xref: /freebsd/sys/dev/hyperv/vmbus/vmbus_chan.c (revision 273c26a3c3bea87a241d6879abd4f991db180bf0)
1 /*-
2  * Copyright (c) 2009-2012,2016 Microsoft Corp.
3  * Copyright (c) 2012 NetApp Inc.
4  * Copyright (c) 2012 Citrix Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include <sys/param.h>
33 #include <sys/kernel.h>
34 #include <sys/lock.h>
35 #include <sys/malloc.h>
36 #include <sys/mutex.h>
37 #include <sys/smp.h>
38 #include <sys/sysctl.h>
39 #include <sys/systm.h>
40 
41 #include <machine/atomic.h>
42 
43 #include <dev/hyperv/include/hyperv_busdma.h>
44 #include <dev/hyperv/vmbus/hyperv_var.h>
45 #include <dev/hyperv/vmbus/vmbus_reg.h>
46 #include <dev/hyperv/vmbus/vmbus_var.h>
47 #include <dev/hyperv/vmbus/vmbus_brvar.h>
48 #include <dev/hyperv/vmbus/vmbus_chanvar.h>
49 
50 static void			vmbus_chan_update_evtflagcnt(
51 				    struct vmbus_softc *,
52 				    const struct vmbus_channel *);
53 static void			vmbus_chan_close_internal(
54 				    struct vmbus_channel *);
55 static int			vmbus_chan_sysctl_mnf(SYSCTL_HANDLER_ARGS);
56 static void			vmbus_chan_sysctl_create(
57 				    struct vmbus_channel *);
58 static struct vmbus_channel	*vmbus_chan_alloc(struct vmbus_softc *);
59 static void			vmbus_chan_free(struct vmbus_channel *);
60 static int			vmbus_chan_add(struct vmbus_channel *);
61 static void			vmbus_chan_cpu_default(struct vmbus_channel *);
62 static int			vmbus_chan_release(struct vmbus_channel *);
63 static void			vmbus_chan_set_chmap(struct vmbus_channel *);
64 static void			vmbus_chan_clear_chmap(struct vmbus_channel *);
65 
66 static void			vmbus_chan_ins_prilist(struct vmbus_softc *,
67 				    struct vmbus_channel *);
68 static void			vmbus_chan_rem_prilist(struct vmbus_softc *,
69 				    struct vmbus_channel *);
70 static void			vmbus_chan_ins_list(struct vmbus_softc *,
71 				    struct vmbus_channel *);
72 static void			vmbus_chan_rem_list(struct vmbus_softc *,
73 				    struct vmbus_channel *);
74 static void			vmbus_chan_ins_sublist(struct vmbus_channel *,
75 				    struct vmbus_channel *);
76 static void			vmbus_chan_rem_sublist(struct vmbus_channel *,
77 				    struct vmbus_channel *);
78 
79 static void			vmbus_chan_task(void *, int);
80 static void			vmbus_chan_task_nobatch(void *, int);
81 static void			vmbus_chan_clrchmap_task(void *, int);
82 static void			vmbus_prichan_attach_task(void *, int);
83 static void			vmbus_subchan_attach_task(void *, int);
84 static void			vmbus_prichan_detach_task(void *, int);
85 static void			vmbus_subchan_detach_task(void *, int);
86 
87 static void			vmbus_chan_msgproc_choffer(struct vmbus_softc *,
88 				    const struct vmbus_message *);
89 static void			vmbus_chan_msgproc_chrescind(
90 				    struct vmbus_softc *,
91 				    const struct vmbus_message *);
92 
93 /*
94  * Vmbus channel message processing.
95  */
96 static const vmbus_chanmsg_proc_t
97 vmbus_chan_msgprocs[VMBUS_CHANMSG_TYPE_MAX] = {
98 	VMBUS_CHANMSG_PROC(CHOFFER,	vmbus_chan_msgproc_choffer),
99 	VMBUS_CHANMSG_PROC(CHRESCIND,	vmbus_chan_msgproc_chrescind),
100 
101 	VMBUS_CHANMSG_PROC_WAKEUP(CHOPEN_RESP),
102 	VMBUS_CHANMSG_PROC_WAKEUP(GPADL_CONNRESP),
103 	VMBUS_CHANMSG_PROC_WAKEUP(GPADL_DISCONNRESP)
104 };
105 
106 /*
107  * Notify host that there are data pending on our TX bufring.
108  */
109 static __inline void
110 vmbus_chan_signal_tx(const struct vmbus_channel *chan)
111 {
112 	atomic_set_long(chan->ch_evtflag, chan->ch_evtflag_mask);
113 	if (chan->ch_txflags & VMBUS_CHAN_TXF_HASMNF)
114 		atomic_set_int(chan->ch_montrig, chan->ch_montrig_mask);
115 	else
116 		hypercall_signal_event(chan->ch_monprm_dma.hv_paddr);
117 }
118 
119 static void
120 vmbus_chan_ins_prilist(struct vmbus_softc *sc, struct vmbus_channel *chan)
121 {
122 
123 	mtx_assert(&sc->vmbus_prichan_lock, MA_OWNED);
124 	if (atomic_testandset_int(&chan->ch_stflags,
125 	    VMBUS_CHAN_ST_ONPRIL_SHIFT))
126 		panic("channel is already on the prilist");
127 	TAILQ_INSERT_TAIL(&sc->vmbus_prichans, chan, ch_prilink);
128 }
129 
130 static void
131 vmbus_chan_rem_prilist(struct vmbus_softc *sc, struct vmbus_channel *chan)
132 {
133 
134 	mtx_assert(&sc->vmbus_prichan_lock, MA_OWNED);
135 	if (atomic_testandclear_int(&chan->ch_stflags,
136 	    VMBUS_CHAN_ST_ONPRIL_SHIFT) == 0)
137 		panic("channel is not on the prilist");
138 	TAILQ_REMOVE(&sc->vmbus_prichans, chan, ch_prilink);
139 }
140 
141 static void
142 vmbus_chan_ins_sublist(struct vmbus_channel *prichan,
143     struct vmbus_channel *chan)
144 {
145 
146 	mtx_assert(&prichan->ch_subchan_lock, MA_OWNED);
147 
148 	if (atomic_testandset_int(&chan->ch_stflags,
149 	    VMBUS_CHAN_ST_ONSUBL_SHIFT))
150 		panic("channel is already on the sublist");
151 	TAILQ_INSERT_TAIL(&prichan->ch_subchans, chan, ch_sublink);
152 
153 	/* Bump sub-channel count. */
154 	prichan->ch_subchan_cnt++;
155 }
156 
157 static void
158 vmbus_chan_rem_sublist(struct vmbus_channel *prichan,
159     struct vmbus_channel *chan)
160 {
161 
162 	mtx_assert(&prichan->ch_subchan_lock, MA_OWNED);
163 
164 	KASSERT(prichan->ch_subchan_cnt > 0,
165 	    ("invalid subchan_cnt %d", prichan->ch_subchan_cnt));
166 	prichan->ch_subchan_cnt--;
167 
168 	if (atomic_testandclear_int(&chan->ch_stflags,
169 	    VMBUS_CHAN_ST_ONSUBL_SHIFT) == 0)
170 		panic("channel is not on the sublist");
171 	TAILQ_REMOVE(&prichan->ch_subchans, chan, ch_sublink);
172 }
173 
174 static void
175 vmbus_chan_ins_list(struct vmbus_softc *sc, struct vmbus_channel *chan)
176 {
177 
178 	mtx_assert(&sc->vmbus_chan_lock, MA_OWNED);
179 	if (atomic_testandset_int(&chan->ch_stflags,
180 	    VMBUS_CHAN_ST_ONLIST_SHIFT))
181 		panic("channel is already on the list");
182 	TAILQ_INSERT_TAIL(&sc->vmbus_chans, chan, ch_link);
183 }
184 
185 static void
186 vmbus_chan_rem_list(struct vmbus_softc *sc, struct vmbus_channel *chan)
187 {
188 
189 	mtx_assert(&sc->vmbus_chan_lock, MA_OWNED);
190 	if (atomic_testandclear_int(&chan->ch_stflags,
191 	    VMBUS_CHAN_ST_ONLIST_SHIFT) == 0)
192 		panic("channel is not on the list");
193 	TAILQ_REMOVE(&sc->vmbus_chans, chan, ch_link);
194 }
195 
196 static int
197 vmbus_chan_sysctl_mnf(SYSCTL_HANDLER_ARGS)
198 {
199 	struct vmbus_channel *chan = arg1;
200 	int mnf = 0;
201 
202 	if (chan->ch_txflags & VMBUS_CHAN_TXF_HASMNF)
203 		mnf = 1;
204 	return sysctl_handle_int(oidp, &mnf, 0, req);
205 }
206 
207 static void
208 vmbus_chan_sysctl_create(struct vmbus_channel *chan)
209 {
210 	struct sysctl_oid *ch_tree, *chid_tree, *br_tree;
211 	struct sysctl_ctx_list *ctx;
212 	uint32_t ch_id;
213 	char name[16];
214 
215 	/*
216 	 * Add sysctl nodes related to this channel to this
217 	 * channel's sysctl ctx, so that they can be destroyed
218 	 * independently upon close of this channel, which can
219 	 * happen even if the device is not detached.
220 	 */
221 	ctx = &chan->ch_sysctl_ctx;
222 	sysctl_ctx_init(ctx);
223 
224 	/*
225 	 * Create dev.NAME.UNIT.channel tree.
226 	 */
227 	ch_tree = SYSCTL_ADD_NODE(ctx,
228 	    SYSCTL_CHILDREN(device_get_sysctl_tree(chan->ch_dev)),
229 	    OID_AUTO, "channel", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
230 	if (ch_tree == NULL)
231 		return;
232 
233 	/*
234 	 * Create dev.NAME.UNIT.channel.CHANID tree.
235 	 */
236 	if (VMBUS_CHAN_ISPRIMARY(chan))
237 		ch_id = chan->ch_id;
238 	else
239 		ch_id = chan->ch_prichan->ch_id;
240 	snprintf(name, sizeof(name), "%d", ch_id);
241 	chid_tree = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(ch_tree),
242 	    OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
243 	if (chid_tree == NULL)
244 		return;
245 
246 	if (!VMBUS_CHAN_ISPRIMARY(chan)) {
247 		/*
248 		 * Create dev.NAME.UNIT.channel.CHANID.sub tree.
249 		 */
250 		ch_tree = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(chid_tree),
251 		    OID_AUTO, "sub", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
252 		if (ch_tree == NULL)
253 			return;
254 
255 		/*
256 		 * Create dev.NAME.UNIT.channel.CHANID.sub.SUBIDX tree.
257 		 *
258 		 * NOTE:
259 		 * chid_tree is changed to this new sysctl tree.
260 		 */
261 		snprintf(name, sizeof(name), "%d", chan->ch_subidx);
262 		chid_tree = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(ch_tree),
263 		    OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
264 		if (chid_tree == NULL)
265 			return;
266 
267 		SYSCTL_ADD_UINT(ctx, SYSCTL_CHILDREN(chid_tree), OID_AUTO,
268 		    "chanid", CTLFLAG_RD, &chan->ch_id, 0, "channel id");
269 	}
270 
271 	SYSCTL_ADD_UINT(ctx, SYSCTL_CHILDREN(chid_tree), OID_AUTO,
272 	    "cpu", CTLFLAG_RD, &chan->ch_cpuid, 0, "owner CPU id");
273 	SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(chid_tree), OID_AUTO,
274 	    "mnf", CTLTYPE_INT | CTLFLAG_RD | CTLFLAG_MPSAFE,
275 	    chan, 0, vmbus_chan_sysctl_mnf, "I",
276 	    "has monitor notification facilities");
277 
278 	br_tree = SYSCTL_ADD_NODE(ctx, SYSCTL_CHILDREN(chid_tree), OID_AUTO,
279 	    "br", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
280 	if (br_tree != NULL) {
281 		/*
282 		 * Create sysctl tree for RX bufring.
283 		 */
284 		vmbus_br_sysctl_create(ctx, br_tree, &chan->ch_rxbr.rxbr, "rx");
285 		/*
286 		 * Create sysctl tree for TX bufring.
287 		 */
288 		vmbus_br_sysctl_create(ctx, br_tree, &chan->ch_txbr.txbr, "tx");
289 	}
290 }
291 
292 int
293 vmbus_chan_open(struct vmbus_channel *chan, int txbr_size, int rxbr_size,
294     const void *udata, int udlen, vmbus_chan_callback_t cb, void *cbarg)
295 {
296 	struct vmbus_chan_br cbr;
297 	int error;
298 
299 	/*
300 	 * Allocate the TX+RX bufrings.
301 	 */
302 	KASSERT(chan->ch_bufring == NULL, ("bufrings are allocated"));
303 	chan->ch_bufring = hyperv_dmamem_alloc(bus_get_dma_tag(chan->ch_dev),
304 	    PAGE_SIZE, 0, txbr_size + rxbr_size, &chan->ch_bufring_dma,
305 	    BUS_DMA_WAITOK);
306 	if (chan->ch_bufring == NULL) {
307 		device_printf(chan->ch_dev, "bufring allocation failed\n");
308 		return (ENOMEM);
309 	}
310 
311 	cbr.cbr = chan->ch_bufring;
312 	cbr.cbr_paddr = chan->ch_bufring_dma.hv_paddr;
313 	cbr.cbr_txsz = txbr_size;
314 	cbr.cbr_rxsz = rxbr_size;
315 
316 	error = vmbus_chan_open_br(chan, &cbr, udata, udlen, cb, cbarg);
317 	if (error) {
318 		hyperv_dmamem_free(&chan->ch_bufring_dma, chan->ch_bufring);
319 		chan->ch_bufring = NULL;
320 	}
321 	return (error);
322 }
323 
324 int
325 vmbus_chan_open_br(struct vmbus_channel *chan, const struct vmbus_chan_br *cbr,
326     const void *udata, int udlen, vmbus_chan_callback_t cb, void *cbarg)
327 {
328 	struct vmbus_softc *sc = chan->ch_vmbus;
329 	const struct vmbus_chanmsg_chopen_resp *resp;
330 	const struct vmbus_message *msg;
331 	struct vmbus_chanmsg_chopen *req;
332 	struct vmbus_msghc *mh;
333 	uint32_t status;
334 	int error, txbr_size, rxbr_size;
335 	task_fn_t *task_fn;
336 	uint8_t *br;
337 
338 	if (udlen > VMBUS_CHANMSG_CHOPEN_UDATA_SIZE) {
339 		device_printf(sc->vmbus_dev,
340 		    "invalid udata len %d for chan%u\n", udlen, chan->ch_id);
341 		return EINVAL;
342 	}
343 
344 	br = cbr->cbr;
345 	txbr_size = cbr->cbr_txsz;
346 	rxbr_size = cbr->cbr_rxsz;
347 	KASSERT((txbr_size & PAGE_MASK) == 0,
348 	    ("send bufring size is not multiple page"));
349 	KASSERT((rxbr_size & PAGE_MASK) == 0,
350 	    ("recv bufring size is not multiple page"));
351 	KASSERT((cbr->cbr_paddr & PAGE_MASK) == 0,
352 	    ("bufring is not page aligned"));
353 
354 	/*
355 	 * Zero out the TX/RX bufrings, in case that they were used before.
356 	 */
357 	memset(br, 0, txbr_size + rxbr_size);
358 
359 	if (atomic_testandset_int(&chan->ch_stflags,
360 	    VMBUS_CHAN_ST_OPENED_SHIFT))
361 		panic("double-open chan%u", chan->ch_id);
362 
363 	chan->ch_cb = cb;
364 	chan->ch_cbarg = cbarg;
365 
366 	vmbus_chan_update_evtflagcnt(sc, chan);
367 
368 	chan->ch_tq = VMBUS_PCPU_GET(chan->ch_vmbus, event_tq, chan->ch_cpuid);
369 	if (chan->ch_flags & VMBUS_CHAN_FLAG_BATCHREAD)
370 		task_fn = vmbus_chan_task;
371 	else
372 		task_fn = vmbus_chan_task_nobatch;
373 	TASK_INIT(&chan->ch_task, 0, task_fn, chan);
374 
375 	/* TX bufring comes first */
376 	vmbus_txbr_setup(&chan->ch_txbr, br, txbr_size);
377 	/* RX bufring immediately follows TX bufring */
378 	vmbus_rxbr_setup(&chan->ch_rxbr, br + txbr_size, rxbr_size);
379 
380 	/* Create sysctl tree for this channel */
381 	vmbus_chan_sysctl_create(chan);
382 
383 	/*
384 	 * Connect the bufrings, both RX and TX, to this channel.
385 	 */
386 	error = vmbus_chan_gpadl_connect(chan, cbr->cbr_paddr,
387 	    txbr_size + rxbr_size, &chan->ch_bufring_gpadl);
388 	if (error) {
389 		device_printf(sc->vmbus_dev,
390 		    "failed to connect bufring GPADL to chan%u\n", chan->ch_id);
391 		goto failed;
392 	}
393 
394 	/*
395 	 * Install this channel, before it is opened, but after everything
396 	 * else has been setup.
397 	 */
398 	vmbus_chan_set_chmap(chan);
399 
400 	/*
401 	 * Open channel w/ the bufring GPADL on the target CPU.
402 	 */
403 	mh = vmbus_msghc_get(sc, sizeof(*req));
404 	if (mh == NULL) {
405 		device_printf(sc->vmbus_dev,
406 		    "can not get msg hypercall for chopen(chan%u)\n",
407 		    chan->ch_id);
408 		error = ENXIO;
409 		goto failed;
410 	}
411 
412 	req = vmbus_msghc_dataptr(mh);
413 	req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_CHOPEN;
414 	req->chm_chanid = chan->ch_id;
415 	req->chm_openid = chan->ch_id;
416 	req->chm_gpadl = chan->ch_bufring_gpadl;
417 	req->chm_vcpuid = chan->ch_vcpuid;
418 	req->chm_txbr_pgcnt = txbr_size >> PAGE_SHIFT;
419 	if (udlen > 0)
420 		memcpy(req->chm_udata, udata, udlen);
421 
422 	error = vmbus_msghc_exec(sc, mh);
423 	if (error) {
424 		device_printf(sc->vmbus_dev,
425 		    "chopen(chan%u) msg hypercall exec failed: %d\n",
426 		    chan->ch_id, error);
427 		vmbus_msghc_put(sc, mh);
428 		goto failed;
429 	}
430 
431 	msg = vmbus_msghc_wait_result(sc, mh);
432 	resp = (const struct vmbus_chanmsg_chopen_resp *)msg->msg_data;
433 	status = resp->chm_status;
434 
435 	vmbus_msghc_put(sc, mh);
436 
437 	if (status == 0) {
438 		if (bootverbose) {
439 			device_printf(sc->vmbus_dev, "chan%u opened\n",
440 			    chan->ch_id);
441 		}
442 		return 0;
443 	}
444 
445 	device_printf(sc->vmbus_dev, "failed to open chan%u\n", chan->ch_id);
446 	error = ENXIO;
447 
448 failed:
449 	vmbus_chan_clear_chmap(chan);
450 	if (chan->ch_bufring_gpadl) {
451 		vmbus_chan_gpadl_disconnect(chan, chan->ch_bufring_gpadl);
452 		chan->ch_bufring_gpadl = 0;
453 	}
454 	atomic_clear_int(&chan->ch_stflags, VMBUS_CHAN_ST_OPENED);
455 	return error;
456 }
457 
458 int
459 vmbus_chan_gpadl_connect(struct vmbus_channel *chan, bus_addr_t paddr,
460     int size, uint32_t *gpadl0)
461 {
462 	struct vmbus_softc *sc = chan->ch_vmbus;
463 	struct vmbus_msghc *mh;
464 	struct vmbus_chanmsg_gpadl_conn *req;
465 	const struct vmbus_message *msg;
466 	size_t reqsz;
467 	uint32_t gpadl, status;
468 	int page_count, range_len, i, cnt, error;
469 	uint64_t page_id;
470 
471 	/*
472 	 * Preliminary checks.
473 	 */
474 
475 	KASSERT((size & PAGE_MASK) == 0,
476 	    ("invalid GPA size %d, not multiple page size", size));
477 	page_count = size >> PAGE_SHIFT;
478 
479 	KASSERT((paddr & PAGE_MASK) == 0,
480 	    ("GPA is not page aligned %jx", (uintmax_t)paddr));
481 	page_id = paddr >> PAGE_SHIFT;
482 
483 	range_len = __offsetof(struct vmbus_gpa_range, gpa_page[page_count]);
484 	/*
485 	 * We don't support multiple GPA ranges.
486 	 */
487 	if (range_len > UINT16_MAX) {
488 		device_printf(sc->vmbus_dev, "GPA too large, %d pages\n",
489 		    page_count);
490 		return EOPNOTSUPP;
491 	}
492 
493 	/*
494 	 * Allocate GPADL id.
495 	 */
496 	gpadl = vmbus_gpadl_alloc(sc);
497 	*gpadl0 = gpadl;
498 
499 	/*
500 	 * Connect this GPADL to the target channel.
501 	 *
502 	 * NOTE:
503 	 * Since each message can only hold small set of page
504 	 * addresses, several messages may be required to
505 	 * complete the connection.
506 	 */
507 	if (page_count > VMBUS_CHANMSG_GPADL_CONN_PGMAX)
508 		cnt = VMBUS_CHANMSG_GPADL_CONN_PGMAX;
509 	else
510 		cnt = page_count;
511 	page_count -= cnt;
512 
513 	reqsz = __offsetof(struct vmbus_chanmsg_gpadl_conn,
514 	    chm_range.gpa_page[cnt]);
515 	mh = vmbus_msghc_get(sc, reqsz);
516 	if (mh == NULL) {
517 		device_printf(sc->vmbus_dev,
518 		    "can not get msg hypercall for gpadl->chan%u\n",
519 		    chan->ch_id);
520 		return EIO;
521 	}
522 
523 	req = vmbus_msghc_dataptr(mh);
524 	req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_GPADL_CONN;
525 	req->chm_chanid = chan->ch_id;
526 	req->chm_gpadl = gpadl;
527 	req->chm_range_len = range_len;
528 	req->chm_range_cnt = 1;
529 	req->chm_range.gpa_len = size;
530 	req->chm_range.gpa_ofs = 0;
531 	for (i = 0; i < cnt; ++i)
532 		req->chm_range.gpa_page[i] = page_id++;
533 
534 	error = vmbus_msghc_exec(sc, mh);
535 	if (error) {
536 		device_printf(sc->vmbus_dev,
537 		    "gpadl->chan%u msg hypercall exec failed: %d\n",
538 		    chan->ch_id, error);
539 		vmbus_msghc_put(sc, mh);
540 		return error;
541 	}
542 
543 	while (page_count > 0) {
544 		struct vmbus_chanmsg_gpadl_subconn *subreq;
545 
546 		if (page_count > VMBUS_CHANMSG_GPADL_SUBCONN_PGMAX)
547 			cnt = VMBUS_CHANMSG_GPADL_SUBCONN_PGMAX;
548 		else
549 			cnt = page_count;
550 		page_count -= cnt;
551 
552 		reqsz = __offsetof(struct vmbus_chanmsg_gpadl_subconn,
553 		    chm_gpa_page[cnt]);
554 		vmbus_msghc_reset(mh, reqsz);
555 
556 		subreq = vmbus_msghc_dataptr(mh);
557 		subreq->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_GPADL_SUBCONN;
558 		subreq->chm_gpadl = gpadl;
559 		for (i = 0; i < cnt; ++i)
560 			subreq->chm_gpa_page[i] = page_id++;
561 
562 		vmbus_msghc_exec_noresult(mh);
563 	}
564 	KASSERT(page_count == 0, ("invalid page count %d", page_count));
565 
566 	msg = vmbus_msghc_wait_result(sc, mh);
567 	status = ((const struct vmbus_chanmsg_gpadl_connresp *)
568 	    msg->msg_data)->chm_status;
569 
570 	vmbus_msghc_put(sc, mh);
571 
572 	if (status != 0) {
573 		device_printf(sc->vmbus_dev, "gpadl->chan%u failed: "
574 		    "status %u\n", chan->ch_id, status);
575 		return EIO;
576 	} else {
577 		if (bootverbose) {
578 			device_printf(sc->vmbus_dev, "gpadl->chan%u "
579 			    "succeeded\n", chan->ch_id);
580 		}
581 	}
582 	return 0;
583 }
584 
585 /*
586  * Disconnect the GPA from the target channel
587  */
588 int
589 vmbus_chan_gpadl_disconnect(struct vmbus_channel *chan, uint32_t gpadl)
590 {
591 	struct vmbus_softc *sc = chan->ch_vmbus;
592 	struct vmbus_msghc *mh;
593 	struct vmbus_chanmsg_gpadl_disconn *req;
594 	int error;
595 
596 	mh = vmbus_msghc_get(sc, sizeof(*req));
597 	if (mh == NULL) {
598 		device_printf(sc->vmbus_dev,
599 		    "can not get msg hypercall for gpa x->chan%u\n",
600 		    chan->ch_id);
601 		return EBUSY;
602 	}
603 
604 	req = vmbus_msghc_dataptr(mh);
605 	req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_GPADL_DISCONN;
606 	req->chm_chanid = chan->ch_id;
607 	req->chm_gpadl = gpadl;
608 
609 	error = vmbus_msghc_exec(sc, mh);
610 	if (error) {
611 		device_printf(sc->vmbus_dev,
612 		    "gpa x->chan%u msg hypercall exec failed: %d\n",
613 		    chan->ch_id, error);
614 		vmbus_msghc_put(sc, mh);
615 		return error;
616 	}
617 
618 	vmbus_msghc_wait_result(sc, mh);
619 	/* Discard result; no useful information */
620 	vmbus_msghc_put(sc, mh);
621 
622 	return 0;
623 }
624 
625 static void
626 vmbus_chan_clrchmap_task(void *xchan, int pending __unused)
627 {
628 	struct vmbus_channel *chan = xchan;
629 
630 	critical_enter();
631 	chan->ch_vmbus->vmbus_chmap[chan->ch_id] = NULL;
632 	critical_exit();
633 }
634 
635 static void
636 vmbus_chan_clear_chmap(struct vmbus_channel *chan)
637 {
638 	struct task chmap_task;
639 
640 	TASK_INIT(&chmap_task, 0, vmbus_chan_clrchmap_task, chan);
641 	taskqueue_enqueue(chan->ch_tq, &chmap_task);
642 	taskqueue_drain(chan->ch_tq, &chmap_task);
643 }
644 
645 static void
646 vmbus_chan_set_chmap(struct vmbus_channel *chan)
647 {
648 	__compiler_membar();
649 	chan->ch_vmbus->vmbus_chmap[chan->ch_id] = chan;
650 }
651 
652 static void
653 vmbus_chan_close_internal(struct vmbus_channel *chan)
654 {
655 	struct vmbus_softc *sc = chan->ch_vmbus;
656 	struct vmbus_msghc *mh;
657 	struct vmbus_chanmsg_chclose *req;
658 	int error;
659 
660 	/* TODO: stringent check */
661 	atomic_clear_int(&chan->ch_stflags, VMBUS_CHAN_ST_OPENED);
662 
663 	/*
664 	 * Free this channel's sysctl tree attached to its device's
665 	 * sysctl tree.
666 	 */
667 	sysctl_ctx_free(&chan->ch_sysctl_ctx);
668 
669 	/*
670 	 * NOTE:
671 	 * Order is critical.  This channel _must_ be uninstalled first,
672 	 * else the channel task may be enqueued by the IDT after it has
673 	 * been drained.
674 	 */
675 	vmbus_chan_clear_chmap(chan);
676 	taskqueue_drain(chan->ch_tq, &chan->ch_task);
677 	chan->ch_tq = NULL;
678 
679 	/*
680 	 * Close this channel.
681 	 */
682 	mh = vmbus_msghc_get(sc, sizeof(*req));
683 	if (mh == NULL) {
684 		device_printf(sc->vmbus_dev,
685 		    "can not get msg hypercall for chclose(chan%u)\n",
686 		    chan->ch_id);
687 		return;
688 	}
689 
690 	req = vmbus_msghc_dataptr(mh);
691 	req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_CHCLOSE;
692 	req->chm_chanid = chan->ch_id;
693 
694 	error = vmbus_msghc_exec_noresult(mh);
695 	vmbus_msghc_put(sc, mh);
696 
697 	if (error) {
698 		device_printf(sc->vmbus_dev,
699 		    "chclose(chan%u) msg hypercall exec failed: %d\n",
700 		    chan->ch_id, error);
701 		return;
702 	} else if (bootverbose) {
703 		device_printf(sc->vmbus_dev, "close chan%u\n", chan->ch_id);
704 	}
705 
706 	/*
707 	 * Disconnect the TX+RX bufrings from this channel.
708 	 */
709 	if (chan->ch_bufring_gpadl) {
710 		vmbus_chan_gpadl_disconnect(chan, chan->ch_bufring_gpadl);
711 		chan->ch_bufring_gpadl = 0;
712 	}
713 
714 	/*
715 	 * Destroy the TX+RX bufrings.
716 	 */
717 	if (chan->ch_bufring != NULL) {
718 		hyperv_dmamem_free(&chan->ch_bufring_dma, chan->ch_bufring);
719 		chan->ch_bufring = NULL;
720 	}
721 }
722 
723 /*
724  * Caller should make sure that all sub-channels have
725  * been added to 'chan' and all to-be-closed channels
726  * are not being opened.
727  */
728 void
729 vmbus_chan_close(struct vmbus_channel *chan)
730 {
731 	int subchan_cnt;
732 
733 	if (!VMBUS_CHAN_ISPRIMARY(chan)) {
734 		/*
735 		 * Sub-channel is closed when its primary channel
736 		 * is closed; done.
737 		 */
738 		return;
739 	}
740 
741 	/*
742 	 * Close all sub-channels, if any.
743 	 */
744 	subchan_cnt = chan->ch_subchan_cnt;
745 	if (subchan_cnt > 0) {
746 		struct vmbus_channel **subchan;
747 		int i;
748 
749 		subchan = vmbus_subchan_get(chan, subchan_cnt);
750 		for (i = 0; i < subchan_cnt; ++i)
751 			vmbus_chan_close_internal(subchan[i]);
752 		vmbus_subchan_rel(subchan, subchan_cnt);
753 	}
754 
755 	/* Then close the primary channel. */
756 	vmbus_chan_close_internal(chan);
757 }
758 
759 void
760 vmbus_chan_intr_drain(struct vmbus_channel *chan)
761 {
762 
763 	taskqueue_drain(chan->ch_tq, &chan->ch_task);
764 }
765 
766 int
767 vmbus_chan_send(struct vmbus_channel *chan, uint16_t type, uint16_t flags,
768     void *data, int dlen, uint64_t xactid)
769 {
770 	struct vmbus_chanpkt pkt;
771 	int pktlen, pad_pktlen, hlen, error;
772 	uint64_t pad = 0;
773 	struct iovec iov[3];
774 	boolean_t send_evt;
775 
776 	hlen = sizeof(pkt);
777 	pktlen = hlen + dlen;
778 	pad_pktlen = VMBUS_CHANPKT_TOTLEN(pktlen);
779 	KASSERT(pad_pktlen <= vmbus_txbr_maxpktsz(&chan->ch_txbr),
780 	    ("invalid packet size %d", pad_pktlen));
781 
782 	pkt.cp_hdr.cph_type = type;
783 	pkt.cp_hdr.cph_flags = flags;
784 	VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_hlen, hlen);
785 	VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_tlen, pad_pktlen);
786 	pkt.cp_hdr.cph_xactid = xactid;
787 
788 	iov[0].iov_base = &pkt;
789 	iov[0].iov_len = hlen;
790 	iov[1].iov_base = data;
791 	iov[1].iov_len = dlen;
792 	iov[2].iov_base = &pad;
793 	iov[2].iov_len = pad_pktlen - pktlen;
794 
795 	error = vmbus_txbr_write(&chan->ch_txbr, iov, 3, &send_evt);
796 	if (!error && send_evt)
797 		vmbus_chan_signal_tx(chan);
798 	return error;
799 }
800 
801 int
802 vmbus_chan_send_sglist(struct vmbus_channel *chan,
803     struct vmbus_gpa sg[], int sglen, void *data, int dlen, uint64_t xactid)
804 {
805 	struct vmbus_chanpkt_sglist pkt;
806 	int pktlen, pad_pktlen, hlen, error;
807 	struct iovec iov[4];
808 	boolean_t send_evt;
809 	uint64_t pad = 0;
810 
811 	hlen = __offsetof(struct vmbus_chanpkt_sglist, cp_gpa[sglen]);
812 	pktlen = hlen + dlen;
813 	pad_pktlen = VMBUS_CHANPKT_TOTLEN(pktlen);
814 	KASSERT(pad_pktlen <= vmbus_txbr_maxpktsz(&chan->ch_txbr),
815 	    ("invalid packet size %d", pad_pktlen));
816 
817 	pkt.cp_hdr.cph_type = VMBUS_CHANPKT_TYPE_GPA;
818 	pkt.cp_hdr.cph_flags = VMBUS_CHANPKT_FLAG_RC;
819 	VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_hlen, hlen);
820 	VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_tlen, pad_pktlen);
821 	pkt.cp_hdr.cph_xactid = xactid;
822 	pkt.cp_rsvd = 0;
823 	pkt.cp_gpa_cnt = sglen;
824 
825 	iov[0].iov_base = &pkt;
826 	iov[0].iov_len = sizeof(pkt);
827 	iov[1].iov_base = sg;
828 	iov[1].iov_len = sizeof(struct vmbus_gpa) * sglen;
829 	iov[2].iov_base = data;
830 	iov[2].iov_len = dlen;
831 	iov[3].iov_base = &pad;
832 	iov[3].iov_len = pad_pktlen - pktlen;
833 
834 	error = vmbus_txbr_write(&chan->ch_txbr, iov, 4, &send_evt);
835 	if (!error && send_evt)
836 		vmbus_chan_signal_tx(chan);
837 	return error;
838 }
839 
840 int
841 vmbus_chan_send_prplist(struct vmbus_channel *chan,
842     struct vmbus_gpa_range *prp, int prp_cnt, void *data, int dlen,
843     uint64_t xactid)
844 {
845 	struct vmbus_chanpkt_prplist pkt;
846 	int pktlen, pad_pktlen, hlen, error;
847 	struct iovec iov[4];
848 	boolean_t send_evt;
849 	uint64_t pad = 0;
850 
851 	hlen = __offsetof(struct vmbus_chanpkt_prplist,
852 	    cp_range[0].gpa_page[prp_cnt]);
853 	pktlen = hlen + dlen;
854 	pad_pktlen = VMBUS_CHANPKT_TOTLEN(pktlen);
855 	KASSERT(pad_pktlen <= vmbus_txbr_maxpktsz(&chan->ch_txbr),
856 	    ("invalid packet size %d", pad_pktlen));
857 
858 	pkt.cp_hdr.cph_type = VMBUS_CHANPKT_TYPE_GPA;
859 	pkt.cp_hdr.cph_flags = VMBUS_CHANPKT_FLAG_RC;
860 	VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_hlen, hlen);
861 	VMBUS_CHANPKT_SETLEN(pkt.cp_hdr.cph_tlen, pad_pktlen);
862 	pkt.cp_hdr.cph_xactid = xactid;
863 	pkt.cp_rsvd = 0;
864 	pkt.cp_range_cnt = 1;
865 
866 	iov[0].iov_base = &pkt;
867 	iov[0].iov_len = sizeof(pkt);
868 	iov[1].iov_base = prp;
869 	iov[1].iov_len = __offsetof(struct vmbus_gpa_range, gpa_page[prp_cnt]);
870 	iov[2].iov_base = data;
871 	iov[2].iov_len = dlen;
872 	iov[3].iov_base = &pad;
873 	iov[3].iov_len = pad_pktlen - pktlen;
874 
875 	error = vmbus_txbr_write(&chan->ch_txbr, iov, 4, &send_evt);
876 	if (!error && send_evt)
877 		vmbus_chan_signal_tx(chan);
878 	return error;
879 }
880 
881 int
882 vmbus_chan_recv(struct vmbus_channel *chan, void *data, int *dlen0,
883     uint64_t *xactid)
884 {
885 	struct vmbus_chanpkt_hdr pkt;
886 	int error, dlen, hlen;
887 
888 	error = vmbus_rxbr_peek(&chan->ch_rxbr, &pkt, sizeof(pkt));
889 	if (error)
890 		return (error);
891 
892 	if (__predict_false(pkt.cph_hlen < VMBUS_CHANPKT_HLEN_MIN)) {
893 		device_printf(chan->ch_dev, "invalid hlen %u\n",
894 		    pkt.cph_hlen);
895 		/* XXX this channel is dead actually. */
896 		return (EIO);
897 	}
898 	if (__predict_false(pkt.cph_hlen > pkt.cph_tlen)) {
899 		device_printf(chan->ch_dev, "invalid hlen %u and tlen %u\n",
900 		    pkt.cph_hlen, pkt.cph_tlen);
901 		/* XXX this channel is dead actually. */
902 		return (EIO);
903 	}
904 
905 	hlen = VMBUS_CHANPKT_GETLEN(pkt.cph_hlen);
906 	dlen = VMBUS_CHANPKT_GETLEN(pkt.cph_tlen) - hlen;
907 
908 	if (*dlen0 < dlen) {
909 		/* Return the size of this packet's data. */
910 		*dlen0 = dlen;
911 		return (ENOBUFS);
912 	}
913 
914 	*xactid = pkt.cph_xactid;
915 	*dlen0 = dlen;
916 
917 	/* Skip packet header */
918 	error = vmbus_rxbr_read(&chan->ch_rxbr, data, dlen, hlen);
919 	KASSERT(!error, ("vmbus_rxbr_read failed"));
920 
921 	return (0);
922 }
923 
924 int
925 vmbus_chan_recv_pkt(struct vmbus_channel *chan,
926     struct vmbus_chanpkt_hdr *pkt0, int *pktlen0)
927 {
928 	struct vmbus_chanpkt_hdr pkt;
929 	int error, pktlen;
930 
931 	error = vmbus_rxbr_peek(&chan->ch_rxbr, &pkt, sizeof(pkt));
932 	if (error)
933 		return (error);
934 
935 	if (__predict_false(pkt.cph_hlen < VMBUS_CHANPKT_HLEN_MIN)) {
936 		device_printf(chan->ch_dev, "invalid hlen %u\n",
937 		    pkt.cph_hlen);
938 		/* XXX this channel is dead actually. */
939 		return (EIO);
940 	}
941 	if (__predict_false(pkt.cph_hlen > pkt.cph_tlen)) {
942 		device_printf(chan->ch_dev, "invalid hlen %u and tlen %u\n",
943 		    pkt.cph_hlen, pkt.cph_tlen);
944 		/* XXX this channel is dead actually. */
945 		return (EIO);
946 	}
947 
948 	pktlen = VMBUS_CHANPKT_GETLEN(pkt.cph_tlen);
949 	if (*pktlen0 < pktlen) {
950 		/* Return the size of this packet. */
951 		*pktlen0 = pktlen;
952 		return (ENOBUFS);
953 	}
954 	*pktlen0 = pktlen;
955 
956 	/* Include packet header */
957 	error = vmbus_rxbr_read(&chan->ch_rxbr, pkt0, pktlen, 0);
958 	KASSERT(!error, ("vmbus_rxbr_read failed"));
959 
960 	return (0);
961 }
962 
963 static void
964 vmbus_chan_task(void *xchan, int pending __unused)
965 {
966 	struct vmbus_channel *chan = xchan;
967 	vmbus_chan_callback_t cb = chan->ch_cb;
968 	void *cbarg = chan->ch_cbarg;
969 
970 	/*
971 	 * Optimize host to guest signaling by ensuring:
972 	 * 1. While reading the channel, we disable interrupts from
973 	 *    host.
974 	 * 2. Ensure that we process all posted messages from the host
975 	 *    before returning from this callback.
976 	 * 3. Once we return, enable signaling from the host. Once this
977 	 *    state is set we check to see if additional packets are
978 	 *    available to read. In this case we repeat the process.
979 	 *
980 	 * NOTE: Interrupt has been disabled in the ISR.
981 	 */
982 	for (;;) {
983 		uint32_t left;
984 
985 		cb(chan, cbarg);
986 
987 		left = vmbus_rxbr_intr_unmask(&chan->ch_rxbr);
988 		if (left == 0) {
989 			/* No more data in RX bufring; done */
990 			break;
991 		}
992 		vmbus_rxbr_intr_mask(&chan->ch_rxbr);
993 	}
994 }
995 
996 static void
997 vmbus_chan_task_nobatch(void *xchan, int pending __unused)
998 {
999 	struct vmbus_channel *chan = xchan;
1000 
1001 	chan->ch_cb(chan, chan->ch_cbarg);
1002 }
1003 
1004 static __inline void
1005 vmbus_event_flags_proc(struct vmbus_softc *sc, volatile u_long *event_flags,
1006     int flag_cnt)
1007 {
1008 	int f;
1009 
1010 	for (f = 0; f < flag_cnt; ++f) {
1011 		uint32_t chid_base;
1012 		u_long flags;
1013 		int chid_ofs;
1014 
1015 		if (event_flags[f] == 0)
1016 			continue;
1017 
1018 		flags = atomic_swap_long(&event_flags[f], 0);
1019 		chid_base = f << VMBUS_EVTFLAG_SHIFT;
1020 
1021 		while ((chid_ofs = ffsl(flags)) != 0) {
1022 			struct vmbus_channel *chan;
1023 
1024 			--chid_ofs; /* NOTE: ffsl is 1-based */
1025 			flags &= ~(1UL << chid_ofs);
1026 
1027 			chan = sc->vmbus_chmap[chid_base + chid_ofs];
1028 			if (__predict_false(chan == NULL)) {
1029 				/* Channel is closed. */
1030 				continue;
1031 			}
1032 			__compiler_membar();
1033 
1034 			if (chan->ch_flags & VMBUS_CHAN_FLAG_BATCHREAD)
1035 				vmbus_rxbr_intr_mask(&chan->ch_rxbr);
1036 			taskqueue_enqueue(chan->ch_tq, &chan->ch_task);
1037 		}
1038 	}
1039 }
1040 
1041 void
1042 vmbus_event_proc(struct vmbus_softc *sc, int cpu)
1043 {
1044 	struct vmbus_evtflags *eventf;
1045 
1046 	/*
1047 	 * On Host with Win8 or above, the event page can be checked directly
1048 	 * to get the id of the channel that has the pending interrupt.
1049 	 */
1050 	eventf = VMBUS_PCPU_GET(sc, event_flags, cpu) + VMBUS_SINT_MESSAGE;
1051 	vmbus_event_flags_proc(sc, eventf->evt_flags,
1052 	    VMBUS_PCPU_GET(sc, event_flags_cnt, cpu));
1053 }
1054 
1055 void
1056 vmbus_event_proc_compat(struct vmbus_softc *sc, int cpu)
1057 {
1058 	struct vmbus_evtflags *eventf;
1059 
1060 	eventf = VMBUS_PCPU_GET(sc, event_flags, cpu) + VMBUS_SINT_MESSAGE;
1061 	if (atomic_testandclear_long(&eventf->evt_flags[0], 0)) {
1062 		vmbus_event_flags_proc(sc, sc->vmbus_rx_evtflags,
1063 		    VMBUS_CHAN_MAX_COMPAT >> VMBUS_EVTFLAG_SHIFT);
1064 	}
1065 }
1066 
1067 static void
1068 vmbus_chan_update_evtflagcnt(struct vmbus_softc *sc,
1069     const struct vmbus_channel *chan)
1070 {
1071 	volatile int *flag_cnt_ptr;
1072 	int flag_cnt;
1073 
1074 	flag_cnt = (chan->ch_id / VMBUS_EVTFLAG_LEN) + 1;
1075 	flag_cnt_ptr = VMBUS_PCPU_PTR(sc, event_flags_cnt, chan->ch_cpuid);
1076 
1077 	for (;;) {
1078 		int old_flag_cnt;
1079 
1080 		old_flag_cnt = *flag_cnt_ptr;
1081 		if (old_flag_cnt >= flag_cnt)
1082 			break;
1083 		if (atomic_cmpset_int(flag_cnt_ptr, old_flag_cnt, flag_cnt)) {
1084 			if (bootverbose) {
1085 				device_printf(sc->vmbus_dev,
1086 				    "channel%u update cpu%d flag_cnt to %d\n",
1087 				    chan->ch_id, chan->ch_cpuid, flag_cnt);
1088 			}
1089 			break;
1090 		}
1091 	}
1092 }
1093 
1094 static struct vmbus_channel *
1095 vmbus_chan_alloc(struct vmbus_softc *sc)
1096 {
1097 	struct vmbus_channel *chan;
1098 
1099 	chan = malloc(sizeof(*chan), M_DEVBUF, M_WAITOK | M_ZERO);
1100 
1101 	chan->ch_monprm = hyperv_dmamem_alloc(bus_get_dma_tag(sc->vmbus_dev),
1102 	    HYPERCALL_PARAM_ALIGN, 0, sizeof(struct hyperv_mon_param),
1103 	    &chan->ch_monprm_dma, BUS_DMA_WAITOK | BUS_DMA_ZERO);
1104 	if (chan->ch_monprm == NULL) {
1105 		device_printf(sc->vmbus_dev, "monprm alloc failed\n");
1106 		free(chan, M_DEVBUF);
1107 		return NULL;
1108 	}
1109 
1110 	chan->ch_vmbus = sc;
1111 	mtx_init(&chan->ch_subchan_lock, "vmbus subchan", NULL, MTX_DEF);
1112 	TAILQ_INIT(&chan->ch_subchans);
1113 	vmbus_rxbr_init(&chan->ch_rxbr);
1114 	vmbus_txbr_init(&chan->ch_txbr);
1115 
1116 	return chan;
1117 }
1118 
1119 static void
1120 vmbus_chan_free(struct vmbus_channel *chan)
1121 {
1122 
1123 	KASSERT(TAILQ_EMPTY(&chan->ch_subchans) && chan->ch_subchan_cnt == 0,
1124 	    ("still owns sub-channels"));
1125 	KASSERT((chan->ch_stflags &
1126 	    (VMBUS_CHAN_ST_OPENED |
1127 	     VMBUS_CHAN_ST_ONPRIL |
1128 	     VMBUS_CHAN_ST_ONSUBL |
1129 	     VMBUS_CHAN_ST_ONLIST)) == 0, ("free busy channel"));
1130 	hyperv_dmamem_free(&chan->ch_monprm_dma, chan->ch_monprm);
1131 	mtx_destroy(&chan->ch_subchan_lock);
1132 	vmbus_rxbr_deinit(&chan->ch_rxbr);
1133 	vmbus_txbr_deinit(&chan->ch_txbr);
1134 	free(chan, M_DEVBUF);
1135 }
1136 
1137 static int
1138 vmbus_chan_add(struct vmbus_channel *newchan)
1139 {
1140 	struct vmbus_softc *sc = newchan->ch_vmbus;
1141 	struct vmbus_channel *prichan;
1142 
1143 	if (newchan->ch_id == 0) {
1144 		/*
1145 		 * XXX
1146 		 * Chan0 will neither be processed nor should be offered;
1147 		 * skip it.
1148 		 */
1149 		device_printf(sc->vmbus_dev, "got chan0 offer, discard\n");
1150 		return EINVAL;
1151 	} else if (newchan->ch_id >= VMBUS_CHAN_MAX) {
1152 		device_printf(sc->vmbus_dev, "invalid chan%u offer\n",
1153 		    newchan->ch_id);
1154 		return EINVAL;
1155 	}
1156 
1157 	if (bootverbose) {
1158 		device_printf(sc->vmbus_dev, "chan%u subidx%u offer\n",
1159 		    newchan->ch_id, newchan->ch_subidx);
1160 	}
1161 
1162 	mtx_lock(&sc->vmbus_prichan_lock);
1163 	TAILQ_FOREACH(prichan, &sc->vmbus_prichans, ch_prilink) {
1164 		/*
1165 		 * Sub-channel will have the same type GUID and instance
1166 		 * GUID as its primary channel.
1167 		 */
1168 		if (memcmp(&prichan->ch_guid_type, &newchan->ch_guid_type,
1169 		    sizeof(struct hyperv_guid)) == 0 &&
1170 		    memcmp(&prichan->ch_guid_inst, &newchan->ch_guid_inst,
1171 		    sizeof(struct hyperv_guid)) == 0)
1172 			break;
1173 	}
1174 	if (VMBUS_CHAN_ISPRIMARY(newchan)) {
1175 		if (prichan == NULL) {
1176 			/* Install the new primary channel */
1177 			vmbus_chan_ins_prilist(sc, newchan);
1178 			mtx_unlock(&sc->vmbus_prichan_lock);
1179 			goto done;
1180 		} else {
1181 			mtx_unlock(&sc->vmbus_prichan_lock);
1182 			device_printf(sc->vmbus_dev, "duplicated primary "
1183 			    "chan%u\n", newchan->ch_id);
1184 			return EINVAL;
1185 		}
1186 	} else { /* Sub-channel */
1187 		if (prichan == NULL) {
1188 			mtx_unlock(&sc->vmbus_prichan_lock);
1189 			device_printf(sc->vmbus_dev, "no primary chan for "
1190 			    "chan%u\n", newchan->ch_id);
1191 			return EINVAL;
1192 		}
1193 		/*
1194 		 * Found the primary channel for this sub-channel and
1195 		 * move on.
1196 		 *
1197 		 * XXX refcnt prichan
1198 		 */
1199 	}
1200 	mtx_unlock(&sc->vmbus_prichan_lock);
1201 
1202 	/*
1203 	 * This is a sub-channel; link it with the primary channel.
1204 	 */
1205 	KASSERT(!VMBUS_CHAN_ISPRIMARY(newchan),
1206 	    ("new channel is not sub-channel"));
1207 	KASSERT(prichan != NULL, ("no primary channel"));
1208 
1209 	newchan->ch_prichan = prichan;
1210 	newchan->ch_dev = prichan->ch_dev;
1211 
1212 	mtx_lock(&prichan->ch_subchan_lock);
1213 	vmbus_chan_ins_sublist(prichan, newchan);
1214 	mtx_unlock(&prichan->ch_subchan_lock);
1215 	/*
1216 	 * Notify anyone that is interested in this sub-channel,
1217 	 * after this sub-channel is setup.
1218 	 */
1219 	wakeup(prichan);
1220 done:
1221 	/*
1222 	 * Hook this channel up for later rescind.
1223 	 */
1224 	mtx_lock(&sc->vmbus_chan_lock);
1225 	vmbus_chan_ins_list(sc, newchan);
1226 	mtx_unlock(&sc->vmbus_chan_lock);
1227 	return 0;
1228 }
1229 
1230 void
1231 vmbus_chan_cpu_set(struct vmbus_channel *chan, int cpu)
1232 {
1233 	KASSERT(cpu >= 0 && cpu < mp_ncpus, ("invalid cpu %d", cpu));
1234 
1235 	if (chan->ch_vmbus->vmbus_version == VMBUS_VERSION_WS2008 ||
1236 	    chan->ch_vmbus->vmbus_version == VMBUS_VERSION_WIN7) {
1237 		/* Only cpu0 is supported */
1238 		cpu = 0;
1239 	}
1240 
1241 	chan->ch_cpuid = cpu;
1242 	chan->ch_vcpuid = VMBUS_PCPU_GET(chan->ch_vmbus, vcpuid, cpu);
1243 
1244 	if (bootverbose) {
1245 		printf("vmbus_chan%u: assigned to cpu%u [vcpu%u]\n",
1246 		    chan->ch_id, chan->ch_cpuid, chan->ch_vcpuid);
1247 	}
1248 }
1249 
1250 void
1251 vmbus_chan_cpu_rr(struct vmbus_channel *chan)
1252 {
1253 	static uint32_t vmbus_chan_nextcpu;
1254 	int cpu;
1255 
1256 	cpu = atomic_fetchadd_int(&vmbus_chan_nextcpu, 1) % mp_ncpus;
1257 	vmbus_chan_cpu_set(chan, cpu);
1258 }
1259 
1260 static void
1261 vmbus_chan_cpu_default(struct vmbus_channel *chan)
1262 {
1263 	/*
1264 	 * By default, pin the channel to cpu0.  Devices having
1265 	 * special channel-cpu mapping requirement should call
1266 	 * vmbus_chan_cpu_{set,rr}().
1267 	 */
1268 	vmbus_chan_cpu_set(chan, 0);
1269 }
1270 
1271 static void
1272 vmbus_chan_msgproc_choffer(struct vmbus_softc *sc,
1273     const struct vmbus_message *msg)
1274 {
1275 	const struct vmbus_chanmsg_choffer *offer;
1276 	struct vmbus_channel *chan;
1277 	task_fn_t *detach_fn, *attach_fn;
1278 	int error;
1279 
1280 	offer = (const struct vmbus_chanmsg_choffer *)msg->msg_data;
1281 
1282 	chan = vmbus_chan_alloc(sc);
1283 	if (chan == NULL) {
1284 		device_printf(sc->vmbus_dev, "allocate chan%u failed\n",
1285 		    offer->chm_chanid);
1286 		return;
1287 	}
1288 
1289 	chan->ch_id = offer->chm_chanid;
1290 	chan->ch_subidx = offer->chm_subidx;
1291 	chan->ch_guid_type = offer->chm_chtype;
1292 	chan->ch_guid_inst = offer->chm_chinst;
1293 
1294 	/* Batch reading is on by default */
1295 	chan->ch_flags |= VMBUS_CHAN_FLAG_BATCHREAD;
1296 
1297 	chan->ch_monprm->mp_connid = VMBUS_CONNID_EVENT;
1298 	if (sc->vmbus_version != VMBUS_VERSION_WS2008)
1299 		chan->ch_monprm->mp_connid = offer->chm_connid;
1300 
1301 	if (offer->chm_flags1 & VMBUS_CHOFFER_FLAG1_HASMNF) {
1302 		int trig_idx;
1303 
1304 		/*
1305 		 * Setup MNF stuffs.
1306 		 */
1307 		chan->ch_txflags |= VMBUS_CHAN_TXF_HASMNF;
1308 
1309 		trig_idx = offer->chm_montrig / VMBUS_MONTRIG_LEN;
1310 		if (trig_idx >= VMBUS_MONTRIGS_MAX)
1311 			panic("invalid monitor trigger %u", offer->chm_montrig);
1312 		chan->ch_montrig =
1313 		    &sc->vmbus_mnf2->mnf_trigs[trig_idx].mt_pending;
1314 
1315 		chan->ch_montrig_mask =
1316 		    1 << (offer->chm_montrig % VMBUS_MONTRIG_LEN);
1317 	}
1318 
1319 	/*
1320 	 * Setup event flag.
1321 	 */
1322 	chan->ch_evtflag =
1323 	    &sc->vmbus_tx_evtflags[chan->ch_id >> VMBUS_EVTFLAG_SHIFT];
1324 	chan->ch_evtflag_mask = 1UL << (chan->ch_id & VMBUS_EVTFLAG_MASK);
1325 
1326 	/*
1327 	 * Setup attach and detach tasks.
1328 	 */
1329 	if (VMBUS_CHAN_ISPRIMARY(chan)) {
1330 		chan->ch_mgmt_tq = sc->vmbus_devtq;
1331 		attach_fn = vmbus_prichan_attach_task;
1332 		detach_fn = vmbus_prichan_detach_task;
1333 	} else {
1334 		chan->ch_mgmt_tq = sc->vmbus_subchtq;
1335 		attach_fn = vmbus_subchan_attach_task;
1336 		detach_fn = vmbus_subchan_detach_task;
1337 	}
1338 	TASK_INIT(&chan->ch_attach_task, 0, attach_fn, chan);
1339 	TASK_INIT(&chan->ch_detach_task, 0, detach_fn, chan);
1340 
1341 	/* Select default cpu for this channel. */
1342 	vmbus_chan_cpu_default(chan);
1343 
1344 	error = vmbus_chan_add(chan);
1345 	if (error) {
1346 		device_printf(sc->vmbus_dev, "add chan%u failed: %d\n",
1347 		    chan->ch_id, error);
1348 		vmbus_chan_free(chan);
1349 		return;
1350 	}
1351 	taskqueue_enqueue(chan->ch_mgmt_tq, &chan->ch_attach_task);
1352 }
1353 
1354 static void
1355 vmbus_chan_msgproc_chrescind(struct vmbus_softc *sc,
1356     const struct vmbus_message *msg)
1357 {
1358 	const struct vmbus_chanmsg_chrescind *note;
1359 	struct vmbus_channel *chan;
1360 
1361 	note = (const struct vmbus_chanmsg_chrescind *)msg->msg_data;
1362 	if (note->chm_chanid > VMBUS_CHAN_MAX) {
1363 		device_printf(sc->vmbus_dev, "invalid rescinded chan%u\n",
1364 		    note->chm_chanid);
1365 		return;
1366 	}
1367 
1368 	if (bootverbose) {
1369 		device_printf(sc->vmbus_dev, "chan%u rescinded\n",
1370 		    note->chm_chanid);
1371 	}
1372 
1373 	/*
1374 	 * Find and remove the target channel from the channel list.
1375 	 */
1376 	mtx_lock(&sc->vmbus_chan_lock);
1377 	TAILQ_FOREACH(chan, &sc->vmbus_chans, ch_link) {
1378 		if (chan->ch_id == note->chm_chanid)
1379 			break;
1380 	}
1381 	if (chan == NULL) {
1382 		mtx_unlock(&sc->vmbus_chan_lock);
1383 		device_printf(sc->vmbus_dev, "chan%u is not offered\n",
1384 		    note->chm_chanid);
1385 		return;
1386 	}
1387 	vmbus_chan_rem_list(sc, chan);
1388 	mtx_unlock(&sc->vmbus_chan_lock);
1389 
1390 	if (VMBUS_CHAN_ISPRIMARY(chan)) {
1391 		/*
1392 		 * The target channel is a primary channel; remove the
1393 		 * target channel from the primary channel list now,
1394 		 * instead of later, so that it will not be found by
1395 		 * other sub-channel offers, which are processed in
1396 		 * this thread.
1397 		 */
1398 		mtx_lock(&sc->vmbus_prichan_lock);
1399 		vmbus_chan_rem_prilist(sc, chan);
1400 		mtx_unlock(&sc->vmbus_prichan_lock);
1401 	}
1402 
1403 	/* Detach the target channel. */
1404 	taskqueue_enqueue(chan->ch_mgmt_tq, &chan->ch_detach_task);
1405 }
1406 
1407 static int
1408 vmbus_chan_release(struct vmbus_channel *chan)
1409 {
1410 	struct vmbus_softc *sc = chan->ch_vmbus;
1411 	struct vmbus_chanmsg_chfree *req;
1412 	struct vmbus_msghc *mh;
1413 	int error;
1414 
1415 	mh = vmbus_msghc_get(sc, sizeof(*req));
1416 	if (mh == NULL) {
1417 		device_printf(sc->vmbus_dev, "can not get msg hypercall for "
1418 		    "chfree(chan%u)\n", chan->ch_id);
1419 		return (ENXIO);
1420 	}
1421 
1422 	req = vmbus_msghc_dataptr(mh);
1423 	req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_CHFREE;
1424 	req->chm_chanid = chan->ch_id;
1425 
1426 	error = vmbus_msghc_exec_noresult(mh);
1427 	vmbus_msghc_put(sc, mh);
1428 
1429 	if (error) {
1430 		device_printf(sc->vmbus_dev, "chfree(chan%u) failed: %d",
1431 		    chan->ch_id, error);
1432 	} else {
1433 		if (bootverbose) {
1434 			device_printf(sc->vmbus_dev, "chan%u freed\n",
1435 			    chan->ch_id);
1436 		}
1437 	}
1438 	return (error);
1439 }
1440 
1441 static void
1442 vmbus_prichan_detach_task(void *xchan, int pending __unused)
1443 {
1444 	struct vmbus_channel *chan = xchan;
1445 
1446 	KASSERT(VMBUS_CHAN_ISPRIMARY(chan),
1447 	    ("chan%u is not primary channel", chan->ch_id));
1448 
1449 	/* Delete and detach the device associated with this channel. */
1450 	vmbus_delete_child(chan);
1451 
1452 	/* Release this channel (back to vmbus). */
1453 	vmbus_chan_release(chan);
1454 
1455 	/* Free this channel's resource. */
1456 	vmbus_chan_free(chan);
1457 }
1458 
1459 static void
1460 vmbus_subchan_detach_task(void *xchan, int pending __unused)
1461 {
1462 	struct vmbus_channel *chan = xchan;
1463 	struct vmbus_channel *pri_chan = chan->ch_prichan;
1464 
1465 	KASSERT(!VMBUS_CHAN_ISPRIMARY(chan),
1466 	    ("chan%u is primary channel", chan->ch_id));
1467 
1468 	/* Release this channel (back to vmbus). */
1469 	vmbus_chan_release(chan);
1470 
1471 	/* Unlink from its primary channel's sub-channel list. */
1472 	mtx_lock(&pri_chan->ch_subchan_lock);
1473 	vmbus_chan_rem_sublist(pri_chan, chan);
1474 	mtx_unlock(&pri_chan->ch_subchan_lock);
1475 	/* Notify anyone that is waiting for this sub-channel to vanish. */
1476 	wakeup(pri_chan);
1477 
1478 	/* Free this channel's resource. */
1479 	vmbus_chan_free(chan);
1480 }
1481 
1482 static void
1483 vmbus_prichan_attach_task(void *xchan, int pending __unused)
1484 {
1485 
1486 	/*
1487 	 * Add device for this primary channel.
1488 	 */
1489 	vmbus_add_child(xchan);
1490 }
1491 
1492 static void
1493 vmbus_subchan_attach_task(void *xchan __unused, int pending __unused)
1494 {
1495 
1496 	/* Nothing */
1497 }
1498 
1499 void
1500 vmbus_chan_destroy_all(struct vmbus_softc *sc)
1501 {
1502 
1503 	/*
1504 	 * Detach all devices and destroy the corresponding primary
1505 	 * channels.
1506 	 */
1507 	for (;;) {
1508 		struct vmbus_channel *chan;
1509 
1510 		mtx_lock(&sc->vmbus_chan_lock);
1511 		TAILQ_FOREACH(chan, &sc->vmbus_chans, ch_link) {
1512 			if (VMBUS_CHAN_ISPRIMARY(chan))
1513 				break;
1514 		}
1515 		if (chan == NULL) {
1516 			/* No more primary channels; done. */
1517 			mtx_unlock(&sc->vmbus_chan_lock);
1518 			break;
1519 		}
1520 		vmbus_chan_rem_list(sc, chan);
1521 		mtx_unlock(&sc->vmbus_chan_lock);
1522 
1523 		mtx_lock(&sc->vmbus_prichan_lock);
1524 		vmbus_chan_rem_prilist(sc, chan);
1525 		mtx_unlock(&sc->vmbus_prichan_lock);
1526 
1527 		taskqueue_enqueue(chan->ch_mgmt_tq, &chan->ch_detach_task);
1528 	}
1529 }
1530 
1531 /*
1532  * The channel whose vcpu binding is closest to the currect vcpu will
1533  * be selected.
1534  * If no multi-channel, always select primary channel.
1535  */
1536 struct vmbus_channel *
1537 vmbus_chan_cpu2chan(struct vmbus_channel *prichan, int cpu)
1538 {
1539 	struct vmbus_channel *sel, *chan;
1540 	uint32_t vcpu, sel_dist;
1541 
1542 	KASSERT(cpu >= 0 && cpu < mp_ncpus, ("invalid cpuid %d", cpu));
1543 	if (TAILQ_EMPTY(&prichan->ch_subchans))
1544 		return prichan;
1545 
1546 	vcpu = VMBUS_PCPU_GET(prichan->ch_vmbus, vcpuid, cpu);
1547 
1548 #define CHAN_VCPU_DIST(ch, vcpu)		\
1549 	(((ch)->ch_vcpuid > (vcpu)) ?		\
1550 	 ((ch)->ch_vcpuid - (vcpu)) : ((vcpu) - (ch)->ch_vcpuid))
1551 
1552 #define CHAN_SELECT(ch)				\
1553 do {						\
1554 	sel = ch;				\
1555 	sel_dist = CHAN_VCPU_DIST(ch, vcpu);	\
1556 } while (0)
1557 
1558 	CHAN_SELECT(prichan);
1559 
1560 	mtx_lock(&prichan->ch_subchan_lock);
1561 	TAILQ_FOREACH(chan, &prichan->ch_subchans, ch_sublink) {
1562 		uint32_t dist;
1563 
1564 		KASSERT(chan->ch_stflags & VMBUS_CHAN_ST_OPENED,
1565 		    ("chan%u is not opened", chan->ch_id));
1566 
1567 		if (chan->ch_vcpuid == vcpu) {
1568 			/* Exact match; done */
1569 			CHAN_SELECT(chan);
1570 			break;
1571 		}
1572 
1573 		dist = CHAN_VCPU_DIST(chan, vcpu);
1574 		if (sel_dist <= dist) {
1575 			/* Far or same distance; skip */
1576 			continue;
1577 		}
1578 
1579 		/* Select the closer channel. */
1580 		CHAN_SELECT(chan);
1581 	}
1582 	mtx_unlock(&prichan->ch_subchan_lock);
1583 
1584 #undef CHAN_SELECT
1585 #undef CHAN_VCPU_DIST
1586 
1587 	return sel;
1588 }
1589 
1590 struct vmbus_channel **
1591 vmbus_subchan_get(struct vmbus_channel *pri_chan, int subchan_cnt)
1592 {
1593 	struct vmbus_channel **ret, *chan;
1594 	int i;
1595 
1596 	KASSERT(subchan_cnt > 0, ("invalid sub-channel count %d", subchan_cnt));
1597 
1598 	ret = malloc(subchan_cnt * sizeof(struct vmbus_channel *), M_TEMP,
1599 	    M_WAITOK);
1600 
1601 	mtx_lock(&pri_chan->ch_subchan_lock);
1602 
1603 	while (pri_chan->ch_subchan_cnt < subchan_cnt)
1604 		mtx_sleep(pri_chan, &pri_chan->ch_subchan_lock, 0, "subch", 0);
1605 
1606 	i = 0;
1607 	TAILQ_FOREACH(chan, &pri_chan->ch_subchans, ch_sublink) {
1608 		/* TODO: refcnt chan */
1609 		ret[i] = chan;
1610 
1611 		++i;
1612 		if (i == subchan_cnt)
1613 			break;
1614 	}
1615 	KASSERT(i == subchan_cnt, ("invalid subchan count %d, should be %d",
1616 	    pri_chan->ch_subchan_cnt, subchan_cnt));
1617 
1618 	mtx_unlock(&pri_chan->ch_subchan_lock);
1619 
1620 	return ret;
1621 }
1622 
1623 void
1624 vmbus_subchan_rel(struct vmbus_channel **subchan, int subchan_cnt __unused)
1625 {
1626 
1627 	free(subchan, M_TEMP);
1628 }
1629 
1630 void
1631 vmbus_subchan_drain(struct vmbus_channel *pri_chan)
1632 {
1633 	mtx_lock(&pri_chan->ch_subchan_lock);
1634 	while (pri_chan->ch_subchan_cnt > 0)
1635 		mtx_sleep(pri_chan, &pri_chan->ch_subchan_lock, 0, "dsubch", 0);
1636 	mtx_unlock(&pri_chan->ch_subchan_lock);
1637 }
1638 
1639 void
1640 vmbus_chan_msgproc(struct vmbus_softc *sc, const struct vmbus_message *msg)
1641 {
1642 	vmbus_chanmsg_proc_t msg_proc;
1643 	uint32_t msg_type;
1644 
1645 	msg_type = ((const struct vmbus_chanmsg_hdr *)msg->msg_data)->chm_type;
1646 	KASSERT(msg_type < VMBUS_CHANMSG_TYPE_MAX,
1647 	    ("invalid message type %u", msg_type));
1648 
1649 	msg_proc = vmbus_chan_msgprocs[msg_type];
1650 	if (msg_proc != NULL)
1651 		msg_proc(sc, msg);
1652 }
1653 
1654 void
1655 vmbus_chan_set_readbatch(struct vmbus_channel *chan, bool on)
1656 {
1657 	if (!on)
1658 		chan->ch_flags &= ~VMBUS_CHAN_FLAG_BATCHREAD;
1659 	else
1660 		chan->ch_flags |= VMBUS_CHAN_FLAG_BATCHREAD;
1661 }
1662 
1663 uint32_t
1664 vmbus_chan_id(const struct vmbus_channel *chan)
1665 {
1666 	return chan->ch_id;
1667 }
1668 
1669 uint32_t
1670 vmbus_chan_subidx(const struct vmbus_channel *chan)
1671 {
1672 	return chan->ch_subidx;
1673 }
1674 
1675 bool
1676 vmbus_chan_is_primary(const struct vmbus_channel *chan)
1677 {
1678 	if (VMBUS_CHAN_ISPRIMARY(chan))
1679 		return true;
1680 	else
1681 		return false;
1682 }
1683 
1684 const struct hyperv_guid *
1685 vmbus_chan_guid_inst(const struct vmbus_channel *chan)
1686 {
1687 	return &chan->ch_guid_inst;
1688 }
1689 
1690 int
1691 vmbus_chan_prplist_nelem(int br_size, int prpcnt_max, int dlen_max)
1692 {
1693 	int elem_size;
1694 
1695 	elem_size = __offsetof(struct vmbus_chanpkt_prplist,
1696 	    cp_range[0].gpa_page[prpcnt_max]);
1697 	elem_size += dlen_max;
1698 	elem_size = VMBUS_CHANPKT_TOTLEN(elem_size);
1699 
1700 	return (vmbus_br_nelem(br_size, elem_size));
1701 }
1702 
1703 bool
1704 vmbus_chan_tx_empty(const struct vmbus_channel *chan)
1705 {
1706 
1707 	return (vmbus_txbr_empty(&chan->ch_txbr));
1708 }
1709 
1710 bool
1711 vmbus_chan_rx_empty(const struct vmbus_channel *chan)
1712 {
1713 
1714 	return (vmbus_rxbr_empty(&chan->ch_rxbr));
1715 }
1716