xref: /netbsd-src/sys/dev/hyperv/vmbus.c (revision 7330f729ccf0bd976a06f95fad452fe774fc7fd1)
1 /*	$NetBSD: vmbus.c,v 1.4 2019/07/09 10:07:11 nakayama Exp $	*/
2 /*	$OpenBSD: hyperv.c,v 1.43 2017/06/27 13:56:15 mikeb Exp $	*/
3 
4 /*-
5  * Copyright (c) 2009-2012 Microsoft Corp.
6  * Copyright (c) 2012 NetApp Inc.
7  * Copyright (c) 2012 Citrix Inc.
8  * Copyright (c) 2016 Mike Belopuhov <mike@esdenera.com>
9  * All rights reserved.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice unmodified, this list of conditions, and the following
16  *    disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in the
19  *    documentation and/or other materials provided with the distribution.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
22  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
23  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
24  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
26  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
27  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
28  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
29  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
30  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * The OpenBSD port was done under funding by Esdenera Networks GmbH.
35  */
36 
37 #include <sys/cdefs.h>
38 __KERNEL_RCSID(0, "$NetBSD: vmbus.c,v 1.4 2019/07/09 10:07:11 nakayama Exp $");
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/device.h>
43 #include <sys/atomic.h>
44 #include <sys/bitops.h>
45 #include <sys/bus.h>
46 #include <sys/cpu.h>
47 #include <sys/intr.h>
48 #include <sys/kmem.h>
49 #include <sys/module.h>
50 #include <sys/mutex.h>
51 #include <sys/xcall.h>
52 
53 #include <uvm/uvm_extern.h>
54 
55 #include <dev/hyperv/vmbusvar.h>
56 
57 #define VMBUS_GPADL_START		0xffff /* 0x10000 effectively */
58 
59 /* Command submission flags */
60 #define HCF_SLEEPOK	0x0000
61 #define HCF_NOSLEEP	0x0002	/* M_NOWAIT */
62 #define HCF_NOREPLY	0x0004
63 
64 static void	vmbus_attach_deferred(device_t);
65 static int	vmbus_alloc_dma(struct vmbus_softc *);
66 static void	vmbus_free_dma(struct vmbus_softc *);
67 static int	vmbus_init_interrupts(struct vmbus_softc *);
68 static void	vmbus_deinit_interrupts(struct vmbus_softc *);
69 static void	vmbus_init_synic(void *, void *);
70 static void	vmbus_deinit_synic(void *, void *);
71 
72 static int	vmbus_connect(struct vmbus_softc *);
73 static int	vmbus_cmd(struct vmbus_softc *, void *, size_t, void *, size_t,
74 		    int);
75 static int	vmbus_start(struct vmbus_softc *, struct vmbus_msg *, paddr_t);
76 static int	vmbus_reply(struct vmbus_softc *, struct vmbus_msg *);
77 static void	vmbus_wait(struct vmbus_softc *,
78 		    int (*done)(struct vmbus_softc *, struct vmbus_msg *),
79 		    struct vmbus_msg *, void *, const char *);
80 static uint16_t vmbus_intr_signal(struct vmbus_softc *, paddr_t);
81 static void	vmbus_event_proc(void *, struct cpu_info *);
82 static void	vmbus_event_proc_compat(void *, struct cpu_info *);
83 static void	vmbus_message_proc(void *, struct cpu_info *);
84 static void	vmbus_message_softintr(void *);
85 static void	vmbus_channel_response(struct vmbus_softc *,
86 		    struct vmbus_chanmsg_hdr *);
87 static void	vmbus_channel_offer(struct vmbus_softc *,
88 		    struct vmbus_chanmsg_hdr *);
89 static void	vmbus_channel_rescind(struct vmbus_softc *,
90 		    struct vmbus_chanmsg_hdr *);
91 static void	vmbus_channel_delivered(struct vmbus_softc *,
92 		    struct vmbus_chanmsg_hdr *);
93 static int	vmbus_channel_scan(struct vmbus_softc *);
94 static void	vmbus_channel_cpu_default(struct vmbus_channel *);
95 static void	vmbus_process_offer(struct vmbus_softc *, struct vmbus_offer *);
96 static struct vmbus_channel *
97 		vmbus_channel_lookup(struct vmbus_softc *, uint32_t);
98 static int	vmbus_channel_ring_create(struct vmbus_channel *, uint32_t);
99 static void	vmbus_channel_ring_destroy(struct vmbus_channel *);
100 static void	vmbus_channel_pause(struct vmbus_channel *);
101 static uint32_t	vmbus_channel_unpause(struct vmbus_channel *);
102 static uint32_t	vmbus_channel_ready(struct vmbus_channel *);
103 static int	vmbus_attach_icdevs(struct vmbus_softc *);
104 static int	vmbus_attach_devices(struct vmbus_softc *);
105 
106 static struct vmbus_softc *vmbus_sc;
107 
108 static const struct {
109 	int	hmd_response;
110 	int	hmd_request;
111 	void	(*hmd_handler)(struct vmbus_softc *,
112 		    struct vmbus_chanmsg_hdr *);
113 } vmbus_msg_dispatch[] = {
114 	{ 0,					0, NULL },
115 	{ VMBUS_CHANMSG_CHOFFER,		0, vmbus_channel_offer },
116 	{ VMBUS_CHANMSG_CHRESCIND,		0, vmbus_channel_rescind },
117 	{ VMBUS_CHANMSG_CHREQUEST,		VMBUS_CHANMSG_CHOFFER, NULL },
118 	{ VMBUS_CHANMSG_CHOFFER_DONE,		0, vmbus_channel_delivered },
119 	{ VMBUS_CHANMSG_CHOPEN,			0, NULL },
120 	{ VMBUS_CHANMSG_CHOPEN_RESP,		VMBUS_CHANMSG_CHOPEN,
121 	  vmbus_channel_response },
122 	{ VMBUS_CHANMSG_CHCLOSE,		0, NULL },
123 	{ VMBUS_CHANMSG_GPADL_CONN,		0, NULL },
124 	{ VMBUS_CHANMSG_GPADL_SUBCONN,		0, NULL },
125 	{ VMBUS_CHANMSG_GPADL_CONNRESP,		VMBUS_CHANMSG_GPADL_CONN,
126 	  vmbus_channel_response },
127 	{ VMBUS_CHANMSG_GPADL_DISCONN,		0, NULL },
128 	{ VMBUS_CHANMSG_GPADL_DISCONNRESP,	VMBUS_CHANMSG_GPADL_DISCONN,
129 	  vmbus_channel_response },
130 	{ VMBUS_CHANMSG_CHFREE,			0, NULL },
131 	{ VMBUS_CHANMSG_CONNECT,		0, NULL },
132 	{ VMBUS_CHANMSG_CONNECT_RESP,		VMBUS_CHANMSG_CONNECT,
133 	  vmbus_channel_response },
134 	{ VMBUS_CHANMSG_DISCONNECT,		0, NULL },
135 };
136 
137 const struct hyperv_guid hyperv_guid_network = {
138 	{ 0x63, 0x51, 0x61, 0xf8, 0x3e, 0xdf, 0xc5, 0x46,
139 	  0x91, 0x3f, 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e }
140 };
141 
142 const struct hyperv_guid hyperv_guid_ide = {
143 	{ 0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
144 	  0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5 }
145 };
146 
147 const struct hyperv_guid hyperv_guid_scsi = {
148 	{ 0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
149 	  0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f }
150 };
151 
152 const struct hyperv_guid hyperv_guid_shutdown = {
153 	{ 0x31, 0x60, 0x0b, 0x0e, 0x13, 0x52, 0x34, 0x49,
154 	  0x81, 0x8b, 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb }
155 };
156 
157 const struct hyperv_guid hyperv_guid_timesync = {
158 	{ 0x30, 0xe6, 0x27, 0x95, 0xae, 0xd0, 0x7b, 0x49,
159 	  0xad, 0xce, 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf }
160 };
161 
162 const struct hyperv_guid hyperv_guid_heartbeat = {
163 	{ 0x39, 0x4f, 0x16, 0x57, 0x15, 0x91, 0x78, 0x4e,
164 	  0xab, 0x55, 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d }
165 };
166 
167 const struct hyperv_guid hyperv_guid_kvp = {
168 	{ 0xe7, 0xf4, 0xa0, 0xa9, 0x45, 0x5a, 0x96, 0x4d,
169 	  0xb8, 0x27, 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6 }
170 };
171 
172 const struct hyperv_guid hyperv_guid_vss = {
173 	{ 0x29, 0x2e, 0xfa, 0x35, 0x23, 0xea, 0x36, 0x42,
174 	  0x96, 0xae, 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40 }
175 };
176 
177 const struct hyperv_guid hyperv_guid_dynmem = {
178 	{ 0xdc, 0x74, 0x50, 0x52, 0x85, 0x89, 0xe2, 0x46,
179 	  0x80, 0x57, 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02 }
180 };
181 
182 const struct hyperv_guid hyperv_guid_mouse = {
183 	{ 0x9e, 0xb6, 0xa8, 0xcf, 0x4a, 0x5b, 0xc0, 0x4c,
184 	  0xb9, 0x8b, 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a }
185 };
186 
187 const struct hyperv_guid hyperv_guid_kbd = {
188 	{ 0x6d, 0xad, 0x12, 0xf9, 0x17, 0x2b, 0xea, 0x48,
189 	  0xbd, 0x65, 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84 }
190 };
191 
192 const struct hyperv_guid hyperv_guid_video = {
193 	{ 0x02, 0x78, 0x0a, 0xda, 0x77, 0xe3, 0xac, 0x4a,
194 	  0x8e, 0x77, 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8 }
195 };
196 
197 const struct hyperv_guid hyperv_guid_fc = {
198 	{ 0x4a, 0xcc, 0x9b, 0x2f, 0x69, 0x00, 0xf3, 0x4a,
199 	  0xb7, 0x6b, 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda }
200 };
201 
202 const struct hyperv_guid hyperv_guid_fcopy = {
203 	{ 0xe3, 0x4b, 0xd1, 0x34, 0xe4, 0xde, 0xc8, 0x41,
204 	  0x9a, 0xe7, 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92 }
205 };
206 
207 const struct hyperv_guid hyperv_guid_pcie = {
208 	{ 0x1d, 0xf6, 0xc4, 0x44, 0x44, 0x44, 0x00, 0x44,
209 	  0x9d, 0x52, 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f }
210 };
211 
212 const struct hyperv_guid hyperv_guid_netdir = {
213 	{ 0x3d, 0xaf, 0x2e, 0x8c, 0xa7, 0x32, 0x09, 0x4b,
214 	  0xab, 0x99, 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01 }
215 };
216 
217 const struct hyperv_guid hyperv_guid_rdesktop = {
218 	{ 0xf4, 0xac, 0x6a, 0x27, 0x15, 0xac, 0x6c, 0x42,
219 	  0x98, 0xdd, 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe }
220 };
221 
222 /* Automatic Virtual Machine Activation (AVMA) Services */
223 const struct hyperv_guid hyperv_guid_avma1 = {
224 	{ 0x55, 0xb2, 0x87, 0x44, 0x8c, 0xb8, 0x3f, 0x40,
225 	  0xbb, 0x51, 0xd1, 0xf6, 0x9c, 0xf1, 0x7f, 0x87 }
226 };
227 
228 const struct hyperv_guid hyperv_guid_avma2 = {
229 	{ 0xf4, 0xba, 0x75, 0x33, 0x15, 0x9e, 0x30, 0x4b,
230 	  0xb7, 0x65, 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b }
231 };
232 
233 const struct hyperv_guid hyperv_guid_avma3 = {
234 	{ 0xa0, 0x1f, 0x22, 0x99, 0xad, 0x24, 0xe2, 0x11,
235 	  0xbe, 0x98, 0x00, 0x1a, 0xa0, 0x1b, 0xbf, 0x6e }
236 };
237 
238 const struct hyperv_guid hyperv_guid_avma4 = {
239 	{ 0x16, 0x57, 0xe6, 0xf8, 0xb3, 0x3c, 0x06, 0x4a,
240 	  0x9a, 0x60, 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5 }
241 };
242 
243 int
244 vmbus_match(device_t parent, cfdata_t cf, void *aux)
245 {
246 
247 	if (cf->cf_unit != 0 ||
248 	    !hyperv_hypercall_enabled() ||
249 	    !hyperv_synic_supported())
250 		return 0;
251 
252 	return 1;
253 }
254 
255 int
256 vmbus_attach(struct vmbus_softc *sc)
257 {
258 
259 	aprint_naive("\n");
260 	aprint_normal(": Hyper-V VMBus\n");
261 
262 	vmbus_sc = sc;
263 
264 	sc->sc_msgpool = pool_cache_init(sizeof(struct vmbus_msg), 8, 0, 0,
265 	    "hvmsg", NULL, IPL_NET, NULL, NULL, NULL);
266 	hyperv_set_message_proc(vmbus_message_proc, sc);
267 
268 	if (vmbus_alloc_dma(sc))
269 		goto cleanup;
270 
271 	if (vmbus_init_interrupts(sc))
272 		goto cleanup;
273 
274 	if (vmbus_connect(sc))
275 		goto cleanup;
276 
277 	aprint_normal_dev(sc->sc_dev, "protocol %d.%d\n",
278 	    VMBUS_VERSION_MAJOR(sc->sc_proto),
279 	    VMBUS_VERSION_MINOR(sc->sc_proto));
280 
281 	if (sc->sc_proto == VMBUS_VERSION_WS2008 ||
282 	    sc->sc_proto == VMBUS_VERSION_WIN7) {
283 		hyperv_set_event_proc(vmbus_event_proc_compat, sc);
284 		sc->sc_channel_max = VMBUS_CHAN_MAX_COMPAT;
285 	} else {
286 		hyperv_set_event_proc(vmbus_event_proc, sc);
287 		sc->sc_channel_max = VMBUS_CHAN_MAX;
288 	}
289 
290 	if (vmbus_channel_scan(sc))
291 		goto cleanup;
292 
293 	/* Attach heartbeat, KVP and other "internal" services */
294 	vmbus_attach_icdevs(sc);
295 
296 	/* Attach devices with external drivers */
297 	vmbus_attach_devices(sc);
298 
299 	config_interrupts(sc->sc_dev, vmbus_attach_deferred);
300 
301 	return 0;
302 
303 cleanup:
304 	vmbus_deinit_interrupts(sc);
305 	vmbus_free_dma(sc);
306 	return -1;
307 }
308 
309 static void
310 vmbus_attach_deferred(device_t self)
311 {
312 	struct vmbus_softc *sc = device_private(self);
313 
314 	xc_wait(xc_broadcast(0, vmbus_init_synic, sc, NULL));
315 }
316 
317 int
318 vmbus_detach(struct vmbus_softc *sc, int flags)
319 {
320 
321 	vmbus_deinit_interrupts(sc);
322 	vmbus_free_dma(sc);
323 
324 	return 0;
325 }
326 
327 static int
328 vmbus_alloc_dma(struct vmbus_softc *sc)
329 {
330 	CPU_INFO_ITERATOR cii;
331 	struct cpu_info *ci;
332 	struct vmbus_percpu_data *pd;
333 	int i;
334 
335 	/*
336 	 * Per-CPU messages and event flags.
337 	 */
338 	for (CPU_INFO_FOREACH(cii, ci)) {
339 		pd = &sc->sc_percpu[cpu_index(ci)];
340 
341 		pd->simp = hyperv_dma_alloc(sc->sc_dmat, &pd->simp_dma,
342 		    PAGE_SIZE, PAGE_SIZE, 0, 1);
343 		if (pd->simp == NULL)
344 			return ENOMEM;
345 
346 		pd->siep = hyperv_dma_alloc(sc->sc_dmat, &pd->siep_dma,
347 		    PAGE_SIZE, PAGE_SIZE, 0, 1);
348 		if (pd->siep == NULL)
349 			return ENOMEM;
350 	}
351 
352 	sc->sc_events = hyperv_dma_alloc(sc->sc_dmat, &sc->sc_events_dma,
353 	    PAGE_SIZE, PAGE_SIZE, 0, 1);
354 	if (sc->sc_events == NULL)
355 		return ENOMEM;
356 	sc->sc_wevents = (u_long *)sc->sc_events;
357 	sc->sc_revents = (u_long *)((uint8_t *)sc->sc_events + (PAGE_SIZE / 2));
358 
359 	for (i = 0; i < __arraycount(sc->sc_monitor); i++) {
360 		sc->sc_monitor[i] = hyperv_dma_alloc(sc->sc_dmat,
361 		    &sc->sc_monitor_dma[i], PAGE_SIZE, PAGE_SIZE, 0, 1);
362 		if (sc->sc_monitor[i] == NULL)
363 			return ENOMEM;
364 	}
365 
366 	return 0;
367 }
368 
369 static void
370 vmbus_free_dma(struct vmbus_softc *sc)
371 {
372 	CPU_INFO_ITERATOR cii;
373 	struct cpu_info *ci;
374 	int i;
375 
376 	if (sc->sc_events != NULL) {
377 		sc->sc_events = sc->sc_wevents = sc->sc_revents = NULL;
378 		hyperv_dma_free(sc->sc_dmat, &sc->sc_events_dma);
379 	}
380 
381 	for (i = 0; i < __arraycount(sc->sc_monitor); i++) {
382 		sc->sc_monitor[i] = NULL;
383 		hyperv_dma_free(sc->sc_dmat, &sc->sc_monitor_dma[i]);
384 	}
385 
386 	for (CPU_INFO_FOREACH(cii, ci)) {
387 		struct vmbus_percpu_data *pd = &sc->sc_percpu[cpu_index(ci)];
388 
389 		if (pd->simp != NULL) {
390 			pd->simp = NULL;
391 			hyperv_dma_free(sc->sc_dmat, &pd->simp_dma);
392 		}
393 		if (pd->siep != NULL) {
394 			pd->siep = NULL;
395 			hyperv_dma_free(sc->sc_dmat, &pd->siep_dma);
396 		}
397 	}
398 }
399 
400 static int
401 vmbus_init_interrupts(struct vmbus_softc *sc)
402 {
403 
404 	TAILQ_INIT(&sc->sc_reqs);
405 	mutex_init(&sc->sc_req_lock, MUTEX_DEFAULT, IPL_NET);
406 
407 	TAILQ_INIT(&sc->sc_rsps);
408 	mutex_init(&sc->sc_rsp_lock, MUTEX_DEFAULT, IPL_NET);
409 
410 	sc->sc_proto = VMBUS_VERSION_WS2008;
411 
412 	/* XXX event_tq */
413 
414 	sc->sc_msg_sih = softint_establish(SOFTINT_NET | SOFTINT_MPSAFE,
415 	    vmbus_message_softintr, sc);
416 	if (sc->sc_msg_sih == NULL)
417 		return -1;
418 
419 	vmbus_init_interrupts_md(sc);
420 
421 	kcpuset_create(&sc->sc_intr_cpuset, true);
422 	if (cold) {
423 		/* Initialize other CPUs later. */
424 		vmbus_init_synic(sc, NULL);
425 	} else
426 		xc_wait(xc_broadcast(0, vmbus_init_synic, sc, NULL));
427 	atomic_or_32(&sc->sc_flags, VMBUS_SCFLAG_SYNIC);
428 
429 	return 0;
430 }
431 
432 static void
433 vmbus_deinit_interrupts(struct vmbus_softc *sc)
434 {
435 
436 	if (ISSET(sc->sc_flags, VMBUS_SCFLAG_SYNIC)) {
437 		if (cold)
438 			vmbus_deinit_synic(sc, NULL);
439 		else
440 			xc_wait(xc_broadcast(0, vmbus_deinit_synic, sc, NULL));
441 		atomic_and_32(&sc->sc_flags, (uint32_t)~VMBUS_SCFLAG_SYNIC);
442 	}
443 
444 	/* XXX event_tq */
445 
446 	if (sc->sc_msg_sih != NULL) {
447 		softint_disestablish(sc->sc_msg_sih);
448 		sc->sc_msg_sih = NULL;
449 	}
450 
451 	vmbus_deinit_interrupts_md(sc);
452 }
453 
454 static void
455 vmbus_init_synic(void *arg1, void *arg2)
456 {
457 	struct vmbus_softc *sc = arg1;
458 	cpuid_t cpu;
459 	int s;
460 
461 	s = splhigh();
462 
463 	cpu = cpu_index(curcpu());
464 	if (!kcpuset_isset(sc->sc_intr_cpuset, cpu)) {
465 		kcpuset_atomic_set(sc->sc_intr_cpuset, cpu);
466 		vmbus_init_synic_md(sc, cpu);
467 	}
468 
469 	splx(s);
470 }
471 
472 static void
473 vmbus_deinit_synic(void *arg1, void *arg2)
474 {
475 	struct vmbus_softc *sc = arg1;
476 	cpuid_t cpu;
477 	int s;
478 
479 	s = splhigh();
480 
481 	cpu = cpu_index(curcpu());
482 	if (kcpuset_isset(sc->sc_intr_cpuset, cpu)) {
483 		vmbus_deinit_synic_md(sc, cpu);
484 		kcpuset_atomic_clear(sc->sc_intr_cpuset, cpu);
485 	}
486 
487 	splx(s);
488 }
489 
490 static int
491 vmbus_connect(struct vmbus_softc *sc)
492 {
493 	static const uint32_t versions[] = {
494 		VMBUS_VERSION_WIN8_1,
495 		VMBUS_VERSION_WIN8,
496 		VMBUS_VERSION_WIN7,
497 		VMBUS_VERSION_WS2008
498 	};
499 	struct vmbus_chanmsg_connect cmd;
500 	struct vmbus_chanmsg_connect_resp rsp;
501 	int i, rv;
502 
503 	memset(&cmd, 0, sizeof(cmd));
504 	cmd.chm_hdr.chm_type = VMBUS_CHANMSG_CONNECT;
505 	cmd.chm_evtflags = hyperv_dma_get_paddr(&sc->sc_events_dma);
506 	cmd.chm_mnf1 = hyperv_dma_get_paddr(&sc->sc_monitor_dma[0]);
507 	cmd.chm_mnf2 = hyperv_dma_get_paddr(&sc->sc_monitor_dma[1]);
508 
509 	memset(&rsp, 0, sizeof(rsp));
510 
511 	for (i = 0; i < __arraycount(versions); i++) {
512 		cmd.chm_ver = versions[i];
513 		rv = vmbus_cmd(sc, &cmd, sizeof(cmd), &rsp, sizeof(rsp),
514 		    cold ? HCF_NOSLEEP : HCF_SLEEPOK);
515 		if (rv) {
516 			DPRINTF("%s: CONNECT failed\n",
517 			    device_xname(sc->sc_dev));
518 			return rv;
519 		}
520 		if (rsp.chm_done) {
521 			atomic_or_32(&sc->sc_flags, VMBUS_SCFLAG_CONNECTED);
522 			sc->sc_proto = versions[i];
523 			sc->sc_handle = VMBUS_GPADL_START;
524 			break;
525 		}
526 	}
527 	if (i == __arraycount(versions)) {
528 		device_printf(sc->sc_dev,
529 		    "failed to negotiate protocol version\n");
530 		return ENXIO;
531 	}
532 
533 	return 0;
534 }
535 
536 static int
537 vmbus_cmd(struct vmbus_softc *sc, void *cmd, size_t cmdlen, void *rsp,
538     size_t rsplen, int flags)
539 {
540 	const int prflags = cold ? PR_NOWAIT : PR_WAITOK;
541 	struct vmbus_msg *msg;
542 	paddr_t pa;
543 	int rv;
544 
545 	if (cmdlen > VMBUS_MSG_DSIZE_MAX) {
546 		device_printf(sc->sc_dev, "payload too large (%zu)\n",
547 		    cmdlen);
548 		return EMSGSIZE;
549 	}
550 
551 	msg = pool_cache_get_paddr(sc->sc_msgpool, prflags, &pa);
552 	if (msg == NULL) {
553 		device_printf(sc->sc_dev, "couldn't get msgpool\n");
554 		return ENOMEM;
555 	}
556 	memset(msg, 0, sizeof(*msg));
557 	msg->msg_req.hc_dsize = cmdlen;
558 	memcpy(msg->msg_req.hc_data, cmd, cmdlen);
559 
560 	if (!(flags & HCF_NOREPLY)) {
561 		msg->msg_rsp = rsp;
562 		msg->msg_rsplen = rsplen;
563 	} else
564 		msg->msg_flags |= MSGF_NOQUEUE;
565 
566 	if (flags & HCF_NOSLEEP)
567 		msg->msg_flags |= MSGF_NOSLEEP;
568 
569 	rv = vmbus_start(sc, msg, pa);
570 	if (rv == 0)
571 		rv = vmbus_reply(sc, msg);
572 	pool_cache_put_paddr(sc->sc_msgpool, msg, pa);
573 	return rv;
574 }
575 
576 static int
577 vmbus_start(struct vmbus_softc *sc, struct vmbus_msg *msg, paddr_t msg_pa)
578 {
579 	static const int delays[] = {
580 		100, 100, 100, 500, 500, 5000, 5000, 5000
581 	};
582 	const char *wchan = "hvstart";
583 	uint16_t status;
584 	int i, s;
585 
586 	msg->msg_req.hc_connid = VMBUS_CONNID_MESSAGE;
587 	msg->msg_req.hc_msgtype = 1;
588 
589 	if (!(msg->msg_flags & MSGF_NOQUEUE)) {
590 		mutex_enter(&sc->sc_req_lock);
591 		TAILQ_INSERT_TAIL(&sc->sc_reqs, msg, msg_entry);
592 		mutex_exit(&sc->sc_req_lock);
593 	}
594 
595 	for (i = 0; i < __arraycount(delays); i++) {
596 		status = hyperv_hypercall_post_message(
597 		    msg_pa + offsetof(struct vmbus_msg, msg_req));
598 		if (status == HYPERCALL_STATUS_SUCCESS)
599 			break;
600 
601 		if (msg->msg_flags & MSGF_NOSLEEP) {
602 			delay(delays[i]);
603 			s = splnet();
604 			hyperv_intr();
605 			splx(s);
606 		} else
607 			tsleep(wchan, PRIBIO, wchan, 1);
608 	}
609 	if (status != HYPERCALL_STATUS_SUCCESS) {
610 		device_printf(sc->sc_dev,
611 		    "posting vmbus message failed with %d\n", status);
612 		if (!(msg->msg_flags & MSGF_NOQUEUE)) {
613 			mutex_enter(&sc->sc_req_lock);
614 			TAILQ_REMOVE(&sc->sc_reqs, msg, msg_entry);
615 			mutex_exit(&sc->sc_req_lock);
616 		}
617 		return EIO;
618 	}
619 
620 	return 0;
621 }
622 
623 static int
624 vmbus_reply_done(struct vmbus_softc *sc, struct vmbus_msg *msg)
625 {
626 	struct vmbus_msg *m;
627 
628 	mutex_enter(&sc->sc_rsp_lock);
629 	TAILQ_FOREACH(m, &sc->sc_rsps, msg_entry) {
630 		if (m == msg) {
631 			mutex_exit(&sc->sc_rsp_lock);
632 			return 1;
633 		}
634 	}
635 	mutex_exit(&sc->sc_rsp_lock);
636 	return 0;
637 }
638 
639 static int
640 vmbus_reply(struct vmbus_softc *sc, struct vmbus_msg *msg)
641 {
642 
643 	if (msg->msg_flags & MSGF_NOQUEUE)
644 		return 0;
645 
646 	vmbus_wait(sc, vmbus_reply_done, msg, msg, "hvreply");
647 
648 	mutex_enter(&sc->sc_rsp_lock);
649 	TAILQ_REMOVE(&sc->sc_rsps, msg, msg_entry);
650 	mutex_exit(&sc->sc_rsp_lock);
651 
652 	return 0;
653 }
654 
655 static void
656 vmbus_wait(struct vmbus_softc *sc,
657     int (*cond)(struct vmbus_softc *, struct vmbus_msg *),
658     struct vmbus_msg *msg, void *wchan, const char *wmsg)
659 {
660 	int s;
661 
662 	while (!cond(sc, msg)) {
663 		if (msg->msg_flags & MSGF_NOSLEEP) {
664 			delay(1000);
665 			s = splnet();
666 			hyperv_intr();
667 			splx(s);
668 		} else
669 			tsleep(wchan, PRIBIO, wmsg ? wmsg : "hvwait", 1);
670 	}
671 }
672 
673 static uint16_t
674 vmbus_intr_signal(struct vmbus_softc *sc, paddr_t con_pa)
675 {
676 	uint64_t status;
677 
678 	status = hyperv_hypercall_signal_event(con_pa);
679 	return (uint16_t)status;
680 }
681 
682 #if LONG_BIT == 64
683 #define ffsl(v)	ffs64(v)
684 #elif LONG_BIT == 32
685 #define ffsl(v)	ffs32(v)
686 #else
687 #error unsupport LONG_BIT
688 #endif	/* LONG_BIT */
689 
690 static void
691 vmbus_event_flags_proc(struct vmbus_softc *sc, volatile u_long *revents,
692     int maxrow)
693 {
694 	struct vmbus_channel *ch;
695 	u_long pending;
696 	uint32_t chanid, chanid_base;
697 	int row, chanid_ofs;
698 
699 	for (row = 0; row < maxrow; row++) {
700 		if (revents[row] == 0)
701 			continue;
702 
703 		pending = atomic_swap_ulong(&revents[row], 0);
704 		chanid_base = row * LONG_BIT;
705 
706 		while ((chanid_ofs = ffsl(pending)) != 0) {
707 			chanid_ofs--;	/* NOTE: ffs is 1-based */
708 			pending &= ~(1UL << chanid_ofs);
709 
710 			chanid = chanid_base + chanid_ofs;
711 			/* vmbus channel protocol message */
712 			if (chanid == 0)
713 				continue;
714 
715 			ch = vmbus_channel_lookup(sc, chanid);
716 			if (ch == NULL) {
717 				device_printf(sc->sc_dev,
718 				    "unhandled event on %d\n", chanid);
719 				continue;
720 			}
721 			if (ch->ch_state != VMBUS_CHANSTATE_OPENED) {
722 				device_printf(sc->sc_dev,
723 				    "channel %d is not active\n", chanid);
724 				continue;
725 			}
726 			ch->ch_evcnt.ev_count++;
727 			vmbus_channel_schedule(ch);
728 		}
729 	}
730 }
731 
732 static void
733 vmbus_event_proc(void *arg, struct cpu_info *ci)
734 {
735 	struct vmbus_softc *sc = arg;
736 	struct vmbus_evtflags *evt;
737 
738 	/*
739 	 * On Host with Win8 or above, the event page can be
740 	 * checked directly to get the id of the channel
741 	 * that has the pending interrupt.
742 	 */
743 	evt = (struct vmbus_evtflags *)sc->sc_percpu[cpu_index(ci)].siep +
744 	    VMBUS_SINT_MESSAGE;
745 
746 	vmbus_event_flags_proc(sc, evt->evt_flags,
747 	    __arraycount(evt->evt_flags));
748 }
749 
750 static void
751 vmbus_event_proc_compat(void *arg, struct cpu_info *ci)
752 {
753 	struct vmbus_softc *sc = arg;
754 	struct vmbus_evtflags *evt;
755 
756 	evt = (struct vmbus_evtflags *)sc->sc_percpu[cpu_index(ci)].siep +
757 	    VMBUS_SINT_MESSAGE;
758 
759 	if (test_bit(0, &evt->evt_flags[0])) {
760 		clear_bit(0, &evt->evt_flags[0]);
761 		/*
762 		 * receive size is 1/2 page and divide that by 4 bytes
763 		 */
764 		vmbus_event_flags_proc(sc, sc->sc_revents,
765 		    VMBUS_CHAN_MAX_COMPAT / VMBUS_EVTFLAG_LEN);
766 	}
767 }
768 
769 static void
770 vmbus_message_proc(void *arg, struct cpu_info *ci)
771 {
772 	struct vmbus_softc *sc = arg;
773 	struct vmbus_message *msg;
774 
775 	msg = (struct vmbus_message *)sc->sc_percpu[cpu_index(ci)].simp +
776 	    VMBUS_SINT_MESSAGE;
777 	if (__predict_false(msg->msg_type != HYPERV_MSGTYPE_NONE)) {
778 		if (__predict_true(!cold))
779 			softint_schedule_cpu(sc->sc_msg_sih, ci);
780 		else
781 			vmbus_message_softintr(sc);
782 	}
783 }
784 
785 static void
786 vmbus_message_softintr(void *arg)
787 {
788 	struct vmbus_softc *sc = arg;
789 	struct vmbus_message *msg;
790 	struct vmbus_chanmsg_hdr *hdr;
791 	uint32_t type;
792 	cpuid_t cpu;
793 
794 	cpu = cpu_index(curcpu());
795 
796 	for (;;) {
797 		msg = (struct vmbus_message *)sc->sc_percpu[cpu].simp +
798 		    VMBUS_SINT_MESSAGE;
799 		if (msg->msg_type == HYPERV_MSGTYPE_NONE)
800 			break;
801 
802 		hdr = (struct vmbus_chanmsg_hdr *)msg->msg_data;
803 		type = hdr->chm_type;
804 		if (type >= VMBUS_CHANMSG_COUNT) {
805 			device_printf(sc->sc_dev,
806 			    "unhandled message type %u flags %#x\n", type,
807 			    msg->msg_flags);
808 		} else {
809 			if (vmbus_msg_dispatch[type].hmd_handler) {
810 				vmbus_msg_dispatch[type].hmd_handler(sc, hdr);
811 			} else {
812 				device_printf(sc->sc_dev,
813 				    "unhandled message type %u\n", type);
814 			}
815 		}
816 
817 		msg->msg_type = HYPERV_MSGTYPE_NONE;
818 		membar_sync();
819 		if (msg->msg_flags & VMBUS_MSGFLAG_PENDING)
820 			hyperv_send_eom();
821 	}
822 }
823 
824 static void
825 vmbus_channel_response(struct vmbus_softc *sc, struct vmbus_chanmsg_hdr *rsphdr)
826 {
827 	struct vmbus_msg *msg;
828 	struct vmbus_chanmsg_hdr *reqhdr;
829 	int req;
830 
831 	req = vmbus_msg_dispatch[rsphdr->chm_type].hmd_request;
832 	mutex_enter(&sc->sc_req_lock);
833 	TAILQ_FOREACH(msg, &sc->sc_reqs, msg_entry) {
834 		reqhdr = (struct vmbus_chanmsg_hdr *)&msg->msg_req.hc_data;
835 		if (reqhdr->chm_type == req) {
836 			TAILQ_REMOVE(&sc->sc_reqs, msg, msg_entry);
837 			break;
838 		}
839 	}
840 	mutex_exit(&sc->sc_req_lock);
841 	if (msg != NULL) {
842 		memcpy(msg->msg_rsp, rsphdr, msg->msg_rsplen);
843 		mutex_enter(&sc->sc_rsp_lock);
844 		TAILQ_INSERT_TAIL(&sc->sc_rsps, msg, msg_entry);
845 		mutex_exit(&sc->sc_rsp_lock);
846 		wakeup(msg);
847 	}
848 }
849 
850 static void
851 vmbus_channel_offer(struct vmbus_softc *sc, struct vmbus_chanmsg_hdr *hdr)
852 {
853 	struct vmbus_offer *co;
854 
855 	co = kmem_intr_zalloc(sizeof(*co), KM_NOSLEEP);
856 	if (co == NULL) {
857 		device_printf(sc->sc_dev, "couldn't allocate offer\n");
858 		return;
859 	}
860 
861 	memcpy(&co->co_chan, hdr, sizeof(co->co_chan));
862 
863 	mutex_enter(&sc->sc_offer_lock);
864 	SIMPLEQ_INSERT_TAIL(&sc->sc_offers, co, co_entry);
865 	mutex_exit(&sc->sc_offer_lock);
866 }
867 
868 static void
869 vmbus_channel_rescind(struct vmbus_softc *sc, struct vmbus_chanmsg_hdr *hdr)
870 {
871 	const struct vmbus_chanmsg_chrescind *cmd;
872 
873 	cmd = (const struct vmbus_chanmsg_chrescind *)hdr;
874 	device_printf(sc->sc_dev, "revoking channel %u\n", cmd->chm_chanid);
875 }
876 
877 static void
878 vmbus_channel_delivered(struct vmbus_softc *sc, struct vmbus_chanmsg_hdr *hdr)
879 {
880 
881 	atomic_or_32(&sc->sc_flags, VMBUS_SCFLAG_OFFERS_DELIVERED);
882 	wakeup(&sc->sc_offers);
883 }
884 
885 static void
886 hyperv_guid_sprint(struct hyperv_guid *guid, char *str, size_t size)
887 {
888 	static const struct {
889 		const struct hyperv_guid *guid;
890 		const char *ident;
891 	} map[] = {
892 		{ &hyperv_guid_network,		"network" },
893 		{ &hyperv_guid_ide,		"ide" },
894 		{ &hyperv_guid_scsi,		"scsi" },
895 		{ &hyperv_guid_shutdown,	"shutdown" },
896 		{ &hyperv_guid_timesync,	"timesync" },
897 		{ &hyperv_guid_heartbeat,	"heartbeat" },
898 		{ &hyperv_guid_kvp,		"kvp" },
899 		{ &hyperv_guid_vss,		"vss" },
900 		{ &hyperv_guid_dynmem,		"dynamic-memory" },
901 		{ &hyperv_guid_mouse,		"mouse" },
902 		{ &hyperv_guid_kbd,		"keyboard" },
903 		{ &hyperv_guid_video,		"video" },
904 		{ &hyperv_guid_fc,		"fiber-channel" },
905 		{ &hyperv_guid_fcopy,		"file-copy" },
906 		{ &hyperv_guid_pcie,		"pcie-passthrough" },
907 		{ &hyperv_guid_netdir,		"network-direct" },
908 		{ &hyperv_guid_rdesktop,	"remote-desktop" },
909 		{ &hyperv_guid_avma1,		"avma-1" },
910 		{ &hyperv_guid_avma2,		"avma-2" },
911 		{ &hyperv_guid_avma3,		"avma-3" },
912 		{ &hyperv_guid_avma4,		"avma-4" },
913 	};
914 	int i;
915 
916 	for (i = 0; i < __arraycount(map); i++) {
917 		if (memcmp(guid, map[i].guid, sizeof(*guid)) == 0) {
918 			strlcpy(str, map[i].ident, size);
919 			return;
920 		}
921 	}
922 	hyperv_guid2str(guid, str, size);
923 }
924 
925 static int
926 vmbus_channel_scan_done(struct vmbus_softc *sc, struct vmbus_msg *msg __unused)
927 {
928 
929 	return ISSET(sc->sc_flags, VMBUS_SCFLAG_OFFERS_DELIVERED);
930 }
931 
932 static int
933 vmbus_channel_scan(struct vmbus_softc *sc)
934 {
935 	struct vmbus_chanmsg_hdr hdr;
936 	struct vmbus_chanmsg_choffer rsp;
937 	struct vmbus_offer *co;
938 
939 	SIMPLEQ_INIT(&sc->sc_offers);
940 	mutex_init(&sc->sc_offer_lock, MUTEX_DEFAULT, IPL_NET);
941 
942 	memset(&hdr, 0, sizeof(hdr));
943 	hdr.chm_type = VMBUS_CHANMSG_CHREQUEST;
944 
945 	if (vmbus_cmd(sc, &hdr, sizeof(hdr), &rsp, sizeof(rsp),
946 	    HCF_NOREPLY | (cold ? HCF_NOSLEEP : HCF_SLEEPOK))) {
947 		DPRINTF("%s: CHREQUEST failed\n", device_xname(sc->sc_dev));
948 		return -1;
949 	}
950 
951 	vmbus_wait(sc, vmbus_channel_scan_done, (struct vmbus_msg *)&hdr,
952 	    &sc->sc_offers, "hvscan");
953 
954 	TAILQ_INIT(&sc->sc_channels);
955 	mutex_init(&sc->sc_channel_lock, MUTEX_DEFAULT, IPL_NET);
956 
957 	mutex_enter(&sc->sc_offer_lock);
958 	while (!SIMPLEQ_EMPTY(&sc->sc_offers)) {
959 		co = SIMPLEQ_FIRST(&sc->sc_offers);
960 		SIMPLEQ_REMOVE_HEAD(&sc->sc_offers, co_entry);
961 		mutex_exit(&sc->sc_offer_lock);
962 
963 		vmbus_process_offer(sc, co);
964 		kmem_free(co, sizeof(*co));
965 
966 		mutex_enter(&sc->sc_offer_lock);
967 	}
968 	mutex_exit(&sc->sc_offer_lock);
969 
970 	return 0;
971 }
972 
973 static struct vmbus_channel *
974 vmbus_channel_alloc(struct vmbus_softc *sc)
975 {
976 	struct vmbus_channel *ch;
977 
978 	ch = kmem_zalloc(sizeof(*ch), cold ? KM_NOSLEEP : KM_SLEEP);
979 
980 	ch->ch_monprm = hyperv_dma_alloc(sc->sc_dmat, &ch->ch_monprm_dma,
981 	    sizeof(*ch->ch_monprm), 8, 0, 1);
982 	if (ch->ch_monprm == NULL) {
983 		device_printf(sc->sc_dev, "monprm alloc failed\n");
984 		kmem_free(ch, sizeof(*ch));
985 		return NULL;
986 	}
987 	memset(ch->ch_monprm, 0, sizeof(*ch->ch_monprm));
988 
989 	ch->ch_refs = 1;
990 	ch->ch_sc = sc;
991 	mutex_init(&ch->ch_subchannel_lock, MUTEX_DEFAULT, IPL_NET);
992 	TAILQ_INIT(&ch->ch_subchannels);
993 
994 	ch->ch_state = VMBUS_CHANSTATE_CLOSED;
995 
996 	return ch;
997 }
998 
999 static void
1000 vmbus_channel_free(struct vmbus_channel *ch)
1001 {
1002 	struct vmbus_softc *sc = ch->ch_sc;
1003 
1004 	KASSERTMSG(TAILQ_EMPTY(&ch->ch_subchannels) &&
1005 	    ch->ch_subchannel_count == 0, "still owns sub-channels");
1006 	KASSERTMSG(ch->ch_state == 0 || ch->ch_state == VMBUS_CHANSTATE_CLOSED,
1007 	    "free busy channel");
1008 	KASSERTMSG(ch->ch_refs == 0, "channel %u: invalid refcnt %d",
1009 	    ch->ch_id, ch->ch_refs);
1010 
1011 	hyperv_dma_free(sc->sc_dmat, &ch->ch_monprm_dma);
1012 	mutex_destroy(&ch->ch_subchannel_lock);
1013 	/* XXX ch_evcnt */
1014 	softint_disestablish(ch->ch_taskq);
1015 	kmem_free(ch, sizeof(*ch));
1016 }
1017 
1018 static int
1019 vmbus_channel_add(struct vmbus_channel *nch)
1020 {
1021 	struct vmbus_softc *sc = nch->ch_sc;
1022 	struct vmbus_channel *ch;
1023 	u_int refs __diagused;
1024 
1025 	if (nch->ch_id == 0) {
1026 		device_printf(sc->sc_dev, "got channel 0 offer, discard\n");
1027 		return EINVAL;
1028 	} else if (nch->ch_id >= sc->sc_channel_max) {
1029 		device_printf(sc->sc_dev, "invalid channel %u offer\n",
1030 		    nch->ch_id);
1031 		return EINVAL;
1032 	}
1033 
1034 	mutex_enter(&sc->sc_channel_lock);
1035 	TAILQ_FOREACH(ch, &sc->sc_channels, ch_entry) {
1036 		if (!memcmp(&ch->ch_type, &nch->ch_type, sizeof(ch->ch_type)) &&
1037 		    !memcmp(&ch->ch_inst, &nch->ch_inst, sizeof(ch->ch_inst)))
1038 			break;
1039 	}
1040 	if (VMBUS_CHAN_ISPRIMARY(nch)) {
1041 		if (ch == NULL) {
1042 			TAILQ_INSERT_TAIL(&sc->sc_channels, nch, ch_entry);
1043 			mutex_exit(&sc->sc_channel_lock);
1044 			goto done;
1045 		} else {
1046 			mutex_exit(&sc->sc_channel_lock);
1047 			device_printf(sc->sc_dev,
1048 			    "duplicated primary channel%u\n", nch->ch_id);
1049 			return EINVAL;
1050 		}
1051 	} else {
1052 		if (ch == NULL) {
1053 			mutex_exit(&sc->sc_channel_lock);
1054 			device_printf(sc->sc_dev, "no primary channel%u\n",
1055 			    nch->ch_id);
1056 			return EINVAL;
1057 		}
1058 	}
1059 	mutex_exit(&sc->sc_channel_lock);
1060 
1061 	KASSERT(!VMBUS_CHAN_ISPRIMARY(nch));
1062 	KASSERT(ch != NULL);
1063 
1064 	refs = atomic_add_int_nv(&nch->ch_refs, 1);
1065 	KASSERT(refs == 1);
1066 
1067 	nch->ch_primary_channel = ch;
1068 	nch->ch_dev = ch->ch_dev;
1069 
1070 	mutex_enter(&ch->ch_subchannel_lock);
1071 	TAILQ_INSERT_TAIL(&ch->ch_subchannels, nch, ch_subentry);
1072 	ch->ch_subchannel_count++;
1073 	mutex_exit(&ch->ch_subchannel_lock);
1074 	wakeup(ch);
1075 
1076 done:
1077 	vmbus_channel_cpu_default(nch);
1078 
1079 	return 0;
1080 }
1081 
1082 void
1083 vmbus_channel_cpu_set(struct vmbus_channel *ch, int cpu)
1084 {
1085 	struct vmbus_softc *sc = ch->ch_sc;
1086 
1087 	KASSERTMSG(cpu >= 0 && cpu < ncpu, "invalid cpu %d", cpu);
1088 
1089 	if (sc->sc_proto == VMBUS_VERSION_WS2008 ||
1090 	    sc->sc_proto == VMBUS_VERSION_WIN7) {
1091 		/* Only cpu0 is supported */
1092 		cpu = 0;
1093 	}
1094 
1095 	ch->ch_cpuid = cpu;
1096 	ch->ch_vcpu = sc->sc_percpu[cpu].vcpuid;
1097 }
1098 
1099 void
1100 vmbus_channel_cpu_rr(struct vmbus_channel *ch)
1101 {
1102 	static uint32_t vmbus_channel_nextcpu;
1103 	int cpu;
1104 
1105 	cpu = atomic_add_32_nv(&vmbus_channel_nextcpu, 1) % ncpu;
1106 	vmbus_channel_cpu_set(ch, cpu);
1107 }
1108 
1109 static void
1110 vmbus_channel_cpu_default(struct vmbus_channel *ch)
1111 {
1112 
1113         /*
1114 	 * By default, pin the channel to cpu0.  Devices having
1115 	 * special channel-cpu mapping requirement should call
1116 	 * vmbus_channel_cpu_{set,rr}().
1117 	 */
1118 	vmbus_channel_cpu_set(ch, 0);
1119 }
1120 
1121 static void
1122 vmbus_process_offer(struct vmbus_softc *sc, struct vmbus_offer *co)
1123 {
1124 	struct vmbus_channel *ch;
1125 
1126 	ch = vmbus_channel_alloc(sc);
1127 	if (ch == NULL) {
1128 		device_printf(sc->sc_dev, "allocate channel %u failed\n",
1129 		    co->co_chan.chm_chanid);
1130 		return;
1131 	}
1132 
1133 	/*
1134 	 * By default we setup state to enable batched reading.
1135 	 * A specific service can choose to disable this prior
1136 	 * to opening the channel.
1137 	 */
1138 	ch->ch_flags |= CHF_BATCHED;
1139 
1140 	hyperv_guid_sprint(&co->co_chan.chm_chtype, ch->ch_ident,
1141 	    sizeof(ch->ch_ident));
1142 
1143 	ch->ch_monprm->mp_connid = VMBUS_CONNID_EVENT;
1144 	if (sc->sc_proto > VMBUS_VERSION_WS2008)
1145 		ch->ch_monprm->mp_connid = co->co_chan.chm_connid;
1146 
1147 	if (co->co_chan.chm_flags1 & VMBUS_CHOFFER_FLAG1_HASMNF) {
1148 		ch->ch_mgroup = co->co_chan.chm_montrig / VMBUS_MONTRIG_LEN;
1149 		ch->ch_mindex = co->co_chan.chm_montrig % VMBUS_MONTRIG_LEN;
1150 		ch->ch_flags |= CHF_MONITOR;
1151 	}
1152 
1153 	ch->ch_id = co->co_chan.chm_chanid;
1154 	ch->ch_subidx = co->co_chan.chm_subidx;
1155 
1156 	memcpy(&ch->ch_type, &co->co_chan.chm_chtype, sizeof(ch->ch_type));
1157 	memcpy(&ch->ch_inst, &co->co_chan.chm_chinst, sizeof(ch->ch_inst));
1158 
1159 	if (VMBUS_CHAN_ISPRIMARY(ch)) {
1160 		/* set primary channel mgmt wq */
1161 	} else {
1162 		/* set sub channel mgmt wq */
1163 	}
1164 
1165 	if (vmbus_channel_add(ch) != 0) {
1166 		vmbus_channel_free(ch);
1167 		return;
1168 	}
1169 
1170 	ch->ch_state = VMBUS_CHANSTATE_OFFERED;
1171 
1172 #ifdef HYPERV_DEBUG
1173 	printf("%s: channel %u: \"%s\"", device_xname(sc->sc_dev), ch->ch_id,
1174 	    ch->ch_ident);
1175 	if (ch->ch_flags & CHF_MONITOR)
1176 		printf(", monitor %u\n", co->co_chan.chm_montrig);
1177 	else
1178 		printf("\n");
1179 #endif
1180 }
1181 
1182 static int
1183 vmbus_channel_release(struct vmbus_channel *ch)
1184 {
1185 	struct vmbus_softc *sc = ch->ch_sc;
1186 	struct vmbus_chanmsg_chfree cmd;
1187 	int rv;
1188 
1189 	memset(&cmd, 0, sizeof(cmd));
1190 	cmd.chm_hdr.chm_type = VMBUS_CHANMSG_CHFREE;
1191 	cmd.chm_chanid = ch->ch_id;
1192 
1193 	rv = vmbus_cmd(sc, &cmd, sizeof(cmd), NULL, 0,
1194 	    HCF_NOREPLY | (cold ? HCF_NOSLEEP : HCF_SLEEPOK));
1195 	if (rv) {
1196 		DPRINTF("%s: CHFREE failed with %d\n", device_xname(sc->sc_dev),
1197 		    rv);
1198 	}
1199 	return rv;
1200 }
1201 
1202 struct vmbus_channel **
1203 vmbus_subchannel_get(struct vmbus_channel *prich, int cnt)
1204 {
1205 	struct vmbus_channel **ret, *ch;
1206 	int i;
1207 
1208 	KASSERT(cnt > 0);
1209 
1210 	ret = kmem_alloc(sizeof(struct vmbus_channel *) * cnt,
1211 	    cold ? KM_NOSLEEP : KM_SLEEP);
1212 
1213 	mutex_enter(&prich->ch_subchannel_lock);
1214 
1215 	while (prich->ch_subchannel_count < cnt)
1216 		/* XXX use condvar(9) instead of mtsleep */
1217 		mtsleep(prich, PRIBIO, "hvvmsubch", 0,
1218 		    &prich->ch_subchannel_lock);
1219 
1220 	i = 0;
1221 	TAILQ_FOREACH(ch, &prich->ch_subchannels, ch_subentry) {
1222 		ret[i] = ch;	/* XXX inc refs */
1223 
1224 		if (++i == cnt)
1225 			break;
1226 	}
1227 
1228 	mutex_exit(&prich->ch_subchannel_lock);
1229 
1230 	return ret;
1231 }
1232 
1233 void
1234 vmbus_subchannel_put(struct vmbus_channel **subch, int cnt)
1235 {
1236 
1237 	kmem_free(subch, sizeof(struct vmbus_channel *) * cnt);
1238 }
1239 
1240 static struct vmbus_channel *
1241 vmbus_channel_lookup(struct vmbus_softc *sc, uint32_t relid)
1242 {
1243 	struct vmbus_channel *ch;
1244 
1245 	TAILQ_FOREACH(ch, &sc->sc_channels, ch_entry) {
1246 		if (ch->ch_id == relid)
1247 			return ch;
1248 	}
1249 	return NULL;
1250 }
1251 
1252 static int
1253 vmbus_channel_ring_create(struct vmbus_channel *ch, uint32_t buflen)
1254 {
1255 	struct vmbus_softc *sc = ch->ch_sc;
1256 
1257 	buflen = roundup(buflen, PAGE_SIZE) + sizeof(struct vmbus_bufring);
1258 	ch->ch_ring_size = 2 * buflen;
1259 	ch->ch_ring = hyperv_dma_alloc(sc->sc_dmat, &ch->ch_ring_dma,
1260 	    ch->ch_ring_size, PAGE_SIZE, 0, 1);	/* page aligned memory */
1261 	if (ch->ch_ring == NULL) {
1262 		device_printf(sc->sc_dev,
1263 		    "failed to allocate channel ring\n");
1264 		return ENOMEM;
1265 	}
1266 	memset(ch->ch_ring, 0, ch->ch_ring_size);
1267 
1268 	memset(&ch->ch_wrd, 0, sizeof(ch->ch_wrd));
1269 	ch->ch_wrd.rd_ring = (struct vmbus_bufring *)ch->ch_ring;
1270 	ch->ch_wrd.rd_size = buflen;
1271 	ch->ch_wrd.rd_dsize = buflen - sizeof(struct vmbus_bufring);
1272 	mutex_init(&ch->ch_wrd.rd_lock, MUTEX_DEFAULT, IPL_NET);
1273 
1274 	memset(&ch->ch_rrd, 0, sizeof(ch->ch_rrd));
1275 	ch->ch_rrd.rd_ring = (struct vmbus_bufring *)((uint8_t *)ch->ch_ring +
1276 	    buflen);
1277 	ch->ch_rrd.rd_size = buflen;
1278 	ch->ch_rrd.rd_dsize = buflen - sizeof(struct vmbus_bufring);
1279 	mutex_init(&ch->ch_rrd.rd_lock, MUTEX_DEFAULT, IPL_NET);
1280 
1281 	if (vmbus_handle_alloc(ch, &ch->ch_ring_dma, ch->ch_ring_size,
1282 	    &ch->ch_ring_gpadl)) {
1283 		device_printf(sc->sc_dev,
1284 		    "failed to obtain a PA handle for the ring\n");
1285 		vmbus_channel_ring_destroy(ch);
1286 		return ENOMEM;
1287 	}
1288 
1289 	return 0;
1290 }
1291 
1292 static void
1293 vmbus_channel_ring_destroy(struct vmbus_channel *ch)
1294 {
1295 	struct vmbus_softc *sc = ch->ch_sc;
1296 
1297 	hyperv_dma_free(sc->sc_dmat, &ch->ch_ring_dma);
1298 	ch->ch_ring = NULL;
1299 	vmbus_handle_free(ch, ch->ch_ring_gpadl);
1300 
1301 	mutex_destroy(&ch->ch_wrd.rd_lock);
1302 	memset(&ch->ch_wrd, 0, sizeof(ch->ch_wrd));
1303 	mutex_destroy(&ch->ch_rrd.rd_lock);
1304 	memset(&ch->ch_rrd, 0, sizeof(ch->ch_rrd));
1305 }
1306 
1307 int
1308 vmbus_channel_open(struct vmbus_channel *ch, size_t buflen, void *udata,
1309     size_t udatalen, void (*handler)(void *), void *arg)
1310 {
1311 	struct vmbus_softc *sc = ch->ch_sc;
1312 	struct vmbus_chanmsg_chopen cmd;
1313 	struct vmbus_chanmsg_chopen_resp rsp;
1314 	int rv = EINVAL;
1315 
1316 	if (ch->ch_ring == NULL &&
1317 	    (rv = vmbus_channel_ring_create(ch, buflen))) {
1318 		DPRINTF("%s: failed to create channel ring\n",
1319 		    device_xname(sc->sc_dev));
1320 		return rv;
1321 	}
1322 
1323 	memset(&cmd, 0, sizeof(cmd));
1324 	cmd.chm_hdr.chm_type = VMBUS_CHANMSG_CHOPEN;
1325 	cmd.chm_openid = ch->ch_id;
1326 	cmd.chm_chanid = ch->ch_id;
1327 	cmd.chm_gpadl = ch->ch_ring_gpadl;
1328 	cmd.chm_txbr_pgcnt = atop(ch->ch_wrd.rd_size);
1329 	cmd.chm_vcpuid = ch->ch_vcpu;
1330 	if (udata && udatalen > 0)
1331 		memcpy(cmd.chm_udata, udata, udatalen);
1332 
1333 	memset(&rsp, 0, sizeof(rsp));
1334 
1335 	ch->ch_handler = handler;
1336 	ch->ch_ctx = arg;
1337 	ch->ch_state = VMBUS_CHANSTATE_OPENED;
1338 
1339 	rv = vmbus_cmd(sc, &cmd, sizeof(cmd), &rsp, sizeof(rsp),
1340 	    cold ? HCF_NOSLEEP : HCF_SLEEPOK);
1341 	if (rv) {
1342 		vmbus_channel_ring_destroy(ch);
1343 		DPRINTF("%s: CHOPEN failed with %d\n", device_xname(sc->sc_dev),
1344 		    rv);
1345 		ch->ch_handler = NULL;
1346 		ch->ch_ctx = NULL;
1347 		ch->ch_state = VMBUS_CHANSTATE_OFFERED;
1348 		return rv;
1349 	}
1350 	return 0;
1351 }
1352 
1353 static void
1354 vmbus_channel_detach(struct vmbus_channel *ch)
1355 {
1356 	u_int refs;
1357 
1358 	refs = atomic_add_int_nv(&ch->ch_refs, -1);
1359 	if (refs == 1) {
1360 		/* XXX on workqueue? */
1361 		if (VMBUS_CHAN_ISPRIMARY(ch)) {
1362 			vmbus_channel_release(ch);
1363 			vmbus_channel_free(ch);
1364 		} else {
1365 			struct vmbus_channel *prich = ch->ch_primary_channel;
1366 
1367 			vmbus_channel_release(ch);
1368 
1369 			mutex_enter(&prich->ch_subchannel_lock);
1370 			TAILQ_REMOVE(&prich->ch_subchannels, ch, ch_subentry);
1371 			prich->ch_subchannel_count--;
1372 			mutex_exit(&prich->ch_subchannel_lock);
1373 			wakeup(prich);
1374 
1375 			vmbus_channel_free(ch);
1376 		}
1377 	}
1378 }
1379 
1380 static int
1381 vmbus_channel_close_internal(struct vmbus_channel *ch)
1382 {
1383 	struct vmbus_softc *sc = ch->ch_sc;
1384 	struct vmbus_chanmsg_chclose cmd;
1385 	int rv;
1386 
1387 	memset(&cmd, 0, sizeof(cmd));
1388 	cmd.chm_hdr.chm_type = VMBUS_CHANMSG_CHCLOSE;
1389 	cmd.chm_chanid = ch->ch_id;
1390 
1391 	ch->ch_state = VMBUS_CHANSTATE_CLOSING;
1392 	rv = vmbus_cmd(sc, &cmd, sizeof(cmd), NULL, 0,
1393 	    HCF_NOREPLY | (cold ? HCF_NOSLEEP : HCF_SLEEPOK));
1394 	if (rv) {
1395 		DPRINTF("%s: CHCLOSE failed with %d\n",
1396 		    device_xname(sc->sc_dev), rv);
1397 		return rv;
1398 	}
1399 	ch->ch_state = VMBUS_CHANSTATE_CLOSED;
1400 	vmbus_channel_ring_destroy(ch);
1401 	return 0;
1402 }
1403 
1404 int
1405 vmbus_channel_close_direct(struct vmbus_channel *ch)
1406 {
1407 	int rv;
1408 
1409 	rv = vmbus_channel_close_internal(ch);
1410 	if (!VMBUS_CHAN_ISPRIMARY(ch))
1411 		vmbus_channel_detach(ch);
1412 	return rv;
1413 }
1414 
1415 int
1416 vmbus_channel_close(struct vmbus_channel *ch)
1417 {
1418 	struct vmbus_channel **subch;
1419 	int i, cnt, rv;
1420 
1421 	if (!VMBUS_CHAN_ISPRIMARY(ch))
1422 		return 0;
1423 
1424 	cnt = ch->ch_subchannel_count;
1425 	if (cnt > 0) {
1426 		subch = vmbus_subchannel_get(ch, cnt);
1427 		for (i = 0; i < ch->ch_subchannel_count; i++) {
1428 			rv = vmbus_channel_close_internal(subch[i]);
1429 			(void) rv;	/* XXX */
1430 			vmbus_channel_detach(ch);
1431 		}
1432 		vmbus_subchannel_put(subch, cnt);
1433 	}
1434 
1435 	return vmbus_channel_close_internal(ch);
1436 }
1437 
1438 static inline void
1439 vmbus_channel_setevent(struct vmbus_softc *sc, struct vmbus_channel *ch)
1440 {
1441 	struct vmbus_mon_trig *mtg;
1442 
1443 	/* Each uint32_t represents 32 channels */
1444 	set_bit(ch->ch_id, sc->sc_wevents);
1445 	if (ch->ch_flags & CHF_MONITOR) {
1446 		mtg = &sc->sc_monitor[1]->mnf_trigs[ch->ch_mgroup];
1447 		set_bit(ch->ch_mindex, &mtg->mt_pending);
1448 	} else
1449 		vmbus_intr_signal(sc, hyperv_dma_get_paddr(&ch->ch_monprm_dma));
1450 }
1451 
1452 static void
1453 vmbus_channel_intr(void *arg)
1454 {
1455 	struct vmbus_channel *ch = arg;
1456 
1457 	if (vmbus_channel_ready(ch))
1458 		ch->ch_handler(ch->ch_ctx);
1459 
1460 	if (vmbus_channel_unpause(ch) == 0)
1461 		return;
1462 
1463 	vmbus_channel_pause(ch);
1464 	vmbus_channel_schedule(ch);
1465 }
1466 
1467 int
1468 vmbus_channel_setdeferred(struct vmbus_channel *ch, const char *name)
1469 {
1470 
1471 	ch->ch_taskq = softint_establish(SOFTINT_NET | SOFTINT_MPSAFE,
1472 	    vmbus_channel_intr, ch);
1473 	if (ch->ch_taskq == NULL)
1474 		return -1;
1475 	return 0;
1476 }
1477 
1478 void
1479 vmbus_channel_schedule(struct vmbus_channel *ch)
1480 {
1481 
1482 	if (ch->ch_handler) {
1483 		if (!cold && (ch->ch_flags & CHF_BATCHED)) {
1484 			vmbus_channel_pause(ch);
1485 			softint_schedule(ch->ch_taskq);
1486 		} else
1487 			ch->ch_handler(ch->ch_ctx);
1488 	}
1489 }
1490 
1491 static __inline void
1492 vmbus_ring_put(struct vmbus_ring_data *wrd, uint8_t *data, uint32_t datalen)
1493 {
1494 	int left = MIN(datalen, wrd->rd_dsize - wrd->rd_prod);
1495 
1496 	memcpy(&wrd->rd_ring->br_data[wrd->rd_prod], data, left);
1497 	memcpy(&wrd->rd_ring->br_data[0], data + left, datalen - left);
1498 	wrd->rd_prod += datalen;
1499 	if (wrd->rd_prod >= wrd->rd_dsize)
1500 		wrd->rd_prod -= wrd->rd_dsize;
1501 }
1502 
1503 static inline void
1504 vmbus_ring_get(struct vmbus_ring_data *rrd, uint8_t *data, uint32_t datalen,
1505     int peek)
1506 {
1507 	int left = MIN(datalen, rrd->rd_dsize - rrd->rd_cons);
1508 
1509 	memcpy(data, &rrd->rd_ring->br_data[rrd->rd_cons], left);
1510 	memcpy(data + left, &rrd->rd_ring->br_data[0], datalen - left);
1511 	if (!peek) {
1512 		rrd->rd_cons += datalen;
1513 		if (rrd->rd_cons >= rrd->rd_dsize)
1514 			rrd->rd_cons -= rrd->rd_dsize;
1515 	}
1516 }
1517 
1518 static __inline void
1519 vmbus_ring_avail(struct vmbus_ring_data *rd, uint32_t *towrite,
1520     uint32_t *toread)
1521 {
1522 	uint32_t ridx = rd->rd_ring->br_rindex;
1523 	uint32_t widx = rd->rd_ring->br_windex;
1524 	uint32_t r, w;
1525 
1526 	if (widx >= ridx)
1527 		w = rd->rd_dsize - (widx - ridx);
1528 	else
1529 		w = ridx - widx;
1530 	r = rd->rd_dsize - w;
1531 	if (towrite)
1532 		*towrite = w;
1533 	if (toread)
1534 		*toread = r;
1535 }
1536 
1537 static int
1538 vmbus_ring_write(struct vmbus_ring_data *wrd, struct iovec *iov, int iov_cnt,
1539     int *needsig)
1540 {
1541 	uint64_t indices = 0;
1542 	uint32_t avail, oprod, datalen = sizeof(indices);
1543 	int i;
1544 
1545 	for (i = 0; i < iov_cnt; i++)
1546 		datalen += iov[i].iov_len;
1547 
1548 	KASSERT(datalen <= wrd->rd_dsize);
1549 
1550 	vmbus_ring_avail(wrd, &avail, NULL);
1551 	if (avail <= datalen) {
1552 		DPRINTF("%s: avail %u datalen %u\n", __func__, avail, datalen);
1553 		return EAGAIN;
1554 	}
1555 
1556 	oprod = wrd->rd_prod;
1557 
1558 	for (i = 0; i < iov_cnt; i++)
1559 		vmbus_ring_put(wrd, iov[i].iov_base, iov[i].iov_len);
1560 
1561 	indices = (uint64_t)oprod << 32;
1562 	vmbus_ring_put(wrd, (uint8_t *)&indices, sizeof(indices));
1563 
1564 	membar_sync();
1565 	wrd->rd_ring->br_windex = wrd->rd_prod;
1566 	membar_sync();
1567 
1568 	/* Signal when the ring transitions from being empty to non-empty */
1569 	if (wrd->rd_ring->br_imask == 0 &&
1570 	    wrd->rd_ring->br_rindex == oprod)
1571 		*needsig = 1;
1572 	else
1573 		*needsig = 0;
1574 
1575 	return 0;
1576 }
1577 
1578 int
1579 vmbus_channel_send(struct vmbus_channel *ch, void *data, uint32_t datalen,
1580     uint64_t rid, int type, uint32_t flags)
1581 {
1582 	struct vmbus_softc *sc = ch->ch_sc;
1583 	struct vmbus_chanpkt cp;
1584 	struct iovec iov[3];
1585 	uint32_t pktlen, pktlen_aligned;
1586 	uint64_t zeropad = 0;
1587 	int rv, needsig = 0;
1588 
1589 	pktlen = sizeof(cp) + datalen;
1590 	pktlen_aligned = roundup(pktlen, sizeof(uint64_t));
1591 
1592 	cp.cp_hdr.cph_type = type;
1593 	cp.cp_hdr.cph_flags = flags;
1594 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_hlen, sizeof(cp));
1595 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_tlen, pktlen_aligned);
1596 	cp.cp_hdr.cph_tid = rid;
1597 
1598 	iov[0].iov_base = &cp;
1599 	iov[0].iov_len = sizeof(cp);
1600 
1601 	iov[1].iov_base = data;
1602 	iov[1].iov_len = datalen;
1603 
1604 	iov[2].iov_base = &zeropad;
1605 	iov[2].iov_len = pktlen_aligned - pktlen;
1606 
1607 	mutex_enter(&ch->ch_wrd.rd_lock);
1608 	rv = vmbus_ring_write(&ch->ch_wrd, iov, 3, &needsig);
1609 	mutex_exit(&ch->ch_wrd.rd_lock);
1610 	if (rv == 0 && needsig)
1611 		vmbus_channel_setevent(sc, ch);
1612 
1613 	return rv;
1614 }
1615 
1616 int
1617 vmbus_channel_send_sgl(struct vmbus_channel *ch, struct vmbus_gpa *sgl,
1618     uint32_t nsge, void *data, uint32_t datalen, uint64_t rid)
1619 {
1620 	struct vmbus_softc *sc = ch->ch_sc;
1621 	struct vmbus_chanpkt_sglist cp;
1622 	struct iovec iov[4];
1623 	uint32_t buflen, pktlen, pktlen_aligned;
1624 	uint64_t zeropad = 0;
1625 	int rv, needsig = 0;
1626 
1627 	buflen = sizeof(struct vmbus_gpa) * nsge;
1628 	pktlen = sizeof(cp) + datalen + buflen;
1629 	pktlen_aligned = roundup(pktlen, sizeof(uint64_t));
1630 
1631 	cp.cp_hdr.cph_type = VMBUS_CHANPKT_TYPE_GPA;
1632 	cp.cp_hdr.cph_flags = VMBUS_CHANPKT_FLAG_RC;
1633 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_hlen, sizeof(cp) + buflen);
1634 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_tlen, pktlen_aligned);
1635 	cp.cp_hdr.cph_tid = rid;
1636 	cp.cp_gpa_cnt = nsge;
1637 	cp.cp_rsvd = 0;
1638 
1639 	iov[0].iov_base = &cp;
1640 	iov[0].iov_len = sizeof(cp);
1641 
1642 	iov[1].iov_base = sgl;
1643 	iov[1].iov_len = buflen;
1644 
1645 	iov[2].iov_base = data;
1646 	iov[2].iov_len = datalen;
1647 
1648 	iov[3].iov_base = &zeropad;
1649 	iov[3].iov_len = pktlen_aligned - pktlen;
1650 
1651 	mutex_enter(&ch->ch_wrd.rd_lock);
1652 	rv = vmbus_ring_write(&ch->ch_wrd, iov, 4, &needsig);
1653 	mutex_exit(&ch->ch_wrd.rd_lock);
1654 	if (rv == 0 && needsig)
1655 		vmbus_channel_setevent(sc, ch);
1656 
1657 	return rv;
1658 }
1659 
1660 int
1661 vmbus_channel_send_prpl(struct vmbus_channel *ch, struct vmbus_gpa_range *prpl,
1662     uint32_t nprp, void *data, uint32_t datalen, uint64_t rid)
1663 {
1664 	struct vmbus_softc *sc = ch->ch_sc;
1665 	struct vmbus_chanpkt_prplist cp;
1666 	struct iovec iov[4];
1667 	uint32_t buflen, pktlen, pktlen_aligned;
1668 	uint64_t zeropad = 0;
1669 	int rv, needsig = 0;
1670 
1671 	buflen = sizeof(struct vmbus_gpa_range) * (nprp + 1);
1672 	pktlen = sizeof(cp) + datalen + buflen;
1673 	pktlen_aligned = roundup(pktlen, sizeof(uint64_t));
1674 
1675 	cp.cp_hdr.cph_type = VMBUS_CHANPKT_TYPE_GPA;
1676 	cp.cp_hdr.cph_flags = VMBUS_CHANPKT_FLAG_RC;
1677 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_hlen, sizeof(cp) + buflen);
1678 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_tlen, pktlen_aligned);
1679 	cp.cp_hdr.cph_tid = rid;
1680 	cp.cp_range_cnt = 1;
1681 	cp.cp_rsvd = 0;
1682 
1683 	iov[0].iov_base = &cp;
1684 	iov[0].iov_len = sizeof(cp);
1685 
1686 	iov[1].iov_base = prpl;
1687 	iov[1].iov_len = buflen;
1688 
1689 	iov[2].iov_base = data;
1690 	iov[2].iov_len = datalen;
1691 
1692 	iov[3].iov_base = &zeropad;
1693 	iov[3].iov_len = pktlen_aligned - pktlen;
1694 
1695 	mutex_enter(&ch->ch_wrd.rd_lock);
1696 	rv = vmbus_ring_write(&ch->ch_wrd, iov, 4, &needsig);
1697 	mutex_exit(&ch->ch_wrd.rd_lock);
1698 	if (rv == 0 && needsig)
1699 		vmbus_channel_setevent(sc, ch);
1700 
1701 	return rv;
1702 }
1703 
1704 static int
1705 vmbus_ring_peek(struct vmbus_ring_data *rrd, void *data, uint32_t datalen)
1706 {
1707 	uint32_t avail;
1708 
1709 	KASSERT(datalen <= rrd->rd_dsize);
1710 
1711 	vmbus_ring_avail(rrd, NULL, &avail);
1712 	if (avail < datalen)
1713 		return EAGAIN;
1714 
1715 	vmbus_ring_get(rrd, (uint8_t *)data, datalen, 1);
1716 	return 0;
1717 }
1718 
1719 static int
1720 vmbus_ring_read(struct vmbus_ring_data *rrd, void *data, uint32_t datalen,
1721     uint32_t offset)
1722 {
1723 	uint64_t indices;
1724 	uint32_t avail;
1725 
1726 	KASSERT(datalen <= rrd->rd_dsize);
1727 
1728 	vmbus_ring_avail(rrd, NULL, &avail);
1729 	if (avail < datalen) {
1730 		DPRINTF("%s: avail %u datalen %u\n", __func__, avail, datalen);
1731 		return EAGAIN;
1732 	}
1733 
1734 	if (offset) {
1735 		rrd->rd_cons += offset;
1736 		if (rrd->rd_cons >= rrd->rd_dsize)
1737 			rrd->rd_cons -= rrd->rd_dsize;
1738 	}
1739 
1740 	vmbus_ring_get(rrd, (uint8_t *)data, datalen, 0);
1741 	vmbus_ring_get(rrd, (uint8_t *)&indices, sizeof(indices), 0);
1742 
1743 	membar_sync();
1744 	rrd->rd_ring->br_rindex = rrd->rd_cons;
1745 
1746 	return 0;
1747 }
1748 
1749 int
1750 vmbus_channel_recv(struct vmbus_channel *ch, void *data, uint32_t datalen,
1751     uint32_t *rlen, uint64_t *rid, int raw)
1752 {
1753 	struct vmbus_softc *sc = ch->ch_sc;
1754 	struct vmbus_chanpkt_hdr cph;
1755 	uint32_t offset, pktlen;
1756 	int rv;
1757 
1758 	*rlen = 0;
1759 
1760 	mutex_enter(&ch->ch_rrd.rd_lock);
1761 
1762 	if ((rv = vmbus_ring_peek(&ch->ch_rrd, &cph, sizeof(cph))) != 0) {
1763 		mutex_exit(&ch->ch_rrd.rd_lock);
1764 		return rv;
1765 	}
1766 
1767 	offset = raw ? 0 : VMBUS_CHANPKT_GETLEN(cph.cph_hlen);
1768 	pktlen = VMBUS_CHANPKT_GETLEN(cph.cph_tlen) - offset;
1769 	if (pktlen > datalen) {
1770 		mutex_exit(&ch->ch_rrd.rd_lock);
1771 		device_printf(sc->sc_dev, "%s: pktlen %u datalen %u\n",
1772 		    __func__, pktlen, datalen);
1773 		return EINVAL;
1774 	}
1775 
1776 	rv = vmbus_ring_read(&ch->ch_rrd, data, pktlen, offset);
1777 	if (rv == 0) {
1778 		*rlen = pktlen;
1779 		*rid = cph.cph_tid;
1780 	}
1781 
1782 	mutex_exit(&ch->ch_rrd.rd_lock);
1783 
1784 	return rv;
1785 }
1786 
1787 static inline void
1788 vmbus_ring_mask(struct vmbus_ring_data *rd)
1789 {
1790 
1791 	membar_sync();
1792 	rd->rd_ring->br_imask = 1;
1793 	membar_sync();
1794 }
1795 
1796 static inline void
1797 vmbus_ring_unmask(struct vmbus_ring_data *rd)
1798 {
1799 
1800 	membar_sync();
1801 	rd->rd_ring->br_imask = 0;
1802 	membar_sync();
1803 }
1804 
1805 static void
1806 vmbus_channel_pause(struct vmbus_channel *ch)
1807 {
1808 
1809 	vmbus_ring_mask(&ch->ch_rrd);
1810 }
1811 
1812 static uint32_t
1813 vmbus_channel_unpause(struct vmbus_channel *ch)
1814 {
1815 	uint32_t avail;
1816 
1817 	vmbus_ring_unmask(&ch->ch_rrd);
1818 	vmbus_ring_avail(&ch->ch_rrd, NULL, &avail);
1819 
1820 	return avail;
1821 }
1822 
1823 static uint32_t
1824 vmbus_channel_ready(struct vmbus_channel *ch)
1825 {
1826 	uint32_t avail;
1827 
1828 	vmbus_ring_avail(&ch->ch_rrd, NULL, &avail);
1829 
1830 	return avail;
1831 }
1832 
1833 /* How many PFNs can be referenced by the header */
1834 #define VMBUS_NPFNHDR	((VMBUS_MSG_DSIZE_MAX -	\
1835 	  sizeof(struct vmbus_chanmsg_gpadl_conn)) / sizeof(uint64_t))
1836 
1837 /* How many PFNs can be referenced by the body */
1838 #define VMBUS_NPFNBODY	((VMBUS_MSG_DSIZE_MAX -	\
1839 	  sizeof(struct vmbus_chanmsg_gpadl_subconn)) / sizeof(uint64_t))
1840 
1841 int
1842 vmbus_handle_alloc(struct vmbus_channel *ch, const struct hyperv_dma *dma,
1843     uint32_t buflen, uint32_t *handle)
1844 {
1845 	const int prflags = cold ? PR_NOWAIT : PR_WAITOK;
1846 	const int kmemflags = cold ? KM_NOSLEEP : KM_SLEEP;
1847 	const int msgflags = cold ? MSGF_NOSLEEP : 0;
1848 	const int hcflags = cold ? HCF_NOSLEEP : HCF_SLEEPOK;
1849 	struct vmbus_softc *sc = ch->ch_sc;
1850 	struct vmbus_chanmsg_gpadl_conn *hdr;
1851 	struct vmbus_chanmsg_gpadl_subconn *cmd;
1852 	struct vmbus_chanmsg_gpadl_connresp rsp;
1853 	struct vmbus_msg *msg;
1854 	int i, j, last, left, rv;
1855 	int bodylen = 0, ncmds = 0, pfn = 0;
1856 	uint64_t *frames;
1857 	paddr_t pa;
1858 	uint8_t *body;
1859 	/* Total number of pages to reference */
1860 	int total = atop(buflen);
1861 	/* Number of pages that will fit the header */
1862 	int inhdr = MIN(total, VMBUS_NPFNHDR);
1863 
1864 	KASSERT((buflen & PAGE_MASK) == 0);
1865 	KASSERT(buflen == (uint32_t)dma->map->dm_mapsize);
1866 
1867 	msg = pool_cache_get_paddr(sc->sc_msgpool, prflags, &pa);
1868 	if (msg == NULL)
1869 		return ENOMEM;
1870 
1871 	/* Prepare array of frame addresses */
1872 	frames = kmem_zalloc(total * sizeof(*frames), kmemflags);
1873 	if (frames == NULL) {
1874 		pool_cache_put_paddr(sc->sc_msgpool, msg, pa);
1875 		return ENOMEM;
1876 	}
1877 	for (i = 0, j = 0; i < dma->map->dm_nsegs && j < total; i++) {
1878 		bus_dma_segment_t *seg = &dma->map->dm_segs[i];
1879 		bus_addr_t addr = seg->ds_addr;
1880 
1881 		KASSERT((addr & PAGE_MASK) == 0);
1882 		KASSERT((seg->ds_len & PAGE_MASK) == 0);
1883 
1884 		while (addr < seg->ds_addr + seg->ds_len && j < total) {
1885 			frames[j++] = atop(addr);
1886 			addr += PAGE_SIZE;
1887 		}
1888 	}
1889 
1890 	memset(msg, 0, sizeof(*msg));
1891 	msg->msg_req.hc_dsize = sizeof(struct vmbus_chanmsg_gpadl_conn) +
1892 	    inhdr * sizeof(uint64_t);
1893 	hdr = (struct vmbus_chanmsg_gpadl_conn *)msg->msg_req.hc_data;
1894 	msg->msg_rsp = &rsp;
1895 	msg->msg_rsplen = sizeof(rsp);
1896 	msg->msg_flags = msgflags;
1897 
1898 	left = total - inhdr;
1899 
1900 	/* Allocate additional gpadl_body structures if required */
1901 	if (left > 0) {
1902 		ncmds = MAX(1, left / VMBUS_NPFNBODY + left % VMBUS_NPFNBODY);
1903 		bodylen = ncmds * VMBUS_MSG_DSIZE_MAX;
1904 		body = kmem_zalloc(bodylen, kmemflags);
1905 		if (body == NULL) {
1906 			kmem_free(frames, total * sizeof(*frames));
1907 			pool_cache_put_paddr(sc->sc_msgpool, msg, pa);
1908 			return ENOMEM;
1909 		}
1910 	}
1911 
1912 	*handle = atomic_add_int_nv(&sc->sc_handle, 1);
1913 
1914 	hdr->chm_hdr.chm_type = VMBUS_CHANMSG_GPADL_CONN;
1915 	hdr->chm_chanid = ch->ch_id;
1916 	hdr->chm_gpadl = *handle;
1917 
1918 	/* Single range for a contiguous buffer */
1919 	hdr->chm_range_cnt = 1;
1920 	hdr->chm_range_len = sizeof(struct vmbus_gpa_range) + total *
1921 	    sizeof(uint64_t);
1922 	hdr->chm_range.gpa_ofs = 0;
1923 	hdr->chm_range.gpa_len = buflen;
1924 
1925 	/* Fit as many pages as possible into the header */
1926 	for (i = 0; i < inhdr; i++)
1927 		hdr->chm_range.gpa_page[i] = frames[pfn++];
1928 
1929 	for (i = 0; i < ncmds; i++) {
1930 		cmd = (struct vmbus_chanmsg_gpadl_subconn *)(body +
1931 		    VMBUS_MSG_DSIZE_MAX * i);
1932 		cmd->chm_hdr.chm_type = VMBUS_CHANMSG_GPADL_SUBCONN;
1933 		cmd->chm_gpadl = *handle;
1934 		last = MIN(left, VMBUS_NPFNBODY);
1935 		for (j = 0; j < last; j++)
1936 			cmd->chm_gpa_page[j] = frames[pfn++];
1937 		left -= last;
1938 	}
1939 
1940 	rv = vmbus_start(sc, msg, pa);
1941 	if (rv != 0) {
1942 		DPRINTF("%s: GPADL_CONN failed\n", device_xname(sc->sc_dev));
1943 		goto out;
1944 	}
1945 	for (i = 0; i < ncmds; i++) {
1946 		int cmdlen = sizeof(*cmd);
1947 		cmd = (struct vmbus_chanmsg_gpadl_subconn *)(body +
1948 		    VMBUS_MSG_DSIZE_MAX * i);
1949 		/* Last element can be short */
1950 		if (i == ncmds - 1)
1951 			cmdlen += last * sizeof(uint64_t);
1952 		else
1953 			cmdlen += VMBUS_NPFNBODY * sizeof(uint64_t);
1954 		rv = vmbus_cmd(sc, cmd, cmdlen, NULL, 0, HCF_NOREPLY | hcflags);
1955 		if (rv != 0) {
1956 			DPRINTF("%s: GPADL_SUBCONN (iteration %d/%d) failed "
1957 			    "with %d\n", device_xname(sc->sc_dev), i, ncmds,
1958 			    rv);
1959 			goto out;
1960 		}
1961 	}
1962 	rv = vmbus_reply(sc, msg);
1963 	if (rv != 0) {
1964 		DPRINTF("%s: GPADL allocation failed with %d\n",
1965 		    device_xname(sc->sc_dev), rv);
1966 	}
1967 
1968  out:
1969 	if (bodylen > 0)
1970 		kmem_free(body, bodylen);
1971 	kmem_free(frames, total * sizeof(*frames));
1972 	pool_cache_put_paddr(sc->sc_msgpool, msg, pa);
1973 	if (rv)
1974 		return rv;
1975 
1976 	KASSERT(*handle == rsp.chm_gpadl);
1977 
1978 	return 0;
1979 }
1980 
1981 void
1982 vmbus_handle_free(struct vmbus_channel *ch, uint32_t handle)
1983 {
1984 	struct vmbus_softc *sc = ch->ch_sc;
1985 	struct vmbus_chanmsg_gpadl_disconn cmd;
1986 	struct vmbus_chanmsg_gpadl_disconn rsp;
1987 	int rv;
1988 
1989 	memset(&cmd, 0, sizeof(cmd));
1990 	cmd.chm_hdr.chm_type = VMBUS_CHANMSG_GPADL_DISCONN;
1991 	cmd.chm_chanid = ch->ch_id;
1992 	cmd.chm_gpadl = handle;
1993 
1994 	rv = vmbus_cmd(sc, &cmd, sizeof(cmd), &rsp, sizeof(rsp),
1995 	    cold ? HCF_NOSLEEP : HCF_SLEEPOK);
1996 	if (rv) {
1997 		DPRINTF("%s: GPADL_DISCONN failed with %d\n",
1998 		    device_xname(sc->sc_dev), rv);
1999 	}
2000 }
2001 
2002 static int
2003 vmbus_attach_print(void *aux, const char *name)
2004 {
2005 	struct vmbus_attach_args *aa = aux;
2006 
2007 	if (name)
2008 		printf("\"%s\" at %s", aa->aa_ident, name);
2009 
2010 	return UNCONF;
2011 }
2012 
2013 static int
2014 vmbus_attach_icdevs(struct vmbus_softc *sc)
2015 {
2016 	struct vmbus_dev *dv;
2017 	struct vmbus_channel *ch;
2018 
2019 	SLIST_INIT(&sc->sc_icdevs);
2020 	mutex_init(&sc->sc_icdev_lock, MUTEX_DEFAULT, IPL_NET);
2021 
2022 	TAILQ_FOREACH(ch, &sc->sc_channels, ch_entry) {
2023 		if (ch->ch_state != VMBUS_CHANSTATE_OFFERED)
2024 			continue;
2025 		if (ch->ch_flags & CHF_MONITOR)
2026 			continue;
2027 
2028 		dv = kmem_zalloc(sizeof(*dv), cold ? KM_NOSLEEP : KM_SLEEP);
2029 		if (dv == NULL) {
2030 			device_printf(sc->sc_dev,
2031 			    "failed to allocate ic device object\n");
2032 			return ENOMEM;
2033 		}
2034 		dv->dv_aa.aa_type = &ch->ch_type;
2035 		dv->dv_aa.aa_inst = &ch->ch_inst;
2036 		dv->dv_aa.aa_ident = ch->ch_ident;
2037 		dv->dv_aa.aa_chan = ch;
2038 		dv->dv_aa.aa_iot = sc->sc_iot;
2039 		dv->dv_aa.aa_memt = sc->sc_memt;
2040 		mutex_enter(&sc->sc_icdev_lock);
2041 		SLIST_INSERT_HEAD(&sc->sc_icdevs, dv, dv_entry);
2042 		mutex_exit(&sc->sc_icdev_lock);
2043 		ch->ch_dev = config_found_ia(sc->sc_dev, "hypervvmbus",
2044 		    &dv->dv_aa, vmbus_attach_print);
2045 	}
2046 	return 0;
2047 }
2048 
2049 static int
2050 vmbus_attach_devices(struct vmbus_softc *sc)
2051 {
2052 	struct vmbus_dev *dv;
2053 	struct vmbus_channel *ch;
2054 
2055 	SLIST_INIT(&sc->sc_devs);
2056 	mutex_init(&sc->sc_dev_lock, MUTEX_DEFAULT, IPL_NET);
2057 
2058 	TAILQ_FOREACH(ch, &sc->sc_channels, ch_entry) {
2059 		if (ch->ch_state != VMBUS_CHANSTATE_OFFERED)
2060 			continue;
2061 		if (!(ch->ch_flags & CHF_MONITOR))
2062 			continue;
2063 
2064 		dv = kmem_zalloc(sizeof(*dv), cold ? KM_NOSLEEP : KM_SLEEP);
2065 		if (dv == NULL) {
2066 			device_printf(sc->sc_dev,
2067 			    "failed to allocate device object\n");
2068 			return ENOMEM;
2069 		}
2070 		dv->dv_aa.aa_type = &ch->ch_type;
2071 		dv->dv_aa.aa_inst = &ch->ch_inst;
2072 		dv->dv_aa.aa_ident = ch->ch_ident;
2073 		dv->dv_aa.aa_chan = ch;
2074 		dv->dv_aa.aa_iot = sc->sc_iot;
2075 		dv->dv_aa.aa_memt = sc->sc_memt;
2076 		mutex_enter(&sc->sc_dev_lock);
2077 		SLIST_INSERT_HEAD(&sc->sc_devs, dv, dv_entry);
2078 		mutex_exit(&sc->sc_dev_lock);
2079 		ch->ch_dev = config_found_ia(sc->sc_dev, "hypervvmbus",
2080 		    &dv->dv_aa, vmbus_attach_print);
2081 	}
2082 	return 0;
2083 }
2084 
2085 MODULE(MODULE_CLASS_DRIVER, vmbus, "hyperv");
2086 
2087 #ifdef _MODULE
2088 #include "ioconf.c"
2089 #endif
2090 
2091 static int
2092 vmbus_modcmd(modcmd_t cmd, void *aux)
2093 {
2094 	int rv = 0;
2095 
2096 	switch (cmd) {
2097 	case MODULE_CMD_INIT:
2098 #ifdef _MODULE
2099 		rv = config_init_component(cfdriver_ioconf_vmbus,
2100 		    cfattach_ioconf_vmbus, cfdata_ioconf_vmbus);
2101 #endif
2102 		break;
2103 
2104 	case MODULE_CMD_FINI:
2105 #ifdef _MODULE
2106 		rv = config_fini_component(cfdriver_ioconf_vmbus,
2107 		    cfattach_ioconf_vmbus, cfdata_ioconf_vmbus);
2108 #endif
2109 		break;
2110 
2111 	default:
2112 		rv = ENOTTY;
2113 		break;
2114 	}
2115 
2116 	return rv;
2117 }
2118