xref: /openbsd-src/sys/dev/pv/hyperv.c (revision de8cc8edbc71bd3e3bc7fbffa27ba0e564c37d8b)
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 	.tc_get_timecount = hv_gettime,
145 	.tc_poll_pps = 0,
146 	.tc_counter_mask = 0xffffffff,
147 	.tc_frequency = 10000000,
148 	.tc_name = "hyperv",
149 	.tc_quality = 9001,
150 	.tc_priv = NULL,
151 	.tc_user = 0,
152 };
153 
154 struct cfdriver hyperv_cd = {
155 	NULL, "hyperv", DV_DULL
156 };
157 
158 const struct cfattach hyperv_ca = {
159 	sizeof(struct hv_softc), hv_match, hv_attach
160 };
161 
162 const struct hv_guid hv_guid_network = {
163 	{ 0x63, 0x51, 0x61, 0xf8, 0x3e, 0xdf, 0xc5, 0x46,
164 	  0x91, 0x3f, 0xf2, 0xd2, 0xf9, 0x65, 0xed, 0x0e }
165 };
166 
167 const struct hv_guid hv_guid_ide = {
168 	{ 0x32, 0x26, 0x41, 0x32, 0xcb, 0x86, 0xa2, 0x44,
169 	  0x9b, 0x5c, 0x50, 0xd1, 0x41, 0x73, 0x54, 0xf5 }
170 };
171 
172 const struct hv_guid hv_guid_scsi = {
173 	{ 0xd9, 0x63, 0x61, 0xba, 0xa1, 0x04, 0x29, 0x4d,
174 	  0xb6, 0x05, 0x72, 0xe2, 0xff, 0xb1, 0xdc, 0x7f }
175 };
176 
177 const struct hv_guid hv_guid_shutdown = {
178 	{ 0x31, 0x60, 0x0b, 0x0e, 0x13, 0x52, 0x34, 0x49,
179 	  0x81, 0x8b, 0x38, 0xd9, 0x0c, 0xed, 0x39, 0xdb }
180 };
181 
182 const struct hv_guid hv_guid_timesync = {
183 	{ 0x30, 0xe6, 0x27, 0x95, 0xae, 0xd0, 0x7b, 0x49,
184 	  0xad, 0xce, 0xe8, 0x0a, 0xb0, 0x17, 0x5c, 0xaf }
185 };
186 
187 const struct hv_guid hv_guid_heartbeat = {
188 	{ 0x39, 0x4f, 0x16, 0x57, 0x15, 0x91, 0x78, 0x4e,
189 	  0xab, 0x55, 0x38, 0x2f, 0x3b, 0xd5, 0x42, 0x2d }
190 };
191 
192 const struct hv_guid hv_guid_kvp = {
193 	{ 0xe7, 0xf4, 0xa0, 0xa9, 0x45, 0x5a, 0x96, 0x4d,
194 	  0xb8, 0x27, 0x8a, 0x84, 0x1e, 0x8c, 0x03, 0xe6 }
195 };
196 
197 #ifdef HYPERV_DEBUG
198 const struct hv_guid hv_guid_vss = {
199 	{ 0x29, 0x2e, 0xfa, 0x35, 0x23, 0xea, 0x36, 0x42,
200 	  0x96, 0xae, 0x3a, 0x6e, 0xba, 0xcb, 0xa4, 0x40 }
201 };
202 
203 const struct hv_guid hv_guid_dynmem = {
204 	{ 0xdc, 0x74, 0x50, 0x52, 0x85, 0x89, 0xe2, 0x46,
205 	  0x80, 0x57, 0xa3, 0x07, 0xdc, 0x18, 0xa5, 0x02 }
206 };
207 
208 const struct hv_guid hv_guid_mouse = {
209 	{ 0x9e, 0xb6, 0xa8, 0xcf, 0x4a, 0x5b, 0xc0, 0x4c,
210 	  0xb9, 0x8b, 0x8b, 0xa1, 0xa1, 0xf3, 0xf9, 0x5a }
211 };
212 
213 const struct hv_guid hv_guid_kbd = {
214 	{ 0x6d, 0xad, 0x12, 0xf9, 0x17, 0x2b, 0xea, 0x48,
215 	  0xbd, 0x65, 0xf9, 0x27, 0xa6, 0x1c, 0x76, 0x84 }
216 };
217 
218 const struct hv_guid hv_guid_video = {
219 	{ 0x02, 0x78, 0x0a, 0xda, 0x77, 0xe3, 0xac, 0x4a,
220 	  0x8e, 0x77, 0x05, 0x58, 0xeb, 0x10, 0x73, 0xf8 }
221 };
222 
223 const struct hv_guid hv_guid_fc = {
224 	{ 0x4a, 0xcc, 0x9b, 0x2f, 0x69, 0x00, 0xf3, 0x4a,
225 	  0xb7, 0x6b, 0x6f, 0xd0, 0xbe, 0x52, 0x8c, 0xda }
226 };
227 
228 const struct hv_guid hv_guid_fcopy = {
229 	{ 0xe3, 0x4b, 0xd1, 0x34, 0xe4, 0xde, 0xc8, 0x41,
230 	  0x9a, 0xe7, 0x6b, 0x17, 0x49, 0x77, 0xc1, 0x92 }
231 };
232 
233 const struct hv_guid hv_guid_pcie = {
234 	{ 0x1d, 0xf6, 0xc4, 0x44, 0x44, 0x44, 0x00, 0x44,
235 	  0x9d, 0x52, 0x80, 0x2e, 0x27, 0xed, 0xe1, 0x9f }
236 };
237 
238 const struct hv_guid hv_guid_netdir = {
239 	{ 0x3d, 0xaf, 0x2e, 0x8c, 0xa7, 0x32, 0x09, 0x4b,
240 	  0xab, 0x99, 0xbd, 0x1f, 0x1c, 0x86, 0xb5, 0x01 }
241 };
242 
243 const struct hv_guid hv_guid_rdesktop = {
244 	{ 0xf4, 0xac, 0x6a, 0x27, 0x15, 0xac, 0x6c, 0x42,
245 	  0x98, 0xdd, 0x75, 0x21, 0xad, 0x3f, 0x01, 0xfe }
246 };
247 
248 /* Automatic Virtual Machine Activation (AVMA) Services */
249 const struct hv_guid hv_guid_avma1 = {
250 	{ 0x55, 0xb2, 0x87, 0x44, 0x8c, 0xb8, 0x3f, 0x40,
251 	  0xbb, 0x51, 0xd1, 0xf6, 0x9c, 0xf1, 0x7f, 0x87 }
252 };
253 
254 const struct hv_guid hv_guid_avma2 = {
255 	{ 0xf4, 0xba, 0x75, 0x33, 0x15, 0x9e, 0x30, 0x4b,
256 	  0xb7, 0x65, 0x67, 0xac, 0xb1, 0x0d, 0x60, 0x7b }
257 };
258 
259 const struct hv_guid hv_guid_avma3 = {
260 	{ 0xa0, 0x1f, 0x22, 0x99, 0xad, 0x24, 0xe2, 0x11,
261 	  0xbe, 0x98, 0x00, 0x1a, 0xa0, 0x1b, 0xbf, 0x6e }
262 };
263 
264 const struct hv_guid hv_guid_avma4 = {
265 	{ 0x16, 0x57, 0xe6, 0xf8, 0xb3, 0x3c, 0x06, 0x4a,
266 	  0x9a, 0x60, 0x18, 0x89, 0xc5, 0xcc, 0xca, 0xb5 }
267 };
268 #endif	/* HYPERV_DEBUG */
269 
270 int
271 hv_match(struct device *parent, void *match, void *aux)
272 {
273 	struct pv_attach_args *pva = aux;
274 	struct pvbus_hv *hv = &pva->pva_hv[PVBUS_HYPERV];
275 
276 	if ((hv->hv_major == 0 && hv->hv_minor == 0) || hv->hv_base == 0)
277 		return (0);
278 
279 	return (1);
280 }
281 
282 void
283 hv_attach(struct device *parent, struct device *self, void *aux)
284 {
285 	struct hv_softc *sc = (struct hv_softc *)self;
286 	struct pv_attach_args *pva = aux;
287 	struct pvbus_hv *hv = &pva->pva_hv[PVBUS_HYPERV];
288 
289 	sc->sc_pvbus = hv;
290 	sc->sc_dmat = pva->pva_dmat;
291 
292 	if (!(hv->hv_features & CPUID_HV_MSR_HYPERCALL) ||
293 	    !(hv->hv_features & CPUID_HV_MSR_SYNIC)) {
294 		printf(": not functional\n");
295 		return;
296 	}
297 
298 	DPRINTF("\n");
299 
300 	hv_set_version(sc);
301 
302 	if (hv->hv_features & CPUID_HV_MSR_TIME_REFCNT)
303 		tc_init(&hv_timecounter);
304 
305 	if (hv_init_hypercall(sc))
306 		return;
307 
308 	/* Wire it up to the global */
309 	hv_sc = sc;
310 
311 	if (hv_init_interrupts(sc))
312 		return;
313 
314 	if (hv_vmbus_connect(sc))
315 		return;
316 
317 	DPRINTF("%s", sc->sc_dev.dv_xname);
318 	printf(": protocol %d.%d, features %#x\n",
319 	    VMBUS_VERSION_MAJOR(sc->sc_proto),
320 	    VMBUS_VERSION_MINOR(sc->sc_proto),
321 	    hv->hv_features);
322 
323 	if (hv_channel_scan(sc))
324 		return;
325 
326 	/* Attach heartbeat, KVP and other "internal" services */
327 	hv_attach_icdevs(sc);
328 
329 	/* Attach devices with external drivers */
330 	hv_attach_devices(sc);
331 }
332 
333 void
334 hv_set_version(struct hv_softc *sc)
335 {
336 	uint64_t ver;
337 
338 	/* OpenBSD build date */
339 	ver = MSR_HV_GUESTID_OSTYPE_OPENBSD;
340 	ver |= (uint64_t)OpenBSD << MSR_HV_GUESTID_VERSION_SHIFT;
341 	wrmsr(MSR_HV_GUEST_OS_ID, ver);
342 }
343 
344 u_int
345 hv_gettime(struct timecounter *tc)
346 {
347 	u_int now = rdmsr(MSR_HV_TIME_REF_COUNT);
348 
349 	return (now);
350 }
351 
352 int
353 hv_init_hypercall(struct hv_softc *sc)
354 {
355 	extern void *hv_hypercall_page;
356 	uint64_t msr;
357 	paddr_t pa;
358 
359 	sc->sc_hc = &hv_hypercall_page;
360 
361 	if (!pmap_extract(pmap_kernel(), (vaddr_t)sc->sc_hc, &pa)) {
362 		printf(": hypercall page PA extraction failed\n");
363 		return (-1);
364 	}
365 
366 	msr = (atop(pa) << MSR_HV_HYPERCALL_PGSHIFT) | MSR_HV_HYPERCALL_ENABLE;
367 	wrmsr(MSR_HV_HYPERCALL, msr);
368 
369 	if (!(rdmsr(MSR_HV_HYPERCALL) & MSR_HV_HYPERCALL_ENABLE)) {
370 		printf(": failed to set up a hypercall page\n");
371 		return (-1);
372 	}
373 
374 	return (0);
375 }
376 
377 uint64_t
378 hv_hypercall(struct hv_softc *sc, uint64_t control, void *input,
379     void *output)
380 {
381 	paddr_t input_pa = 0, output_pa = 0;
382 	uint64_t status = 0;
383 
384 	if (input != NULL &&
385 	    pmap_extract(pmap_kernel(), (vaddr_t)input, &input_pa) == 0) {
386 		printf("%s: hypercall input PA extraction failed\n",
387 		    sc->sc_dev.dv_xname);
388 		return (~HYPERCALL_STATUS_SUCCESS);
389 	}
390 
391 	if (output != NULL &&
392 	    pmap_extract(pmap_kernel(), (vaddr_t)output, &output_pa) == 0) {
393 		printf("%s: hypercall output PA extraction failed\n",
394 		    sc->sc_dev.dv_xname);
395 		return (~HYPERCALL_STATUS_SUCCESS);
396 	}
397 
398 #ifdef __amd64__
399 	__asm__ __volatile__ ("mov %0, %%r8" : : "r" (output_pa) : "r8");
400 	__asm__ __volatile__ ("call *%3" : "=a" (status) : "c" (control),
401 	    "d" (input_pa), "m" (sc->sc_hc));
402 #else  /* __i386__ */
403 	{
404 		uint32_t control_hi = control >> 32;
405 		uint32_t control_lo = control & 0xfffffffff;
406 		uint32_t status_hi = 1;
407 		uint32_t status_lo = 1;
408 
409 		__asm__ __volatile__ ("call *%8" :
410 		    "=d" (status_hi), "=a"(status_lo) :
411 		    "d" (control_hi), "a" (control_lo),
412 		    "b" (0), "c" (input_pa), "D" (0), "S" (output_pa),
413 		    "m" (sc->sc_hc));
414 
415 		status = status_lo | ((uint64_t)status_hi << 32);
416 	}
417 #endif	/* __amd64__ */
418 
419 	return (status);
420 }
421 
422 int
423 hv_init_interrupts(struct hv_softc *sc)
424 {
425 	struct cpu_info *ci = curcpu();
426 	int cpu = CPU_INFO_UNIT(ci);
427 
428 	sc->sc_idtvec = LAPIC_HYPERV_VECTOR;
429 
430 	TAILQ_INIT(&sc->sc_reqs);
431 	mtx_init(&sc->sc_reqlck, IPL_NET);
432 
433 	TAILQ_INIT(&sc->sc_rsps);
434 	mtx_init(&sc->sc_rsplck, IPL_NET);
435 
436 	sc->sc_simp[cpu] = km_alloc(PAGE_SIZE, &kv_any, &kp_zero, &kd_nowait);
437 	if (sc->sc_simp[cpu] == NULL) {
438 		printf(": failed to allocate SIMP\n");
439 		return (-1);
440 	}
441 
442 	sc->sc_siep[cpu] = km_alloc(PAGE_SIZE, &kv_any, &kp_zero, &kd_nowait);
443 	if (sc->sc_siep[cpu] == NULL) {
444 		printf(": failed to allocate SIEP\n");
445 		km_free(sc->sc_simp[cpu], PAGE_SIZE, &kv_any, &kp_zero);
446 		return (-1);
447 	}
448 
449 	sc->sc_proto = VMBUS_VERSION_WS2008;
450 
451 	return (hv_init_synic(sc));
452 }
453 
454 int
455 hv_init_synic(struct hv_softc *sc)
456 {
457 	struct cpu_info *ci = curcpu();
458 	int cpu = CPU_INFO_UNIT(ci);
459 	uint64_t simp, siefp, sctrl, sint;
460 	paddr_t pa;
461 
462 	/*
463 	 * Setup the Synic's message page
464 	 */
465 	if (!pmap_extract(pmap_kernel(), (vaddr_t)sc->sc_simp[cpu], &pa)) {
466 		printf(": SIMP PA extraction failed\n");
467 		return (-1);
468 	}
469 	simp = rdmsr(MSR_HV_SIMP);
470 	simp &= (1 << MSR_HV_SIMP_PGSHIFT) - 1;
471 	simp |= (atop(pa) << MSR_HV_SIMP_PGSHIFT);
472 	simp |= MSR_HV_SIMP_ENABLE;
473 	wrmsr(MSR_HV_SIMP, simp);
474 
475 	/*
476 	 * Setup the Synic's event page
477 	 */
478 	if (!pmap_extract(pmap_kernel(), (vaddr_t)sc->sc_siep[cpu], &pa)) {
479 		printf(": SIEP PA extraction failed\n");
480 		return (-1);
481 	}
482 	siefp = rdmsr(MSR_HV_SIEFP);
483 	siefp &= (1<<MSR_HV_SIEFP_PGSHIFT) - 1;
484 	siefp |= (atop(pa) << MSR_HV_SIEFP_PGSHIFT);
485 	siefp |= MSR_HV_SIEFP_ENABLE;
486 	wrmsr(MSR_HV_SIEFP, siefp);
487 
488 	/*
489 	 * Configure and unmask SINT for message and event flags
490 	 */
491 	sint = rdmsr(MSR_HV_SINT0 + VMBUS_SINT_MESSAGE);
492 	sint = sc->sc_idtvec | MSR_HV_SINT_AUTOEOI |
493 	    (sint & MSR_HV_SINT_RSVD_MASK);
494 	wrmsr(MSR_HV_SINT0 + VMBUS_SINT_MESSAGE, sint);
495 
496 	/* Enable the global synic bit */
497 	sctrl = rdmsr(MSR_HV_SCONTROL);
498 	sctrl |= MSR_HV_SCTRL_ENABLE;
499 	wrmsr(MSR_HV_SCONTROL, sctrl);
500 
501 	sc->sc_vcpus[cpu] = rdmsr(MSR_HV_VP_INDEX);
502 
503 	DPRINTF("vcpu%u: SIMP %#llx SIEFP %#llx SCTRL %#llx\n",
504 	    sc->sc_vcpus[cpu], simp, siefp, sctrl);
505 
506 	return (0);
507 }
508 
509 int
510 hv_cmd(struct hv_softc *sc, void *cmd, size_t cmdlen, void *rsp,
511     size_t rsplen, int flags)
512 {
513 	struct hv_msg msg;
514 	int rv;
515 
516 	if (cmdlen > VMBUS_MSG_DSIZE_MAX) {
517 		printf("%s: payload too large (%lu)\n", sc->sc_dev.dv_xname,
518 		    cmdlen);
519 		return (EMSGSIZE);
520 	}
521 
522 	memset(&msg, 0, sizeof(msg));
523 
524 	msg.msg_req.hc_dsize = cmdlen;
525 	memcpy(msg.msg_req.hc_data, cmd, cmdlen);
526 
527 	if (!(flags & HCF_NOREPLY)) {
528 		msg.msg_rsp = rsp;
529 		msg.msg_rsplen = rsplen;
530 	} else
531 		msg.msg_flags |= MSGF_NOQUEUE;
532 
533 	if (flags & HCF_NOSLEEP)
534 		msg.msg_flags |= MSGF_NOSLEEP;
535 
536 	if ((rv = hv_start(sc, &msg)) != 0)
537 		return (rv);
538 	return (hv_reply(sc, &msg));
539 }
540 
541 int
542 hv_start(struct hv_softc *sc, struct hv_msg *msg)
543 {
544 	const int delays[] = { 100, 100, 100, 500, 500, 5000, 5000, 5000 };
545 	const char *wchan = "hvstart";
546 	uint16_t status;
547 	int i, s;
548 
549 	msg->msg_req.hc_connid = VMBUS_CONNID_MESSAGE;
550 	msg->msg_req.hc_msgtype = 1;
551 
552 	if (!(msg->msg_flags & MSGF_NOQUEUE)) {
553 		mtx_enter(&sc->sc_reqlck);
554 		TAILQ_INSERT_TAIL(&sc->sc_reqs, msg, msg_entry);
555 		mtx_leave(&sc->sc_reqlck);
556 	}
557 
558 	for (i = 0; i < nitems(delays); i++) {
559 		status = hv_hypercall(sc, HYPERCALL_POST_MESSAGE,
560 		    &msg->msg_req, NULL);
561 		if (status == HYPERCALL_STATUS_SUCCESS)
562 			break;
563 		if (msg->msg_flags & MSGF_NOSLEEP) {
564 			delay(delays[i]);
565 			s = splnet();
566 			hv_intr();
567 			splx(s);
568 		} else {
569 			tsleep_nsec(wchan, PRIBIO, wchan,
570 			    USEC_TO_NSEC(delays[i]));
571 		}
572 	}
573 	if (status != 0) {
574 		printf("%s: posting vmbus message failed with %d\n",
575 		    sc->sc_dev.dv_xname, status);
576 		if (!(msg->msg_flags & MSGF_NOQUEUE)) {
577 			mtx_enter(&sc->sc_reqlck);
578 			TAILQ_REMOVE(&sc->sc_reqs, msg, msg_entry);
579 			mtx_leave(&sc->sc_reqlck);
580 		}
581 		return (EIO);
582 	}
583 
584 	return (0);
585 }
586 
587 static int
588 hv_reply_done(struct hv_softc *sc, struct hv_msg *msg)
589 {
590 	struct hv_msg *m;
591 
592 	mtx_enter(&sc->sc_rsplck);
593 	TAILQ_FOREACH(m, &sc->sc_rsps, msg_entry) {
594 		if (m == msg) {
595 			mtx_leave(&sc->sc_rsplck);
596 			return (1);
597 		}
598 	}
599 	mtx_leave(&sc->sc_rsplck);
600 	return (0);
601 }
602 
603 int
604 hv_reply(struct hv_softc *sc, struct hv_msg *msg)
605 {
606 	if (msg->msg_flags & MSGF_NOQUEUE)
607 		return (0);
608 
609 	hv_wait(sc, hv_reply_done, msg, msg, "hvreply");
610 
611 	mtx_enter(&sc->sc_rsplck);
612 	TAILQ_REMOVE(&sc->sc_rsps, msg, msg_entry);
613 	mtx_leave(&sc->sc_rsplck);
614 
615 	return (0);
616 }
617 
618 void
619 hv_wait(struct hv_softc *sc, int (*cond)(struct hv_softc *, struct hv_msg *),
620     struct hv_msg *msg, void *wchan, const char *wmsg)
621 {
622 	int s;
623 
624 	KASSERT(cold ? msg->msg_flags & MSGF_NOSLEEP : 1);
625 
626 	while (!cond(sc, msg)) {
627 		if (msg->msg_flags & MSGF_NOSLEEP) {
628 			delay(1000);
629 			s = splnet();
630 			hv_intr();
631 			splx(s);
632 		} else {
633 			tsleep_nsec(wchan, PRIBIO, wmsg ? wmsg : "hvwait",
634 			    USEC_TO_NSEC(1000));
635 		}
636 	}
637 }
638 
639 uint16_t
640 hv_intr_signal(struct hv_softc *sc, void *con)
641 {
642 	uint64_t status;
643 
644 	status = hv_hypercall(sc, HYPERCALL_SIGNAL_EVENT, con, NULL);
645 	return ((uint16_t)status);
646 }
647 
648 void
649 hv_intr(void)
650 {
651 	struct hv_softc *sc = hv_sc;
652 
653 	hv_event_intr(sc);
654 	hv_message_intr(sc);
655 }
656 
657 void
658 hv_event_intr(struct hv_softc *sc)
659 {
660 	struct vmbus_evtflags *evt;
661 	struct cpu_info *ci = curcpu();
662 	int cpu = CPU_INFO_UNIT(ci);
663 	int bit, row, maxrow, chanid;
664 	struct hv_channel *ch;
665 	u_long *revents, pending;
666 
667 	evt = (struct vmbus_evtflags *)sc->sc_siep[cpu] +
668 	    VMBUS_SINT_MESSAGE;
669 	if ((sc->sc_proto == VMBUS_VERSION_WS2008) ||
670 	    (sc->sc_proto == VMBUS_VERSION_WIN7)) {
671 		if (!test_bit(0, &evt->evt_flags[0]))
672 			return;
673 		clear_bit(0, &evt->evt_flags[0]);
674 		maxrow = VMBUS_CHAN_MAX_COMPAT / VMBUS_EVTFLAG_LEN;
675 		/*
676 		 * receive size is 1/2 page and divide that by 4 bytes
677 		 */
678 		revents = sc->sc_revents;
679 	} else {
680 		maxrow = nitems(evt->evt_flags);
681 		/*
682 		 * On Host with Win8 or above, the event page can be
683 		 * checked directly to get the id of the channel
684 		 * that has the pending interrupt.
685 		 */
686 		revents = &evt->evt_flags[0];
687 	}
688 
689 	for (row = 0; row < maxrow; row++) {
690 		if (revents[row] == 0)
691 			continue;
692 		pending = atomic_swap_ulong(&revents[row], 0);
693 		for (bit = 0; pending > 0; pending >>= 1, bit++) {
694 			if ((pending & 1) == 0)
695 				continue;
696 			chanid = (row * LONG_BIT) + bit;
697 			/* vmbus channel protocol message */
698 			if (chanid == 0)
699 				continue;
700 			ch = hv_channel_lookup(sc, chanid);
701 			if (ch == NULL) {
702 				printf("%s: unhandled event on %d\n",
703 				    sc->sc_dev.dv_xname, chanid);
704 				continue;
705 			}
706 			if (ch->ch_state != HV_CHANSTATE_OPENED) {
707 				printf("%s: channel %d is not active\n",
708 				    sc->sc_dev.dv_xname, chanid);
709 				continue;
710 			}
711 			ch->ch_evcnt.ec_count++;
712 			hv_channel_schedule(ch);
713 		}
714 	}
715 }
716 
717 void
718 hv_message_intr(struct hv_softc *sc)
719 {
720 	struct vmbus_message *msg;
721 	struct vmbus_chanmsg_hdr *hdr;
722 	struct cpu_info *ci = curcpu();
723 	int cpu = CPU_INFO_UNIT(ci);
724 
725 	for (;;) {
726 		msg = (struct vmbus_message *)sc->sc_simp[cpu] +
727 		    VMBUS_SINT_MESSAGE;
728 		if (msg->msg_type == VMBUS_MSGTYPE_NONE)
729 			break;
730 
731 		hdr = (struct vmbus_chanmsg_hdr *)msg->msg_data;
732 		if (hdr->chm_type >= VMBUS_CHANMSG_COUNT) {
733 			printf("%s: unhandled message type %u flags %#x\n",
734 			    sc->sc_dev.dv_xname, hdr->chm_type,
735 			    msg->msg_flags);
736 			goto skip;
737 		}
738 		if (hv_msg_dispatch[hdr->chm_type].hmd_handler)
739 			hv_msg_dispatch[hdr->chm_type].hmd_handler(sc, hdr);
740 		else
741 			printf("%s: unhandled message type %u\n",
742 			    sc->sc_dev.dv_xname, hdr->chm_type);
743  skip:
744 		msg->msg_type = VMBUS_MSGTYPE_NONE;
745 		virtio_membar_sync();
746 		if (msg->msg_flags & VMBUS_MSGFLAG_PENDING)
747 			wrmsr(MSR_HV_EOM, 0);
748 	}
749 }
750 
751 void
752 hv_channel_response(struct hv_softc *sc, struct vmbus_chanmsg_hdr *rsphdr)
753 {
754 	struct hv_msg *msg;
755 	struct vmbus_chanmsg_hdr *reqhdr;
756 	int req;
757 
758 	req = hv_msg_dispatch[rsphdr->chm_type].hmd_request;
759 	mtx_enter(&sc->sc_reqlck);
760 	TAILQ_FOREACH(msg, &sc->sc_reqs, msg_entry) {
761 		reqhdr = (struct vmbus_chanmsg_hdr *)&msg->msg_req.hc_data;
762 		if (reqhdr->chm_type == req) {
763 			TAILQ_REMOVE(&sc->sc_reqs, msg, msg_entry);
764 			break;
765 		}
766 	}
767 	mtx_leave(&sc->sc_reqlck);
768 	if (msg != NULL) {
769 		memcpy(msg->msg_rsp, rsphdr, msg->msg_rsplen);
770 		mtx_enter(&sc->sc_rsplck);
771 		TAILQ_INSERT_TAIL(&sc->sc_rsps, msg, msg_entry);
772 		mtx_leave(&sc->sc_rsplck);
773 		wakeup(msg);
774 	}
775 }
776 
777 void
778 hv_channel_offer(struct hv_softc *sc, struct vmbus_chanmsg_hdr *hdr)
779 {
780 	struct hv_offer *co;
781 
782 	co = malloc(sizeof(*co), M_DEVBUF, M_NOWAIT | M_ZERO);
783 	if (co == NULL) {
784 		printf("%s: failed to allocate an offer object\n",
785 		    sc->sc_dev.dv_xname);
786 		return;
787 	}
788 
789 	memcpy(&co->co_chan, hdr, sizeof(co->co_chan));
790 
791 	mtx_enter(&sc->sc_offerlck);
792 	SIMPLEQ_INSERT_TAIL(&sc->sc_offers, co, co_entry);
793 	mtx_leave(&sc->sc_offerlck);
794 }
795 
796 void
797 hv_channel_rescind(struct hv_softc *sc, struct vmbus_chanmsg_hdr *hdr)
798 {
799 	const struct vmbus_chanmsg_chrescind *cmd;
800 
801 	cmd = (const struct vmbus_chanmsg_chrescind *)hdr;
802 	printf("%s: revoking channel %u\n", sc->sc_dev.dv_xname,
803 	    cmd->chm_chanid);
804 }
805 
806 void
807 hv_channel_delivered(struct hv_softc *sc, struct vmbus_chanmsg_hdr *hdr)
808 {
809 	atomic_setbits_int(&sc->sc_flags, HSF_OFFERS_DELIVERED);
810 	wakeup(&sc->sc_offers);
811 }
812 
813 int
814 hv_vmbus_connect(struct hv_softc *sc)
815 {
816 	const uint32_t versions[] = {
817 		VMBUS_VERSION_WIN10,
818 		VMBUS_VERSION_WIN8_1, VMBUS_VERSION_WIN8,
819 		VMBUS_VERSION_WIN7, VMBUS_VERSION_WS2008
820 	};
821 	struct vmbus_chanmsg_connect cmd;
822 	struct vmbus_chanmsg_connect_resp rsp;
823 	paddr_t epa, mpa1, mpa2;
824 	int i;
825 
826 	sc->sc_events = km_alloc(PAGE_SIZE, &kv_any, &kp_zero, &kd_nowait);
827 	if (sc->sc_events == NULL) {
828 		printf(": failed to allocate channel port events page\n");
829 		goto errout;
830 	}
831 	if (!pmap_extract(pmap_kernel(), (vaddr_t)sc->sc_events, &epa)) {
832 		printf(": channel port events page PA extraction failed\n");
833 		goto errout;
834 	}
835 
836 	sc->sc_wevents = (u_long *)sc->sc_events;
837 	sc->sc_revents = (u_long *)((caddr_t)sc->sc_events + (PAGE_SIZE >> 1));
838 
839 	sc->sc_monitor[0] = km_alloc(PAGE_SIZE, &kv_any, &kp_zero, &kd_nowait);
840 	if (sc->sc_monitor[0] == NULL) {
841 		printf(": failed to allocate monitor page 1\n");
842 		goto errout;
843 	}
844 	if (!pmap_extract(pmap_kernel(), (vaddr_t)sc->sc_monitor[0], &mpa1)) {
845 		printf(": monitor page 1 PA extraction failed\n");
846 		goto errout;
847 	}
848 
849 	sc->sc_monitor[1] = km_alloc(PAGE_SIZE, &kv_any, &kp_zero, &kd_nowait);
850 	if (sc->sc_monitor[1] == NULL) {
851 		printf(": failed to allocate monitor page 2\n");
852 		goto errout;
853 	}
854 	if (!pmap_extract(pmap_kernel(), (vaddr_t)sc->sc_monitor[1], &mpa2)) {
855 		printf(": monitor page 2 PA extraction failed\n");
856 		goto errout;
857 	}
858 
859 	memset(&cmd, 0, sizeof(cmd));
860 	cmd.chm_hdr.chm_type = VMBUS_CHANMSG_CONNECT;
861 	cmd.chm_evtflags = (uint64_t)epa;
862 	cmd.chm_mnf1 = (uint64_t)mpa1;
863 	cmd.chm_mnf2 = (uint64_t)mpa2;
864 
865 	memset(&rsp, 0, sizeof(rsp));
866 
867 	for (i = 0; i < nitems(versions); i++) {
868 		cmd.chm_ver = versions[i];
869 		if (hv_cmd(sc, &cmd, sizeof(cmd), &rsp, sizeof(rsp),
870 		    HCF_NOSLEEP)) {
871 			DPRINTF("%s: CONNECT failed\n",
872 			    sc->sc_dev.dv_xname);
873 			goto errout;
874 		}
875 		if (rsp.chm_done) {
876 			sc->sc_flags |= HSF_CONNECTED;
877 			sc->sc_proto = versions[i];
878 			sc->sc_handle = VMBUS_GPADL_START;
879 			break;
880 		}
881 	}
882 	if (i == nitems(versions)) {
883 		printf("%s: failed to negotiate protocol version\n",
884 		    sc->sc_dev.dv_xname);
885 		goto errout;
886 	}
887 
888 	return (0);
889 
890  errout:
891 	if (sc->sc_events) {
892 		km_free(sc->sc_events, PAGE_SIZE, &kv_any, &kp_zero);
893 		sc->sc_events = NULL;
894 		sc->sc_wevents = NULL;
895 		sc->sc_revents = NULL;
896 	}
897 	if (sc->sc_monitor[0]) {
898 		km_free(sc->sc_monitor[0], PAGE_SIZE, &kv_any, &kp_zero);
899 		sc->sc_monitor[0] = NULL;
900 	}
901 	if (sc->sc_monitor[1]) {
902 		km_free(sc->sc_monitor[1], PAGE_SIZE, &kv_any, &kp_zero);
903 		sc->sc_monitor[1] = NULL;
904 	}
905 	return (-1);
906 }
907 
908 #ifdef HYPERV_DEBUG
909 static inline char *
910 guidprint(struct hv_guid *a)
911 {
912 	/* 3     0  5  4 7 6  8 9  10        15 */
913 	/* 33221100-5544-7766-9988-FFEEDDCCBBAA */
914 	static char buf[16 * 2 + 4 + 1];
915 	int i, j = 0;
916 
917 	for (i = 3; i != -1; i -= 1, j += 2)
918 		snprintf(&buf[j], 3, "%02x", (uint8_t)a->data[i]);
919 	buf[j++] = '-';
920 	for (i = 5; i != 3; i -= 1, j += 2)
921 		snprintf(&buf[j], 3, "%02x", (uint8_t)a->data[i]);
922 	buf[j++] = '-';
923 	for (i = 7; i != 5; i -= 1, j += 2)
924 		snprintf(&buf[j], 3, "%02x", (uint8_t)a->data[i]);
925 	buf[j++] = '-';
926 	for (i = 8; i < 10; i += 1, j += 2)
927 		snprintf(&buf[j], 3, "%02x", (uint8_t)a->data[i]);
928 	buf[j++] = '-';
929 	for (i = 10; i < 16; i += 1, j += 2)
930 		snprintf(&buf[j], 3, "%02x", (uint8_t)a->data[i]);
931 	return (&buf[0]);
932 }
933 #endif	/* HYPERV_DEBUG */
934 
935 void
936 hv_guid_sprint(struct hv_guid *guid, char *str, size_t size)
937 {
938 	const struct {
939 		const struct hv_guid	*guid;
940 		const char		*ident;
941 	} map[] = {
942 		{ &hv_guid_network,	"network" },
943 		{ &hv_guid_ide,		"ide" },
944 		{ &hv_guid_scsi,	"scsi" },
945 		{ &hv_guid_shutdown,	"shutdown" },
946 		{ &hv_guid_timesync,	"timesync" },
947 		{ &hv_guid_heartbeat,	"heartbeat" },
948 		{ &hv_guid_kvp,		"kvp" },
949 #ifdef HYPERV_DEBUG
950 		{ &hv_guid_vss,		"vss" },
951 		{ &hv_guid_dynmem,	"dynamic-memory" },
952 		{ &hv_guid_mouse,	"mouse" },
953 		{ &hv_guid_kbd,		"keyboard" },
954 		{ &hv_guid_video,	"video" },
955 		{ &hv_guid_fc,		"fiber-channel" },
956 		{ &hv_guid_fcopy,	"file-copy" },
957 		{ &hv_guid_pcie,	"pcie-passthrough" },
958 		{ &hv_guid_netdir,	"network-direct" },
959 		{ &hv_guid_rdesktop,	"remote-desktop" },
960 		{ &hv_guid_avma1,	"avma-1" },
961 		{ &hv_guid_avma2,	"avma-2" },
962 		{ &hv_guid_avma3,	"avma-3" },
963 		{ &hv_guid_avma4,	"avma-4" },
964 #endif
965 	};
966 	int i;
967 
968 	for (i = 0; i < nitems(map); i++) {
969 		if (memcmp(guid, map[i].guid, sizeof(*guid)) == 0) {
970 			strlcpy(str, map[i].ident, size);
971 			return;
972 		}
973 	}
974 #ifdef HYPERV_DEBUG
975 	strlcpy(str, guidprint(guid), size);
976 #endif
977 }
978 
979 static int
980 hv_channel_scan_done(struct hv_softc *sc, struct hv_msg *msg __unused)
981 {
982 	return (sc->sc_flags & HSF_OFFERS_DELIVERED);
983 }
984 
985 int
986 hv_channel_scan(struct hv_softc *sc)
987 {
988 	struct vmbus_chanmsg_hdr hdr;
989 	struct vmbus_chanmsg_choffer rsp;
990 	struct hv_offer *co;
991 
992 	SIMPLEQ_INIT(&sc->sc_offers);
993 	mtx_init(&sc->sc_offerlck, IPL_NET);
994 
995 	memset(&hdr, 0, sizeof(hdr));
996 	hdr.chm_type = VMBUS_CHANMSG_CHREQUEST;
997 
998 	if (hv_cmd(sc, &hdr, sizeof(hdr), &rsp, sizeof(rsp),
999 	    HCF_NOSLEEP | HCF_NOREPLY)) {
1000 		DPRINTF("%s: CHREQUEST failed\n", sc->sc_dev.dv_xname);
1001 		return (-1);
1002 	}
1003 
1004 	hv_wait(sc, hv_channel_scan_done, (struct hv_msg *)&hdr,
1005 	    &sc->sc_offers, "hvscan");
1006 
1007 	TAILQ_INIT(&sc->sc_channels);
1008 	mtx_init(&sc->sc_channelck, IPL_NET);
1009 
1010 	mtx_enter(&sc->sc_offerlck);
1011 	while (!SIMPLEQ_EMPTY(&sc->sc_offers)) {
1012 		co = SIMPLEQ_FIRST(&sc->sc_offers);
1013 		SIMPLEQ_REMOVE_HEAD(&sc->sc_offers, co_entry);
1014 		mtx_leave(&sc->sc_offerlck);
1015 
1016 		hv_process_offer(sc, co);
1017 		free(co, M_DEVBUF, sizeof(*co));
1018 
1019 		mtx_enter(&sc->sc_offerlck);
1020 	}
1021 	mtx_leave(&sc->sc_offerlck);
1022 
1023 	return (0);
1024 }
1025 
1026 void
1027 hv_process_offer(struct hv_softc *sc, struct hv_offer *co)
1028 {
1029 	struct hv_channel *ch, *nch;
1030 
1031 	nch = malloc(sizeof(*nch), M_DEVBUF, M_ZERO | M_NOWAIT);
1032 	if (nch == NULL) {
1033 		printf("%s: failed to allocate memory for the channel\n",
1034 		    sc->sc_dev.dv_xname);
1035 		return;
1036 	}
1037 	nch->ch_sc = sc;
1038 	hv_guid_sprint(&co->co_chan.chm_chtype, nch->ch_ident,
1039 	    sizeof(nch->ch_ident));
1040 
1041 	/*
1042 	 * By default we setup state to enable batched reading.
1043 	 * A specific service can choose to disable this prior
1044 	 * to opening the channel.
1045 	 */
1046 	nch->ch_flags |= CHF_BATCHED;
1047 
1048 	KASSERT((((vaddr_t)&nch->ch_monprm) & 0x7) == 0);
1049 	memset(&nch->ch_monprm, 0, sizeof(nch->ch_monprm));
1050 	nch->ch_monprm.mp_connid = VMBUS_CONNID_EVENT;
1051 
1052 	if (sc->sc_proto != VMBUS_VERSION_WS2008)
1053 		nch->ch_monprm.mp_connid = co->co_chan.chm_connid;
1054 
1055 	if (co->co_chan.chm_flags1 & VMBUS_CHOFFER_FLAG1_HASMNF) {
1056 		nch->ch_mgroup = co->co_chan.chm_montrig / VMBUS_MONTRIG_LEN;
1057 		nch->ch_mindex = co->co_chan.chm_montrig % VMBUS_MONTRIG_LEN;
1058 		nch->ch_flags |= CHF_MONITOR;
1059 	}
1060 
1061 	nch->ch_id = co->co_chan.chm_chanid;
1062 
1063 	memcpy(&nch->ch_type, &co->co_chan.chm_chtype, sizeof(ch->ch_type));
1064 	memcpy(&nch->ch_inst, &co->co_chan.chm_chinst, sizeof(ch->ch_inst));
1065 
1066 	mtx_enter(&sc->sc_channelck);
1067 	TAILQ_FOREACH(ch, &sc->sc_channels, ch_entry) {
1068 		if (!memcmp(&ch->ch_type, &nch->ch_type, sizeof(ch->ch_type)) &&
1069 		    !memcmp(&ch->ch_inst, &nch->ch_inst, sizeof(ch->ch_inst)))
1070 			break;
1071 	}
1072 	if (ch != NULL) {
1073 		if (co->co_chan.chm_subidx == 0) {
1074 			printf("%s: unknown offer \"%s\"\n",
1075 			    sc->sc_dev.dv_xname, nch->ch_ident);
1076 			mtx_leave(&sc->sc_channelck);
1077 			free(nch, M_DEVBUF, sizeof(*nch));
1078 			return;
1079 		}
1080 #ifdef HYPERV_DEBUG
1081 		printf("%s: subchannel %u for \"%s\"\n", sc->sc_dev.dv_xname,
1082 		    co->co_chan.chm_subidx, ch->ch_ident);
1083 #endif
1084 		mtx_leave(&sc->sc_channelck);
1085 		free(nch, M_DEVBUF, sizeof(*nch));
1086 		return;
1087 	}
1088 
1089 	nch->ch_state = HV_CHANSTATE_OFFERED;
1090 
1091 	TAILQ_INSERT_TAIL(&sc->sc_channels, nch, ch_entry);
1092 	mtx_leave(&sc->sc_channelck);
1093 
1094 #ifdef HYPERV_DEBUG
1095 	printf("%s: channel %u: \"%s\"", sc->sc_dev.dv_xname, nch->ch_id,
1096 	    nch->ch_ident);
1097 	if (nch->ch_flags & CHF_MONITOR)
1098 		printf(", monitor %u\n", co->co_chan.chm_montrig);
1099 	else
1100 		printf("\n");
1101 #endif
1102 }
1103 
1104 struct hv_channel *
1105 hv_channel_lookup(struct hv_softc *sc, uint32_t relid)
1106 {
1107 	struct hv_channel *ch;
1108 
1109 	TAILQ_FOREACH(ch, &sc->sc_channels, ch_entry) {
1110 		if (ch->ch_id == relid)
1111 			return (ch);
1112 	}
1113 	return (NULL);
1114 }
1115 
1116 int
1117 hv_channel_ring_create(struct hv_channel *ch, uint32_t buflen)
1118 {
1119 	struct hv_softc *sc = ch->ch_sc;
1120 
1121 	buflen = roundup(buflen, PAGE_SIZE) + sizeof(struct vmbus_bufring);
1122 	ch->ch_ring = km_alloc(2 * buflen, &kv_any, &kp_zero, cold ?
1123 	    &kd_nowait : &kd_waitok);
1124 	if (ch->ch_ring == NULL) {
1125 		printf("%s: failed to allocate channel ring\n",
1126 		    sc->sc_dev.dv_xname);
1127 		return (-1);
1128 	}
1129 	ch->ch_ring_size = 2 * buflen;
1130 
1131 	memset(&ch->ch_wrd, 0, sizeof(ch->ch_wrd));
1132 	ch->ch_wrd.rd_ring = (struct vmbus_bufring *)ch->ch_ring;
1133 	ch->ch_wrd.rd_size = buflen;
1134 	ch->ch_wrd.rd_dsize = buflen - sizeof(struct vmbus_bufring);
1135 	mtx_init(&ch->ch_wrd.rd_lock, IPL_NET);
1136 
1137 	memset(&ch->ch_rrd, 0, sizeof(ch->ch_rrd));
1138 	ch->ch_rrd.rd_ring = (struct vmbus_bufring *)((uint8_t *)ch->ch_ring +
1139 	    buflen);
1140 	ch->ch_rrd.rd_size = buflen;
1141 	ch->ch_rrd.rd_dsize = buflen - sizeof(struct vmbus_bufring);
1142 	mtx_init(&ch->ch_rrd.rd_lock, IPL_NET);
1143 
1144 	if (hv_handle_alloc(ch, ch->ch_ring, 2 * buflen, &ch->ch_ring_gpadl)) {
1145 		printf("%s: failed to obtain a PA handle for the ring\n",
1146 		    sc->sc_dev.dv_xname);
1147 		hv_channel_ring_destroy(ch);
1148 		return (-1);
1149 	}
1150 
1151 	return (0);
1152 }
1153 
1154 void
1155 hv_channel_ring_destroy(struct hv_channel *ch)
1156 {
1157 	km_free(ch->ch_ring, ch->ch_ring_size, &kv_any, &kp_zero);
1158 	ch->ch_ring = NULL;
1159 	hv_handle_free(ch, ch->ch_ring_gpadl);
1160 
1161 	memset(&ch->ch_wrd, 0, sizeof(ch->ch_wrd));
1162 	memset(&ch->ch_rrd, 0, sizeof(ch->ch_rrd));
1163 }
1164 
1165 int
1166 hv_channel_open(struct hv_channel *ch, size_t buflen, void *udata,
1167     size_t udatalen, void (*handler)(void *), void *arg)
1168 {
1169 	struct hv_softc *sc = ch->ch_sc;
1170 	struct vmbus_chanmsg_chopen cmd;
1171 	struct vmbus_chanmsg_chopen_resp rsp;
1172 	int rv;
1173 
1174 	if (ch->ch_ring == NULL &&
1175 	    hv_channel_ring_create(ch, buflen)) {
1176 		DPRINTF("%s: failed to create channel ring\n",
1177 		    sc->sc_dev.dv_xname);
1178 		return (-1);
1179 	}
1180 
1181 	memset(&cmd, 0, sizeof(cmd));
1182 	cmd.chm_hdr.chm_type = VMBUS_CHANMSG_CHOPEN;
1183 	cmd.chm_openid = ch->ch_id;
1184 	cmd.chm_chanid = ch->ch_id;
1185 	cmd.chm_gpadl = ch->ch_ring_gpadl;
1186 	cmd.chm_txbr_pgcnt = ch->ch_wrd.rd_size >> PAGE_SHIFT;
1187 	cmd.chm_vcpuid = ch->ch_vcpu;
1188 
1189 	if (udata && udatalen > 0)
1190 		memcpy(cmd.chm_udata, udata, udatalen);
1191 
1192 	memset(&rsp, 0, sizeof(rsp));
1193 
1194 	ch->ch_handler = handler;
1195 	ch->ch_ctx = arg;
1196 
1197 	ch->ch_state = HV_CHANSTATE_OPENED;
1198 
1199 	rv = hv_cmd(sc, &cmd, sizeof(cmd), &rsp, sizeof(rsp),
1200 	    cold ? HCF_NOSLEEP : HCF_SLEEPOK);
1201 	if (rv) {
1202 		hv_channel_ring_destroy(ch);
1203 		DPRINTF("%s: CHOPEN failed with %d\n",
1204 		    sc->sc_dev.dv_xname, rv);
1205 		ch->ch_handler = NULL;
1206 		ch->ch_ctx = NULL;
1207 		ch->ch_state = HV_CHANSTATE_OFFERED;
1208 		return (-1);
1209 	}
1210 
1211 	return (0);
1212 }
1213 
1214 int
1215 hv_channel_close(struct hv_channel *ch)
1216 {
1217 	struct hv_softc *sc = ch->ch_sc;
1218 	struct vmbus_chanmsg_chclose cmd;
1219 	int rv;
1220 
1221 	memset(&cmd, 0, sizeof(cmd));
1222 	cmd.chm_hdr.chm_type = VMBUS_CHANMSG_CHCLOSE;
1223 	cmd.chm_chanid = ch->ch_id;
1224 
1225 	ch->ch_state = HV_CHANSTATE_CLOSING;
1226 	rv = hv_cmd(sc, &cmd, sizeof(cmd), NULL, 0, HCF_NOREPLY);
1227 	if (rv) {
1228 		DPRINTF("%s: CHCLOSE failed with %d\n",
1229 		    sc->sc_dev.dv_xname, rv);
1230 		return (-1);
1231 	}
1232 	ch->ch_state = HV_CHANSTATE_CLOSED;
1233 	hv_channel_ring_destroy(ch);
1234 	return (0);
1235 }
1236 
1237 static inline void
1238 hv_channel_setevent(struct hv_softc *sc, struct hv_channel *ch)
1239 {
1240 	struct vmbus_mon_trig *mtg;
1241 
1242 	/* Each uint32_t represents 32 channels */
1243 	set_bit(ch->ch_id, sc->sc_wevents);
1244 	if (ch->ch_flags & CHF_MONITOR) {
1245 		mtg = &sc->sc_monitor[1]->mnf_trigs[ch->ch_mgroup];
1246 		set_bit(ch->ch_mindex, &mtg->mt_pending);
1247 	} else
1248 		hv_intr_signal(sc, &ch->ch_monprm);
1249 }
1250 
1251 void
1252 hv_channel_intr(void *arg)
1253 {
1254 	struct hv_channel *ch = arg;
1255 
1256 	if (hv_channel_ready(ch))
1257 		ch->ch_handler(ch->ch_ctx);
1258 
1259 	if (hv_channel_unpause(ch) == 0)
1260 		return;
1261 
1262 	hv_channel_pause(ch);
1263 	hv_channel_schedule(ch);
1264 }
1265 
1266 int
1267 hv_channel_setdeferred(struct hv_channel *ch, const char *name)
1268 {
1269 	ch->ch_taskq = taskq_create(name, 1, IPL_NET, TASKQ_MPSAFE);
1270 	if (ch->ch_taskq == NULL)
1271 		return (-1);
1272 	task_set(&ch->ch_task, hv_channel_intr, ch);
1273 	return (0);
1274 }
1275 
1276 void
1277 hv_channel_schedule(struct hv_channel *ch)
1278 {
1279 	if (ch->ch_handler) {
1280 		if (!cold && (ch->ch_flags & CHF_BATCHED)) {
1281 			hv_channel_pause(ch);
1282 			task_add(ch->ch_taskq, &ch->ch_task);
1283 		} else
1284 			ch->ch_handler(ch->ch_ctx);
1285 	}
1286 }
1287 
1288 static inline void
1289 hv_ring_put(struct hv_ring_data *wrd, uint8_t *data, uint32_t datalen)
1290 {
1291 	int left = MIN(datalen, wrd->rd_dsize - wrd->rd_prod);
1292 
1293 	memcpy(&wrd->rd_ring->br_data[wrd->rd_prod], data, left);
1294 	memcpy(&wrd->rd_ring->br_data[0], data + left, datalen - left);
1295 	wrd->rd_prod += datalen;
1296 	if (wrd->rd_prod >= wrd->rd_dsize)
1297 		wrd->rd_prod -= wrd->rd_dsize;
1298 }
1299 
1300 static inline void
1301 hv_ring_get(struct hv_ring_data *rrd, uint8_t *data, uint32_t datalen,
1302     int peek)
1303 {
1304 	int left = MIN(datalen, rrd->rd_dsize - rrd->rd_cons);
1305 
1306 	memcpy(data, &rrd->rd_ring->br_data[rrd->rd_cons], left);
1307 	memcpy(data + left, &rrd->rd_ring->br_data[0], datalen - left);
1308 	if (!peek) {
1309 		rrd->rd_cons += datalen;
1310 		if (rrd->rd_cons >= rrd->rd_dsize)
1311 			rrd->rd_cons -= rrd->rd_dsize;
1312 	}
1313 }
1314 
1315 static inline void
1316 hv_ring_avail(struct hv_ring_data *rd, uint32_t *towrite, uint32_t *toread)
1317 {
1318 	uint32_t ridx = rd->rd_ring->br_rindex;
1319 	uint32_t widx = rd->rd_ring->br_windex;
1320 	uint32_t r, w;
1321 
1322 	if (widx >= ridx)
1323 		w = rd->rd_dsize - (widx - ridx);
1324 	else
1325 		w = ridx - widx;
1326 	r = rd->rd_dsize - w;
1327 	if (towrite)
1328 		*towrite = w;
1329 	if (toread)
1330 		*toread = r;
1331 }
1332 
1333 int
1334 hv_ring_write(struct hv_ring_data *wrd, struct iovec *iov, int iov_cnt,
1335     int *needsig)
1336 {
1337 	uint64_t indices = 0;
1338 	uint32_t avail, oprod, datalen = sizeof(indices);
1339 	int i;
1340 
1341 	for (i = 0; i < iov_cnt; i++)
1342 		datalen += iov[i].iov_len;
1343 
1344 	KASSERT(datalen <= wrd->rd_dsize);
1345 
1346 	hv_ring_avail(wrd, &avail, NULL);
1347 	if (avail <= datalen) {
1348 		DPRINTF("%s: avail %u datalen %u\n", __func__, avail, datalen);
1349 		return (EAGAIN);
1350 	}
1351 
1352 	oprod = wrd->rd_prod;
1353 
1354 	for (i = 0; i < iov_cnt; i++)
1355 		hv_ring_put(wrd, iov[i].iov_base, iov[i].iov_len);
1356 
1357 	indices = (uint64_t)oprod << 32;
1358 	hv_ring_put(wrd, (uint8_t *)&indices, sizeof(indices));
1359 
1360 	virtio_membar_sync();
1361 	wrd->rd_ring->br_windex = wrd->rd_prod;
1362 	virtio_membar_sync();
1363 
1364 	/* Signal when the ring transitions from being empty to non-empty */
1365 	if (wrd->rd_ring->br_imask == 0 &&
1366 	    wrd->rd_ring->br_rindex == oprod)
1367 		*needsig = 1;
1368 	else
1369 		*needsig = 0;
1370 
1371 	return (0);
1372 }
1373 
1374 int
1375 hv_channel_send(struct hv_channel *ch, void *data, uint32_t datalen,
1376     uint64_t rid, int type, uint32_t flags)
1377 {
1378 	struct hv_softc *sc = ch->ch_sc;
1379 	struct vmbus_chanpkt cp;
1380 	struct iovec iov[3];
1381 	uint32_t pktlen, pktlen_aligned;
1382 	uint64_t zeropad = 0;
1383 	int rv, needsig = 0;
1384 
1385 	pktlen = sizeof(cp) + datalen;
1386 	pktlen_aligned = roundup(pktlen, sizeof(uint64_t));
1387 
1388 	cp.cp_hdr.cph_type = type;
1389 	cp.cp_hdr.cph_flags = flags;
1390 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_hlen, sizeof(cp));
1391 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_tlen, pktlen_aligned);
1392 	cp.cp_hdr.cph_tid = rid;
1393 
1394 	iov[0].iov_base = &cp;
1395 	iov[0].iov_len = sizeof(cp);
1396 
1397 	iov[1].iov_base = data;
1398 	iov[1].iov_len = datalen;
1399 
1400 	iov[2].iov_base = &zeropad;
1401 	iov[2].iov_len = pktlen_aligned - pktlen;
1402 
1403 	mtx_enter(&ch->ch_wrd.rd_lock);
1404 	rv = hv_ring_write(&ch->ch_wrd, iov, 3, &needsig);
1405 	mtx_leave(&ch->ch_wrd.rd_lock);
1406 	if (rv == 0 && needsig)
1407 		hv_channel_setevent(sc, ch);
1408 
1409 	return (rv);
1410 }
1411 
1412 int
1413 hv_channel_send_sgl(struct hv_channel *ch, struct vmbus_gpa *sgl,
1414     uint32_t nsge, void *data, uint32_t datalen, uint64_t rid)
1415 {
1416 	struct hv_softc *sc = ch->ch_sc;
1417 	struct vmbus_chanpkt_sglist cp;
1418 	struct iovec iov[4];
1419 	uint32_t buflen, pktlen, pktlen_aligned;
1420 	uint64_t zeropad = 0;
1421 	int rv, needsig = 0;
1422 
1423 	buflen = sizeof(struct vmbus_gpa) * nsge;
1424 	pktlen = sizeof(cp) + datalen + buflen;
1425 	pktlen_aligned = roundup(pktlen, sizeof(uint64_t));
1426 
1427 	cp.cp_hdr.cph_type = VMBUS_CHANPKT_TYPE_GPA;
1428 	cp.cp_hdr.cph_flags = VMBUS_CHANPKT_FLAG_RC;
1429 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_hlen, sizeof(cp) + buflen);
1430 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_tlen, pktlen_aligned);
1431 	cp.cp_hdr.cph_tid = rid;
1432 	cp.cp_gpa_cnt = nsge;
1433 	cp.cp_rsvd = 0;
1434 
1435 	iov[0].iov_base = &cp;
1436 	iov[0].iov_len = sizeof(cp);
1437 
1438 	iov[1].iov_base = sgl;
1439 	iov[1].iov_len = buflen;
1440 
1441 	iov[2].iov_base = data;
1442 	iov[2].iov_len = datalen;
1443 
1444 	iov[3].iov_base = &zeropad;
1445 	iov[3].iov_len = pktlen_aligned - pktlen;
1446 
1447 	mtx_enter(&ch->ch_wrd.rd_lock);
1448 	rv = hv_ring_write(&ch->ch_wrd, iov, 4, &needsig);
1449 	mtx_leave(&ch->ch_wrd.rd_lock);
1450 	if (rv == 0 && needsig)
1451 		hv_channel_setevent(sc, ch);
1452 
1453 	return (rv);
1454 }
1455 
1456 int
1457 hv_channel_send_prpl(struct hv_channel *ch, struct vmbus_gpa_range *prpl,
1458     uint32_t nprp, void *data, uint32_t datalen, uint64_t rid)
1459 {
1460 	struct hv_softc *sc = ch->ch_sc;
1461 	struct vmbus_chanpkt_prplist cp;
1462 	struct iovec iov[4];
1463 	uint32_t buflen, pktlen, pktlen_aligned;
1464 	uint64_t zeropad = 0;
1465 	int rv, needsig = 0;
1466 
1467 	buflen = sizeof(struct vmbus_gpa_range) * (nprp + 1);
1468 	pktlen = sizeof(cp) + datalen + buflen;
1469 	pktlen_aligned = roundup(pktlen, sizeof(uint64_t));
1470 
1471 	cp.cp_hdr.cph_type = VMBUS_CHANPKT_TYPE_GPA;
1472 	cp.cp_hdr.cph_flags = VMBUS_CHANPKT_FLAG_RC;
1473 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_hlen, sizeof(cp) + buflen);
1474 	VMBUS_CHANPKT_SETLEN(cp.cp_hdr.cph_tlen, pktlen_aligned);
1475 	cp.cp_hdr.cph_tid = rid;
1476 	cp.cp_range_cnt = 1;
1477 	cp.cp_rsvd = 0;
1478 
1479 	iov[0].iov_base = &cp;
1480 	iov[0].iov_len = sizeof(cp);
1481 
1482 	iov[1].iov_base = prpl;
1483 	iov[1].iov_len = buflen;
1484 
1485 	iov[2].iov_base = data;
1486 	iov[2].iov_len = datalen;
1487 
1488 	iov[3].iov_base = &zeropad;
1489 	iov[3].iov_len = pktlen_aligned - pktlen;
1490 
1491 	mtx_enter(&ch->ch_wrd.rd_lock);
1492 	rv = hv_ring_write(&ch->ch_wrd, iov, 4, &needsig);
1493 	mtx_leave(&ch->ch_wrd.rd_lock);
1494 	if (rv == 0 && needsig)
1495 		hv_channel_setevent(sc, ch);
1496 
1497 	return (rv);
1498 }
1499 
1500 int
1501 hv_ring_peek(struct hv_ring_data *rrd, void *data, uint32_t datalen)
1502 {
1503 	uint32_t avail;
1504 
1505 	KASSERT(datalen <= rrd->rd_dsize);
1506 
1507 	hv_ring_avail(rrd, NULL, &avail);
1508 	if (avail < datalen)
1509 		return (EAGAIN);
1510 
1511 	hv_ring_get(rrd, (uint8_t *)data, datalen, 1);
1512 	return (0);
1513 }
1514 
1515 int
1516 hv_ring_read(struct hv_ring_data *rrd, void *data, uint32_t datalen,
1517     uint32_t offset)
1518 {
1519 	uint64_t indices;
1520 	uint32_t avail;
1521 
1522 	KASSERT(datalen <= rrd->rd_dsize);
1523 
1524 	hv_ring_avail(rrd, NULL, &avail);
1525 	if (avail < datalen) {
1526 		DPRINTF("%s: avail %u datalen %u\n", __func__, avail, datalen);
1527 		return (EAGAIN);
1528 	}
1529 
1530 	if (offset) {
1531 		rrd->rd_cons += offset;
1532 		if (rrd->rd_cons >= rrd->rd_dsize)
1533 			rrd->rd_cons -= rrd->rd_dsize;
1534 	}
1535 
1536 	hv_ring_get(rrd, (uint8_t *)data, datalen, 0);
1537 	hv_ring_get(rrd, (uint8_t *)&indices, sizeof(indices), 0);
1538 
1539 	virtio_membar_sync();
1540 	rrd->rd_ring->br_rindex = rrd->rd_cons;
1541 
1542 	return (0);
1543 }
1544 
1545 int
1546 hv_channel_recv(struct hv_channel *ch, void *data, uint32_t datalen,
1547     uint32_t *rlen, uint64_t *rid, int raw)
1548 {
1549 	struct vmbus_chanpkt_hdr cph;
1550 	uint32_t offset, pktlen;
1551 	int rv;
1552 
1553 	*rlen = 0;
1554 
1555 	mtx_enter(&ch->ch_rrd.rd_lock);
1556 
1557 	if ((rv = hv_ring_peek(&ch->ch_rrd, &cph, sizeof(cph))) != 0) {
1558 		mtx_leave(&ch->ch_rrd.rd_lock);
1559 		return (rv);
1560 	}
1561 
1562 	offset = raw ? 0 : VMBUS_CHANPKT_GETLEN(cph.cph_hlen);
1563 	pktlen = VMBUS_CHANPKT_GETLEN(cph.cph_tlen) - offset;
1564 	if (pktlen > datalen) {
1565 		mtx_leave(&ch->ch_rrd.rd_lock);
1566 		printf("%s: pktlen %u datalen %u\n", __func__, pktlen, datalen);
1567 		return (EINVAL);
1568 	}
1569 
1570 	rv = hv_ring_read(&ch->ch_rrd, data, pktlen, offset);
1571 	if (rv == 0) {
1572 		*rlen = pktlen;
1573 		*rid = cph.cph_tid;
1574 	}
1575 
1576 	mtx_leave(&ch->ch_rrd.rd_lock);
1577 
1578 	return (rv);
1579 }
1580 
1581 static inline void
1582 hv_ring_mask(struct hv_ring_data *rd)
1583 {
1584 	virtio_membar_sync();
1585 	rd->rd_ring->br_imask = 1;
1586 	virtio_membar_sync();
1587 }
1588 
1589 static inline void
1590 hv_ring_unmask(struct hv_ring_data *rd)
1591 {
1592 	virtio_membar_sync();
1593 	rd->rd_ring->br_imask = 0;
1594 	virtio_membar_sync();
1595 }
1596 
1597 void
1598 hv_channel_pause(struct hv_channel *ch)
1599 {
1600 	hv_ring_mask(&ch->ch_rrd);
1601 }
1602 
1603 uint
1604 hv_channel_unpause(struct hv_channel *ch)
1605 {
1606 	uint32_t avail;
1607 
1608 	hv_ring_unmask(&ch->ch_rrd);
1609 	hv_ring_avail(&ch->ch_rrd, NULL, &avail);
1610 
1611 	return (avail);
1612 }
1613 
1614 uint
1615 hv_channel_ready(struct hv_channel *ch)
1616 {
1617 	uint32_t avail;
1618 
1619 	hv_ring_avail(&ch->ch_rrd, NULL, &avail);
1620 
1621 	return (avail);
1622 }
1623 
1624 /* How many PFNs can be referenced by the header */
1625 #define HV_NPFNHDR	((VMBUS_MSG_DSIZE_MAX -	\
1626 	  sizeof(struct vmbus_chanmsg_gpadl_conn)) / sizeof(uint64_t))
1627 
1628 /* How many PFNs can be referenced by the body */
1629 #define HV_NPFNBODY	((VMBUS_MSG_DSIZE_MAX -	\
1630 	  sizeof(struct vmbus_chanmsg_gpadl_subconn)) / sizeof(uint64_t))
1631 
1632 int
1633 hv_handle_alloc(struct hv_channel *ch, void *buffer, uint32_t buflen,
1634     uint32_t *handle)
1635 {
1636 	struct hv_softc *sc = ch->ch_sc;
1637 	struct vmbus_chanmsg_gpadl_conn *hdr;
1638 	struct vmbus_chanmsg_gpadl_subconn *cmd;
1639 	struct vmbus_chanmsg_gpadl_connresp rsp;
1640 	struct hv_msg *msg;
1641 	int i, j, last, left, rv;
1642 	int bodylen = 0, ncmds = 0, pfn = 0;
1643 	int waitflag = cold ? M_NOWAIT : M_WAITOK;
1644 	uint64_t *frames;
1645 	paddr_t pa;
1646 	caddr_t body;
1647 	/* Total number of pages to reference */
1648 	int total = atop(buflen);
1649 	/* Number of pages that will fit the header */
1650 	int inhdr = MIN(total, HV_NPFNHDR);
1651 
1652 	KASSERT((buflen & (PAGE_SIZE - 1)) == 0);
1653 
1654 	if ((msg = malloc(sizeof(*msg), M_DEVBUF, M_ZERO | waitflag)) == NULL)
1655 		return (ENOMEM);
1656 
1657 	/* Prepare array of frame addresses */
1658 	if ((frames = mallocarray(total, sizeof(*frames), M_DEVBUF, M_ZERO |
1659 	    waitflag)) == NULL) {
1660 		free(msg, M_DEVBUF, sizeof(*msg));
1661 		return (ENOMEM);
1662 	}
1663 	for (i = 0; i < total; i++) {
1664 		if (!pmap_extract(pmap_kernel(), (vaddr_t)buffer +
1665 		    PAGE_SIZE * i, &pa)) {
1666 			free(msg, M_DEVBUF, sizeof(*msg));
1667 			free(frames, M_DEVBUF, total * sizeof(*frames));
1668 			return (EFAULT);
1669 		}
1670 		frames[i] = atop(pa);
1671 	}
1672 
1673 	msg->msg_req.hc_dsize = sizeof(struct vmbus_chanmsg_gpadl_conn) +
1674 	    inhdr * sizeof(uint64_t);
1675 	hdr = (struct vmbus_chanmsg_gpadl_conn *)msg->msg_req.hc_data;
1676 	msg->msg_rsp = &rsp;
1677 	msg->msg_rsplen = sizeof(rsp);
1678 	if (waitflag == M_NOWAIT)
1679 		msg->msg_flags = MSGF_NOSLEEP;
1680 
1681 	left = total - inhdr;
1682 
1683 	/* Allocate additional gpadl_body structures if required */
1684 	if (left > 0) {
1685 		ncmds = MAX(1, left / HV_NPFNBODY + left % HV_NPFNBODY);
1686 		bodylen = ncmds * VMBUS_MSG_DSIZE_MAX;
1687 		body = malloc(bodylen, M_DEVBUF, M_ZERO | waitflag);
1688 		if (body == NULL) {
1689 			free(msg, M_DEVBUF, sizeof(*msg));
1690 			free(frames, M_DEVBUF, atop(buflen) * sizeof(*frames));
1691 			return (ENOMEM);
1692 		}
1693 	}
1694 
1695 	*handle = atomic_inc_int_nv(&sc->sc_handle);
1696 
1697 	hdr->chm_hdr.chm_type = VMBUS_CHANMSG_GPADL_CONN;
1698 	hdr->chm_chanid = ch->ch_id;
1699 	hdr->chm_gpadl = *handle;
1700 
1701 	/* Single range for a contiguous buffer */
1702 	hdr->chm_range_cnt = 1;
1703 	hdr->chm_range_len = sizeof(struct vmbus_gpa_range) + total *
1704 	    sizeof(uint64_t);
1705 	hdr->chm_range.gpa_ofs = 0;
1706 	hdr->chm_range.gpa_len = buflen;
1707 
1708 	/* Fit as many pages as possible into the header */
1709 	for (i = 0; i < inhdr; i++)
1710 		hdr->chm_range.gpa_page[i] = frames[pfn++];
1711 
1712 	for (i = 0; i < ncmds; i++) {
1713 		cmd = (struct vmbus_chanmsg_gpadl_subconn *)(body +
1714 		    VMBUS_MSG_DSIZE_MAX * i);
1715 		cmd->chm_hdr.chm_type = VMBUS_CHANMSG_GPADL_SUBCONN;
1716 		cmd->chm_gpadl = *handle;
1717 		last = MIN(left, HV_NPFNBODY);
1718 		for (j = 0; j < last; j++)
1719 			cmd->chm_gpa_page[j] = frames[pfn++];
1720 		left -= last;
1721 	}
1722 
1723 	rv = hv_start(sc, msg);
1724 	if (rv != 0) {
1725 		DPRINTF("%s: GPADL_CONN failed\n", sc->sc_dev.dv_xname);
1726 		goto out;
1727 	}
1728 	for (i = 0; i < ncmds; i++) {
1729 		int cmdlen = sizeof(*cmd);
1730 		cmd = (struct vmbus_chanmsg_gpadl_subconn *)(body +
1731 		    VMBUS_MSG_DSIZE_MAX * i);
1732 		/* Last element can be short */
1733 		if (i == ncmds - 1)
1734 			cmdlen += last * sizeof(uint64_t);
1735 		else
1736 			cmdlen += HV_NPFNBODY * sizeof(uint64_t);
1737 		rv = hv_cmd(sc, cmd, cmdlen, NULL, 0, waitflag | HCF_NOREPLY);
1738 		if (rv != 0) {
1739 			DPRINTF("%s: GPADL_SUBCONN (iteration %d/%d) failed "
1740 			    "with %d\n", sc->sc_dev.dv_xname, i, ncmds, rv);
1741 			goto out;
1742 		}
1743 	}
1744 	rv = hv_reply(sc, msg);
1745 	if (rv != 0)
1746 		DPRINTF("%s: GPADL allocation failed with %d\n",
1747 		    sc->sc_dev.dv_xname, rv);
1748 
1749  out:
1750 	free(msg, M_DEVBUF, sizeof(*msg));
1751 	free(frames, M_DEVBUF, total * sizeof(*frames));
1752 	if (bodylen > 0)
1753 		free(body, M_DEVBUF, bodylen);
1754 	if (rv != 0)
1755 		return (rv);
1756 
1757 	KASSERT(*handle == rsp.chm_gpadl);
1758 
1759 	return (0);
1760 }
1761 
1762 void
1763 hv_handle_free(struct hv_channel *ch, uint32_t handle)
1764 {
1765 	struct hv_softc *sc = ch->ch_sc;
1766 	struct vmbus_chanmsg_gpadl_disconn cmd;
1767 	struct vmbus_chanmsg_gpadl_disconn rsp;
1768 	int rv;
1769 
1770 	memset(&cmd, 0, sizeof(cmd));
1771 	cmd.chm_hdr.chm_type = VMBUS_CHANMSG_GPADL_DISCONN;
1772 	cmd.chm_chanid = ch->ch_id;
1773 	cmd.chm_gpadl = handle;
1774 
1775 	rv = hv_cmd(sc, &cmd, sizeof(cmd), &rsp, sizeof(rsp), cold ?
1776 	    HCF_NOSLEEP : 0);
1777 	if (rv)
1778 		DPRINTF("%s: GPADL_DISCONN failed with %d\n",
1779 		    sc->sc_dev.dv_xname, rv);
1780 }
1781 
1782 static int
1783 hv_attach_print(void *aux, const char *name)
1784 {
1785 	struct hv_attach_args *aa = aux;
1786 
1787 	if (name)
1788 		printf("\"%s\" at %s", aa->aa_ident, name);
1789 
1790 	return (UNCONF);
1791 }
1792 
1793 int
1794 hv_attach_devices(struct hv_softc *sc)
1795 {
1796 	struct hv_dev *dv;
1797 	struct hv_channel *ch;
1798 
1799 	SLIST_INIT(&sc->sc_devs);
1800 	mtx_init(&sc->sc_devlck, IPL_NET);
1801 
1802 	TAILQ_FOREACH(ch, &sc->sc_channels, ch_entry) {
1803 		if (ch->ch_state != HV_CHANSTATE_OFFERED)
1804 			continue;
1805 		if (!(ch->ch_flags & CHF_MONITOR))
1806 			continue;
1807 		dv = malloc(sizeof(*dv), M_DEVBUF, M_ZERO | M_NOWAIT);
1808 		if (dv == NULL) {
1809 			printf("%s: failed to allocate device object\n",
1810 			    sc->sc_dev.dv_xname);
1811 			return (-1);
1812 		}
1813 		dv->dv_aa.aa_parent = sc;
1814 		dv->dv_aa.aa_type = &ch->ch_type;
1815 		dv->dv_aa.aa_inst = &ch->ch_inst;
1816 		dv->dv_aa.aa_ident = ch->ch_ident;
1817 		dv->dv_aa.aa_chan = ch;
1818 		dv->dv_aa.aa_dmat = sc->sc_dmat;
1819 		mtx_enter(&sc->sc_devlck);
1820 		SLIST_INSERT_HEAD(&sc->sc_devs, dv, dv_entry);
1821 		mtx_leave(&sc->sc_devlck);
1822 		config_found((struct device *)sc, &dv->dv_aa, hv_attach_print);
1823 	}
1824 	return (0);
1825 }
1826 
1827 void
1828 hv_evcount_attach(struct hv_channel *ch, const char *name)
1829 {
1830 	struct hv_softc *sc = ch->ch_sc;
1831 
1832 	evcount_attach(&ch->ch_evcnt, name, &sc->sc_idtvec);
1833 }
1834