xref: /openbsd-src/sys/dev/pv/hyperv.c (revision 46035553bfdd96e63c94e32da0210227ec2e3cf1)
1 /*-
2  * Copyright (c) 2009-2012 Microsoft Corp.
3  * Copyright (c) 2012 NetApp Inc.
4  * Copyright (c) 2012 Citrix Inc.
5  * Copyright (c) 2016 Mike Belopuhov <mike@esdenera.com>
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice unmodified, this list of conditions, and the following
13  *    disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
22  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
23  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
24  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
25  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
26  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
27  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
28  */
29 
30 /*
31  * The OpenBSD port was done under funding by Esdenera Networks GmbH.
32  */
33 
34 #include <sys/param.h>
35 
36 /* Hyperv requires locked atomic operations */
37 #ifndef MULTIPROCESSOR
38 #define _HYPERVMPATOMICS
39 #define MULTIPROCESSOR
40 #endif
41 #include <sys/atomic.h>
42 #ifdef _HYPERVMPATOMICS
43 #undef MULTIPROCESSOR
44 #undef _HYPERVMPATOMICS
45 #endif
46 
47 #include <sys/systm.h>
48 #include <sys/proc.h>
49 #include <sys/signal.h>
50 #include <sys/signalvar.h>
51 #include <sys/malloc.h>
52 #include <sys/kernel.h>
53 #include <sys/device.h>
54 #include <sys/timetc.h>
55 #include <sys/task.h>
56 #include <sys/syslog.h>
57 
58 #include <machine/bus.h>
59 #include <machine/cpu.h>
60 #include <machine/cpufunc.h>
61 
62 #include <uvm/uvm_extern.h>
63 
64 #include <machine/i82489var.h>
65 
66 #include <dev/pv/pvvar.h>
67 #include <dev/pv/pvreg.h>
68 #include <dev/pv/hypervreg.h>
69 #include <dev/pv/hypervvar.h>
70 
71 /* Command submission flags */
72 #define HCF_SLEEPOK	0x0001	/* M_WAITOK */
73 #define HCF_NOSLEEP	0x0002	/* M_NOWAIT */
74 #define HCF_NOREPLY	0x0004
75 
76 struct hv_softc *hv_sc;
77 
78 int 	hv_match(struct device *, void *, void *);
79 void	hv_attach(struct device *, struct device *, void *);
80 void	hv_set_version(struct hv_softc *);
81 u_int	hv_gettime(struct timecounter *);
82 int	hv_init_hypercall(struct hv_softc *);
83 uint64_t hv_hypercall(struct hv_softc *, uint64_t, void *, void *);
84 int	hv_init_interrupts(struct hv_softc *);
85 int	hv_init_synic(struct hv_softc *);
86 int	hv_cmd(struct hv_softc *, void *, size_t, void *, size_t, int);
87 int	hv_start(struct hv_softc *, struct hv_msg *);
88 int	hv_reply(struct hv_softc *, struct hv_msg *);
89 void	hv_wait(struct hv_softc *, int (*done)(struct hv_softc *,
90 	    struct hv_msg *), struct hv_msg *, void *, const char *);
91 uint16_t hv_intr_signal(struct hv_softc *, void *);
92 void	hv_intr(void);
93 void	hv_event_intr(struct hv_softc *);
94 void	hv_message_intr(struct hv_softc *);
95 int	hv_vmbus_connect(struct hv_softc *);
96 void	hv_channel_response(struct hv_softc *, struct vmbus_chanmsg_hdr *);
97 void	hv_channel_offer(struct hv_softc *, struct vmbus_chanmsg_hdr *);
98 void	hv_channel_rescind(struct hv_softc *, struct vmbus_chanmsg_hdr *);
99 void	hv_channel_delivered(struct hv_softc *, struct vmbus_chanmsg_hdr *);
100 int	hv_channel_scan(struct hv_softc *);
101 void	hv_process_offer(struct hv_softc *, struct hv_offer *);
102 struct hv_channel *
103 	hv_channel_lookup(struct hv_softc *, uint32_t);
104 int	hv_channel_ring_create(struct hv_channel *, uint32_t);
105 void	hv_channel_ring_destroy(struct hv_channel *);
106 void	hv_channel_pause(struct hv_channel *);
107 uint	hv_channel_unpause(struct hv_channel *);
108 uint	hv_channel_ready(struct hv_channel *);
109 extern void hv_attach_icdevs(struct hv_softc *);
110 int	hv_attach_devices(struct hv_softc *);
111 
112 struct {
113 	int		  hmd_response;
114 	int		  hmd_request;
115 	void		(*hmd_handler)(struct hv_softc *,
116 			    struct vmbus_chanmsg_hdr *);
117 } hv_msg_dispatch[] = {
118 	{ 0,					0, NULL },
119 	{ VMBUS_CHANMSG_CHOFFER,		0, hv_channel_offer },
120 	{ VMBUS_CHANMSG_CHRESCIND,		0, hv_channel_rescind },
121 	{ VMBUS_CHANMSG_CHREQUEST,		VMBUS_CHANMSG_CHOFFER,
122 	  NULL },
123 	{ VMBUS_CHANMSG_CHOFFER_DONE,		0,
124 	  hv_channel_delivered },
125 	{ VMBUS_CHANMSG_CHOPEN,			0, NULL },
126 	{ VMBUS_CHANMSG_CHOPEN_RESP,		VMBUS_CHANMSG_CHOPEN,
127 	  hv_channel_response },
128 	{ VMBUS_CHANMSG_CHCLOSE,		0, NULL },
129 	{ VMBUS_CHANMSG_GPADL_CONN,		0, NULL },
130 	{ VMBUS_CHANMSG_GPADL_SUBCONN,		0, NULL },
131 	{ VMBUS_CHANMSG_GPADL_CONNRESP,		VMBUS_CHANMSG_GPADL_CONN,
132 	  hv_channel_response },
133 	{ VMBUS_CHANMSG_GPADL_DISCONN,		0, NULL },
134 	{ VMBUS_CHANMSG_GPADL_DISCONNRESP,	VMBUS_CHANMSG_GPADL_DISCONN,
135 	  hv_channel_response },
136 	{ VMBUS_CHANMSG_CHFREE,			0, NULL },
137 	{ VMBUS_CHANMSG_CONNECT,		0, NULL },
138 	{ VMBUS_CHANMSG_CONNECT_RESP,		VMBUS_CHANMSG_CONNECT,
139 	  hv_channel_response },
140 	{ VMBUS_CHANMSG_DISCONNECT,		0, NULL },
141 };
142 
143 struct timecounter hv_timecounter = {
144 	hv_gettime, 0, 0xffffffff, 10000000, "hyperv", 9001, NULL, 0
145 };
146 
147 struct cfdriver hyperv_cd = {
148 	NULL, "hyperv", DV_DULL
149 };
150 
151 const struct cfattach hyperv_ca = {
152 	sizeof(struct hv_softc), hv_match, hv_attach
153 };
154 
155 const struct hv_guid hv_guid_network = {
156 	{ 0x63, 0x51, 0x61, 0xf8, 0x3e, 0xdf, 0xc5, 0x46,
157 	  0x91, 0x3f, 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e }
158 };
159 
160 const struct hv_guid hv_guid_ide = {
161 	{ 0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
162 	  0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5 }
163 };
164 
165 const struct hv_guid hv_guid_scsi = {
166 	{ 0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
167 	  0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f }
168 };
169 
170 const struct hv_guid hv_guid_shutdown = {
171 	{ 0x31, 0x60, 0x0b, 0x0e, 0x13, 0x52, 0x34, 0x49,
172 	  0x81, 0x8b, 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb }
173 };
174 
175 const struct hv_guid hv_guid_timesync = {
176 	{ 0x30, 0xe6, 0x27, 0x95, 0xae, 0xd0, 0x7b, 0x49,
177 	  0xad, 0xce, 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf }
178 };
179 
180 const struct hv_guid hv_guid_heartbeat = {
181 	{ 0x39, 0x4f, 0x16, 0x57, 0x15, 0x91, 0x78, 0x4e,
182 	  0xab, 0x55, 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d }
183 };
184 
185 const struct hv_guid hv_guid_kvp = {
186 	{ 0xe7, 0xf4, 0xa0, 0xa9, 0x45, 0x5a, 0x96, 0x4d,
187 	  0xb8, 0x27, 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6 }
188 };
189 
190 #ifdef HYPERV_DEBUG
191 const struct hv_guid hv_guid_vss = {
192 	{ 0x29, 0x2e, 0xfa, 0x35, 0x23, 0xea, 0x36, 0x42,
193 	  0x96, 0xae, 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40 }
194 };
195 
196 const struct hv_guid hv_guid_dynmem = {
197 	{ 0xdc, 0x74, 0x50, 0x52, 0x85, 0x89, 0xe2, 0x46,
198 	  0x80, 0x57, 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02 }
199 };
200 
201 const struct hv_guid hv_guid_mouse = {
202 	{ 0x9e, 0xb6, 0xa8, 0xcf, 0x4a, 0x5b, 0xc0, 0x4c,
203 	  0xb9, 0x8b, 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a }
204 };
205 
206 const struct hv_guid hv_guid_kbd = {
207 	{ 0x6d, 0xad, 0x12, 0xf9, 0x17, 0x2b, 0xea, 0x48,
208 	  0xbd, 0x65, 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84 }
209 };
210 
211 const struct hv_guid hv_guid_video = {
212 	{ 0x02, 0x78, 0x0a, 0xda, 0x77, 0xe3, 0xac, 0x4a,
213 	  0x8e, 0x77, 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8 }
214 };
215 
216 const struct hv_guid hv_guid_fc = {
217 	{ 0x4a, 0xcc, 0x9b, 0x2f, 0x69, 0x00, 0xf3, 0x4a,
218 	  0xb7, 0x6b, 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda }
219 };
220 
221 const struct hv_guid hv_guid_fcopy = {
222 	{ 0xe3, 0x4b, 0xd1, 0x34, 0xe4, 0xde, 0xc8, 0x41,
223 	  0x9a, 0xe7, 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92 }
224 };
225 
226 const struct hv_guid hv_guid_pcie = {
227 	{ 0x1d, 0xf6, 0xc4, 0x44, 0x44, 0x44, 0x00, 0x44,
228 	  0x9d, 0x52, 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f }
229 };
230 
231 const struct hv_guid hv_guid_netdir = {
232 	{ 0x3d, 0xaf, 0x2e, 0x8c, 0xa7, 0x32, 0x09, 0x4b,
233 	  0xab, 0x99, 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01 }
234 };
235 
236 const struct hv_guid hv_guid_rdesktop = {
237 	{ 0xf4, 0xac, 0x6a, 0x27, 0x15, 0xac, 0x6c, 0x42,
238 	  0x98, 0xdd, 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe }
239 };
240 
241 /* Automatic Virtual Machine Activation (AVMA) Services */
242 const struct hv_guid hv_guid_avma1 = {
243 	{ 0x55, 0xb2, 0x87, 0x44, 0x8c, 0xb8, 0x3f, 0x40,
244 	  0xbb, 0x51, 0xd1, 0xf6, 0x9c, 0xf1, 0x7f, 0x87 }
245 };
246 
247 const struct hv_guid hv_guid_avma2 = {
248 	{ 0xf4, 0xba, 0x75, 0x33, 0x15, 0x9e, 0x30, 0x4b,
249 	  0xb7, 0x65, 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b }
250 };
251 
252 const struct hv_guid hv_guid_avma3 = {
253 	{ 0xa0, 0x1f, 0x22, 0x99, 0xad, 0x24, 0xe2, 0x11,
254 	  0xbe, 0x98, 0x00, 0x1a, 0xa0, 0x1b, 0xbf, 0x6e }
255 };
256 
257 const struct hv_guid hv_guid_avma4 = {
258 	{ 0x16, 0x57, 0xe6, 0xf8, 0xb3, 0x3c, 0x06, 0x4a,
259 	  0x9a, 0x60, 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5 }
260 };
261 #endif	/* HYPERV_DEBUG */
262 
263 int
264 hv_match(struct device *parent, void *match, void *aux)
265 {
266 	struct pv_attach_args *pva = aux;
267 	struct pvbus_hv *hv = &pva->pva_hv[PVBUS_HYPERV];
268 
269 	if ((hv->hv_major == 0 && hv->hv_minor == 0) || hv->hv_base == 0)
270 		return (0);
271 
272 	return (1);
273 }
274 
275 void
276 hv_attach(struct device *parent, struct device *self, void *aux)
277 {
278 	struct hv_softc *sc = (struct hv_softc *)self;
279 	struct pv_attach_args *pva = aux;
280 	struct pvbus_hv *hv = &pva->pva_hv[PVBUS_HYPERV];
281 
282 	sc->sc_pvbus = hv;
283 	sc->sc_dmat = pva->pva_dmat;
284 
285 	if (!(hv->hv_features & CPUID_HV_MSR_HYPERCALL) ||
286 	    !(hv->hv_features & CPUID_HV_MSR_SYNIC)) {
287 		printf(": not functional\n");
288 		return;
289 	}
290 
291 	DPRINTF("\n");
292 
293 	hv_set_version(sc);
294 
295 	if (hv->hv_features & CPUID_HV_MSR_TIME_REFCNT)
296 		tc_init(&hv_timecounter);
297 
298 	if (hv_init_hypercall(sc))
299 		return;
300 
301 	/* Wire it up to the global */
302 	hv_sc = sc;
303 
304 	if (hv_init_interrupts(sc))
305 		return;
306 
307 	if (hv_vmbus_connect(sc))
308 		return;
309 
310 	DPRINTF("%s", sc->sc_dev.dv_xname);
311 	printf(": protocol %d.%d, features %#x\n",
312 	    VMBUS_VERSION_MAJOR(sc->sc_proto),
313 	    VMBUS_VERSION_MINOR(sc->sc_proto),
314 	    hv->hv_features);
315 
316 	if (hv_channel_scan(sc))
317 		return;
318 
319 	/* Attach heartbeat, KVP and other "internal" services */
320 	hv_attach_icdevs(sc);
321 
322 	/* Attach devices with external drivers */
323 	hv_attach_devices(sc);
324 }
325 
326 void
327 hv_set_version(struct hv_softc *sc)
328 {
329 	uint64_t ver;
330 
331 	/* OpenBSD build date */
332 	ver = MSR_HV_GUESTID_OSTYPE_OPENBSD;
333 	ver |= (uint64_t)OpenBSD << MSR_HV_GUESTID_VERSION_SHIFT;
334 	wrmsr(MSR_HV_GUEST_OS_ID, ver);
335 }
336 
337 u_int
338 hv_gettime(struct timecounter *tc)
339 {
340 	u_int now = rdmsr(MSR_HV_TIME_REF_COUNT);
341 
342 	return (now);
343 }
344 
345 int
346 hv_init_hypercall(struct hv_softc *sc)
347 {
348 	extern void *hv_hypercall_page;
349 	uint64_t msr;
350 	paddr_t pa;
351 
352 	sc->sc_hc = &hv_hypercall_page;
353 
354 	if (!pmap_extract(pmap_kernel(), (vaddr_t)sc->sc_hc, &pa)) {
355 		printf(": hypercall page PA extraction failed\n");
356 		return (-1);
357 	}
358 
359 	msr = (atop(pa) << MSR_HV_HYPERCALL_PGSHIFT) | MSR_HV_HYPERCALL_ENABLE;
360 	wrmsr(MSR_HV_HYPERCALL, msr);
361 
362 	if (!(rdmsr(MSR_HV_HYPERCALL) & MSR_HV_HYPERCALL_ENABLE)) {
363 		printf(": failed to set up a hypercall page\n");
364 		return (-1);
365 	}
366 
367 	return (0);
368 }
369 
370 uint64_t
371 hv_hypercall(struct hv_softc *sc, uint64_t control, void *input,
372     void *output)
373 {
374 	paddr_t input_pa = 0, output_pa = 0;
375 	uint64_t status = 0;
376 
377 	if (input != NULL &&
378 	    pmap_extract(pmap_kernel(), (vaddr_t)input, &input_pa) == 0) {
379 		printf("%s: hypercall input PA extraction failed\n",
380 		    sc->sc_dev.dv_xname);
381 		return (~HYPERCALL_STATUS_SUCCESS);
382 	}
383 
384 	if (output != NULL &&
385 	    pmap_extract(pmap_kernel(), (vaddr_t)output, &output_pa) == 0) {
386 		printf("%s: hypercall output PA extraction failed\n",
387 		    sc->sc_dev.dv_xname);
388 		return (~HYPERCALL_STATUS_SUCCESS);
389 	}
390 
391 #ifdef __amd64__
392 	__asm__ __volatile__ ("mov %0, %%r8" : : "r" (output_pa) : "r8");
393 	__asm__ __volatile__ ("call *%3" : "=a" (status) : "c" (control),
394 	    "d" (input_pa), "m" (sc->sc_hc));
395 #else  /* __i386__ */
396 	{
397 		uint32_t control_hi = control >> 32;
398 		uint32_t control_lo = control & 0xfffffffff;
399 		uint32_t status_hi = 1;
400 		uint32_t status_lo = 1;
401 
402 		__asm__ __volatile__ ("call *%8" :
403 		    "=d" (status_hi), "=a"(status_lo) :
404 		    "d" (control_hi), "a" (control_lo),
405 		    "b" (0), "c" (input_pa), "D" (0), "S" (output_pa),
406 		    "m" (sc->sc_hc));
407 
408 		status = status_lo | ((uint64_t)status_hi << 32);
409 	}
410 #endif	/* __amd64__ */
411 
412 	return (status);
413 }
414 
415 int
416 hv_init_interrupts(struct hv_softc *sc)
417 {
418 	struct cpu_info *ci = curcpu();
419 	int cpu = CPU_INFO_UNIT(ci);
420 
421 	sc->sc_idtvec = LAPIC_HYPERV_VECTOR;
422 
423 	TAILQ_INIT(&sc->sc_reqs);
424 	mtx_init(&sc->sc_reqlck, IPL_NET);
425 
426 	TAILQ_INIT(&sc->sc_rsps);
427 	mtx_init(&sc->sc_rsplck, IPL_NET);
428 
429 	sc->sc_simp[cpu] = km_alloc(PAGE_SIZE, &kv_any, &kp_zero, &kd_nowait);
430 	if (sc->sc_simp[cpu] == NULL) {
431 		printf(": failed to allocate SIMP\n");
432 		return (-1);
433 	}
434 
435 	sc->sc_siep[cpu] = km_alloc(PAGE_SIZE, &kv_any, &kp_zero, &kd_nowait);
436 	if (sc->sc_siep[cpu] == NULL) {
437 		printf(": failed to allocate SIEP\n");
438 		km_free(sc->sc_simp[cpu], PAGE_SIZE, &kv_any, &kp_zero);
439 		return (-1);
440 	}
441 
442 	sc->sc_proto = VMBUS_VERSION_WS2008;
443 
444 	return (hv_init_synic(sc));
445 }
446 
447 int
448 hv_init_synic(struct hv_softc *sc)
449 {
450 	struct cpu_info *ci = curcpu();
451 	int cpu = CPU_INFO_UNIT(ci);
452 	uint64_t simp, siefp, sctrl, sint;
453 	paddr_t pa;
454 
455 	/*
456 	 * Setup the Synic's message page
457 	 */
458 	if (!pmap_extract(pmap_kernel(), (vaddr_t)sc->sc_simp[cpu], &pa)) {
459 		printf(": SIMP PA extraction failed\n");
460 		return (-1);
461 	}
462 	simp = rdmsr(MSR_HV_SIMP);
463 	simp &= (1 << MSR_HV_SIMP_PGSHIFT) - 1;
464 	simp |= (atop(pa) << MSR_HV_SIMP_PGSHIFT);
465 	simp |= MSR_HV_SIMP_ENABLE;
466 	wrmsr(MSR_HV_SIMP, simp);
467 
468 	/*
469 	 * Setup the Synic's event page
470 	 */
471 	if (!pmap_extract(pmap_kernel(), (vaddr_t)sc->sc_siep[cpu], &pa)) {
472 		printf(": SIEP PA extraction failed\n");
473 		return (-1);
474 	}
475 	siefp = rdmsr(MSR_HV_SIEFP);
476 	siefp &= (1<<MSR_HV_SIEFP_PGSHIFT) - 1;
477 	siefp |= (atop(pa) << MSR_HV_SIEFP_PGSHIFT);
478 	siefp |= MSR_HV_SIEFP_ENABLE;
479 	wrmsr(MSR_HV_SIEFP, siefp);
480 
481 	/*
482 	 * Configure and unmask SINT for message and event flags
483 	 */
484 	sint = rdmsr(MSR_HV_SINT0 + VMBUS_SINT_MESSAGE);
485 	sint = sc->sc_idtvec | MSR_HV_SINT_AUTOEOI |
486 	    (sint & MSR_HV_SINT_RSVD_MASK);
487 	wrmsr(MSR_HV_SINT0 + VMBUS_SINT_MESSAGE, sint);
488 
489 	/* Enable the global synic bit */
490 	sctrl = rdmsr(MSR_HV_SCONTROL);
491 	sctrl |= MSR_HV_SCTRL_ENABLE;
492 	wrmsr(MSR_HV_SCONTROL, sctrl);
493 
494 	sc->sc_vcpus[cpu] = rdmsr(MSR_HV_VP_INDEX);
495 
496 	DPRINTF("vcpu%u: SIMP %#llx SIEFP %#llx SCTRL %#llx\n",
497 	    sc->sc_vcpus[cpu], simp, siefp, sctrl);
498 
499 	return (0);
500 }
501 
502 int
503 hv_cmd(struct hv_softc *sc, void *cmd, size_t cmdlen, void *rsp,
504     size_t rsplen, int flags)
505 {
506 	struct hv_msg msg;
507 	int rv;
508 
509 	if (cmdlen > VMBUS_MSG_DSIZE_MAX) {
510 		printf("%s: payload too large (%lu)\n", sc->sc_dev.dv_xname,
511 		    cmdlen);
512 		return (EMSGSIZE);
513 	}
514 
515 	memset(&msg, 0, sizeof(msg));
516 
517 	msg.msg_req.hc_dsize = cmdlen;
518 	memcpy(msg.msg_req.hc_data, cmd, cmdlen);
519 
520 	if (!(flags & HCF_NOREPLY)) {
521 		msg.msg_rsp = rsp;
522 		msg.msg_rsplen = rsplen;
523 	} else
524 		msg.msg_flags |= MSGF_NOQUEUE;
525 
526 	if (flags & HCF_NOSLEEP)
527 		msg.msg_flags |= MSGF_NOSLEEP;
528 
529 	if ((rv = hv_start(sc, &msg)) != 0)
530 		return (rv);
531 	return (hv_reply(sc, &msg));
532 }
533 
534 int
535 hv_start(struct hv_softc *sc, struct hv_msg *msg)
536 {
537 	const int delays[] = { 100, 100, 100, 500, 500, 5000, 5000, 5000 };
538 	const char *wchan = "hvstart";
539 	uint16_t status;
540 	int i, s;
541 
542 	msg->msg_req.hc_connid = VMBUS_CONNID_MESSAGE;
543 	msg->msg_req.hc_msgtype = 1;
544 
545 	if (!(msg->msg_flags & MSGF_NOQUEUE)) {
546 		mtx_enter(&sc->sc_reqlck);
547 		TAILQ_INSERT_TAIL(&sc->sc_reqs, msg, msg_entry);
548 		mtx_leave(&sc->sc_reqlck);
549 	}
550 
551 	for (i = 0; i < nitems(delays); i++) {
552 		status = hv_hypercall(sc, HYPERCALL_POST_MESSAGE,
553 		    &msg->msg_req, NULL);
554 		if (status == HYPERCALL_STATUS_SUCCESS)
555 			break;
556 		if (msg->msg_flags & MSGF_NOSLEEP) {
557 			delay(delays[i]);
558 			s = splnet();
559 			hv_intr();
560 			splx(s);
561 		} else {
562 			tsleep_nsec(wchan, PRIBIO, wchan,
563 			    USEC_TO_NSEC(delays[i]));
564 		}
565 	}
566 	if (status != 0) {
567 		printf("%s: posting vmbus message failed with %d\n",
568 		    sc->sc_dev.dv_xname, status);
569 		if (!(msg->msg_flags & MSGF_NOQUEUE)) {
570 			mtx_enter(&sc->sc_reqlck);
571 			TAILQ_REMOVE(&sc->sc_reqs, msg, msg_entry);
572 			mtx_leave(&sc->sc_reqlck);
573 		}
574 		return (EIO);
575 	}
576 
577 	return (0);
578 }
579 
580 static int
581 hv_reply_done(struct hv_softc *sc, struct hv_msg *msg)
582 {
583 	struct hv_msg *m;
584 
585 	mtx_enter(&sc->sc_rsplck);
586 	TAILQ_FOREACH(m, &sc->sc_rsps, msg_entry) {
587 		if (m == msg) {
588 			mtx_leave(&sc->sc_rsplck);
589 			return (1);
590 		}
591 	}
592 	mtx_leave(&sc->sc_rsplck);
593 	return (0);
594 }
595 
596 int
597 hv_reply(struct hv_softc *sc, struct hv_msg *msg)
598 {
599 	if (msg->msg_flags & MSGF_NOQUEUE)
600 		return (0);
601 
602 	hv_wait(sc, hv_reply_done, msg, msg, "hvreply");
603 
604 	mtx_enter(&sc->sc_rsplck);
605 	TAILQ_REMOVE(&sc->sc_rsps, msg, msg_entry);
606 	mtx_leave(&sc->sc_rsplck);
607 
608 	return (0);
609 }
610 
611 void
612 hv_wait(struct hv_softc *sc, int (*cond)(struct hv_softc *, struct hv_msg *),
613     struct hv_msg *msg, void *wchan, const char *wmsg)
614 {
615 	int s;
616 
617 	KASSERT(cold ? msg->msg_flags & MSGF_NOSLEEP : 1);
618 
619 	while (!cond(sc, msg)) {
620 		if (msg->msg_flags & MSGF_NOSLEEP) {
621 			delay(1000);
622 			s = splnet();
623 			hv_intr();
624 			splx(s);
625 		} else {
626 			tsleep_nsec(wchan, PRIBIO, wmsg ? wmsg : "hvwait",
627 			    USEC_TO_NSEC(1000));
628 		}
629 	}
630 }
631 
632 uint16_t
633 hv_intr_signal(struct hv_softc *sc, void *con)
634 {
635 	uint64_t status;
636 
637 	status = hv_hypercall(sc, HYPERCALL_SIGNAL_EVENT, con, NULL);
638 	return ((uint16_t)status);
639 }
640 
641 void
642 hv_intr(void)
643 {
644 	struct hv_softc *sc = hv_sc;
645 
646 	hv_event_intr(sc);
647 	hv_message_intr(sc);
648 }
649 
650 void
651 hv_event_intr(struct hv_softc *sc)
652 {
653 	struct vmbus_evtflags *evt;
654 	struct cpu_info *ci = curcpu();
655 	int cpu = CPU_INFO_UNIT(ci);
656 	int bit, row, maxrow, chanid;
657 	struct hv_channel *ch;
658 	u_long *revents, pending;
659 
660 	evt = (struct vmbus_evtflags *)sc->sc_siep[cpu] +
661 	    VMBUS_SINT_MESSAGE;
662 	if ((sc->sc_proto == VMBUS_VERSION_WS2008) ||
663 	    (sc->sc_proto == VMBUS_VERSION_WIN7)) {
664 		if (!test_bit(0, &evt->evt_flags[0]))
665 			return;
666 		clear_bit(0, &evt->evt_flags[0]);
667 		maxrow = VMBUS_CHAN_MAX_COMPAT / VMBUS_EVTFLAG_LEN;
668 		/*
669 		 * receive size is 1/2 page and divide that by 4 bytes
670 		 */
671 		revents = sc->sc_revents;
672 	} else {
673 		maxrow = nitems(evt->evt_flags);
674 		/*
675 		 * On Host with Win8 or above, the event page can be
676 		 * checked directly to get the id of the channel
677 		 * that has the pending interrupt.
678 		 */
679 		revents = &evt->evt_flags[0];
680 	}
681 
682 	for (row = 0; row < maxrow; row++) {
683 		if (revents[row] == 0)
684 			continue;
685 		pending = atomic_swap_ulong(&revents[row], 0);
686 		for (bit = 0; pending > 0; pending >>= 1, bit++) {
687 			if ((pending & 1) == 0)
688 				continue;
689 			chanid = (row * LONG_BIT) + bit;
690 			/* vmbus channel protocol message */
691 			if (chanid == 0)
692 				continue;
693 			ch = hv_channel_lookup(sc, chanid);
694 			if (ch == NULL) {
695 				printf("%s: unhandled event on %d\n",
696 				    sc->sc_dev.dv_xname, chanid);
697 				continue;
698 			}
699 			if (ch->ch_state != HV_CHANSTATE_OPENED) {
700 				printf("%s: channel %d is not active\n",
701 				    sc->sc_dev.dv_xname, chanid);
702 				continue;
703 			}
704 			ch->ch_evcnt.ec_count++;
705 			hv_channel_schedule(ch);
706 		}
707 	}
708 }
709 
710 void
711 hv_message_intr(struct hv_softc *sc)
712 {
713 	struct vmbus_message *msg;
714 	struct vmbus_chanmsg_hdr *hdr;
715 	struct cpu_info *ci = curcpu();
716 	int cpu = CPU_INFO_UNIT(ci);
717 
718 	for (;;) {
719 		msg = (struct vmbus_message *)sc->sc_simp[cpu] +
720 		    VMBUS_SINT_MESSAGE;
721 		if (msg->msg_type == VMBUS_MSGTYPE_NONE)
722 			break;
723 
724 		hdr = (struct vmbus_chanmsg_hdr *)msg->msg_data;
725 		if (hdr->chm_type >= VMBUS_CHANMSG_COUNT) {
726 			printf("%s: unhandled message type %u flags %#x\n",
727 			    sc->sc_dev.dv_xname, hdr->chm_type,
728 			    msg->msg_flags);
729 			goto skip;
730 		}
731 		if (hv_msg_dispatch[hdr->chm_type].hmd_handler)
732 			hv_msg_dispatch[hdr->chm_type].hmd_handler(sc, hdr);
733 		else
734 			printf("%s: unhandled message type %u\n",
735 			    sc->sc_dev.dv_xname, hdr->chm_type);
736  skip:
737 		msg->msg_type = VMBUS_MSGTYPE_NONE;
738 		virtio_membar_sync();
739 		if (msg->msg_flags & VMBUS_MSGFLAG_PENDING)
740 			wrmsr(MSR_HV_EOM, 0);
741 	}
742 }
743 
744 void
745 hv_channel_response(struct hv_softc *sc, struct vmbus_chanmsg_hdr *rsphdr)
746 {
747 	struct hv_msg *msg;
748 	struct vmbus_chanmsg_hdr *reqhdr;
749 	int req;
750 
751 	req = hv_msg_dispatch[rsphdr->chm_type].hmd_request;
752 	mtx_enter(&sc->sc_reqlck);
753 	TAILQ_FOREACH(msg, &sc->sc_reqs, msg_entry) {
754 		reqhdr = (struct vmbus_chanmsg_hdr *)&msg->msg_req.hc_data;
755 		if (reqhdr->chm_type == req) {
756 			TAILQ_REMOVE(&sc->sc_reqs, msg, msg_entry);
757 			break;
758 		}
759 	}
760 	mtx_leave(&sc->sc_reqlck);
761 	if (msg != NULL) {
762 		memcpy(msg->msg_rsp, rsphdr, msg->msg_rsplen);
763 		mtx_enter(&sc->sc_rsplck);
764 		TAILQ_INSERT_TAIL(&sc->sc_rsps, msg, msg_entry);
765 		mtx_leave(&sc->sc_rsplck);
766 		wakeup(msg);
767 	}
768 }
769 
770 void
771 hv_channel_offer(struct hv_softc *sc, struct vmbus_chanmsg_hdr *hdr)
772 {
773 	struct hv_offer *co;
774 
775 	co = malloc(sizeof(*co), M_DEVBUF, M_NOWAIT | M_ZERO);
776 	if (co == NULL) {
777 		printf("%s: failed to allocate an offer object\n",
778 		    sc->sc_dev.dv_xname);
779 		return;
780 	}
781 
782 	memcpy(&co->co_chan, hdr, sizeof(co->co_chan));
783 
784 	mtx_enter(&sc->sc_offerlck);
785 	SIMPLEQ_INSERT_TAIL(&sc->sc_offers, co, co_entry);
786 	mtx_leave(&sc->sc_offerlck);
787 }
788 
789 void
790 hv_channel_rescind(struct hv_softc *sc, struct vmbus_chanmsg_hdr *hdr)
791 {
792 	const struct vmbus_chanmsg_chrescind *cmd;
793 
794 	cmd = (const struct vmbus_chanmsg_chrescind *)hdr;
795 	printf("%s: revoking channel %u\n", sc->sc_dev.dv_xname,
796 	    cmd->chm_chanid);
797 }
798 
799 void
800 hv_channel_delivered(struct hv_softc *sc, struct vmbus_chanmsg_hdr *hdr)
801 {
802 	atomic_setbits_int(&sc->sc_flags, HSF_OFFERS_DELIVERED);
803 	wakeup(&sc->sc_offers);
804 }
805 
806 int
807 hv_vmbus_connect(struct hv_softc *sc)
808 {
809 	const uint32_t versions[] = {
810 		VMBUS_VERSION_WIN10,
811 		VMBUS_VERSION_WIN8_1, VMBUS_VERSION_WIN8,
812 		VMBUS_VERSION_WIN7, VMBUS_VERSION_WS2008
813 	};
814 	struct vmbus_chanmsg_connect cmd;
815 	struct vmbus_chanmsg_connect_resp rsp;
816 	paddr_t epa, mpa1, mpa2;
817 	int i;
818 
819 	sc->sc_events = km_alloc(PAGE_SIZE, &kv_any, &kp_zero, &kd_nowait);
820 	if (sc->sc_events == NULL) {
821 		printf(": failed to allocate channel port events page\n");
822 		goto errout;
823 	}
824 	if (!pmap_extract(pmap_kernel(), (vaddr_t)sc->sc_events, &epa)) {
825 		printf(": channel port events page PA extraction failed\n");
826 		goto errout;
827 	}
828 
829 	sc->sc_wevents = (u_long *)sc->sc_events;
830 	sc->sc_revents = (u_long *)((caddr_t)sc->sc_events + (PAGE_SIZE >> 1));
831 
832 	sc->sc_monitor[0] = km_alloc(PAGE_SIZE, &kv_any, &kp_zero, &kd_nowait);
833 	if (sc->sc_monitor[0] == NULL) {
834 		printf(": failed to allocate monitor page 1\n");
835 		goto errout;
836 	}
837 	if (!pmap_extract(pmap_kernel(), (vaddr_t)sc->sc_monitor[0], &mpa1)) {
838 		printf(": monitor page 1 PA extraction failed\n");
839 		goto errout;
840 	}
841 
842 	sc->sc_monitor[1] = km_alloc(PAGE_SIZE, &kv_any, &kp_zero, &kd_nowait);
843 	if (sc->sc_monitor[1] == NULL) {
844 		printf(": failed to allocate monitor page 2\n");
845 		goto errout;
846 	}
847 	if (!pmap_extract(pmap_kernel(), (vaddr_t)sc->sc_monitor[1], &mpa2)) {
848 		printf(": monitor page 2 PA extraction failed\n");
849 		goto errout;
850 	}
851 
852 	memset(&cmd, 0, sizeof(cmd));
853 	cmd.chm_hdr.chm_type = VMBUS_CHANMSG_CONNECT;
854 	cmd.chm_evtflags = (uint64_t)epa;
855 	cmd.chm_mnf1 = (uint64_t)mpa1;
856 	cmd.chm_mnf2 = (uint64_t)mpa2;
857 
858 	memset(&rsp, 0, sizeof(rsp));
859 
860 	for (i = 0; i < nitems(versions); i++) {
861 		cmd.chm_ver = versions[i];
862 		if (hv_cmd(sc, &cmd, sizeof(cmd), &rsp, sizeof(rsp),
863 		    HCF_NOSLEEP)) {
864 			DPRINTF("%s: CONNECT failed\n",
865 			    sc->sc_dev.dv_xname);
866 			goto errout;
867 		}
868 		if (rsp.chm_done) {
869 			sc->sc_flags |= HSF_CONNECTED;
870 			sc->sc_proto = versions[i];
871 			sc->sc_handle = VMBUS_GPADL_START;
872 			break;
873 		}
874 	}
875 	if (i == nitems(versions)) {
876 		printf("%s: failed to negotiate protocol version\n",
877 		    sc->sc_dev.dv_xname);
878 		goto errout;
879 	}
880 
881 	return (0);
882 
883  errout:
884 	if (sc->sc_events) {
885 		km_free(sc->sc_events, PAGE_SIZE, &kv_any, &kp_zero);
886 		sc->sc_events = NULL;
887 		sc->sc_wevents = NULL;
888 		sc->sc_revents = NULL;
889 	}
890 	if (sc->sc_monitor[0]) {
891 		km_free(sc->sc_monitor[0], PAGE_SIZE, &kv_any, &kp_zero);
892 		sc->sc_monitor[0] = NULL;
893 	}
894 	if (sc->sc_monitor[1]) {
895 		km_free(sc->sc_monitor[1], PAGE_SIZE, &kv_any, &kp_zero);
896 		sc->sc_monitor[1] = NULL;
897 	}
898 	return (-1);
899 }
900 
901 #ifdef HYPERV_DEBUG
902 static inline char *
903 guidprint(struct hv_guid *a)
904 {
905 	/* 3     0  5  4 7 6  8 9  10        15 */
906 	/* 33221100-5544-7766-9988-FFEEDDCCBBAA */
907 	static char buf[16 * 2 + 4 + 1];
908 	int i, j = 0;
909 
910 	for (i = 3; i != -1; i -= 1, j += 2)
911 		snprintf(&buf[j], 3, "%02x", (uint8_t)a->data[i]);
912 	buf[j++] = '-';
913 	for (i = 5; i != 3; i -= 1, j += 2)
914 		snprintf(&buf[j], 3, "%02x", (uint8_t)a->data[i]);
915 	buf[j++] = '-';
916 	for (i = 7; i != 5; i -= 1, j += 2)
917 		snprintf(&buf[j], 3, "%02x", (uint8_t)a->data[i]);
918 	buf[j++] = '-';
919 	for (i = 8; i < 10; i += 1, j += 2)
920 		snprintf(&buf[j], 3, "%02x", (uint8_t)a->data[i]);
921 	buf[j++] = '-';
922 	for (i = 10; i < 16; i += 1, j += 2)
923 		snprintf(&buf[j], 3, "%02x", (uint8_t)a->data[i]);
924 	return (&buf[0]);
925 }
926 #endif	/* HYPERV_DEBUG */
927 
928 void
929 hv_guid_sprint(struct hv_guid *guid, char *str, size_t size)
930 {
931 	const struct {
932 		const struct hv_guid	*guid;
933 		const char		*ident;
934 	} map[] = {
935 		{ &hv_guid_network,	"network" },
936 		{ &hv_guid_ide,		"ide" },
937 		{ &hv_guid_scsi,	"scsi" },
938 		{ &hv_guid_shutdown,	"shutdown" },
939 		{ &hv_guid_timesync,	"timesync" },
940 		{ &hv_guid_heartbeat,	"heartbeat" },
941 		{ &hv_guid_kvp,		"kvp" },
942 #ifdef HYPERV_DEBUG
943 		{ &hv_guid_vss,		"vss" },
944 		{ &hv_guid_dynmem,	"dynamic-memory" },
945 		{ &hv_guid_mouse,	"mouse" },
946 		{ &hv_guid_kbd,		"keyboard" },
947 		{ &hv_guid_video,	"video" },
948 		{ &hv_guid_fc,		"fiber-channel" },
949 		{ &hv_guid_fcopy,	"file-copy" },
950 		{ &hv_guid_pcie,	"pcie-passthrough" },
951 		{ &hv_guid_netdir,	"network-direct" },
952 		{ &hv_guid_rdesktop,	"remote-desktop" },
953 		{ &hv_guid_avma1,	"avma-1" },
954 		{ &hv_guid_avma2,	"avma-2" },
955 		{ &hv_guid_avma3,	"avma-3" },
956 		{ &hv_guid_avma4,	"avma-4" },
957 #endif
958 	};
959 	int i;
960 
961 	for (i = 0; i < nitems(map); i++) {
962 		if (memcmp(guid, map[i].guid, sizeof(*guid)) == 0) {
963 			strlcpy(str, map[i].ident, size);
964 			return;
965 		}
966 	}
967 #ifdef HYPERV_DEBUG
968 	strlcpy(str, guidprint(guid), size);
969 #endif
970 }
971 
972 static int
973 hv_channel_scan_done(struct hv_softc *sc, struct hv_msg *msg __unused)
974 {
975 	return (sc->sc_flags & HSF_OFFERS_DELIVERED);
976 }
977 
978 int
979 hv_channel_scan(struct hv_softc *sc)
980 {
981 	struct vmbus_chanmsg_hdr hdr;
982 	struct vmbus_chanmsg_choffer rsp;
983 	struct hv_offer *co;
984 
985 	SIMPLEQ_INIT(&sc->sc_offers);
986 	mtx_init(&sc->sc_offerlck, IPL_NET);
987 
988 	memset(&hdr, 0, sizeof(hdr));
989 	hdr.chm_type = VMBUS_CHANMSG_CHREQUEST;
990 
991 	if (hv_cmd(sc, &hdr, sizeof(hdr), &rsp, sizeof(rsp),
992 	    HCF_NOSLEEP | HCF_NOREPLY)) {
993 		DPRINTF("%s: CHREQUEST failed\n", sc->sc_dev.dv_xname);
994 		return (-1);
995 	}
996 
997 	hv_wait(sc, hv_channel_scan_done, (struct hv_msg *)&hdr,
998 	    &sc->sc_offers, "hvscan");
999 
1000 	TAILQ_INIT(&sc->sc_channels);
1001 	mtx_init(&sc->sc_channelck, IPL_NET);
1002 
1003 	mtx_enter(&sc->sc_offerlck);
1004 	while (!SIMPLEQ_EMPTY(&sc->sc_offers)) {
1005 		co = SIMPLEQ_FIRST(&sc->sc_offers);
1006 		SIMPLEQ_REMOVE_HEAD(&sc->sc_offers, co_entry);
1007 		mtx_leave(&sc->sc_offerlck);
1008 
1009 		hv_process_offer(sc, co);
1010 		free(co, M_DEVBUF, sizeof(*co));
1011 
1012 		mtx_enter(&sc->sc_offerlck);
1013 	}
1014 	mtx_leave(&sc->sc_offerlck);
1015 
1016 	return (0);
1017 }
1018 
1019 void
1020 hv_process_offer(struct hv_softc *sc, struct hv_offer *co)
1021 {
1022 	struct hv_channel *ch, *nch;
1023 
1024 	nch = malloc(sizeof(*nch), M_DEVBUF, M_ZERO | M_NOWAIT);
1025 	if (nch == NULL) {
1026 		printf("%s: failed to allocate memory for the channel\n",
1027 		    sc->sc_dev.dv_xname);
1028 		return;
1029 	}
1030 	nch->ch_sc = sc;
1031 	hv_guid_sprint(&co->co_chan.chm_chtype, nch->ch_ident,
1032 	    sizeof(nch->ch_ident));
1033 
1034 	/*
1035 	 * By default we setup state to enable batched reading.
1036 	 * A specific service can choose to disable this prior
1037 	 * to opening the channel.
1038 	 */
1039 	nch->ch_flags |= CHF_BATCHED;
1040 
1041 	KASSERT((((vaddr_t)&nch->ch_monprm) & 0x7) == 0);
1042 	memset(&nch->ch_monprm, 0, sizeof(nch->ch_monprm));
1043 	nch->ch_monprm.mp_connid = VMBUS_CONNID_EVENT;
1044 
1045 	if (sc->sc_proto != VMBUS_VERSION_WS2008)
1046 		nch->ch_monprm.mp_connid = co->co_chan.chm_connid;
1047 
1048 	if (co->co_chan.chm_flags1 & VMBUS_CHOFFER_FLAG1_HASMNF) {
1049 		nch->ch_mgroup = co->co_chan.chm_montrig / VMBUS_MONTRIG_LEN;
1050 		nch->ch_mindex = co->co_chan.chm_montrig % VMBUS_MONTRIG_LEN;
1051 		nch->ch_flags |= CHF_MONITOR;
1052 	}
1053 
1054 	nch->ch_id = co->co_chan.chm_chanid;
1055 
1056 	memcpy(&nch->ch_type, &co->co_chan.chm_chtype, sizeof(ch->ch_type));
1057 	memcpy(&nch->ch_inst, &co->co_chan.chm_chinst, sizeof(ch->ch_inst));
1058 
1059 	mtx_enter(&sc->sc_channelck);
1060 	TAILQ_FOREACH(ch, &sc->sc_channels, ch_entry) {
1061 		if (!memcmp(&ch->ch_type, &nch->ch_type, sizeof(ch->ch_type)) &&
1062 		    !memcmp(&ch->ch_inst, &nch->ch_inst, sizeof(ch->ch_inst)))
1063 			break;
1064 	}
1065 	if (ch != NULL) {
1066 		if (co->co_chan.chm_subidx == 0) {
1067 			printf("%s: unknown offer \"%s\"\n",
1068 			    sc->sc_dev.dv_xname, nch->ch_ident);
1069 			mtx_leave(&sc->sc_channelck);
1070 			free(nch, M_DEVBUF, sizeof(*nch));
1071 			return;
1072 		}
1073 #ifdef HYPERV_DEBUG
1074 		printf("%s: subchannel %u for \"%s\"\n", sc->sc_dev.dv_xname,
1075 		    co->co_chan.chm_subidx, ch->ch_ident);
1076 #endif
1077 		mtx_leave(&sc->sc_channelck);
1078 		free(nch, M_DEVBUF, sizeof(*nch));
1079 		return;
1080 	}
1081 
1082 	nch->ch_state = HV_CHANSTATE_OFFERED;
1083 
1084 	TAILQ_INSERT_TAIL(&sc->sc_channels, nch, ch_entry);
1085 	mtx_leave(&sc->sc_channelck);
1086 
1087 #ifdef HYPERV_DEBUG
1088 	printf("%s: channel %u: \"%s\"", sc->sc_dev.dv_xname, nch->ch_id,
1089 	    nch->ch_ident);
1090 	if (nch->ch_flags & CHF_MONITOR)
1091 		printf(", monitor %u\n", co->co_chan.chm_montrig);
1092 	else
1093 		printf("\n");
1094 #endif
1095 }
1096 
1097 struct hv_channel *
1098 hv_channel_lookup(struct hv_softc *sc, uint32_t relid)
1099 {
1100 	struct hv_channel *ch;
1101 
1102 	TAILQ_FOREACH(ch, &sc->sc_channels, ch_entry) {
1103 		if (ch->ch_id == relid)
1104 			return (ch);
1105 	}
1106 	return (NULL);
1107 }
1108 
1109 int
1110 hv_channel_ring_create(struct hv_channel *ch, uint32_t buflen)
1111 {
1112 	struct hv_softc *sc = ch->ch_sc;
1113 
1114 	buflen = roundup(buflen, PAGE_SIZE) + sizeof(struct vmbus_bufring);
1115 	ch->ch_ring = km_alloc(2 * buflen, &kv_any, &kp_zero, cold ?
1116 	    &kd_nowait : &kd_waitok);
1117 	if (ch->ch_ring == NULL) {
1118 		printf("%s: failed to allocate channel ring\n",
1119 		    sc->sc_dev.dv_xname);
1120 		return (-1);
1121 	}
1122 	ch->ch_ring_size = 2 * buflen;
1123 
1124 	memset(&ch->ch_wrd, 0, sizeof(ch->ch_wrd));
1125 	ch->ch_wrd.rd_ring = (struct vmbus_bufring *)ch->ch_ring;
1126 	ch->ch_wrd.rd_size = buflen;
1127 	ch->ch_wrd.rd_dsize = buflen - sizeof(struct vmbus_bufring);
1128 	mtx_init(&ch->ch_wrd.rd_lock, IPL_NET);
1129 
1130 	memset(&ch->ch_rrd, 0, sizeof(ch->ch_rrd));
1131 	ch->ch_rrd.rd_ring = (struct vmbus_bufring *)((uint8_t *)ch->ch_ring +
1132 	    buflen);
1133 	ch->ch_rrd.rd_size = buflen;
1134 	ch->ch_rrd.rd_dsize = buflen - sizeof(struct vmbus_bufring);
1135 	mtx_init(&ch->ch_rrd.rd_lock, IPL_NET);
1136 
1137 	if (hv_handle_alloc(ch, ch->ch_ring, 2 * buflen, &ch->ch_ring_gpadl)) {
1138 		printf("%s: failed to obtain a PA handle for the ring\n",
1139 		    sc->sc_dev.dv_xname);
1140 		hv_channel_ring_destroy(ch);
1141 		return (-1);
1142 	}
1143 
1144 	return (0);
1145 }
1146 
1147 void
1148 hv_channel_ring_destroy(struct hv_channel *ch)
1149 {
1150 	km_free(ch->ch_ring, ch->ch_ring_size, &kv_any, &kp_zero);
1151 	ch->ch_ring = NULL;
1152 	hv_handle_free(ch, ch->ch_ring_gpadl);
1153 
1154 	memset(&ch->ch_wrd, 0, sizeof(ch->ch_wrd));
1155 	memset(&ch->ch_rrd, 0, sizeof(ch->ch_rrd));
1156 }
1157 
1158 int
1159 hv_channel_open(struct hv_channel *ch, size_t buflen, void *udata,
1160     size_t udatalen, void (*handler)(void *), void *arg)
1161 {
1162 	struct hv_softc *sc = ch->ch_sc;
1163 	struct vmbus_chanmsg_chopen cmd;
1164 	struct vmbus_chanmsg_chopen_resp rsp;
1165 	int rv;
1166 
1167 	if (ch->ch_ring == NULL &&
1168 	    hv_channel_ring_create(ch, buflen)) {
1169 		DPRINTF("%s: failed to create channel ring\n",
1170 		    sc->sc_dev.dv_xname);
1171 		return (-1);
1172 	}
1173 
1174 	memset(&cmd, 0, sizeof(cmd));
1175 	cmd.chm_hdr.chm_type = VMBUS_CHANMSG_CHOPEN;
1176 	cmd.chm_openid = ch->ch_id;
1177 	cmd.chm_chanid = ch->ch_id;
1178 	cmd.chm_gpadl = ch->ch_ring_gpadl;
1179 	cmd.chm_txbr_pgcnt = ch->ch_wrd.rd_size >> PAGE_SHIFT;
1180 	cmd.chm_vcpuid = ch->ch_vcpu;
1181 
1182 	if (udata && udatalen > 0)
1183 		memcpy(cmd.chm_udata, udata, udatalen);
1184 
1185 	memset(&rsp, 0, sizeof(rsp));
1186 
1187 	ch->ch_handler = handler;
1188 	ch->ch_ctx = arg;
1189 
1190 	ch->ch_state = HV_CHANSTATE_OPENED;
1191 
1192 	rv = hv_cmd(sc, &cmd, sizeof(cmd), &rsp, sizeof(rsp),
1193 	    cold ? HCF_NOSLEEP : HCF_SLEEPOK);
1194 	if (rv) {
1195 		hv_channel_ring_destroy(ch);
1196 		DPRINTF("%s: CHOPEN failed with %d\n",
1197 		    sc->sc_dev.dv_xname, rv);
1198 		ch->ch_handler = NULL;
1199 		ch->ch_ctx = NULL;
1200 		ch->ch_state = HV_CHANSTATE_OFFERED;
1201 		return (-1);
1202 	}
1203 
1204 	return (0);
1205 }
1206 
1207 int
1208 hv_channel_close(struct hv_channel *ch)
1209 {
1210 	struct hv_softc *sc = ch->ch_sc;
1211 	struct vmbus_chanmsg_chclose cmd;
1212 	int rv;
1213 
1214 	memset(&cmd, 0, sizeof(cmd));
1215 	cmd.chm_hdr.chm_type = VMBUS_CHANMSG_CHCLOSE;
1216 	cmd.chm_chanid = ch->ch_id;
1217 
1218 	ch->ch_state = HV_CHANSTATE_CLOSING;
1219 	rv = hv_cmd(sc, &cmd, sizeof(cmd), NULL, 0, HCF_NOREPLY);
1220 	if (rv) {
1221 		DPRINTF("%s: CHCLOSE failed with %d\n",
1222 		    sc->sc_dev.dv_xname, rv);
1223 		return (-1);
1224 	}
1225 	ch->ch_state = HV_CHANSTATE_CLOSED;
1226 	hv_channel_ring_destroy(ch);
1227 	return (0);
1228 }
1229 
1230 static inline void
1231 hv_channel_setevent(struct hv_softc *sc, struct hv_channel *ch)
1232 {
1233 	struct vmbus_mon_trig *mtg;
1234 
1235 	/* Each uint32_t represents 32 channels */
1236 	set_bit(ch->ch_id, sc->sc_wevents);
1237 	if (ch->ch_flags & CHF_MONITOR) {
1238 		mtg = &sc->sc_monitor[1]->mnf_trigs[ch->ch_mgroup];
1239 		set_bit(ch->ch_mindex, &mtg->mt_pending);
1240 	} else
1241 		hv_intr_signal(sc, &ch->ch_monprm);
1242 }
1243 
1244 void
1245 hv_channel_intr(void *arg)
1246 {
1247 	struct hv_channel *ch = arg;
1248 
1249 	if (hv_channel_ready(ch))
1250 		ch->ch_handler(ch->ch_ctx);
1251 
1252 	if (hv_channel_unpause(ch) == 0)
1253 		return;
1254 
1255 	hv_channel_pause(ch);
1256 	hv_channel_schedule(ch);
1257 }
1258 
1259 int
1260 hv_channel_setdeferred(struct hv_channel *ch, const char *name)
1261 {
1262 	ch->ch_taskq = taskq_create(name, 1, IPL_NET, TASKQ_MPSAFE);
1263 	if (ch->ch_taskq == NULL)
1264 		return (-1);
1265 	task_set(&ch->ch_task, hv_channel_intr, ch);
1266 	return (0);
1267 }
1268 
1269 void
1270 hv_channel_schedule(struct hv_channel *ch)
1271 {
1272 	if (ch->ch_handler) {
1273 		if (!cold && (ch->ch_flags & CHF_BATCHED)) {
1274 			hv_channel_pause(ch);
1275 			task_add(ch->ch_taskq, &ch->ch_task);
1276 		} else
1277 			ch->ch_handler(ch->ch_ctx);
1278 	}
1279 }
1280 
1281 static inline void
1282 hv_ring_put(struct hv_ring_data *wrd, uint8_t *data, uint32_t datalen)
1283 {
1284 	int left = MIN(datalen, wrd->rd_dsize - wrd->rd_prod);
1285 
1286 	memcpy(&wrd->rd_ring->br_data[wrd->rd_prod], data, left);
1287 	memcpy(&wrd->rd_ring->br_data[0], data + left, datalen - left);
1288 	wrd->rd_prod += datalen;
1289 	if (wrd->rd_prod >= wrd->rd_dsize)
1290 		wrd->rd_prod -= wrd->rd_dsize;
1291 }
1292 
1293 static inline void
1294 hv_ring_get(struct hv_ring_data *rrd, uint8_t *data, uint32_t datalen,
1295     int peek)
1296 {
1297 	int left = MIN(datalen, rrd->rd_dsize - rrd->rd_cons);
1298 
1299 	memcpy(data, &rrd->rd_ring->br_data[rrd->rd_cons], left);
1300 	memcpy(data + left, &rrd->rd_ring->br_data[0], datalen - left);
1301 	if (!peek) {
1302 		rrd->rd_cons += datalen;
1303 		if (rrd->rd_cons >= rrd->rd_dsize)
1304 			rrd->rd_cons -= rrd->rd_dsize;
1305 	}
1306 }
1307 
1308 static inline void
1309 hv_ring_avail(struct hv_ring_data *rd, uint32_t *towrite, uint32_t *toread)
1310 {
1311 	uint32_t ridx = rd->rd_ring->br_rindex;
1312 	uint32_t widx = rd->rd_ring->br_windex;
1313 	uint32_t r, w;
1314 
1315 	if (widx >= ridx)
1316 		w = rd->rd_dsize - (widx - ridx);
1317 	else
1318 		w = ridx - widx;
1319 	r = rd->rd_dsize - w;
1320 	if (towrite)
1321 		*towrite = w;
1322 	if (toread)
1323 		*toread = r;
1324 }
1325 
1326 int
1327 hv_ring_write(struct hv_ring_data *wrd, struct iovec *iov, int iov_cnt,
1328     int *needsig)
1329 {
1330 	uint64_t indices = 0;
1331 	uint32_t avail, oprod, datalen = sizeof(indices);
1332 	int i;
1333 
1334 	for (i = 0; i < iov_cnt; i++)
1335 		datalen += iov[i].iov_len;
1336 
1337 	KASSERT(datalen <= wrd->rd_dsize);
1338 
1339 	hv_ring_avail(wrd, &avail, NULL);
1340 	if (avail <= datalen) {
1341 		DPRINTF("%s: avail %u datalen %u\n", __func__, avail, datalen);
1342 		return (EAGAIN);
1343 	}
1344 
1345 	oprod = wrd->rd_prod;
1346 
1347 	for (i = 0; i < iov_cnt; i++)
1348 		hv_ring_put(wrd, iov[i].iov_base, iov[i].iov_len);
1349 
1350 	indices = (uint64_t)oprod << 32;
1351 	hv_ring_put(wrd, (uint8_t *)&indices, sizeof(indices));
1352 
1353 	virtio_membar_sync();
1354 	wrd->rd_ring->br_windex = wrd->rd_prod;
1355 	virtio_membar_sync();
1356 
1357 	/* Signal when the ring transitions from being empty to non-empty */
1358 	if (wrd->rd_ring->br_imask == 0 &&
1359 	    wrd->rd_ring->br_rindex == oprod)
1360 		*needsig = 1;
1361 	else
1362 		*needsig = 0;
1363 
1364 	return (0);
1365 }
1366 
1367 int
1368 hv_channel_send(struct hv_channel *ch, void *data, uint32_t datalen,
1369     uint64_t rid, int type, uint32_t flags)
1370 {
1371 	struct hv_softc *sc = ch->ch_sc;
1372 	struct vmbus_chanpkt cp;
1373 	struct iovec iov[3];
1374 	uint32_t pktlen, pktlen_aligned;
1375 	uint64_t zeropad = 0;
1376 	int rv, needsig = 0;
1377 
1378 	pktlen = sizeof(cp) + datalen;
1379 	pktlen_aligned = roundup(pktlen, sizeof(uint64_t));
1380 
1381 	cp.cp_hdr.cph_type = type;
1382 	cp.cp_hdr.cph_flags = flags;
1383 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_hlen, sizeof(cp));
1384 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_tlen, pktlen_aligned);
1385 	cp.cp_hdr.cph_tid = rid;
1386 
1387 	iov[0].iov_base = &cp;
1388 	iov[0].iov_len = sizeof(cp);
1389 
1390 	iov[1].iov_base = data;
1391 	iov[1].iov_len = datalen;
1392 
1393 	iov[2].iov_base = &zeropad;
1394 	iov[2].iov_len = pktlen_aligned - pktlen;
1395 
1396 	mtx_enter(&ch->ch_wrd.rd_lock);
1397 	rv = hv_ring_write(&ch->ch_wrd, iov, 3, &needsig);
1398 	mtx_leave(&ch->ch_wrd.rd_lock);
1399 	if (rv == 0 && needsig)
1400 		hv_channel_setevent(sc, ch);
1401 
1402 	return (rv);
1403 }
1404 
1405 int
1406 hv_channel_send_sgl(struct hv_channel *ch, struct vmbus_gpa *sgl,
1407     uint32_t nsge, void *data, uint32_t datalen, uint64_t rid)
1408 {
1409 	struct hv_softc *sc = ch->ch_sc;
1410 	struct vmbus_chanpkt_sglist cp;
1411 	struct iovec iov[4];
1412 	uint32_t buflen, pktlen, pktlen_aligned;
1413 	uint64_t zeropad = 0;
1414 	int rv, needsig = 0;
1415 
1416 	buflen = sizeof(struct vmbus_gpa) * nsge;
1417 	pktlen = sizeof(cp) + datalen + buflen;
1418 	pktlen_aligned = roundup(pktlen, sizeof(uint64_t));
1419 
1420 	cp.cp_hdr.cph_type = VMBUS_CHANPKT_TYPE_GPA;
1421 	cp.cp_hdr.cph_flags = VMBUS_CHANPKT_FLAG_RC;
1422 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_hlen, sizeof(cp) + buflen);
1423 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_tlen, pktlen_aligned);
1424 	cp.cp_hdr.cph_tid = rid;
1425 	cp.cp_gpa_cnt = nsge;
1426 	cp.cp_rsvd = 0;
1427 
1428 	iov[0].iov_base = &cp;
1429 	iov[0].iov_len = sizeof(cp);
1430 
1431 	iov[1].iov_base = sgl;
1432 	iov[1].iov_len = buflen;
1433 
1434 	iov[2].iov_base = data;
1435 	iov[2].iov_len = datalen;
1436 
1437 	iov[3].iov_base = &zeropad;
1438 	iov[3].iov_len = pktlen_aligned - pktlen;
1439 
1440 	mtx_enter(&ch->ch_wrd.rd_lock);
1441 	rv = hv_ring_write(&ch->ch_wrd, iov, 4, &needsig);
1442 	mtx_leave(&ch->ch_wrd.rd_lock);
1443 	if (rv == 0 && needsig)
1444 		hv_channel_setevent(sc, ch);
1445 
1446 	return (rv);
1447 }
1448 
1449 int
1450 hv_channel_send_prpl(struct hv_channel *ch, struct vmbus_gpa_range *prpl,
1451     uint32_t nprp, void *data, uint32_t datalen, uint64_t rid)
1452 {
1453 	struct hv_softc *sc = ch->ch_sc;
1454 	struct vmbus_chanpkt_prplist cp;
1455 	struct iovec iov[4];
1456 	uint32_t buflen, pktlen, pktlen_aligned;
1457 	uint64_t zeropad = 0;
1458 	int rv, needsig = 0;
1459 
1460 	buflen = sizeof(struct vmbus_gpa_range) * (nprp + 1);
1461 	pktlen = sizeof(cp) + datalen + buflen;
1462 	pktlen_aligned = roundup(pktlen, sizeof(uint64_t));
1463 
1464 	cp.cp_hdr.cph_type = VMBUS_CHANPKT_TYPE_GPA;
1465 	cp.cp_hdr.cph_flags = VMBUS_CHANPKT_FLAG_RC;
1466 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_hlen, sizeof(cp) + buflen);
1467 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_tlen, pktlen_aligned);
1468 	cp.cp_hdr.cph_tid = rid;
1469 	cp.cp_range_cnt = 1;
1470 	cp.cp_rsvd = 0;
1471 
1472 	iov[0].iov_base = &cp;
1473 	iov[0].iov_len = sizeof(cp);
1474 
1475 	iov[1].iov_base = prpl;
1476 	iov[1].iov_len = buflen;
1477 
1478 	iov[2].iov_base = data;
1479 	iov[2].iov_len = datalen;
1480 
1481 	iov[3].iov_base = &zeropad;
1482 	iov[3].iov_len = pktlen_aligned - pktlen;
1483 
1484 	mtx_enter(&ch->ch_wrd.rd_lock);
1485 	rv = hv_ring_write(&ch->ch_wrd, iov, 4, &needsig);
1486 	mtx_leave(&ch->ch_wrd.rd_lock);
1487 	if (rv == 0 && needsig)
1488 		hv_channel_setevent(sc, ch);
1489 
1490 	return (rv);
1491 }
1492 
1493 int
1494 hv_ring_peek(struct hv_ring_data *rrd, void *data, uint32_t datalen)
1495 {
1496 	uint32_t avail;
1497 
1498 	KASSERT(datalen <= rrd->rd_dsize);
1499 
1500 	hv_ring_avail(rrd, NULL, &avail);
1501 	if (avail < datalen)
1502 		return (EAGAIN);
1503 
1504 	hv_ring_get(rrd, (uint8_t *)data, datalen, 1);
1505 	return (0);
1506 }
1507 
1508 int
1509 hv_ring_read(struct hv_ring_data *rrd, void *data, uint32_t datalen,
1510     uint32_t offset)
1511 {
1512 	uint64_t indices;
1513 	uint32_t avail;
1514 
1515 	KASSERT(datalen <= rrd->rd_dsize);
1516 
1517 	hv_ring_avail(rrd, NULL, &avail);
1518 	if (avail < datalen) {
1519 		DPRINTF("%s: avail %u datalen %u\n", __func__, avail, datalen);
1520 		return (EAGAIN);
1521 	}
1522 
1523 	if (offset) {
1524 		rrd->rd_cons += offset;
1525 		if (rrd->rd_cons >= rrd->rd_dsize)
1526 			rrd->rd_cons -= rrd->rd_dsize;
1527 	}
1528 
1529 	hv_ring_get(rrd, (uint8_t *)data, datalen, 0);
1530 	hv_ring_get(rrd, (uint8_t *)&indices, sizeof(indices), 0);
1531 
1532 	virtio_membar_sync();
1533 	rrd->rd_ring->br_rindex = rrd->rd_cons;
1534 
1535 	return (0);
1536 }
1537 
1538 int
1539 hv_channel_recv(struct hv_channel *ch, void *data, uint32_t datalen,
1540     uint32_t *rlen, uint64_t *rid, int raw)
1541 {
1542 	struct vmbus_chanpkt_hdr cph;
1543 	uint32_t offset, pktlen;
1544 	int rv;
1545 
1546 	*rlen = 0;
1547 
1548 	mtx_enter(&ch->ch_rrd.rd_lock);
1549 
1550 	if ((rv = hv_ring_peek(&ch->ch_rrd, &cph, sizeof(cph))) != 0) {
1551 		mtx_leave(&ch->ch_rrd.rd_lock);
1552 		return (rv);
1553 	}
1554 
1555 	offset = raw ? 0 : VMBUS_CHANPKT_GETLEN(cph.cph_hlen);
1556 	pktlen = VMBUS_CHANPKT_GETLEN(cph.cph_tlen) - offset;
1557 	if (pktlen > datalen) {
1558 		mtx_leave(&ch->ch_rrd.rd_lock);
1559 		printf("%s: pktlen %u datalen %u\n", __func__, pktlen, datalen);
1560 		return (EINVAL);
1561 	}
1562 
1563 	rv = hv_ring_read(&ch->ch_rrd, data, pktlen, offset);
1564 	if (rv == 0) {
1565 		*rlen = pktlen;
1566 		*rid = cph.cph_tid;
1567 	}
1568 
1569 	mtx_leave(&ch->ch_rrd.rd_lock);
1570 
1571 	return (rv);
1572 }
1573 
1574 static inline void
1575 hv_ring_mask(struct hv_ring_data *rd)
1576 {
1577 	virtio_membar_sync();
1578 	rd->rd_ring->br_imask = 1;
1579 	virtio_membar_sync();
1580 }
1581 
1582 static inline void
1583 hv_ring_unmask(struct hv_ring_data *rd)
1584 {
1585 	virtio_membar_sync();
1586 	rd->rd_ring->br_imask = 0;
1587 	virtio_membar_sync();
1588 }
1589 
1590 void
1591 hv_channel_pause(struct hv_channel *ch)
1592 {
1593 	hv_ring_mask(&ch->ch_rrd);
1594 }
1595 
1596 uint
1597 hv_channel_unpause(struct hv_channel *ch)
1598 {
1599 	uint32_t avail;
1600 
1601 	hv_ring_unmask(&ch->ch_rrd);
1602 	hv_ring_avail(&ch->ch_rrd, NULL, &avail);
1603 
1604 	return (avail);
1605 }
1606 
1607 uint
1608 hv_channel_ready(struct hv_channel *ch)
1609 {
1610 	uint32_t avail;
1611 
1612 	hv_ring_avail(&ch->ch_rrd, NULL, &avail);
1613 
1614 	return (avail);
1615 }
1616 
1617 /* How many PFNs can be referenced by the header */
1618 #define HV_NPFNHDR	((VMBUS_MSG_DSIZE_MAX -	\
1619 	  sizeof(struct vmbus_chanmsg_gpadl_conn)) / sizeof(uint64_t))
1620 
1621 /* How many PFNs can be referenced by the body */
1622 #define HV_NPFNBODY	((VMBUS_MSG_DSIZE_MAX -	\
1623 	  sizeof(struct vmbus_chanmsg_gpadl_subconn)) / sizeof(uint64_t))
1624 
1625 int
1626 hv_handle_alloc(struct hv_channel *ch, void *buffer, uint32_t buflen,
1627     uint32_t *handle)
1628 {
1629 	struct hv_softc *sc = ch->ch_sc;
1630 	struct vmbus_chanmsg_gpadl_conn *hdr;
1631 	struct vmbus_chanmsg_gpadl_subconn *cmd;
1632 	struct vmbus_chanmsg_gpadl_connresp rsp;
1633 	struct hv_msg *msg;
1634 	int i, j, last, left, rv;
1635 	int bodylen = 0, ncmds = 0, pfn = 0;
1636 	int waitflag = cold ? M_NOWAIT : M_WAITOK;
1637 	uint64_t *frames;
1638 	paddr_t pa;
1639 	caddr_t body;
1640 	/* Total number of pages to reference */
1641 	int total = atop(buflen);
1642 	/* Number of pages that will fit the header */
1643 	int inhdr = MIN(total, HV_NPFNHDR);
1644 
1645 	KASSERT((buflen & (PAGE_SIZE - 1)) == 0);
1646 
1647 	if ((msg = malloc(sizeof(*msg), M_DEVBUF, M_ZERO | waitflag)) == NULL)
1648 		return (ENOMEM);
1649 
1650 	/* Prepare array of frame addresses */
1651 	if ((frames = mallocarray(total, sizeof(*frames), M_DEVBUF, M_ZERO |
1652 	    waitflag)) == NULL) {
1653 		free(msg, M_DEVBUF, sizeof(*msg));
1654 		return (ENOMEM);
1655 	}
1656 	for (i = 0; i < total; i++) {
1657 		if (!pmap_extract(pmap_kernel(), (vaddr_t)buffer +
1658 		    PAGE_SIZE * i, &pa)) {
1659 			free(msg, M_DEVBUF, sizeof(*msg));
1660 			free(frames, M_DEVBUF, total * sizeof(*frames));
1661 			return (EFAULT);
1662 		}
1663 		frames[i] = atop(pa);
1664 	}
1665 
1666 	msg->msg_req.hc_dsize = sizeof(struct vmbus_chanmsg_gpadl_conn) +
1667 	    inhdr * sizeof(uint64_t);
1668 	hdr = (struct vmbus_chanmsg_gpadl_conn *)msg->msg_req.hc_data;
1669 	msg->msg_rsp = &rsp;
1670 	msg->msg_rsplen = sizeof(rsp);
1671 	if (waitflag == M_NOWAIT)
1672 		msg->msg_flags = MSGF_NOSLEEP;
1673 
1674 	left = total - inhdr;
1675 
1676 	/* Allocate additional gpadl_body structures if required */
1677 	if (left > 0) {
1678 		ncmds = MAX(1, left / HV_NPFNBODY + left % HV_NPFNBODY);
1679 		bodylen = ncmds * VMBUS_MSG_DSIZE_MAX;
1680 		body = malloc(bodylen, M_DEVBUF, M_ZERO | waitflag);
1681 		if (body == NULL) {
1682 			free(msg, M_DEVBUF, sizeof(*msg));
1683 			free(frames, M_DEVBUF, atop(buflen) * sizeof(*frames));
1684 			return (ENOMEM);
1685 		}
1686 	}
1687 
1688 	*handle = atomic_inc_int_nv(&sc->sc_handle);
1689 
1690 	hdr->chm_hdr.chm_type = VMBUS_CHANMSG_GPADL_CONN;
1691 	hdr->chm_chanid = ch->ch_id;
1692 	hdr->chm_gpadl = *handle;
1693 
1694 	/* Single range for a contiguous buffer */
1695 	hdr->chm_range_cnt = 1;
1696 	hdr->chm_range_len = sizeof(struct vmbus_gpa_range) + total *
1697 	    sizeof(uint64_t);
1698 	hdr->chm_range.gpa_ofs = 0;
1699 	hdr->chm_range.gpa_len = buflen;
1700 
1701 	/* Fit as many pages as possible into the header */
1702 	for (i = 0; i < inhdr; i++)
1703 		hdr->chm_range.gpa_page[i] = frames[pfn++];
1704 
1705 	for (i = 0; i < ncmds; i++) {
1706 		cmd = (struct vmbus_chanmsg_gpadl_subconn *)(body +
1707 		    VMBUS_MSG_DSIZE_MAX * i);
1708 		cmd->chm_hdr.chm_type = VMBUS_CHANMSG_GPADL_SUBCONN;
1709 		cmd->chm_gpadl = *handle;
1710 		last = MIN(left, HV_NPFNBODY);
1711 		for (j = 0; j < last; j++)
1712 			cmd->chm_gpa_page[j] = frames[pfn++];
1713 		left -= last;
1714 	}
1715 
1716 	rv = hv_start(sc, msg);
1717 	if (rv != 0) {
1718 		DPRINTF("%s: GPADL_CONN failed\n", sc->sc_dev.dv_xname);
1719 		goto out;
1720 	}
1721 	for (i = 0; i < ncmds; i++) {
1722 		int cmdlen = sizeof(*cmd);
1723 		cmd = (struct vmbus_chanmsg_gpadl_subconn *)(body +
1724 		    VMBUS_MSG_DSIZE_MAX * i);
1725 		/* Last element can be short */
1726 		if (i == ncmds - 1)
1727 			cmdlen += last * sizeof(uint64_t);
1728 		else
1729 			cmdlen += HV_NPFNBODY * sizeof(uint64_t);
1730 		rv = hv_cmd(sc, cmd, cmdlen, NULL, 0, waitflag | HCF_NOREPLY);
1731 		if (rv != 0) {
1732 			DPRINTF("%s: GPADL_SUBCONN (iteration %d/%d) failed "
1733 			    "with %d\n", sc->sc_dev.dv_xname, i, ncmds, rv);
1734 			goto out;
1735 		}
1736 	}
1737 	rv = hv_reply(sc, msg);
1738 	if (rv != 0)
1739 		DPRINTF("%s: GPADL allocation failed with %d\n",
1740 		    sc->sc_dev.dv_xname, rv);
1741 
1742  out:
1743 	free(msg, M_DEVBUF, sizeof(*msg));
1744 	free(frames, M_DEVBUF, total * sizeof(*frames));
1745 	if (bodylen > 0)
1746 		free(body, M_DEVBUF, bodylen);
1747 	if (rv != 0)
1748 		return (rv);
1749 
1750 	KASSERT(*handle == rsp.chm_gpadl);
1751 
1752 	return (0);
1753 }
1754 
1755 void
1756 hv_handle_free(struct hv_channel *ch, uint32_t handle)
1757 {
1758 	struct hv_softc *sc = ch->ch_sc;
1759 	struct vmbus_chanmsg_gpadl_disconn cmd;
1760 	struct vmbus_chanmsg_gpadl_disconn rsp;
1761 	int rv;
1762 
1763 	memset(&cmd, 0, sizeof(cmd));
1764 	cmd.chm_hdr.chm_type = VMBUS_CHANMSG_GPADL_DISCONN;
1765 	cmd.chm_chanid = ch->ch_id;
1766 	cmd.chm_gpadl = handle;
1767 
1768 	rv = hv_cmd(sc, &cmd, sizeof(cmd), &rsp, sizeof(rsp), cold ?
1769 	    HCF_NOSLEEP : 0);
1770 	if (rv)
1771 		DPRINTF("%s: GPADL_DISCONN failed with %d\n",
1772 		    sc->sc_dev.dv_xname, rv);
1773 }
1774 
1775 static int
1776 hv_attach_print(void *aux, const char *name)
1777 {
1778 	struct hv_attach_args *aa = aux;
1779 
1780 	if (name)
1781 		printf("\"%s\" at %s", aa->aa_ident, name);
1782 
1783 	return (UNCONF);
1784 }
1785 
1786 int
1787 hv_attach_devices(struct hv_softc *sc)
1788 {
1789 	struct hv_dev *dv;
1790 	struct hv_channel *ch;
1791 
1792 	SLIST_INIT(&sc->sc_devs);
1793 	mtx_init(&sc->sc_devlck, IPL_NET);
1794 
1795 	TAILQ_FOREACH(ch, &sc->sc_channels, ch_entry) {
1796 		if (ch->ch_state != HV_CHANSTATE_OFFERED)
1797 			continue;
1798 		if (!(ch->ch_flags & CHF_MONITOR))
1799 			continue;
1800 		dv = malloc(sizeof(*dv), M_DEVBUF, M_ZERO | M_NOWAIT);
1801 		if (dv == NULL) {
1802 			printf("%s: failed to allocate device object\n",
1803 			    sc->sc_dev.dv_xname);
1804 			return (-1);
1805 		}
1806 		dv->dv_aa.aa_parent = sc;
1807 		dv->dv_aa.aa_type = &ch->ch_type;
1808 		dv->dv_aa.aa_inst = &ch->ch_inst;
1809 		dv->dv_aa.aa_ident = ch->ch_ident;
1810 		dv->dv_aa.aa_chan = ch;
1811 		dv->dv_aa.aa_dmat = sc->sc_dmat;
1812 		mtx_enter(&sc->sc_devlck);
1813 		SLIST_INSERT_HEAD(&sc->sc_devs, dv, dv_entry);
1814 		mtx_leave(&sc->sc_devlck);
1815 		config_found((struct device *)sc, &dv->dv_aa, hv_attach_print);
1816 	}
1817 	return (0);
1818 }
1819 
1820 void
1821 hv_evcount_attach(struct hv_channel *ch, const char *name)
1822 {
1823 	struct hv_softc *sc = ch->ch_sc;
1824 
1825 	evcount_attach(&ch->ch_evcnt, name, &sc->sc_idtvec);
1826 }
1827