xref: /dragonfly/sys/dev/virtual/hyperv/vmbus/vmbus.c (revision 2b7dbe20)
1 /*-
2  * Copyright (c) 2009-2012,2016 Microsoft Corp.
3  * Copyright (c) 2012 NetApp Inc.
4  * Copyright (c) 2012 Citrix Inc.
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #include "opt_acpi.h"
30 
31 #include <sys/param.h>
32 #include <sys/bus.h>
33 #include <sys/kernel.h>
34 #include <sys/malloc.h>
35 #include <sys/module.h>
36 #include <sys/rman.h>
37 #include <sys/systimer.h>
38 #include <sys/thread.h>
39 #include <sys/thread2.h>
40 
41 #include <machine/intr_machdep.h>
42 #include <machine/smp.h>
43 
44 #include <dev/virtual/hyperv/hyperv_busdma.h>
45 #include <dev/virtual/hyperv/hyperv_machdep.h>
46 #include <dev/virtual/hyperv/hyperv_reg.h>
47 #include <dev/virtual/hyperv/hyperv_var.h>
48 #include <dev/virtual/hyperv/vmbus/vmbus_reg.h>
49 #include <dev/virtual/hyperv/vmbus/vmbus_var.h>
50 
51 #include "acpi.h"
52 #include "acpi_if.h"
53 #include "pcib_if.h"
54 
55 #define MSR_HV_STIMER0_CFG_SINT		\
56 	((((uint64_t)VMBUS_SINT_TIMER) << MSR_HV_STIMER_CFG_SINT_SHIFT) & \
57 	 MSR_HV_STIMER_CFG_SINT_MASK)
58 
59 /*
60  * Additionally required feature:
61  * - SynIC is needed for interrupt generation.
62  */
63 #define CPUID_HV_TIMER_MASK		(CPUID_HV_MSR_SYNIC |		\
64 					 CPUID_HV_MSR_SYNTIMER)
65 
66 /*
67  * NOTE: DO NOT CHANGE THIS.
68  */
69 #define VMBUS_SINT_MESSAGE		2
70 /*
71  * NOTE:
72  * - DO NOT set it to the same value as VMBUS_SINT_MESSAGE.
73  * - DO NOT set it to 0.
74  */
75 #define VMBUS_SINT_TIMER		4
76 
77 /*
78  * NOTE: DO NOT CHANGE THESE
79  */
80 #define VMBUS_CONNID_MESSAGE		1
81 #define VMBUS_CONNID_EVENT		2
82 
83 struct vmbus_msghc {
84 	struct hypercall_postmsg_in	*mh_inprm;
85 	struct hypercall_postmsg_in	mh_inprm_save;
86 	struct hyperv_dma		mh_inprm_dma;
87 
88 	struct vmbus_message		*mh_resp;
89 	struct vmbus_message		mh_resp0;
90 };
91 
92 struct vmbus_msghc_ctx {
93 	struct vmbus_msghc		*mhc_free;
94 	struct lwkt_token		mhc_free_token;
95 	uint32_t			mhc_flags;
96 
97 	struct vmbus_msghc		*mhc_active;
98 	struct lwkt_token		mhc_active_token;
99 };
100 
101 #define VMBUS_MSGHC_CTXF_DESTROY	0x0001
102 
103 static int			vmbus_probe(device_t);
104 static int			vmbus_attach(device_t);
105 static int			vmbus_detach(device_t);
106 static void			vmbus_intr(void *);
107 static void			vmbus_timer_intr_reload(struct cputimer_intr *,
108 				    sysclock_t);
109 static void			vmbus_timer_intr_pcpuhand(
110 				    struct cputimer_intr *);
111 static void			vmbus_timer_intr_restart(
112 				    struct cputimer_intr *);
113 
114 static int			vmbus_dma_alloc(struct vmbus_softc *);
115 static void			vmbus_dma_free(struct vmbus_softc *);
116 static int			vmbus_intr_setup(struct vmbus_softc *);
117 static void			vmbus_intr_teardown(struct vmbus_softc *);
118 static void			vmbus_synic_setup(void *);
119 static void			vmbus_synic_teardown(void *);
120 static void			vmbus_timer_stop(void *);
121 static void			vmbus_timer_config(void *);
122 static int			vmbus_init(struct vmbus_softc *);
123 static int			vmbus_init_contact(struct vmbus_softc *,
124 				    uint32_t);
125 static void			vmbus_timer_restart(void *);
126 static void			vmbus_timer_msgintr(struct vmbus_pcpu_data *);
127 
128 static void			vmbus_chan_msgproc(struct vmbus_softc *,
129 				    const struct vmbus_message *);
130 
131 static struct vmbus_msghc_ctx	*vmbus_msghc_ctx_create(bus_dma_tag_t);
132 static void			vmbus_msghc_ctx_destroy(
133 				    struct vmbus_msghc_ctx *);
134 static void			vmbus_msghc_ctx_free(struct vmbus_msghc_ctx *);
135 static struct vmbus_msghc	*vmbus_msghc_alloc(bus_dma_tag_t);
136 static void			vmbus_msghc_free(struct vmbus_msghc *);
137 static struct vmbus_msghc	*vmbus_msghc_get1(struct vmbus_msghc_ctx *,
138 				    uint32_t);
139 
140 static device_method_t vmbus_methods[] = {
141 	/* Device interface */
142 	DEVMETHOD(device_probe,			vmbus_probe),
143 	DEVMETHOD(device_attach,		vmbus_attach),
144 	DEVMETHOD(device_detach,		vmbus_detach),
145 	DEVMETHOD(device_shutdown,		bus_generic_shutdown),
146 	DEVMETHOD(device_suspend,		bus_generic_suspend),
147 	DEVMETHOD(device_resume,		bus_generic_resume),
148 
149 	DEVMETHOD_END
150 };
151 
152 static driver_t vmbus_driver = {
153 	"vmbus",
154 	vmbus_methods,
155 	sizeof(struct vmbus_softc)
156 };
157 
158 static devclass_t vmbus_devclass;
159 
160 DRIVER_MODULE(vmbus, acpi, vmbus_driver, vmbus_devclass, NULL, NULL);
161 MODULE_DEPEND(vmbus, acpi, 1, 1, 1);
162 MODULE_VERSION(vmbus, 1);
163 
164 static struct cputimer_intr vmbus_cputimer_intr = {
165 	.freq = HYPERV_TIMER_FREQ,
166 	.reload = vmbus_timer_intr_reload,
167 	.enable = cputimer_intr_default_enable,
168 	.config = cputimer_intr_default_config,
169 	.restart = vmbus_timer_intr_restart,
170 	.pmfixup = cputimer_intr_default_pmfixup,
171 	.initclock = cputimer_intr_default_initclock,
172 	.pcpuhand = vmbus_timer_intr_pcpuhand,
173 	.next = SLIST_ENTRY_INITIALIZER,
174 	.name = "hyperv",
175 	.type = CPUTIMER_INTR_VMM,
176 	.prio = CPUTIMER_INTR_PRIO_VMM,
177 	.caps = CPUTIMER_INTR_CAP_PS,
178 	.priv = NULL
179 };
180 
181 static const uint32_t	vmbus_version[] = {
182 	VMBUS_VERSION_WIN8_1,
183 	VMBUS_VERSION_WIN8,
184 	VMBUS_VERSION_WIN7,
185 	VMBUS_VERSION_WS2008
186 };
187 
188 static int		vmbus_timer_intr_enable = 1;
189 TUNABLE_INT("hw.vmbus.timer_intr.enable", &vmbus_timer_intr_enable);
190 
191 static int
192 vmbus_probe(device_t dev)
193 {
194 	char *id[] = { "VMBUS", NULL };
195 
196 	if (ACPI_ID_PROBE(device_get_parent(dev), dev, id) == NULL ||
197 	    device_get_unit(dev) != 0 || vmm_guest != VMM_GUEST_HYPERV ||
198 	    (hyperv_features & CPUID_HV_MSR_SYNIC) == 0)
199 		return (ENXIO);
200 
201 	device_set_desc(dev, "Hyper-V vmbus");
202 
203 	return (0);
204 }
205 
206 static int
207 vmbus_attach(device_t dev)
208 {
209 	struct vmbus_softc *sc = device_get_softc(dev);
210 	int error, cpu, use_timer;
211 
212 	/*
213 	 * Basic setup.
214 	 */
215 	sc->vmbus_dev = dev;
216 	for (cpu = 0; cpu < ncpus; ++cpu) {
217 		struct vmbus_pcpu_data *psc = VMBUS_PCPU(sc, cpu);
218 
219 		psc->sc = sc;
220 		psc->cpuid = cpu;
221 		psc->timer_last = UINT64_MAX;
222 	}
223 
224 	/*
225 	 * Should we use interrupt timer?
226 	 */
227 	use_timer = 0;
228 	if (device_get_unit(dev) == 0 &&
229 	    (hyperv_features & CPUID_HV_TIMER_MASK) == CPUID_HV_TIMER_MASK &&
230 	    hyperv_tc64 != NULL)
231 		use_timer = 1;
232 
233 	/*
234 	 * Create context for "post message" Hypercalls
235 	 */
236 	sc->vmbus_msg_hc = vmbus_msghc_ctx_create(
237 	    bus_get_dma_tag(sc->vmbus_dev));
238 	if (sc->vmbus_msg_hc == NULL)
239 		return ENXIO;
240 
241 	/*
242 	 * Allocate DMA stuffs.
243 	 */
244 	error = vmbus_dma_alloc(sc);
245 	if (error)
246 		goto failed;
247 
248 	/*
249 	 * Setup interrupt.
250 	 */
251 	error = vmbus_intr_setup(sc);
252 	if (error)
253 		goto failed;
254 
255 	if (use_timer) {
256 		/*
257 		 * Make sure that interrupt timer is stopped.
258 		 */
259 		lwkt_cpusync_simple(smp_active_mask, vmbus_timer_stop, sc);
260 	}
261 
262 	/*
263 	 * Setup SynIC.
264 	 */
265 	lwkt_cpusync_simple(smp_active_mask, vmbus_synic_setup, sc);
266 	sc->vmbus_flags |= VMBUS_FLAG_SYNIC;
267 
268 	/*
269 	 * Initialize vmbus.
270 	 */
271 	error = vmbus_init(sc);
272 	if (error)
273 		goto failed;
274 
275 	if (use_timer) {
276 		/*
277 		 * Configure and register vmbus interrupt timer.
278 		 */
279 		lwkt_cpusync_simple(smp_active_mask, vmbus_timer_config, sc);
280 		vmbus_cputimer_intr.priv = sc;
281 		cputimer_intr_register(&vmbus_cputimer_intr);
282 		if (vmbus_timer_intr_enable)
283 			cputimer_intr_select(&vmbus_cputimer_intr, 0);
284 	}
285 
286 	return 0;
287 failed:
288 	vmbus_detach(dev);
289 	return error;
290 }
291 
292 static int
293 vmbus_detach(device_t dev)
294 {
295 	struct vmbus_softc *sc = device_get_softc(dev);
296 
297 	/* TODO: uninitialize vmbus. */
298 	/* TODO: stop and deregister timer */
299 
300 	if (sc->vmbus_flags & VMBUS_FLAG_SYNIC)
301 		lwkt_cpusync_simple(smp_active_mask, vmbus_synic_teardown, sc);
302 	vmbus_intr_teardown(sc);
303 	vmbus_dma_free(sc);
304 
305 	if (sc->vmbus_msg_hc != NULL) {
306 		vmbus_msghc_ctx_destroy(sc->vmbus_msg_hc);
307 		sc->vmbus_msg_hc = NULL;
308 	}
309 	return (0);
310 }
311 
312 static __inline void
313 vmbus_msg_reset(volatile struct vmbus_message *msg)
314 {
315 	msg->msg_type = HYPERV_MSGTYPE_NONE;
316 	/*
317 	 * Make sure that the write to msg_type (i.e. set to
318 	 * HYPERV_MSGTYPE_NONE) happens before we read the
319 	 * msg_flags and send EOM to the hypervisor.
320 	 */
321 	cpu_mfence();
322 	if (msg->msg_flags & VMBUS_MSGFLAG_PENDING) {
323 		/*
324 		 * Ask the hypervisor to rescan message queue,
325 		 * and deliver new message if any.
326 		 */
327 		wrmsr(MSR_HV_EOM, 0);
328 	}
329 }
330 
331 static void
332 vmbus_intr(void *xpsc)
333 {
334 	struct vmbus_pcpu_data *psc = xpsc;
335 	volatile struct vmbus_message *msg;
336 
337 	msg = psc->message + VMBUS_SINT_MESSAGE;
338 	while (__predict_false(msg->msg_type != HYPERV_MSGTYPE_NONE)) {
339 		if (msg->msg_type == HYPERV_MSGTYPE_CHANNEL) {
340 			/* Channel message */
341 			vmbus_chan_msgproc(psc->sc,
342 			    __DEVOLATILE(const struct vmbus_message *, msg));
343 		}
344 		vmbus_msg_reset(msg);
345 	}
346 }
347 
348 static __inline void
349 vmbus_timer_oneshot(struct vmbus_pcpu_data *psc, uint64_t current)
350 {
351 	psc->timer_last = current;
352 	wrmsr(MSR_HV_STIMER0_COUNT, current);
353 }
354 
355 static void
356 vmbus_timer_intr_reload(struct cputimer_intr *cti, sysclock_t reload)
357 {
358 	struct globaldata *gd = mycpu;
359 	struct vmbus_softc *sc = cti->priv;
360 	struct vmbus_pcpu_data *psc = VMBUS_PCPU(sc, gd->gd_cpuid);
361 	uint64_t current;
362 
363 	if ((ssysclock_t)reload < 0)		/* neg value */
364 		reload = 1;
365 	reload = muldivu64(reload, cti->freq, sys_cputimer->freq);
366 	current = hyperv_tc64() + reload;
367 
368 	if (gd->gd_timer_running) {
369 		if (current < psc->timer_last)
370 			vmbus_timer_oneshot(psc, current);
371 	} else {
372 		gd->gd_timer_running = 1;
373 		vmbus_timer_oneshot(psc, current);
374 	}
375 }
376 
377 static void
378 vmbus_timer_intr_pcpuhand(struct cputimer_intr *cti)
379 {
380 	struct vmbus_softc *sc = cti->priv;
381 	struct vmbus_pcpu_data *psc = VMBUS_PCPU(sc, mycpuid);
382 
383 	vmbus_timer_msgintr(psc);
384 }
385 
386 static void
387 vmbus_timer_intr_restart(struct cputimer_intr *cti)
388 {
389 	lwkt_send_ipiq_mask(smp_active_mask, vmbus_timer_restart, cti->priv);
390 }
391 
392 static struct vmbus_msghc *
393 vmbus_msghc_alloc(bus_dma_tag_t parent_dtag)
394 {
395 	struct vmbus_msghc *mh;
396 
397 	mh = kmalloc(sizeof(*mh), M_DEVBUF, M_WAITOK | M_ZERO);
398 
399 	mh->mh_inprm = hyperv_dmamem_alloc(parent_dtag,
400 	    HYPERCALL_POSTMSGIN_ALIGN, 0, HYPERCALL_POSTMSGIN_SIZE,
401 	    &mh->mh_inprm_dma, BUS_DMA_WAITOK);
402 	if (mh->mh_inprm == NULL) {
403 		kfree(mh, M_DEVBUF);
404 		return NULL;
405 	}
406 	return mh;
407 }
408 
409 static void
410 vmbus_msghc_free(struct vmbus_msghc *mh)
411 {
412 	hyperv_dmamem_free(&mh->mh_inprm_dma, mh->mh_inprm);
413 	kfree(mh, M_DEVBUF);
414 }
415 
416 static void
417 vmbus_msghc_ctx_free(struct vmbus_msghc_ctx *mhc)
418 {
419 	KASSERT(mhc->mhc_active == NULL, ("still have active msg hypercall"));
420 	KASSERT(mhc->mhc_free == NULL, ("still have hypercall msg"));
421 
422 	lwkt_token_uninit(&mhc->mhc_free_token);
423 	lwkt_token_uninit(&mhc->mhc_active_token);
424 	kfree(mhc, M_DEVBUF);
425 }
426 
427 static struct vmbus_msghc_ctx *
428 vmbus_msghc_ctx_create(bus_dma_tag_t parent_dtag)
429 {
430 	struct vmbus_msghc_ctx *mhc;
431 
432 	mhc = kmalloc(sizeof(*mhc), M_DEVBUF, M_WAITOK | M_ZERO);
433 	lwkt_token_init(&mhc->mhc_free_token, "msghcf");
434 	lwkt_token_init(&mhc->mhc_active_token, "msghca");
435 
436 	mhc->mhc_free = vmbus_msghc_alloc(parent_dtag);
437 	if (mhc->mhc_free == NULL) {
438 		vmbus_msghc_ctx_free(mhc);
439 		return NULL;
440 	}
441 	return mhc;
442 }
443 
444 static struct vmbus_msghc *
445 vmbus_msghc_get1(struct vmbus_msghc_ctx *mhc, uint32_t dtor_flag)
446 {
447 	struct vmbus_msghc *mh;
448 
449 	lwkt_gettoken(&mhc->mhc_free_token);
450 
451 	while ((mhc->mhc_flags & dtor_flag) == 0 && mhc->mhc_free == NULL)
452 		tsleep(&mhc->mhc_free, 0, "gmsghc", 0);
453 	if (mhc->mhc_flags & dtor_flag) {
454 		/* Being destroyed */
455 		mh = NULL;
456 	} else {
457 		mh = mhc->mhc_free;
458 		KASSERT(mh != NULL, ("no free hypercall msg"));
459 		KASSERT(mh->mh_resp == NULL,
460 		    ("hypercall msg has pending response"));
461 		mhc->mhc_free = NULL;
462 	}
463 
464 	lwkt_reltoken(&mhc->mhc_free_token);
465 
466 	return mh;
467 }
468 
469 struct vmbus_msghc *
470 vmbus_msghc_get(struct vmbus_softc *sc, size_t dsize)
471 {
472 	struct hypercall_postmsg_in *inprm;
473 	struct vmbus_msghc *mh;
474 
475 	if (dsize > HYPERCALL_POSTMSGIN_DSIZE_MAX)
476 		return NULL;
477 
478 	mh = vmbus_msghc_get1(sc->vmbus_msg_hc, VMBUS_MSGHC_CTXF_DESTROY);
479 	if (mh == NULL)
480 		return NULL;
481 
482 	inprm = mh->mh_inprm;
483 	memset(inprm, 0, HYPERCALL_POSTMSGIN_SIZE);
484 	inprm->hc_connid = VMBUS_CONNID_MESSAGE;
485 	inprm->hc_msgtype = HYPERV_MSGTYPE_CHANNEL;
486 	inprm->hc_dsize = dsize;
487 
488 	return mh;
489 }
490 
491 void
492 vmbus_msghc_put(struct vmbus_softc *sc, struct vmbus_msghc *mh)
493 {
494 	struct vmbus_msghc_ctx *mhc = sc->vmbus_msg_hc;
495 
496 	KASSERT(mhc->mhc_active == NULL, ("msg hypercall is active"));
497 	mh->mh_resp = NULL;
498 
499 	lwkt_gettoken(&mhc->mhc_free_token);
500 	KASSERT(mhc->mhc_free == NULL, ("has free hypercall msg"));
501 	mhc->mhc_free = mh;
502 	lwkt_reltoken(&mhc->mhc_free_token);
503 	wakeup(&mhc->mhc_free);
504 }
505 
506 void *
507 vmbus_msghc_dataptr(struct vmbus_msghc *mh)
508 {
509 	return mh->mh_inprm->hc_data;
510 }
511 
512 static void
513 vmbus_msghc_ctx_destroy(struct vmbus_msghc_ctx *mhc)
514 {
515 	struct vmbus_msghc *mh;
516 
517 	lwkt_gettoken(&mhc->mhc_free_token);
518 	mhc->mhc_flags |= VMBUS_MSGHC_CTXF_DESTROY;
519 	lwkt_reltoken(&mhc->mhc_free_token);
520 	wakeup(&mhc->mhc_free);
521 
522 	mh = vmbus_msghc_get1(mhc, 0);
523 	if (mh == NULL)
524 		panic("can't get msghc");
525 
526 	vmbus_msghc_free(mh);
527 	vmbus_msghc_ctx_free(mhc);
528 }
529 
530 int
531 vmbus_msghc_exec_noresult(struct vmbus_msghc *mh)
532 {
533 	int i, wait_ticks = 1;
534 
535 	/*
536 	 * Save the input parameter so that we could restore the input
537 	 * parameter if the Hypercall failed.
538 	 *
539 	 * XXX
540 	 * Is this really necessary?!  i.e. Will the Hypercall ever
541 	 * overwrite the input parameter?
542 	 */
543 	memcpy(&mh->mh_inprm_save, mh->mh_inprm, HYPERCALL_POSTMSGIN_SIZE);
544 
545 	/*
546 	 * In order to cope with transient failures, e.g. insufficient
547 	 * resources on host side, we retry the post message Hypercall
548 	 * several times.  20 retries seem sufficient.
549 	 */
550 #define HC_RETRY_MAX	20
551 
552 	for (i = 0; i < HC_RETRY_MAX; ++i) {
553 		uint64_t status;
554 
555 		status = hypercall_post_message(mh->mh_inprm_dma.hv_paddr);
556 		if (status == HYPERCALL_STATUS_SUCCESS)
557 			return 0;
558 
559 		tsleep(&status, 0, "hcpmsg", wait_ticks);
560 		if (wait_ticks < hz)
561 			wait_ticks *= 2;
562 
563 		/* Restore input parameter and try again */
564 		memcpy(mh->mh_inprm, &mh->mh_inprm_save,
565 		    HYPERCALL_POSTMSGIN_SIZE);
566 	}
567 
568 #undef HC_RETRY_MAX
569 
570 	return EIO;
571 }
572 
573 int
574 vmbus_msghc_exec(struct vmbus_softc *sc, struct vmbus_msghc *mh)
575 {
576 	struct vmbus_msghc_ctx *mhc = sc->vmbus_msg_hc;
577 	int error;
578 
579 	KASSERT(mh->mh_resp == NULL, ("hypercall msg has pending response"));
580 
581 	lwkt_gettoken(&mhc->mhc_active_token);
582 	KASSERT(mhc->mhc_active == NULL, ("pending active msg hypercall"));
583 	mhc->mhc_active = mh;
584 	lwkt_reltoken(&mhc->mhc_active_token);
585 
586 	error = vmbus_msghc_exec_noresult(mh);
587 	if (error) {
588 		lwkt_gettoken(&mhc->mhc_active_token);
589 		KASSERT(mhc->mhc_active == mh, ("msghc mismatch"));
590 		mhc->mhc_active = NULL;
591 		lwkt_reltoken(&mhc->mhc_active_token);
592 	}
593 	return error;
594 }
595 
596 const struct vmbus_message *
597 vmbus_msghc_wait_result(struct vmbus_softc *sc, struct vmbus_msghc *mh)
598 {
599 	struct vmbus_msghc_ctx *mhc = sc->vmbus_msg_hc;
600 
601 	lwkt_gettoken(&mhc->mhc_active_token);
602 
603 	KASSERT(mhc->mhc_active == mh, ("msghc mismatch"));
604 	while (mh->mh_resp == NULL)
605 		tsleep(&mhc->mhc_active, 0, "wmsghc", 0);
606 	mhc->mhc_active = NULL;
607 
608 	lwkt_reltoken(&mhc->mhc_active_token);
609 
610 	return mh->mh_resp;
611 }
612 
613 void
614 vmbus_msghc_wakeup(struct vmbus_softc *sc, const struct vmbus_message *msg)
615 {
616 	struct vmbus_msghc_ctx *mhc = sc->vmbus_msg_hc;
617 	struct vmbus_msghc *mh;
618 
619 	lwkt_gettoken(&mhc->mhc_active_token);
620 
621 	mh = mhc->mhc_active;
622 	KASSERT(mh != NULL, ("no pending msg hypercall"));
623 	memcpy(&mh->mh_resp0, msg, sizeof(mh->mh_resp0));
624 	mh->mh_resp = &mh->mh_resp0;
625 
626 	lwkt_reltoken(&mhc->mhc_active_token);
627 	wakeup(&mhc->mhc_active);
628 }
629 
630 static int
631 vmbus_dma_alloc(struct vmbus_softc *sc)
632 {
633 	bus_dma_tag_t parent_dtag;
634 	uint8_t *evtflags;
635 	int cpu;
636 
637 	parent_dtag = bus_get_dma_tag(sc->vmbus_dev);
638 	for (cpu = 0; cpu < ncpus; ++cpu) {
639 		struct vmbus_pcpu_data *psc = VMBUS_PCPU(sc, cpu);
640 
641 		/*
642 		 * Per-cpu messages and event flags.
643 		 */
644 		psc->message = hyperv_dmamem_alloc(parent_dtag,
645 		    PAGE_SIZE, 0, PAGE_SIZE, &psc->message_dma,
646 		    BUS_DMA_WAITOK | BUS_DMA_ZERO);
647 		if (psc->message == NULL)
648 			return ENOMEM;
649 
650 		psc->event_flags = hyperv_dmamem_alloc(parent_dtag,
651 		    PAGE_SIZE, 0, PAGE_SIZE, &psc->event_flags_dma,
652 		    BUS_DMA_WAITOK | BUS_DMA_ZERO);
653 		if (psc->event_flags == NULL)
654 			return ENOMEM;
655 	}
656 
657 	evtflags = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0,
658 	    PAGE_SIZE, &sc->vmbus_evtflags_dma, BUS_DMA_WAITOK | BUS_DMA_ZERO);
659 	if (evtflags == NULL)
660 		return ENOMEM;
661 	sc->vmbus_rx_evtflags = (u_long *)evtflags;
662 	sc->vmbus_tx_evtflags = (u_long *)(evtflags + (PAGE_SIZE / 2));
663 	sc->vmbus_evtflags = evtflags;
664 
665 	sc->vmbus_mnf1 = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0,
666 	    PAGE_SIZE, &sc->vmbus_mnf1_dma, BUS_DMA_WAITOK | BUS_DMA_ZERO);
667 	if (sc->vmbus_mnf1 == NULL)
668 		return ENOMEM;
669 
670 	sc->vmbus_mnf2 = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0,
671 	    PAGE_SIZE, &sc->vmbus_mnf2_dma, BUS_DMA_WAITOK | BUS_DMA_ZERO);
672 	if (sc->vmbus_mnf2 == NULL)
673 		return ENOMEM;
674 
675 	return 0;
676 }
677 
678 static void
679 vmbus_dma_free(struct vmbus_softc *sc)
680 {
681 	int cpu;
682 
683 	if (sc->vmbus_evtflags != NULL) {
684 		hyperv_dmamem_free(&sc->vmbus_evtflags_dma, sc->vmbus_evtflags);
685 		sc->vmbus_evtflags = NULL;
686 		sc->vmbus_rx_evtflags = NULL;
687 		sc->vmbus_tx_evtflags = NULL;
688 	}
689 	if (sc->vmbus_mnf1 != NULL) {
690 		hyperv_dmamem_free(&sc->vmbus_mnf1_dma, sc->vmbus_mnf1);
691 		sc->vmbus_mnf1 = NULL;
692 	}
693 	if (sc->vmbus_mnf2 != NULL) {
694 		hyperv_dmamem_free(&sc->vmbus_mnf2_dma, sc->vmbus_mnf2);
695 		sc->vmbus_mnf2 = NULL;
696 	}
697 
698 	for (cpu = 0; cpu < ncpus; ++cpu) {
699 		struct vmbus_pcpu_data *psc = VMBUS_PCPU(sc, cpu);
700 
701 		if (psc->message != NULL) {
702 			hyperv_dmamem_free(&psc->message_dma, psc->message);
703 			psc->message = NULL;
704 		}
705 		if (psc->event_flags != NULL) {
706 			hyperv_dmamem_free(&psc->event_flags_dma,
707 			    psc->event_flags);
708 			psc->event_flags = NULL;
709 		}
710 	}
711 }
712 
713 static int
714 vmbus_intr_setup(struct vmbus_softc *sc)
715 {
716 	device_t dev = sc->vmbus_dev;
717 	device_t bus = device_get_parent(device_get_parent(dev));
718 	int rid, cpu;
719 
720 	rid = 0;
721 	for (cpu = 0; cpu < ncpus; ++cpu) {
722 		struct vmbus_pcpu_data *psc = VMBUS_PCPU(sc, cpu);
723 		uint64_t msi_addr;
724 		uint32_t msi_data;
725 		int error;
726 
727 		error = PCIB_ALLOC_MSIX(bus, dev, &psc->intr_irq, cpu);
728 		if (error) {
729 			device_printf(dev, "alloc vector on cpu%d failed: %d\n",
730 			    cpu, error);
731 			return ENXIO;
732 		}
733 		psc->intr_rid = ++rid;
734 
735 		psc->intr_res = BUS_ALLOC_RESOURCE(bus, dev, SYS_RES_IRQ,
736 		    &psc->intr_rid, psc->intr_irq, psc->intr_irq, 1,
737 		    RF_ACTIVE, cpu);
738 		if (psc->intr_res == NULL) {
739 			device_printf(dev, "alloc irq on cpu%d failed: %d\n",
740 			    cpu, error);
741 			return ENXIO;
742 		}
743 
744 		error = PCIB_MAP_MSI(bus, dev, rman_get_start(psc->intr_res),
745 		    &msi_addr, &msi_data, cpu);
746 		if (error) {
747 			device_printf(dev, "map irq on cpu%d failed: %d\n",
748 			    cpu, error);
749 			return ENXIO;
750 		}
751 		psc->intr_vec = hyperv_msi2vector(msi_addr, msi_data);
752 
753 		if (bootverbose) {
754 			device_printf(dev, "vector %d irq %d on cpu%d\n",
755 			    psc->intr_vec, psc->intr_irq, cpu);
756 		}
757 
758 		ksnprintf(psc->intr_desc, sizeof(psc->intr_desc), "%s cpu%d",
759 		    device_get_nameunit(dev), cpu);
760 		error = bus_setup_intr_descr(dev, psc->intr_res, INTR_MPSAFE,
761 		    vmbus_intr, psc, &psc->intr_hand, NULL, psc->intr_desc);
762 		if (error) {
763 			device_printf(dev, "setup intr on cpu%d failed: %d\n",
764 			    cpu, error);
765 			return ENXIO;
766 		}
767 	}
768 	return 0;
769 }
770 
771 static void
772 vmbus_intr_teardown(struct vmbus_softc *sc)
773 {
774 	device_t dev = sc->vmbus_dev;
775 	device_t bus = device_get_parent(device_get_parent(dev));
776 	int cpu;
777 
778 	for (cpu = 0; cpu < ncpus; ++cpu) {
779 		struct vmbus_pcpu_data *psc = VMBUS_PCPU(sc, cpu);
780 
781 		if (psc->intr_hand != NULL) {
782 			bus_teardown_intr(dev, psc->intr_res, psc->intr_hand);
783 			psc->intr_hand = NULL;
784 		}
785 
786 		if (psc->intr_res != NULL) {
787 			BUS_RELEASE_RESOURCE(bus, dev, SYS_RES_IRQ,
788 			    psc->intr_rid, psc->intr_res);
789 			psc->intr_res = NULL;
790 		}
791 
792 		if (psc->intr_rid != 0) {
793 			PCIB_RELEASE_MSIX(bus, dev, psc->intr_irq, psc->cpuid);
794 			psc->intr_rid = 0;
795 		}
796 	}
797 }
798 
799 static void
800 vmbus_synic_setup(void *xsc)
801 {
802 	struct vmbus_softc *sc = xsc;
803 	struct vmbus_pcpu_data *psc = VMBUS_PCPU(sc, mycpuid);
804 	uint64_t val, orig;
805 	uint32_t sint;
806 
807 	if (hyperv_features & CPUID_HV_MSR_VP_INDEX) {
808 		/*
809 		 * Save virtual processor id.
810 		 */
811 		psc->vcpuid = rdmsr(MSR_HV_VP_INDEX);
812 	} else {
813 		/*
814 		 * XXX
815 		 * Virtual processoor id is only used by a pretty broken
816 		 * channel selection code from storvsc.  It's nothing
817 		 * critical even if CPUID_HV_MSR_VP_INDEX is not set; keep
818 		 * moving on.
819 		 */
820 		psc->vcpuid = mycpuid;
821 	}
822 
823 	/*
824 	 * Setup the SynIC message.
825 	 */
826 	orig = rdmsr(MSR_HV_SIMP);
827 	val = MSR_HV_SIMP_ENABLE | (orig & MSR_HV_SIMP_RSVD_MASK) |
828 	    ((psc->message_dma.hv_paddr >> PAGE_SHIFT) << MSR_HV_SIMP_PGSHIFT);
829 	wrmsr(MSR_HV_SIMP, val);
830 
831 	/*
832 	 * Setup the SynIC event flags.
833 	 */
834 	orig = rdmsr(MSR_HV_SIEFP);
835 	val = MSR_HV_SIEFP_ENABLE | (orig & MSR_HV_SIEFP_RSVD_MASK) |
836 	    ((psc->event_flags_dma.hv_paddr >> PAGE_SHIFT) <<
837 	     MSR_HV_SIEFP_PGSHIFT);
838 	wrmsr(MSR_HV_SIEFP, val);
839 
840 
841 	/*
842 	 * Configure and unmask SINT for message and event flags.
843 	 */
844 	sint = MSR_HV_SINT0 + VMBUS_SINT_MESSAGE;
845 	orig = rdmsr(sint);
846 	val = psc->intr_vec | /* MSR_HV_SINT_AUTOEOI | notyet */
847 	    (orig & MSR_HV_SINT_RSVD_MASK);
848 	wrmsr(sint, val);
849 
850 	/*
851 	 * Configure and unmask SINT for timer.
852 	 */
853 	sint = MSR_HV_SINT0 + VMBUS_SINT_TIMER;
854 	orig = rdmsr(sint);
855 	val = XTIMER_OFFSET | /* MSR_HV_SINT_AUTOEOI | notyet */
856 	    (orig & MSR_HV_SINT_RSVD_MASK);
857 	wrmsr(sint, val);
858 
859 	/*
860 	 * All done; enable SynIC.
861 	 */
862 	orig = rdmsr(MSR_HV_SCONTROL);
863 	val = MSR_HV_SCTRL_ENABLE | (orig & MSR_HV_SCTRL_RSVD_MASK);
864 	wrmsr(MSR_HV_SCONTROL, val);
865 }
866 
867 static void
868 vmbus_timer_stop(void *arg __unused)
869 {
870 	for (;;) {
871 		uint64_t val;
872 
873 		/* Stop counting, and this also implies disabling STIMER0 */
874 		wrmsr(MSR_HV_STIMER0_COUNT, 0);
875 
876 		val = rdmsr(MSR_HV_STIMER0_CONFIG);
877 		if ((val & MSR_HV_STIMER_CFG_ENABLE) == 0)
878 			break;
879 		cpu_pause();
880 	}
881 }
882 
883 static void
884 vmbus_timer_config(void *arg __unused)
885 {
886 	/*
887 	 * Make sure that STIMER0 is really disabled before writing
888 	 * to STIMER0_CONFIG.
889 	 *
890 	 * "Writing to the configuration register of a timer that
891 	 *  is already enabled may result in undefined behaviour."
892 	 */
893 	vmbus_timer_stop(arg);
894 	wrmsr(MSR_HV_STIMER0_CONFIG,
895 	    MSR_HV_STIMER_CFG_AUTOEN | MSR_HV_STIMER0_CFG_SINT);
896 }
897 
898 static void
899 vmbus_timer_msgintr(struct vmbus_pcpu_data *psc)
900 {
901 	volatile struct vmbus_message *msg;
902 
903 	msg = psc->message + VMBUS_SINT_TIMER;
904 	if (msg->msg_type == HYPERV_MSGTYPE_TIMER_EXPIRED)
905 		vmbus_msg_reset(msg);
906 }
907 
908 static void
909 vmbus_timer_restart(void *xsc)
910 {
911 	struct vmbus_softc *sc = xsc;
912 	struct vmbus_pcpu_data *psc = VMBUS_PCPU(sc, mycpuid);
913 
914 	crit_enter();
915 	vmbus_timer_msgintr(psc);
916 	vmbus_timer_oneshot(psc, hyperv_tc64() + 1);
917 	crit_exit();
918 }
919 
920 static void
921 vmbus_synic_teardown(void *arg __unused)
922 {
923 	uint64_t orig;
924 	uint32_t sint;
925 
926 	/*
927 	 * Disable SynIC.
928 	 */
929 	orig = rdmsr(MSR_HV_SCONTROL);
930 	wrmsr(MSR_HV_SCONTROL, (orig & MSR_HV_SCTRL_RSVD_MASK));
931 
932 	/*
933 	 * Mask message and event flags SINT.
934 	 */
935 	sint = MSR_HV_SINT0 + VMBUS_SINT_MESSAGE;
936 	orig = rdmsr(sint);
937 	wrmsr(sint, orig | MSR_HV_SINT_MASKED);
938 
939 	/*
940 	 * Mask timer SINT.
941 	 */
942 	sint = MSR_HV_SINT0 + VMBUS_SINT_TIMER;
943 	orig = rdmsr(sint);
944 	wrmsr(sint, orig | MSR_HV_SINT_MASKED);
945 
946 	/*
947 	 * Teardown SynIC message.
948 	 */
949 	orig = rdmsr(MSR_HV_SIMP);
950 	wrmsr(MSR_HV_SIMP, (orig & MSR_HV_SIMP_RSVD_MASK));
951 
952 	/*
953 	 * Teardown SynIC event flags.
954 	 */
955 	orig = rdmsr(MSR_HV_SIEFP);
956 	wrmsr(MSR_HV_SIEFP, (orig & MSR_HV_SIEFP_RSVD_MASK));
957 }
958 
959 static int
960 vmbus_init_contact(struct vmbus_softc *sc, uint32_t version)
961 {
962 	struct vmbus_chanmsg_init_contact *req;
963 	const struct vmbus_chanmsg_version_resp *resp;
964 	const struct vmbus_message *msg;
965 	struct vmbus_msghc *mh;
966 	int error, supp = 0;
967 
968 	mh = vmbus_msghc_get(sc, sizeof(*req));
969 	if (mh == NULL)
970 		return ENXIO;
971 
972 	req = vmbus_msghc_dataptr(mh);
973 	req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_INIT_CONTACT;
974 	req->chm_ver = version;
975 	req->chm_evtflags = sc->vmbus_evtflags_dma.hv_paddr;
976 	req->chm_mnf1 = sc->vmbus_mnf1_dma.hv_paddr;
977 	req->chm_mnf2 = sc->vmbus_mnf2_dma.hv_paddr;
978 
979 	error = vmbus_msghc_exec(sc, mh);
980 	if (error) {
981 		vmbus_msghc_put(sc, mh);
982 		return error;
983 	}
984 
985 	msg = vmbus_msghc_wait_result(sc, mh);
986 	resp = (const struct vmbus_chanmsg_version_resp *)msg->msg_data;
987 	supp = resp->chm_supp;
988 
989 	vmbus_msghc_put(sc, mh);
990 
991 	return (supp ? 0 : EOPNOTSUPP);
992 }
993 
994 static int
995 vmbus_init(struct vmbus_softc *sc)
996 {
997 	int i;
998 
999 	for (i = 0; i < nitems(vmbus_version); ++i) {
1000 		int error;
1001 
1002 		error = vmbus_init_contact(sc, vmbus_version[i]);
1003 		if (!error) {
1004 			sc->vmbus_version = vmbus_version[i];
1005 			device_printf(sc->vmbus_dev, "version %u.%u\n",
1006 			    (sc->vmbus_version >> 16),
1007 			    (sc->vmbus_version & 0xffff));
1008 			return 0;
1009 		}
1010 	}
1011 	return ENXIO;
1012 }
1013 
1014 static void
1015 vmbus_chan_msgproc(struct vmbus_softc *sc, const struct vmbus_message *msg)
1016 {
1017 	const struct vmbus_chanmsg_hdr *hdr;
1018 
1019 	hdr = (const struct vmbus_chanmsg_hdr *)msg->msg_data;
1020 
1021 	/* TODO */
1022 	if (hdr->chm_type == VMBUS_CHANMSG_TYPE_VERSION_RESP)
1023 		vmbus_msghc_wakeup(sc, msg);
1024 }
1025