 	caddr_t		local_vaddr;
5517 	size_t		copy_size;
5518 
5519 	if (mhandle == NULL) {
5520 		DWARN(DBG_ALL_LDCS,
5521 		    "i_ldc_mem_acquire_release: invalid memory handle\n");
5522 		return (EINVAL);
5523 	}
5524 	mhdl = (ldc_mhdl_t *)mhandle;
5525 
5526 	mutex_enter(&mhdl->lock);
5527 
5528 	if (mhdl->status != LDC_MAPPED || mhdl->ldcp == NULL) {
5529 		DWARN(DBG_ALL_LDCS,
5530 		    "i_ldc_mem_acquire_release: not mapped memory\n");
5531 		mutex_exit(&mhdl->lock);
5532 		return (EINVAL);
5533 	}
5534 
5535 	/* do nothing for direct map */
5536 	if (mhdl->mtype == LDC_DIRECT_MAP) {
5537 		mutex_exit(&mhdl->lock);
5538 		return (0);
5539 	}
5540 
5541 	/* do nothing if COPY_IN+MEM_W and COPY_OUT+MEM_R */
5542 	if ((direction == LDC_COPY_IN && (mhdl->perm & LDC_MEM_R) == 0) ||
5543 	    (direction == LDC_COPY_OUT && (mhdl->perm & LDC_MEM_W) == 0)) {
5544 		mutex_exit(&mhdl->lock);
5545 		return (0);
5546 	}
5547 
5548 	if (offset >= mhdl->memseg->size ||
5549 	    (offset + size) > mhdl->memseg->size) {
5550 		DWARN(DBG_ALL_LDCS,
5551 		    "i_ldc_mem_acquire_release: memory out of range\n");
5552 		mutex_exit(&mhdl->lock);
5553 		return (EINVAL);
5554 	}
5555 
5556 	/* get the channel handle and memory segment */
5557 	ldcp = mhdl->ldcp;
5558 	memseg = mhdl->memseg;
5559 
5560 	if (mhdl->mtype == LDC_SHADOW_MAP) {
5561 
5562 		local_vaddr = memseg->vaddr + offset;
5563 		copy_size = size;
5564 
5565 		/* copy to/from remote from/to local memory */
5566 		err = ldc_mem_copy((ldc_handle_t)ldcp, local_vaddr, offset,
5567 		    &copy_size, memseg->cookies, memseg->ncookies,
5568 		    direction);
5569 		if (err || copy_size != size) {
5570 			cmn_err(CE_WARN,
5571 			    "i_ldc_mem_acquire_release: copy failed\n");
5572 			mutex_exit(&mhdl->lock);
5573 			return (err);
5574 		}
5575 	}
5576 
5577 	mutex_exit(&mhdl->lock);
5578 
5579 	return (0);
5580 }
5581 
5582 /*
5583  * Ensure that the contents in the remote memory seg are consistent
5584  * with the contents if of local segment
5585  */
5586 int
5587 ldc_mem_acquire(ldc_mem_handle_t mhandle, uint64_t offset, uint64_t size)
5588 {
5589 	return (i_ldc_mem_acquire_release(mhandle, LDC_COPY_IN, offset, size));
5590 }
5591 
5592 
5593 /*
5594  * Ensure that the contents in the local memory seg are consistent
5595  * with the contents if of remote segment
5596  */
5597 int
5598 ldc_mem_release(ldc_mem_handle_t mhandle, uint64_t offset, uint64_t size)
5599 {
5600 	return (i_ldc_mem_acquire_release(mhandle, LDC_COPY_OUT, offset, size));
5601 }
5602 
5603 /*
5604  * Allocate a descriptor ring. The size of each each descriptor
5605  * must be 8-byte aligned and the entire ring should be a multiple
5606  * of MMU_PAGESIZE.
5607  */
5608 int
5609 ldc_mem_dring_create(uint32_t len, uint32_t dsize, ldc_dring_handle_t *dhandle)
5610 {
5611 	ldc_dring_t *dringp;
5612 	size_t size = (dsize * len);
5613 
5614 	D1(DBG_ALL_LDCS, "ldc_mem_dring_create: len=0x%x, size=0x%x\n",
5615 	    len, dsize);
5616 
5617 	if (dhandle == NULL) {
5618 		DWARN(DBG_ALL_LDCS, "ldc_mem_dring_create: invalid dhandle\n");
5619 		return (EINVAL);
5620 	}
5621 
5622 	if (len == 0) {
5623 		DWARN(DBG_ALL_LDCS, "ldc_mem_dring_create: invalid length\n");
5624 		return (EINVAL);
5625 	}
5626 
5627 	/* descriptor size should be 8-byte aligned */
5628 	if (dsize == 0 || (dsize & 0x7)) {
5629 		DWARN(DBG_ALL_LDCS, "ldc_mem_dring_create: invalid size\n");
5630 		return (EINVAL);
5631 	}
5632 
5633 	*dhandle = 0;
5634 
5635 	/* Allocate a desc ring structure */
5636 	dringp = kmem_zalloc(sizeof (ldc_dring_t), KM_SLEEP);
5637 
5638 	/* Initialize dring */
5639 	dringp->length = len;
5640 	dringp->dsize = dsize;
5641 
5642 	/* round off to multiple of pagesize */
5643 	dringp->size = (size & MMU_PAGEMASK);
5644 	if (size & MMU_PAGEOFFSET)
5645 		dringp->size += MMU_PAGESIZE;
5646 
5647 	dringp->status = LDC_UNBOUND;
5648 
5649 	/* allocate descriptor ring memory */
5650 	dringp->base = kmem_zalloc(dringp->size, KM_SLEEP);
5651 
5652 	/* initialize the desc ring lock */
5653 	mutex_init(&dringp->lock, NULL, MUTEX_DRIVER, NULL);
5654 
5655 	/* Add descriptor ring to the head of global list */
5656 	mutex_enter(&ldcssp->lock);
5657 	dringp->next = ldcssp->dring_list;
5658 	ldcssp->dring_list = dringp;
5659 	mutex_exit(&ldcssp->lock);
5660 
5661 	*dhandle = (ldc_dring_handle_t)dringp;
5662 
5663 	D1(DBG_ALL_LDCS, "ldc_mem_dring_create: dring allocated\n");
5664 
5665 	return (0);
5666 }
5667 
5668 
5669 /*
5670  * Destroy a descriptor ring.
5671  */
5672 int
5673 ldc_mem_dring_destroy(ldc_dring_handle_t dhandle)
5674 {
5675 	ldc_dring_t *dringp;
5676 	ldc_dring_t *tmp_dringp;
5677 
5678 	D1(DBG_ALL_LDCS, "ldc_mem_dring_destroy: entered\n");
5679 
5680 	if (dhandle == NULL) {
5681 		DWARN(DBG_ALL_LDCS,
5682 		    "ldc_mem_dring_destroy: invalid desc ring handle\n");
5683 		return (EINVAL);
5684 	}
5685 	dringp = (ldc_dring_t *)dhandle;
5686 
5687 	if (dringp->status == LDC_BOUND) {
5688 		DWARN(DBG_ALL_LDCS,
5689 		    "ldc_mem_dring_destroy: desc ring is bound\n");
5690 		return (EACCES);
5691 	}
5692 
5693 	mutex_enter(&dringp->lock);
5694 	mutex_enter(&ldcssp->lock);
5695 
5696 	/* remove from linked list - if not bound */
5697 	tmp_dringp = ldcssp->dring_list;
5698 	if (tmp_dringp == dringp) {
5699 		ldcssp->dring_list = dringp->next;
5700 		dringp->next = NULL;
5701 
5702 	} else {
5703 		while (tmp_dringp != NULL) {
5704 			if (tmp_dringp->next == dringp) {
5705 				tmp_dringp->next = dringp->next;
5706 				dringp->next = NULL;
5707 				break;
5708 			}
5709 			tmp_dringp = tmp_dringp->next;
5710 		}
5711 		if (tmp_dringp == NULL) {
5712 			DWARN(DBG_ALL_LDCS,
5713 			    "ldc_mem_dring_destroy: invalid descriptor\n");
5714 			mutex_exit(&ldcssp->lock);
5715 			mutex_exit(&dringp->lock);
5716 			return (EINVAL);
5717 		}
5718 	}
5719 
5720 	mutex_exit(&ldcssp->lock);
5721 
5722 	/* free the descriptor ring */
5723 	kmem_free(dringp->base, dringp->size);
5724 
5725 	mutex_exit(&dringp->lock);
5726 
5727 	/* destroy dring lock */
5728 	mutex_destroy(&dringp->lock);
5729 
5730 	/* free desc ring object */
5731 	kmem_free(dringp, sizeof (ldc_dring_t));
5732 
5733 	return (0);
5734 }
5735 
5736 /*
5737  * Bind a previously allocated dring to a channel. The channel should
5738  * be OPEN in order to bind the ring to the channel. Returns back a
5739  * descriptor ring cookie. The descriptor ring is exported for remote
5740  * access by the client at the other end of the channel. An entry for
5741  * dring pages is stored in map table (via call to ldc_mem_bind_handle).
5742  */
5743 int
5744 ldc_mem_dring_bind(ldc_handle_t handle, ldc_dring_handle_t dhandle,
5745     uint8_t mtype, uint8_t perm, ldc_mem_cookie_t *cookie, uint32_t *ccount)
5746 {
5747 	int		err;
5748 	ldc_chan_t 	*ldcp;
5749 	ldc_dring_t	*dringp;
5750 	ldc_mem_handle_t mhandle;
5751 
5752 	/* check to see if channel is initalized */
5753 	if (handle == NULL) {
5754 		DWARN(DBG_ALL_LDCS,
5755 		    "ldc_mem_dring_bind: invalid channel handle\n");
5756 		return (EINVAL);
5757 	}
5758 	ldcp = (ldc_chan_t *)handle;
5759 
5760 	if (dhandle == NULL) {
5761 		DWARN(DBG_ALL_LDCS,
5762 		    "ldc_mem_dring_bind: invalid desc ring handle\n");
5763 		return (EINVAL);
5764 	}
5765 	dringp = (ldc_dring_t *)dhandle;
5766 
5767 	if (cookie == NULL) {
5768 		DWARN(ldcp->id,
5769 		    "ldc_mem_dring_bind: invalid cookie arg\n");
5770 		return (EINVAL);
5771 	}
5772 
5773 	mutex_enter(&dringp->lock);
5774 
5775 	if (dringp->status == LDC_BOUND) {
5776 		DWARN(DBG_ALL_LDCS,
5777 		    "ldc_mem_dring_bind: (0x%llx) descriptor ring is bound\n",
5778 		    ldcp->id);
5779 		mutex_exit(&dringp->lock);
5780 		return (EINVAL);
5781 	}
5782 
5783 	if ((perm & LDC_MEM_RW) == 0) {
5784 		DWARN(DBG_ALL_LDCS,
5785 		    "ldc_mem_dring_bind: invalid permissions\n");
5786 		mutex_exit(&dringp->lock);
5787 		return (EINVAL);
5788 	}
5789 
5790 	if ((mtype & (LDC_SHADOW_MAP|LDC_DIRECT_MAP|LDC_IO_MAP)) == 0) {
5791 		DWARN(DBG_ALL_LDCS, "ldc_mem_dring_bind: invalid type\n");
5792 		mutex_exit(&dringp->lock);
5793 		return (EINVAL);
5794 	}
5795 
5796 	dringp->ldcp = ldcp;
5797 
5798 	/* create an memory handle */
5799 	err = ldc_mem_alloc_handle(handle, &mhandle);
5800 	if (err || mhandle == NULL) {
5801 		DWARN(DBG_ALL_LDCS,
5802 		    "ldc_mem_dring_bind: (0x%llx) error allocating mhandle\n",
5803 		    ldcp->id);
5804 		mutex_exit(&dringp->lock);
5805 		return (err);
5806 	}
5807 	dringp->mhdl = mhandle;
5808 
5809 	/* bind the descriptor ring to channel */
5810 	err = ldc_mem_bind_handle(mhandle, dringp->base, dringp->size,
5811 	    mtype, perm, cookie, ccount);
5812 	if (err) {
5813 		DWARN(ldcp->id,
5814 		    "ldc_mem_dring_bind: (0x%llx) error binding mhandle\n",
5815 		    ldcp->id);
5816 		mutex_exit(&dringp->lock);
5817 		return (err);
5818 	}
5819 
5820 	/*
5821 	 * For now return error if we get more than one cookie
5822 	 * FUTURE: Return multiple cookies ..
5823 	 */
5824 	if (*ccount > 1) {
5825 		(void) ldc_mem_unbind_handle(mhandle);
5826 		(void) ldc_mem_free_handle(mhandle);
5827 
5828 		dringp->ldcp = NULL;
5829 		dringp->mhdl = NULL;
5830 		*ccount = 0;
5831 
5832 		mutex_exit(&dringp->lock);
5833 		return (EAGAIN);
5834 	}
5835 
5836 	/* Add descriptor ring to channel's exported dring list */
5837 	mutex_enter(&ldcp->exp_dlist_lock);
5838 	dringp->ch_next = ldcp->exp_dring_list;
5839 	ldcp->exp_dring_list = dringp;
5840 	mutex_exit(&ldcp->exp_dlist_lock);
5841 
5842 	dringp->status = LDC_BOUND;
5843 
5844 	mutex_exit(&dringp->lock);
5845 
5846 	return (0);
5847 }
5848 
5849 /*
5850  * Return the next cookie associated with the specified dring handle
5851  */
5852 int
5853 ldc_mem_dring_nextcookie(ldc_dring_handle_t dhandle, ldc_mem_cookie_t *cookie)
5854 {
5855 	int		rv = 0;
5856 	ldc_dring_t 	*dringp;
5857 	ldc_chan_t	*ldcp;
5858 
5859 	if (dhandle == NULL) {
5860 		DWARN(DBG_ALL_LDCS,
5861 		    "ldc_mem_dring_nextcookie: invalid desc ring handle\n");
5862 		return (EINVAL);
5863 	}
5864 	dringp = (ldc_dring_t *)dhandle;
5865 	mutex_enter(&dringp->lock);
5866 
5867 	if (dringp->status != LDC_BOUND) {
5868 		DWARN(DBG_ALL_LDCS,
5869 		    "ldc_mem_dring_nextcookie: descriptor ring 0x%llx "
5870 		    "is not bound\n", dringp);
5871 		mutex_exit(&dringp->lock);
5872 		return (EINVAL);
5873 	}
5874 
5875 	ldcp = dringp->ldcp;
5876 
5877 	if (cookie == NULL) {
5878 		DWARN(ldcp->id,
5879 		    "ldc_mem_dring_nextcookie:(0x%llx) invalid cookie arg\n",
5880 		    ldcp->id);
5881 		mutex_exit(&dringp->lock);
5882 		return (EINVAL);
5883 	}
5884 
5885 	rv = ldc_mem_nextcookie((ldc_mem_handle_t)dringp->mhdl, cookie);
5886 	mutex_exit(&dringp->lock);
5887 
5888 	return (rv);
5889 }
5890 /*
5891  * Unbind a previously bound dring from a channel.
5892  */
5893 int
5894 ldc_mem_dring_unbind(ldc_dring_handle_t dhandle)
5895 {
5896 	ldc_dring_t 	*dringp;
5897 	ldc_dring_t	*tmp_dringp;
5898 	ldc_chan_t	*ldcp;
5899 
5900 	if (dhandle == NULL) {
5901 		DWARN(DBG_ALL_LDCS,
5902 		    "ldc_mem_dring_unbind: invalid desc ring handle\n");
5903 		return (EINVAL);
5904 	}
5905 	dringp = (ldc_dring_t *)dhandle;
5906 
5907 	mutex_enter(&dringp->lock);
5908 
5909 	if (dringp->status == LDC_UNBOUND) {
5910 		DWARN(DBG_ALL_LDCS,
5911 		    "ldc_mem_dring_bind: descriptor ring 0x%llx is unbound\n",
5912 		    dringp);
5913 		mutex_exit(&dringp->lock);
5914 		return (EINVAL);
5915 	}
5916 	ldcp = dringp->ldcp;
5917 
5918 	mutex_enter(&ldcp->exp_dlist_lock);
5919 
5920 	tmp_dringp = ldcp->exp_dring_list;
5921 	if (tmp_dringp == dringp) {
5922 		ldcp->exp_dring_list = dringp->ch_next;
5923 		dringp->ch_next = NULL;
5924 
5925 	} else {
5926 		while (tmp_dringp != NULL) {
5927 			if (tmp_dringp->ch_next == dringp) {
5928 				tmp_dringp->ch_next = dringp->ch_next;
5929 				dringp->ch_next = NULL;
5930 				break;
5931 			}
5932 			tmp_dringp = tmp_dringp->ch_next;
5933 		}
5934 		if (tmp_dringp == NULL) {
5935 			DWARN(DBG_ALL_LDCS,
5936 			    "ldc_mem_dring_unbind: invalid descriptor\n");
5937 			mutex_exit(&ldcp->exp_dlist_lock);
5938 			mutex_exit(&dringp->lock);
5939 			return (EINVAL);
5940 		}
5941 	}
5942 
5943 	mutex_exit(&ldcp->exp_dlist_lock);
5944 
5945 	(void) ldc_mem_unbind_handle((ldc_mem_handle_t)dringp->mhdl);
5946 	(void) ldc_mem_free_handle((ldc_mem_handle_t)dringp->mhdl);
5947 
5948 	dringp->ldcp = NULL;
5949 	dringp->mhdl = NULL;
5950 	dringp->status = LDC_UNBOUND;
5951 
5952 	mutex_exit(&dringp->lock);
5953 
5954 	return (0);
5955 }
5956 
5957 /*
5958  * Get information about the dring. The base address of the descriptor
5959  * ring along with the type and permission are returned back.
5960  */
5961 int
5962 ldc_mem_dring_info(ldc_dring_handle_t dhandle, ldc_mem_info_t *minfo)
5963 {
5964 	ldc_dring_t	*dringp;
5965 	int		rv;
5966 
5967 	if (dhandle == NULL) {
5968 		DWARN(DBG_ALL_LDCS,
5969 		    "ldc_mem_dring_info: invalid desc ring handle\n");
5970 		return (EINVAL);
5971 	}
5972 	dringp = (ldc_dring_t *)dhandle;
5973 
5974 	mutex_enter(&dringp->lock);
5975 
5976 	if (dringp->mhdl) {
5977 		rv = ldc_mem_info(dringp->mhdl, minfo);
5978 		if (rv) {
5979 			DWARN(DBG_ALL_LDCS,
5980 			    "ldc_mem_dring_info: error reading mem info\n");
5981 			mutex_exit(&dringp->lock);
5982 			return (rv);
5983 		}
5984 	} else {
5985 		minfo->vaddr = dringp->base;
5986 		minfo->raddr = NULL;
5987 		minfo->status = dringp->status;
5988 	}
5989 
5990 	mutex_exit(&dringp->lock);
5991 
5992 	return (0);
5993 }
5994 
5995 /*
5996  * Map an exported descriptor ring into the local address space. If the
5997  * descriptor ring was exported for direct map access, a HV call is made
5998  * to allocate a RA range. If the map is done via a shadow copy, local
5999  * shadow memory is allocated.
6000  */
6001 int
6002 ldc_mem_dring_map(ldc_handle_t handle, ldc_mem_cookie_t *cookie,
6003     uint32_t ccount, uint32_t len, uint32_t dsize, uint8_t mtype,
6004     ldc_dring_handle_t *dhandle)
6005 {
6006 	int		err;
6007 	ldc_chan_t 	*ldcp = (ldc_chan_t *)handle;
6008 	ldc_mem_handle_t mhandle;
6009 	ldc_dring_t	*dringp;
6010 	size_t		dring_size;
6011 
6012 	if (dhandle == NULL) {
6013 		DWARN(DBG_ALL_LDCS,
6014 		    "ldc_mem_dring_map: invalid dhandle\n");
6015 		return (EINVAL);
6016 	}
6017 
6018 	/* check to see if channel is initalized */
6019 	if (handle == NULL) {
6020 		DWARN(DBG_ALL_LDCS,
6021 		    "ldc_mem_dring_map: invalid channel handle\n");
6022 		return (EINVAL);
6023 	}
6024 	ldcp = (ldc_chan_t *)handle;
6025 
6026 	if (cookie == NULL) {
6027 		DWARN(ldcp->id,
6028 		    "ldc_mem_dring_map: (0x%llx) invalid cookie\n",
6029 		    ldcp->id);
6030 		return (EINVAL);
6031 	}
6032 
6033 	/* FUTURE: For now we support only one cookie per dring */
6034 	ASSERT(ccount == 1);
6035 
6036 	if (cookie->size < (dsize * len)) {
6037 		DWARN(ldcp->id,
6038 		    "ldc_mem_dring_map: (0x%llx) invalid dsize/len\n",
6039 		    ldcp->id);
6040 		return (EINVAL);
6041 	}
6042 
6043 	*dhandle = 0;
6044 
6045 	/* Allocate an dring structure */
6046 	dringp = kmem_zalloc(sizeof (ldc_dring_t), KM_SLEEP);
6047 
6048 	D1(ldcp->id,
6049 	    "ldc_mem_dring_map: 0x%x,0x%x,0x%x,0x%llx,0x%llx\n",
6050 	    mtype, len, dsize, cookie->addr, cookie->size);
6051 
6052 	/* Initialize dring */
6053 	dringp->length = len;
6054 	dringp->dsize = dsize;
6055 
6056 	/* round of to multiple of page size */
6057 	dring_size = len * dsize;
6058 	dringp->size = (dring_size & MMU_PAGEMASK);
6059 	if (dring_size & MMU_PAGEOFFSET)
6060 		dringp->size += MMU_PAGESIZE;
6061 
6062 	dringp->ldcp = ldcp;
6063 
6064 	/* create an memory handle */
6065 	err = ldc_mem_alloc_handle(handle, &mhandle);
6066 	if (err || mhandle == NULL) {
6067 		DWARN(DBG_ALL_LDCS,
6068 		    "ldc_mem_dring_map: cannot alloc hdl err=%d\n",
6069 		    err);
6070 		kmem_free(dringp, sizeof (ldc_dring_t));
6071 		return (ENOMEM);
6072 	}
6073 
6074 	dringp->mhdl = mhandle;
6075 	dringp->base = NULL;
6076 
6077 	/* map the dring into local memory */
6078 	err = ldc_mem_map(mhandle, cookie, ccount, mtype, LDC_MEM_RW,
6079 	    &(dringp->base), NULL);
6080 	if (err || dringp->base == NULL) {
6081 		cmn_err(CE_WARN,
6082 		    "ldc_mem_dring_map: cannot map desc ring err=%d\n", err);
6083 		(void) ldc_mem_free_handle(mhandle);
6084 		kmem_free(dringp, sizeof (ldc_dring_t));
6085 		return (ENOMEM);
6086 	}
6087 
6088 	/* initialize the desc ring lock */
6089 	mutex_init(&dringp->lock, NULL, MUTEX_DRIVER, NULL);
6090 
6091 	/* Add descriptor ring to channel's imported dring list */
6092 	mutex_enter(&ldcp->imp_dlist_lock);
6093 	dringp->ch_next = ldcp->imp_dring_list;
6094 	ldcp->imp_dring_list = dringp;
6095 	mutex_exit(&ldcp->imp_dlist_lock);
6096 
6097 	dringp->status = LDC_MAPPED;
6098 
6099 	*dhandle = (ldc_dring_handle_t)dringp;
6100 
6101 	return (0);
6102 }
6103 
6104 /*
6105  * Unmap a descriptor ring. Free shadow memory (if any).
6106  */
6107 int
6108 ldc_mem_dring_unmap(ldc_dring_handle_t dhandle)
6109 {
6110 	ldc_dring_t 	*dringp;
6111 	ldc_dring_t	*tmp_dringp;
6112 	ldc_chan_t	*ldcp;
6113 
6114 	if (dhandle == NULL) {
6115 		DWARN(DBG_ALL_LDCS,
6116 		    "ldc_mem_dring_unmap: invalid desc ring handle\n");
6117 		return (EINVAL);
6118 	}
6119 	dringp = (ldc_dring_t *)dhandle;
6120 
6121 	if (dringp->status != LDC_MAPPED) {
6122 		DWARN(DBG_ALL_LDCS,
6123 		    "ldc_mem_dring_unmap: not a mapped desc ring\n");
6124 		return (EINVAL);
6125 	}
6126 
6127 	mutex_enter(&dringp->lock);
6128 
6129 	ldcp = dringp->ldcp;
6130 
6131 	mutex_enter(&ldcp->imp_dlist_lock);
6132 
6133 	/* find and unlink the desc ring from channel import list */
6134 	tmp_dringp = ldcp->imp_dring_list;
6135 	if (tmp_dringp == dringp) {
6136 		ldcp->imp_dring_list = dringp->ch_next;
6137 		dringp->ch_next = NULL;
6138 
6139 	} else {
6140 		while (tmp_dringp != NULL) {
6141 			if (tmp_dringp->ch_next == dringp) {
6142 				tmp_dringp->ch_next = dringp->ch_next;
6143 				dringp->ch_next = NULL;
6144 				break;
6145 			}
6146 			tmp_dringp = tmp_dringp->ch_next;
6147 		}
6148 		if (tmp_dringp == NULL) {
6149 			DWARN(DBG_ALL_LDCS,
6150 			    "ldc_mem_dring_unmap: invalid descriptor\n");
6151 			mutex_exit(&ldcp->imp_dlist_lock);
6152 			mutex_exit(&dringp->lock);
6153 			return (EINVAL);
6154 		}
6155 	}
6156 
6157 	mutex_exit(&ldcp->imp_dlist_lock);
6158 
6159 	/* do a LDC memory handle unmap and free */
6160 	(void) ldc_mem_unmap(dringp->mhdl);
6161 	(void) ldc_mem_free_handle((ldc_mem_handle_t)dringp->mhdl);
6162 
6163 	dringp->status = 0;
6164 	dringp->ldcp = NULL;
6165 
6166 	mutex_exit(&dringp->lock);
6167 
6168 	/* destroy dring lock */
6169 	mutex_destroy(&dringp->lock);
6170 
6171 	/* free desc ring object */
6172 	kmem_free(dringp, sizeof (ldc_dring_t));
6173 
6174 	return (0);
6175 }
6176 
6177 /*
6178  * Internal entry point for descriptor ring access entry consistency
6179  * semantics. Acquire copies the contents of the remote descriptor ring
6180  * into the local shadow copy. The release operation copies the local
6181  * contents into the remote dring. The start and end locations specify
6182  * bounds for the entries being synchronized.
6183  */
6184 static int
6185 i_ldc_dring_acquire_release(ldc_dring_handle_t dhandle,
6186     uint8_t direction, uint64_t start, uint64_t end)
6187 {
6188 	int 			err;
6189 	ldc_dring_t		*dringp;
6190 	ldc_chan_t		*ldcp;
6191 	uint64_t		soff;
6192 	size_t			copy_size;
6193 
6194 	if (dhandle == NULL) {
6195 		DWARN(DBG_ALL_LDCS,
6196 		    "i_ldc_dring_acquire_release: invalid desc ring handle\n");
6197 		return (EINVAL);
6198 	}
6199 	dringp = (ldc_dring_t *)dhandle;
6200 	mutex_enter(&dringp->lock);
6201 
6202 	if (dringp->status != LDC_MAPPED || dringp->ldcp == NULL) {
6203 		DWARN(DBG_ALL_LDCS,
6204 		    "i_ldc_dring_acquire_release: not a mapped desc ring\n");
6205 		mutex_exit(&dringp->lock);
6206 		return (EINVAL);
6207 	}
6208 
6209 	if (start >= dringp->length || end >= dringp->length) {
6210 		DWARN(DBG_ALL_LDCS,
6211 		    "i_ldc_dring_acquire_release: index out of range\n");
6212 		mutex_exit(&dringp->lock);
6213 		return (EINVAL);
6214 	}
6215 
6216 	/* get the channel handle */
6217 	ldcp = dringp->ldcp;
6218 
6219 	copy_size = (start <= end) ? (((end - start) + 1) * dringp->dsize) :
6220 		((dringp->length - start) * dringp->dsize);
6221 
6222 	/* Calculate the relative offset for the first desc */
6223 	soff = (start * dringp->dsize);
6224 
6225 	/* copy to/from remote from/to local memory */
6226 	D1(ldcp->id, "i_ldc_dring_acquire_release: c1 off=0x%llx sz=0x%llx\n",
6227 	    soff, copy_size);
6228 	err = i_ldc_mem_acquire_release((ldc_mem_handle_t)dringp->mhdl,
6229 	    direction, soff, copy_size);
6230 	if (err) {
6231 		DWARN(ldcp->id,
6232 		    "i_ldc_dring_acquire_release: copy failed\n");
6233 		mutex_exit(&dringp->lock);
6234 		return (err);
6235 	}
6236 
6237 	/* do the balance */
6238 	if (start > end) {
6239 		copy_size = ((end + 1) * dringp->dsize);
6240 		soff = 0;
6241 
6242 		/* copy to/from remote from/to local memory */
6243 		D1(ldcp->id, "i_ldc_dring_acquire_release: c2 "
6244 		    "off=0x%llx sz=0x%llx\n", soff, copy_size);
6245 		err = i_ldc_mem_acquire_release((ldc_mem_handle_t)dringp->mhdl,
6246 		    direction, soff, copy_size);
6247 		if (err) {
6248 			DWARN(ldcp->id,
6249 			    "i_ldc_dring_acquire_release: copy failed\n");
6250 			mutex_exit(&dringp->lock);
6251 			return (err);
6252 		}
6253 	}
6254 
6255 	mutex_exit(&dringp->lock);
6256 
6257 	return (0);
6258 }
6259 
6260 /*
6261  * Ensure that the contents in the local dring are consistent
6262  * with the contents if of remote dring
6263  */
6264 int
6265 ldc_mem_dring_acquire(ldc_dring_handle_t dhandle, uint64_t start, uint64_t end)
6266 {
6267 	return (i_ldc_dring_acquire_release(dhandle, LDC_COPY_IN, start, end));
6268 }
6269 
6270 /*
6271  * Ensure that the contents in the remote dring are consistent
6272  * with the contents if of local dring
6273  */
6274 int
6275 ldc_mem_dring_release(ldc_dring_handle_t dhandle, uint64_t start, uint64_t end)
6276 {
6277 	return (i_ldc_dring_acquire_release(dhandle, LDC_COPY_OUT, start, end));
6278 }
6279 
6280 
6281 /* ------------------------------------------------------------------------- */
6282