xref: /freebsd/sys/dev/hyperv/vmbus/vmbus.c (revision d9a42747950146bf03cda7f6e25d219253f8a57a)
1 /*-
2  * Copyright (c) 2009-2012,2016-2017 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 /*
30  * VM Bus Driver Implementation
31  */
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include <sys/param.h>
36 #include <sys/bus.h>
37 #include <sys/kernel.h>
38 #include <sys/linker.h>
39 #include <sys/lock.h>
40 #include <sys/malloc.h>
41 #include <sys/module.h>
42 #include <sys/mutex.h>
43 #include <sys/sbuf.h>
44 #include <sys/smp.h>
45 #include <sys/sysctl.h>
46 #include <sys/systm.h>
47 #include <sys/taskqueue.h>
48 
49 #include <vm/vm.h>
50 #include <vm/vm_param.h>
51 #include <vm/pmap.h>
52 
53 #include <machine/bus.h>
54 #if defined(__aarch64__)
55 #include <dev/psci/smccc.h>
56 #include <dev/hyperv/vmbus/aarch64/hyperv_machdep.h>
57 #include <dev/hyperv/vmbus/aarch64/hyperv_reg.h>
58 #else
59 #include <dev/hyperv/vmbus/x86/hyperv_machdep.h>
60 #include <dev/hyperv/vmbus/x86/hyperv_reg.h>
61 #include <machine/intr_machdep.h>
62 #include <x86/include/apicvar.h>
63 #endif
64 #include <machine/metadata.h>
65 #include <machine/md_var.h>
66 #include <machine/resource.h>
67 #include <contrib/dev/acpica/include/acpi.h>
68 #include <dev/acpica/acpivar.h>
69 
70 #include <dev/hyperv/include/hyperv.h>
71 #include <dev/hyperv/include/vmbus_xact.h>
72 #include <dev/hyperv/vmbus/hyperv_var.h>
73 #include <dev/hyperv/vmbus/vmbus_reg.h>
74 #include <dev/hyperv/vmbus/vmbus_var.h>
75 #include <dev/hyperv/vmbus/vmbus_chanvar.h>
76 #include <dev/hyperv/vmbus/hyperv_common_reg.h>
77 #include "acpi_if.h"
78 #include "pcib_if.h"
79 #include "vmbus_if.h"
80 
81 #define VMBUS_GPADL_START		0xe1e10
82 
83 struct vmbus_msghc {
84 	struct vmbus_xact		*mh_xact;
85 	struct hypercall_postmsg_in	mh_inprm_save;
86 };
87 
88 static void			vmbus_identify(driver_t *, device_t);
89 static int			vmbus_probe(device_t);
90 static int			vmbus_attach(device_t);
91 static int			vmbus_detach(device_t);
92 static int			vmbus_read_ivar(device_t, device_t, int,
93 				    uintptr_t *);
94 static int			vmbus_child_pnpinfo(device_t, device_t, struct sbuf *);
95 static struct resource		*vmbus_alloc_resource(device_t dev,
96 				    device_t child, int type, int *rid,
97 				    rman_res_t start, rman_res_t end,
98 				    rman_res_t count, u_int flags);
99 static int			vmbus_alloc_msi(device_t bus, device_t dev,
100 				    int count, int maxcount, int *irqs);
101 static int			vmbus_release_msi(device_t bus, device_t dev,
102 				    int count, int *irqs);
103 static int			vmbus_alloc_msix(device_t bus, device_t dev,
104 				    int *irq);
105 static int			vmbus_release_msix(device_t bus, device_t dev,
106 				    int irq);
107 static int			vmbus_map_msi(device_t bus, device_t dev,
108 				    int irq, uint64_t *addr, uint32_t *data);
109 static uint32_t			vmbus_get_version_method(device_t, device_t);
110 static int			vmbus_probe_guid_method(device_t, device_t,
111 				    const struct hyperv_guid *);
112 static uint32_t			vmbus_get_vcpu_id_method(device_t bus,
113 				    device_t dev, int cpu);
114 static struct taskqueue		*vmbus_get_eventtq_method(device_t, device_t,
115 				    int);
116 #if defined(EARLY_AP_STARTUP) || defined(__aarch64__)
117 static void			vmbus_intrhook(void *);
118 #endif
119 
120 static int			vmbus_init(struct vmbus_softc *);
121 static int			vmbus_connect(struct vmbus_softc *, uint32_t);
122 static int			vmbus_req_channels(struct vmbus_softc *sc);
123 static void			vmbus_disconnect(struct vmbus_softc *);
124 static int			vmbus_scan(struct vmbus_softc *);
125 static void			vmbus_scan_teardown(struct vmbus_softc *);
126 static void			vmbus_scan_done(struct vmbus_softc *,
127 				    const struct vmbus_message *);
128 static void			vmbus_chanmsg_handle(struct vmbus_softc *,
129 				    const struct vmbus_message *);
130 static void			vmbus_msg_task(void *, int);
131 static void			vmbus_synic_setup(void *);
132 static void			vmbus_synic_teardown(void *);
133 static int			vmbus_sysctl_version(SYSCTL_HANDLER_ARGS);
134 static int			vmbus_dma_alloc(struct vmbus_softc *);
135 static void			vmbus_dma_free(struct vmbus_softc *);
136 static int			vmbus_intr_setup(struct vmbus_softc *);
137 static void			vmbus_intr_teardown(struct vmbus_softc *);
138 static int			vmbus_doattach(struct vmbus_softc *);
139 static void			vmbus_event_proc_dummy(struct vmbus_softc *,
140 				    int);
141 static struct vmbus_softc	*vmbus_sc;
142 
143 SYSCTL_NODE(_hw, OID_AUTO, vmbus, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
144     "Hyper-V vmbus");
145 
146 static int			vmbus_pin_evttask = 1;
147 SYSCTL_INT(_hw_vmbus, OID_AUTO, pin_evttask, CTLFLAG_RDTUN,
148     &vmbus_pin_evttask, 0, "Pin event tasks to their respective CPU");
149 uint32_t			vmbus_current_version;
150 
151 static const uint32_t		vmbus_version[] = {
152 	VMBUS_VERSION_WIN10,
153 	VMBUS_VERSION_WIN8_1,
154 	VMBUS_VERSION_WIN8,
155 	VMBUS_VERSION_WIN7,
156 	VMBUS_VERSION_WS2008
157 };
158 
159 static const vmbus_chanmsg_proc_t
160 vmbus_chanmsg_handlers[VMBUS_CHANMSG_TYPE_MAX] = {
161 	VMBUS_CHANMSG_PROC(CHOFFER_DONE, vmbus_scan_done),
162 	VMBUS_CHANMSG_PROC_WAKEUP(CONNECT_RESP)
163 };
164 
165 static device_method_t vmbus_methods[] = {
166 	/* Device interface */
167 	DEVMETHOD(device_identify,		vmbus_identify),
168 	DEVMETHOD(device_probe,			vmbus_probe),
169 	DEVMETHOD(device_attach,		vmbus_attach),
170 	DEVMETHOD(device_detach,		vmbus_detach),
171 	DEVMETHOD(device_shutdown,		bus_generic_shutdown),
172 	DEVMETHOD(device_suspend,		bus_generic_suspend),
173 	DEVMETHOD(device_resume,		bus_generic_resume),
174 
175 	/* Bus interface */
176 	DEVMETHOD(bus_add_child,		bus_generic_add_child),
177 	DEVMETHOD(bus_print_child,		bus_generic_print_child),
178 	DEVMETHOD(bus_read_ivar,		vmbus_read_ivar),
179 	DEVMETHOD(bus_child_pnpinfo,		vmbus_child_pnpinfo),
180 	DEVMETHOD(bus_alloc_resource,		vmbus_alloc_resource),
181 	DEVMETHOD(bus_release_resource,		bus_generic_release_resource),
182 	DEVMETHOD(bus_activate_resource,	bus_generic_activate_resource),
183 	DEVMETHOD(bus_deactivate_resource,	bus_generic_deactivate_resource),
184 	DEVMETHOD(bus_setup_intr,		bus_generic_setup_intr),
185 	DEVMETHOD(bus_teardown_intr,		bus_generic_teardown_intr),
186 	DEVMETHOD(bus_get_cpus,			bus_generic_get_cpus),
187 
188 	/* pcib interface */
189 	DEVMETHOD(pcib_alloc_msi,		vmbus_alloc_msi),
190 	DEVMETHOD(pcib_release_msi,		vmbus_release_msi),
191 	DEVMETHOD(pcib_alloc_msix,		vmbus_alloc_msix),
192 	DEVMETHOD(pcib_release_msix,		vmbus_release_msix),
193 	DEVMETHOD(pcib_map_msi,			vmbus_map_msi),
194 
195 	/* Vmbus interface */
196 	DEVMETHOD(vmbus_get_version,		vmbus_get_version_method),
197 	DEVMETHOD(vmbus_probe_guid,		vmbus_probe_guid_method),
198 	DEVMETHOD(vmbus_get_vcpu_id,		vmbus_get_vcpu_id_method),
199 	DEVMETHOD(vmbus_get_event_taskq,	vmbus_get_eventtq_method),
200 
201 	DEVMETHOD_END
202 };
203 
204 static driver_t vmbus_driver = {
205 	"vmbus",
206 	vmbus_methods,
207 	sizeof(struct vmbus_softc)
208 };
209 
210 DRIVER_MODULE(vmbus, pcib, vmbus_driver, NULL, NULL);
211 DRIVER_MODULE(vmbus, acpi_syscontainer, vmbus_driver, NULL, NULL);
212 
213 MODULE_DEPEND(vmbus, acpi, 1, 1, 1);
214 MODULE_DEPEND(vmbus, pci, 1, 1, 1);
215 MODULE_VERSION(vmbus, 1);
216 
217 static __inline struct vmbus_softc *
218 vmbus_get_softc(void)
219 {
220 	return vmbus_sc;
221 }
222 
223 void
224 vmbus_msghc_reset(struct vmbus_msghc *mh, size_t dsize)
225 {
226 	struct hypercall_postmsg_in *inprm;
227 
228 	if (dsize > HYPERCALL_POSTMSGIN_DSIZE_MAX)
229 		panic("invalid data size %zu", dsize);
230 
231 	inprm = vmbus_xact_req_data(mh->mh_xact);
232 	memset(inprm, 0, HYPERCALL_POSTMSGIN_SIZE);
233 	inprm->hc_connid = VMBUS_CONNID_MESSAGE;
234 	inprm->hc_msgtype = HYPERV_MSGTYPE_CHANNEL;
235 	inprm->hc_dsize = dsize;
236 }
237 
238 struct vmbus_msghc *
239 vmbus_msghc_get(struct vmbus_softc *sc, size_t dsize)
240 {
241 	struct vmbus_msghc *mh;
242 	struct vmbus_xact *xact;
243 
244 	if (dsize > HYPERCALL_POSTMSGIN_DSIZE_MAX)
245 		panic("invalid data size %zu", dsize);
246 
247 	xact = vmbus_xact_get(sc->vmbus_xc,
248 	    dsize + __offsetof(struct hypercall_postmsg_in, hc_data[0]));
249 	if (xact == NULL)
250 		return (NULL);
251 
252 	mh = vmbus_xact_priv(xact, sizeof(*mh));
253 	mh->mh_xact = xact;
254 
255 	vmbus_msghc_reset(mh, dsize);
256 	return (mh);
257 }
258 
259 void
260 vmbus_msghc_put(struct vmbus_softc *sc __unused, struct vmbus_msghc *mh)
261 {
262 
263 	vmbus_xact_put(mh->mh_xact);
264 }
265 
266 void *
267 vmbus_msghc_dataptr(struct vmbus_msghc *mh)
268 {
269 	struct hypercall_postmsg_in *inprm;
270 
271 	inprm = vmbus_xact_req_data(mh->mh_xact);
272 	return (inprm->hc_data);
273 }
274 
275 int
276 vmbus_msghc_exec_noresult(struct vmbus_msghc *mh)
277 {
278 	sbintime_t time = SBT_1MS;
279 	struct hypercall_postmsg_in *inprm;
280 	bus_addr_t inprm_paddr;
281 	int i;
282 
283 	inprm = vmbus_xact_req_data(mh->mh_xact);
284 	inprm_paddr = vmbus_xact_req_paddr(mh->mh_xact);
285 
286 	/*
287 	 * Save the input parameter so that we could restore the input
288 	 * parameter if the Hypercall failed.
289 	 *
290 	 * XXX
291 	 * Is this really necessary?!  i.e. Will the Hypercall ever
292 	 * overwrite the input parameter?
293 	 */
294 	memcpy(&mh->mh_inprm_save, inprm, HYPERCALL_POSTMSGIN_SIZE);
295 
296 	/*
297 	 * In order to cope with transient failures, e.g. insufficient
298 	 * resources on host side, we retry the post message Hypercall
299 	 * several times.  20 retries seem sufficient.
300 	 */
301 #define HC_RETRY_MAX	20
302 
303 	for (i = 0; i < HC_RETRY_MAX; ++i) {
304 		uint64_t status;
305 
306 		status = hypercall_post_message(inprm_paddr);
307 		if (status == HYPERCALL_STATUS_SUCCESS)
308 			return 0;
309 
310 		pause_sbt("hcpmsg", time, 0, C_HARDCLOCK);
311 		if (time < SBT_1S * 2)
312 			time *= 2;
313 
314 		/* Restore input parameter and try again */
315 		memcpy(inprm, &mh->mh_inprm_save, HYPERCALL_POSTMSGIN_SIZE);
316 	}
317 
318 #undef HC_RETRY_MAX
319 
320 	return EIO;
321 }
322 
323 int
324 vmbus_msghc_exec(struct vmbus_softc *sc __unused, struct vmbus_msghc *mh)
325 {
326 	int error;
327 
328 	vmbus_xact_activate(mh->mh_xact);
329 	error = vmbus_msghc_exec_noresult(mh);
330 	if (error)
331 		vmbus_xact_deactivate(mh->mh_xact);
332 	return error;
333 }
334 
335 void
336 vmbus_msghc_exec_cancel(struct vmbus_softc *sc __unused, struct vmbus_msghc *mh)
337 {
338 
339 	vmbus_xact_deactivate(mh->mh_xact);
340 }
341 
342 const struct vmbus_message *
343 vmbus_msghc_wait_result(struct vmbus_softc *sc __unused, struct vmbus_msghc *mh)
344 {
345 	size_t resp_len;
346 
347 	return (vmbus_xact_wait(mh->mh_xact, &resp_len));
348 }
349 
350 const struct vmbus_message *
351 vmbus_msghc_poll_result(struct vmbus_softc *sc __unused, struct vmbus_msghc *mh)
352 {
353 	size_t resp_len;
354 
355 	return (vmbus_xact_poll(mh->mh_xact, &resp_len));
356 }
357 
358 void
359 vmbus_msghc_wakeup(struct vmbus_softc *sc, const struct vmbus_message *msg)
360 {
361 
362 	vmbus_xact_ctx_wakeup(sc->vmbus_xc, msg, sizeof(*msg));
363 }
364 
365 uint32_t
366 vmbus_gpadl_alloc(struct vmbus_softc *sc)
367 {
368 	uint32_t gpadl;
369 
370 again:
371 	gpadl = atomic_fetchadd_int(&sc->vmbus_gpadl, 1);
372 	if (gpadl == 0)
373 		goto again;
374 	return (gpadl);
375 }
376 
377 /* Used for Hyper-V socket when guest client connects to host */
378 int
379 vmbus_req_tl_connect(struct hyperv_guid *guest_srv_id,
380     struct hyperv_guid *host_srv_id)
381 {
382 	struct vmbus_softc *sc = vmbus_get_softc();
383 	struct vmbus_chanmsg_tl_connect *req;
384 	struct vmbus_msghc *mh;
385 	int error;
386 
387 	if (!sc)
388 		return ENXIO;
389 
390 	mh = vmbus_msghc_get(sc, sizeof(*req));
391 	if (mh == NULL) {
392 		device_printf(sc->vmbus_dev,
393 		    "can not get msg hypercall for tl connect\n");
394 		return ENXIO;
395 	}
396 
397 	req = vmbus_msghc_dataptr(mh);
398 	req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_TL_CONN;
399 	req->guest_endpoint_id = *guest_srv_id;
400 	req->host_service_id = *host_srv_id;
401 
402 	error = vmbus_msghc_exec_noresult(mh);
403 	vmbus_msghc_put(sc, mh);
404 
405 	if (error) {
406 		device_printf(sc->vmbus_dev,
407 		    "tl connect msg hypercall failed\n");
408 	}
409 
410 	return error;
411 }
412 
413 static int
414 vmbus_connect(struct vmbus_softc *sc, uint32_t version)
415 {
416 	struct vmbus_chanmsg_connect *req;
417 	const struct vmbus_message *msg;
418 	struct vmbus_msghc *mh;
419 	int error, done = 0;
420 
421 	mh = vmbus_msghc_get(sc, sizeof(*req));
422 	if (mh == NULL)
423 		return ENXIO;
424 
425 	req = vmbus_msghc_dataptr(mh);
426 	req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_CONNECT;
427 	req->chm_ver = version;
428 	req->chm_evtflags = sc->vmbus_evtflags_dma.hv_paddr;
429 	req->chm_mnf1 = sc->vmbus_mnf1_dma.hv_paddr;
430 	req->chm_mnf2 = sc->vmbus_mnf2_dma.hv_paddr;
431 
432 	error = vmbus_msghc_exec(sc, mh);
433 	if (error) {
434 		vmbus_msghc_put(sc, mh);
435 		return error;
436 	}
437 
438 	msg = vmbus_msghc_wait_result(sc, mh);
439 	done = ((const struct vmbus_chanmsg_connect_resp *)
440 	    msg->msg_data)->chm_done;
441 
442 	vmbus_msghc_put(sc, mh);
443 
444 	return (done ? 0 : EOPNOTSUPP);
445 }
446 
447 static int
448 vmbus_init(struct vmbus_softc *sc)
449 {
450 	int i;
451 
452 	for (i = 0; i < nitems(vmbus_version); ++i) {
453 		int error;
454 
455 		error = vmbus_connect(sc, vmbus_version[i]);
456 		if (!error) {
457 			vmbus_current_version = vmbus_version[i];
458 			sc->vmbus_version = vmbus_version[i];
459 			device_printf(sc->vmbus_dev, "version %u.%u\n",
460 			    VMBUS_VERSION_MAJOR(sc->vmbus_version),
461 			    VMBUS_VERSION_MINOR(sc->vmbus_version));
462 			return 0;
463 		}
464 	}
465 	return ENXIO;
466 }
467 
468 static void
469 vmbus_disconnect(struct vmbus_softc *sc)
470 {
471 	struct vmbus_chanmsg_disconnect *req;
472 	struct vmbus_msghc *mh;
473 	int error;
474 
475 	mh = vmbus_msghc_get(sc, sizeof(*req));
476 	if (mh == NULL) {
477 		device_printf(sc->vmbus_dev,
478 		    "can not get msg hypercall for disconnect\n");
479 		return;
480 	}
481 
482 	req = vmbus_msghc_dataptr(mh);
483 	req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_DISCONNECT;
484 
485 	error = vmbus_msghc_exec_noresult(mh);
486 	vmbus_msghc_put(sc, mh);
487 
488 	if (error) {
489 		device_printf(sc->vmbus_dev,
490 		    "disconnect msg hypercall failed\n");
491 	}
492 }
493 
494 static int
495 vmbus_req_channels(struct vmbus_softc *sc)
496 {
497 	struct vmbus_chanmsg_chrequest *req;
498 	struct vmbus_msghc *mh;
499 	int error;
500 
501 	mh = vmbus_msghc_get(sc, sizeof(*req));
502 	if (mh == NULL)
503 		return ENXIO;
504 
505 	req = vmbus_msghc_dataptr(mh);
506 	req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_CHREQUEST;
507 
508 	error = vmbus_msghc_exec_noresult(mh);
509 	vmbus_msghc_put(sc, mh);
510 
511 	return error;
512 }
513 
514 static void
515 vmbus_scan_done_task(void *xsc, int pending __unused)
516 {
517 	struct vmbus_softc *sc = xsc;
518 
519 	bus_topo_lock();
520 	sc->vmbus_scandone = true;
521 	bus_topo_unlock();
522 	wakeup(&sc->vmbus_scandone);
523 }
524 
525 static void
526 vmbus_scan_done(struct vmbus_softc *sc,
527     const struct vmbus_message *msg __unused)
528 {
529 
530 	taskqueue_enqueue(sc->vmbus_devtq, &sc->vmbus_scandone_task);
531 }
532 
533 static int
534 vmbus_scan(struct vmbus_softc *sc)
535 {
536 	int error;
537 
538 	/*
539 	 * Identify, probe and attach for non-channel devices.
540 	 */
541 	bus_generic_probe(sc->vmbus_dev);
542 	bus_generic_attach(sc->vmbus_dev);
543 
544 	/*
545 	 * This taskqueue serializes vmbus devices' attach and detach
546 	 * for channel offer and rescind messages.
547 	 */
548 	sc->vmbus_devtq = taskqueue_create("vmbus dev", M_WAITOK,
549 	    taskqueue_thread_enqueue, &sc->vmbus_devtq);
550 	taskqueue_start_threads(&sc->vmbus_devtq, 1, PI_NET, "vmbusdev");
551 	TASK_INIT(&sc->vmbus_scandone_task, 0, vmbus_scan_done_task, sc);
552 
553 	/*
554 	 * This taskqueue handles sub-channel detach, so that vmbus
555 	 * device's detach running in vmbus_devtq can drain its sub-
556 	 * channels.
557 	 */
558 	sc->vmbus_subchtq = taskqueue_create("vmbus subch", M_WAITOK,
559 	    taskqueue_thread_enqueue, &sc->vmbus_subchtq);
560 	taskqueue_start_threads(&sc->vmbus_subchtq, 1, PI_NET, "vmbussch");
561 
562 	/*
563 	 * Start vmbus scanning.
564 	 */
565 	error = vmbus_req_channels(sc);
566 	if (error) {
567 		device_printf(sc->vmbus_dev, "channel request failed: %d\n",
568 		    error);
569 		return (error);
570 	}
571 
572 	/*
573 	 * Wait for all vmbus devices from the initial channel offers to be
574 	 * attached.
575 	 */
576 	bus_topo_assert();
577 	while (!sc->vmbus_scandone)
578 		mtx_sleep(&sc->vmbus_scandone, bus_topo_mtx(), 0, "vmbusdev", 0);
579 
580 	if (bootverbose) {
581 		device_printf(sc->vmbus_dev, "device scan, probe and attach "
582 		    "done\n");
583 	}
584 	return (0);
585 }
586 
587 static void
588 vmbus_scan_teardown(struct vmbus_softc *sc)
589 {
590 
591 	bus_topo_assert();
592 	if (sc->vmbus_devtq != NULL) {
593 		bus_topo_unlock();
594 		taskqueue_free(sc->vmbus_devtq);
595 		bus_topo_lock();
596 		sc->vmbus_devtq = NULL;
597 	}
598 	if (sc->vmbus_subchtq != NULL) {
599 		bus_topo_unlock();
600 		taskqueue_free(sc->vmbus_subchtq);
601 		bus_topo_lock();
602 		sc->vmbus_subchtq = NULL;
603 	}
604 }
605 
606 static void
607 vmbus_chanmsg_handle(struct vmbus_softc *sc, const struct vmbus_message *msg)
608 {
609 	vmbus_chanmsg_proc_t msg_proc;
610 	uint32_t msg_type;
611 
612 	msg_type = ((const struct vmbus_chanmsg_hdr *)msg->msg_data)->chm_type;
613 	if (msg_type >= VMBUS_CHANMSG_TYPE_MAX) {
614 		device_printf(sc->vmbus_dev, "unknown message type 0x%x\n",
615 		    msg_type);
616 		return;
617 	}
618 
619 	msg_proc = vmbus_chanmsg_handlers[msg_type];
620 	if (msg_proc != NULL)
621 		msg_proc(sc, msg);
622 
623 	/* Channel specific processing */
624 	vmbus_chan_msgproc(sc, msg);
625 }
626 
627 static void
628 vmbus_msg_task(void *xsc, int pending __unused)
629 {
630 	struct vmbus_softc *sc = xsc;
631 	volatile struct vmbus_message *msg;
632 
633 	msg = VMBUS_PCPU_GET(sc, message, curcpu) + VMBUS_SINT_MESSAGE;
634 	for (;;) {
635 		if (msg->msg_type == HYPERV_MSGTYPE_NONE) {
636 			/* No message */
637 			break;
638 		} else if (msg->msg_type == HYPERV_MSGTYPE_CHANNEL) {
639 			/* Channel message */
640 			vmbus_chanmsg_handle(sc,
641 			    __DEVOLATILE(const struct vmbus_message *, msg));
642 		}
643 
644 		msg->msg_type = HYPERV_MSGTYPE_NONE;
645 		/*
646 		 * Make sure the write to msg_type (i.e. set to
647 		 * HYPERV_MSGTYPE_NONE) happens before we read the
648 		 * msg_flags and EOMing. Otherwise, the EOMing will
649 		 * not deliver any more messages since there is no
650 		 * empty slot
651 		 *
652 		 * NOTE:
653 		 * mb() is used here, since atomic_thread_fence_seq_cst()
654 		 * will become compiler fence on UP kernel.
655 		 */
656 		mb();
657 		if (msg->msg_flags & VMBUS_MSGFLAG_PENDING) {
658 			/*
659 			 * This will cause message queue rescan to possibly
660 			 * deliver another msg from the hypervisor
661 			 */
662 			WRMSR(MSR_HV_EOM, 0);
663 		}
664 	}
665 }
666 static __inline int
667 vmbus_handle_intr1(struct vmbus_softc *sc, struct trapframe *frame, int cpu)
668 {
669 	volatile struct vmbus_message *msg;
670 	struct vmbus_message *msg_base;
671 
672 	msg_base = VMBUS_PCPU_GET(sc, message, cpu);
673 
674 	/*
675 	 * Check event timer.
676 	 *
677 	 * TODO: move this to independent IDT vector.
678 	 */
679 	vmbus_handle_timer_intr1(msg_base, frame);
680 	/*
681 	 * Check events.  Hot path for network and storage I/O data; high rate.
682 	 *
683 	 * NOTE:
684 	 * As recommended by the Windows guest fellows, we check events before
685 	 * checking messages.
686 	 */
687 	sc->vmbus_event_proc(sc, cpu);
688 
689 	/*
690 	 * Check messages.  Mainly management stuffs; ultra low rate.
691 	 */
692 	msg = msg_base + VMBUS_SINT_MESSAGE;
693 	if (__predict_false(msg->msg_type != HYPERV_MSGTYPE_NONE)) {
694 		taskqueue_enqueue(VMBUS_PCPU_GET(sc, message_tq, cpu),
695 		    VMBUS_PCPU_PTR(sc, message_task, cpu));
696 	}
697 
698 	return (FILTER_HANDLED);
699 }
700 
701 void
702 vmbus_handle_intr(struct trapframe *trap_frame)
703 {
704 	struct vmbus_softc *sc = vmbus_get_softc();
705 	int cpu = curcpu;
706 
707 	/*
708 	 * Disable preemption.
709 	 */
710 	critical_enter();
711 
712 	/*
713 	 * Do a little interrupt counting. This used x86 specific
714 	 * intrcnt_add function
715 	 */
716 #if !defined(__aarch64__)
717 	(*VMBUS_PCPU_GET(sc, intr_cnt, cpu))++;
718 #endif /* not for aarch64 */
719 	vmbus_handle_intr1(sc, trap_frame, cpu);
720 
721 	/*
722 	 * Enable preemption.
723 	 */
724 	critical_exit();
725 }
726 
727 static void
728 vmbus_synic_setup(void *xsc)
729 {
730 	struct vmbus_softc *sc = xsc;
731 	int cpu = curcpu;
732 	uint64_t val, orig;
733 	uint32_t sint;
734 
735 	if (hyperv_features & CPUID_HV_MSR_VP_INDEX) {
736 		/* Save virtual processor id. */
737 		VMBUS_PCPU_GET(sc, vcpuid, cpu) = RDMSR(MSR_HV_VP_INDEX);
738 	} else {
739 		/* Set virtual processor id to 0 for compatibility. */
740 		VMBUS_PCPU_GET(sc, vcpuid, cpu) = 0;
741 	}
742 
743 	/*
744 	 * Setup the SynIC message.
745 	 */
746 	orig = RDMSR(MSR_HV_SIMP);
747 	val = MSR_HV_SIMP_ENABLE | (orig & MSR_HV_SIMP_RSVD_MASK) |
748 	    ((VMBUS_PCPU_GET(sc, message_dma.hv_paddr, cpu) >> PAGE_SHIFT)
749 		<< MSR_HV_SIMP_PGSHIFT);
750 	WRMSR(MSR_HV_SIMP, val);
751 	/*
752 	 * Setup the SynIC event flags.
753 	 */
754 	orig = RDMSR(MSR_HV_SIEFP);
755 	val = MSR_HV_SIEFP_ENABLE | (orig & MSR_HV_SIEFP_RSVD_MASK) |
756 	    ((VMBUS_PCPU_GET(sc, event_flags_dma.hv_paddr, cpu) >> PAGE_SHIFT)
757 		<< MSR_HV_SIEFP_PGSHIFT);
758 	WRMSR(MSR_HV_SIEFP, val);
759 
760 	/*
761 	 * Configure and unmask SINT for message and event flags.
762 	 */
763 	sint = MSR_HV_SINT0 + VMBUS_SINT_MESSAGE;
764 	orig = RDMSR(sint);
765 	val = sc->vmbus_idtvec | MSR_HV_SINT_AUTOEOI |
766 	    (orig & MSR_HV_SINT_RSVD_MASK);
767 	WRMSR(sint, val);
768 
769 	/*
770 	 * Configure and unmask SINT for timer.
771 	 */
772 	vmbus_synic_setup1(sc);
773 	/*
774 	 * All done; enable SynIC.
775 	 */
776 	orig = RDMSR(MSR_HV_SCONTROL);
777 	val = MSR_HV_SCTRL_ENABLE | (orig & MSR_HV_SCTRL_RSVD_MASK);
778 	WRMSR(MSR_HV_SCONTROL, val);
779 }
780 
781 static void
782 vmbus_synic_teardown(void *arg)
783 {
784 	uint64_t orig;
785 	uint32_t sint;
786 
787 	/*
788 	 * Disable SynIC.
789 	 */
790 	orig = RDMSR(MSR_HV_SCONTROL);
791 	WRMSR(MSR_HV_SCONTROL, (orig & MSR_HV_SCTRL_RSVD_MASK));
792 
793 	/*
794 	 * Mask message and event flags SINT.
795 	 */
796 	sint = MSR_HV_SINT0 + VMBUS_SINT_MESSAGE;
797 	orig = RDMSR(sint);
798 	WRMSR(sint, orig | MSR_HV_SINT_MASKED);
799 
800 	/*
801 	 * Mask timer SINT.
802 	 */
803 	vmbus_synic_teardown1();
804 	/*
805 	 * Teardown SynIC message.
806 	 */
807 	orig = RDMSR(MSR_HV_SIMP);
808 	WRMSR(MSR_HV_SIMP, (orig & MSR_HV_SIMP_RSVD_MASK));
809 
810 	/*
811 	 * Teardown SynIC event flags.
812 	 */
813 	orig = RDMSR(MSR_HV_SIEFP);
814 	WRMSR(MSR_HV_SIEFP, (orig & MSR_HV_SIEFP_RSVD_MASK));
815 }
816 
817 static int
818 vmbus_dma_alloc(struct vmbus_softc *sc)
819 {
820 	bus_dma_tag_t parent_dtag;
821 	uint8_t *evtflags;
822 	int cpu;
823 
824 	parent_dtag = bus_get_dma_tag(sc->vmbus_dev);
825 	CPU_FOREACH(cpu) {
826 		void *ptr;
827 
828 		/*
829 		 * Per-cpu messages and event flags.
830 		 */
831 		ptr = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0,
832 		    PAGE_SIZE, VMBUS_PCPU_PTR(sc, message_dma, cpu),
833 		    BUS_DMA_WAITOK | BUS_DMA_ZERO);
834 		if (ptr == NULL)
835 			return ENOMEM;
836 		VMBUS_PCPU_GET(sc, message, cpu) = ptr;
837 
838 		ptr = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0,
839 		    PAGE_SIZE, VMBUS_PCPU_PTR(sc, event_flags_dma, cpu),
840 		    BUS_DMA_WAITOK | BUS_DMA_ZERO);
841 		if (ptr == NULL)
842 			return ENOMEM;
843 		VMBUS_PCPU_GET(sc, event_flags, cpu) = ptr;
844 	}
845 
846 	evtflags = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0,
847 	    PAGE_SIZE, &sc->vmbus_evtflags_dma, BUS_DMA_WAITOK | BUS_DMA_ZERO);
848 	if (evtflags == NULL)
849 		return ENOMEM;
850 	sc->vmbus_rx_evtflags = (u_long *)evtflags;
851 	sc->vmbus_tx_evtflags = (u_long *)(evtflags + (PAGE_SIZE / 2));
852 	sc->vmbus_evtflags = evtflags;
853 
854 	sc->vmbus_mnf1 = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0,
855 	    PAGE_SIZE, &sc->vmbus_mnf1_dma, BUS_DMA_WAITOK | BUS_DMA_ZERO);
856 	if (sc->vmbus_mnf1 == NULL)
857 		return ENOMEM;
858 
859 	sc->vmbus_mnf2 = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0,
860 	    sizeof(struct vmbus_mnf), &sc->vmbus_mnf2_dma,
861 	    BUS_DMA_WAITOK | BUS_DMA_ZERO);
862 	if (sc->vmbus_mnf2 == NULL)
863 		return ENOMEM;
864 
865 	return 0;
866 }
867 
868 static void
869 vmbus_dma_free(struct vmbus_softc *sc)
870 {
871 	int cpu;
872 
873 	if (sc->vmbus_evtflags != NULL) {
874 		hyperv_dmamem_free(&sc->vmbus_evtflags_dma, sc->vmbus_evtflags);
875 		sc->vmbus_evtflags = NULL;
876 		sc->vmbus_rx_evtflags = NULL;
877 		sc->vmbus_tx_evtflags = NULL;
878 	}
879 	if (sc->vmbus_mnf1 != NULL) {
880 		hyperv_dmamem_free(&sc->vmbus_mnf1_dma, sc->vmbus_mnf1);
881 		sc->vmbus_mnf1 = NULL;
882 	}
883 	if (sc->vmbus_mnf2 != NULL) {
884 		hyperv_dmamem_free(&sc->vmbus_mnf2_dma, sc->vmbus_mnf2);
885 		sc->vmbus_mnf2 = NULL;
886 	}
887 
888 	CPU_FOREACH(cpu) {
889 		if (VMBUS_PCPU_GET(sc, message, cpu) != NULL) {
890 			hyperv_dmamem_free(
891 			    VMBUS_PCPU_PTR(sc, message_dma, cpu),
892 			    VMBUS_PCPU_GET(sc, message, cpu));
893 			VMBUS_PCPU_GET(sc, message, cpu) = NULL;
894 		}
895 		if (VMBUS_PCPU_GET(sc, event_flags, cpu) != NULL) {
896 			hyperv_dmamem_free(
897 			    VMBUS_PCPU_PTR(sc, event_flags_dma, cpu),
898 			    VMBUS_PCPU_GET(sc, event_flags, cpu));
899 			VMBUS_PCPU_GET(sc, event_flags, cpu) = NULL;
900 		}
901 	}
902 }
903 
904 static int
905 vmbus_intr_setup(struct vmbus_softc *sc)
906 {
907 	int cpu;
908 
909 	CPU_FOREACH(cpu) {
910 		char buf[MAXCOMLEN + 1];
911 		cpuset_t cpu_mask;
912 
913 		/* Allocate an interrupt counter for Hyper-V interrupt */
914 		snprintf(buf, sizeof(buf), "cpu%d:hyperv", cpu);
915 #if !defined(__aarch64__)
916 		intrcnt_add(buf, VMBUS_PCPU_PTR(sc, intr_cnt, cpu));
917 #endif /* not for aarch64 */
918 		/*
919 		 * Setup taskqueue to handle events.  Task will be per-
920 		 * channel.
921 		 */
922 		VMBUS_PCPU_GET(sc, event_tq, cpu) = taskqueue_create_fast(
923 		    "hyperv event", M_WAITOK, taskqueue_thread_enqueue,
924 		    VMBUS_PCPU_PTR(sc, event_tq, cpu));
925 		if (vmbus_pin_evttask) {
926 			CPU_SETOF(cpu, &cpu_mask);
927 			taskqueue_start_threads_cpuset(
928 			    VMBUS_PCPU_PTR(sc, event_tq, cpu), 1, PI_NET,
929 			    &cpu_mask, "hvevent%d", cpu);
930 		} else {
931 			taskqueue_start_threads(
932 			    VMBUS_PCPU_PTR(sc, event_tq, cpu), 1, PI_NET,
933 			    "hvevent%d", cpu);
934 		}
935 
936 		/*
937 		 * Setup tasks and taskqueues to handle messages.
938 		 */
939 		VMBUS_PCPU_GET(sc, message_tq, cpu) = taskqueue_create_fast(
940 		    "hyperv msg", M_WAITOK, taskqueue_thread_enqueue,
941 		    VMBUS_PCPU_PTR(sc, message_tq, cpu));
942 		CPU_SETOF(cpu, &cpu_mask);
943 		taskqueue_start_threads_cpuset(
944 		    VMBUS_PCPU_PTR(sc, message_tq, cpu), 1, PI_NET, &cpu_mask,
945 		    "hvmsg%d", cpu);
946 		TASK_INIT(VMBUS_PCPU_PTR(sc, message_task, cpu), 0,
947 		    vmbus_msg_task, sc);
948 	}
949 	return (vmbus_setup_intr1(sc));
950 }
951 static void
952 vmbus_intr_teardown(struct vmbus_softc *sc)
953 {
954 	vmbus_intr_teardown1(sc);
955 }
956 
957 static int
958 vmbus_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
959 {
960 	return (ENOENT);
961 }
962 
963 static int
964 vmbus_child_pnpinfo(device_t dev, device_t child, struct sbuf *sb)
965 {
966 	const struct vmbus_channel *chan;
967 	char guidbuf[HYPERV_GUID_STRLEN];
968 
969 	chan = vmbus_get_channel(child);
970 	if (chan == NULL) {
971 		/* Event timer device, which does not belong to a channel */
972 		return (0);
973 	}
974 
975 	hyperv_guid2str(&chan->ch_guid_type, guidbuf, sizeof(guidbuf));
976 	sbuf_printf(sb, "classid=%s", guidbuf);
977 
978 	hyperv_guid2str(&chan->ch_guid_inst, guidbuf, sizeof(guidbuf));
979 	sbuf_printf(sb, " deviceid=%s", guidbuf);
980 
981 	return (0);
982 }
983 
984 int
985 vmbus_add_child(struct vmbus_channel *chan)
986 {
987 	struct vmbus_softc *sc = chan->ch_vmbus;
988 	device_t parent = sc->vmbus_dev;
989 
990 	bus_topo_lock();
991 	chan->ch_dev = device_add_child(parent, NULL, -1);
992 	if (chan->ch_dev == NULL) {
993 		bus_topo_unlock();
994 		device_printf(parent, "device_add_child for chan%u failed\n",
995 		    chan->ch_id);
996 		return (ENXIO);
997 	}
998 	device_set_ivars(chan->ch_dev, chan);
999 	device_probe_and_attach(chan->ch_dev);
1000 	bus_topo_unlock();
1001 
1002 	return (0);
1003 }
1004 
1005 int
1006 vmbus_delete_child(struct vmbus_channel *chan)
1007 {
1008 	int error = 0;
1009 
1010 	bus_topo_lock();
1011 	if (chan->ch_dev != NULL) {
1012 		error = device_delete_child(chan->ch_vmbus->vmbus_dev,
1013 		    chan->ch_dev);
1014 		chan->ch_dev = NULL;
1015 	}
1016 	bus_topo_unlock();
1017 	return (error);
1018 }
1019 
1020 static int
1021 vmbus_sysctl_version(SYSCTL_HANDLER_ARGS)
1022 {
1023 	struct vmbus_softc *sc = arg1;
1024 	char verstr[16];
1025 
1026 	snprintf(verstr, sizeof(verstr), "%u.%u",
1027 	    VMBUS_VERSION_MAJOR(sc->vmbus_version),
1028 	    VMBUS_VERSION_MINOR(sc->vmbus_version));
1029 	return sysctl_handle_string(oidp, verstr, sizeof(verstr), req);
1030 }
1031 
1032 /*
1033  * We need the function to make sure the MMIO resource is allocated from the
1034  * ranges found in _CRS.
1035  *
1036  * For the release function, we can use bus_generic_release_resource().
1037  */
1038 static struct resource *
1039 vmbus_alloc_resource(device_t dev, device_t child, int type, int *rid,
1040     rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
1041 {
1042 	device_t parent = device_get_parent(dev);
1043 	struct resource *res;
1044 
1045 #ifdef NEW_PCIB
1046 	if (type == SYS_RES_MEMORY) {
1047 		struct vmbus_softc *sc = device_get_softc(dev);
1048 
1049 		res = pcib_host_res_alloc(&sc->vmbus_mmio_res, child, type,
1050 		    rid, start, end, count, flags);
1051 	} else
1052 #endif
1053 	{
1054 		res = BUS_ALLOC_RESOURCE(parent, child, type, rid, start,
1055 		    end, count, flags);
1056 	}
1057 
1058 	return (res);
1059 }
1060 
1061 static int
1062 vmbus_alloc_msi(device_t bus, device_t dev, int count, int maxcount, int *irqs)
1063 {
1064 
1065 	return (PCIB_ALLOC_MSI(device_get_parent(bus), dev, count, maxcount,
1066 	    irqs));
1067 }
1068 
1069 static int
1070 vmbus_release_msi(device_t bus, device_t dev, int count, int *irqs)
1071 {
1072 
1073 	return (PCIB_RELEASE_MSI(device_get_parent(bus), dev, count, irqs));
1074 }
1075 
1076 static int
1077 vmbus_alloc_msix(device_t bus, device_t dev, int *irq)
1078 {
1079 
1080 	return (PCIB_ALLOC_MSIX(device_get_parent(bus), dev, irq));
1081 }
1082 
1083 static int
1084 vmbus_release_msix(device_t bus, device_t dev, int irq)
1085 {
1086 
1087 	return (PCIB_RELEASE_MSIX(device_get_parent(bus), dev, irq));
1088 }
1089 
1090 static int
1091 vmbus_map_msi(device_t bus, device_t dev, int irq, uint64_t *addr,
1092 	uint32_t *data)
1093 {
1094 
1095 	return (PCIB_MAP_MSI(device_get_parent(bus), dev, irq, addr, data));
1096 }
1097 
1098 static uint32_t
1099 vmbus_get_version_method(device_t bus, device_t dev)
1100 {
1101 	struct vmbus_softc *sc = device_get_softc(bus);
1102 
1103 	return sc->vmbus_version;
1104 }
1105 
1106 static int
1107 vmbus_probe_guid_method(device_t bus, device_t dev,
1108     const struct hyperv_guid *guid)
1109 {
1110 	const struct vmbus_channel *chan = vmbus_get_channel(dev);
1111 
1112 	if (memcmp(&chan->ch_guid_type, guid, sizeof(struct hyperv_guid)) == 0)
1113 		return 0;
1114 	return ENXIO;
1115 }
1116 
1117 static uint32_t
1118 vmbus_get_vcpu_id_method(device_t bus, device_t dev, int cpu)
1119 {
1120 	const struct vmbus_softc *sc = device_get_softc(bus);
1121 
1122 	return (VMBUS_PCPU_GET(sc, vcpuid, cpu));
1123 }
1124 
1125 static struct taskqueue *
1126 vmbus_get_eventtq_method(device_t bus, device_t dev __unused, int cpu)
1127 {
1128 	const struct vmbus_softc *sc = device_get_softc(bus);
1129 
1130 	KASSERT(cpu >= 0 && cpu < mp_ncpus, ("invalid cpu%d", cpu));
1131 	return (VMBUS_PCPU_GET(sc, event_tq, cpu));
1132 }
1133 
1134 #ifdef NEW_PCIB
1135 #define VTPM_BASE_ADDR 0xfed40000
1136 #define FOUR_GB (1ULL << 32)
1137 
1138 enum parse_pass { parse_64, parse_32 };
1139 
1140 struct parse_context {
1141 	device_t vmbus_dev;
1142 	enum parse_pass pass;
1143 };
1144 
1145 static ACPI_STATUS
1146 parse_crs(ACPI_RESOURCE *res, void *ctx)
1147 {
1148 	const struct parse_context *pc = ctx;
1149 	device_t vmbus_dev = pc->vmbus_dev;
1150 
1151 	struct vmbus_softc *sc = device_get_softc(vmbus_dev);
1152 	UINT64 start, end;
1153 
1154 	switch (res->Type) {
1155 	case ACPI_RESOURCE_TYPE_ADDRESS32:
1156 		start = res->Data.Address32.Address.Minimum;
1157 		end = res->Data.Address32.Address.Maximum;
1158 		break;
1159 
1160 	case ACPI_RESOURCE_TYPE_ADDRESS64:
1161 		start = res->Data.Address64.Address.Minimum;
1162 		end = res->Data.Address64.Address.Maximum;
1163 		break;
1164 
1165 	default:
1166 		/* Unused types. */
1167 		return (AE_OK);
1168 	}
1169 
1170 	/*
1171 	 * We don't use <1MB addresses.
1172 	 */
1173 	if (end < 0x100000)
1174 		return (AE_OK);
1175 
1176 	/* Don't conflict with vTPM. */
1177 	if (end >= VTPM_BASE_ADDR && start < VTPM_BASE_ADDR)
1178 		end = VTPM_BASE_ADDR - 1;
1179 
1180 	if ((pc->pass == parse_32 && start < FOUR_GB) ||
1181 	    (pc->pass == parse_64 && start >= FOUR_GB))
1182 		pcib_host_res_decodes(&sc->vmbus_mmio_res, SYS_RES_MEMORY,
1183 		    start, end, 0);
1184 
1185 	return (AE_OK);
1186 }
1187 
1188 static void
1189 vmbus_get_crs(device_t dev, device_t vmbus_dev, enum parse_pass pass)
1190 {
1191 	struct parse_context pc;
1192 	ACPI_STATUS status;
1193 
1194 	if (bootverbose)
1195 		device_printf(dev, "walking _CRS, pass=%d\n", pass);
1196 
1197 	pc.vmbus_dev = vmbus_dev;
1198 	pc.pass = pass;
1199 	status = AcpiWalkResources(acpi_get_handle(dev), "_CRS",
1200 			parse_crs, &pc);
1201 
1202 	if (bootverbose && ACPI_FAILURE(status))
1203 		device_printf(dev, "_CRS: not found, pass=%d\n", pass);
1204 }
1205 
1206 static void
1207 vmbus_get_mmio_res_pass(device_t dev, enum parse_pass pass)
1208 {
1209 	device_t acpi0, parent;
1210 
1211 	parent = device_get_parent(dev);
1212 
1213 	acpi0 = device_get_parent(parent);
1214 	if (strcmp("acpi0", device_get_nameunit(acpi0)) == 0) {
1215 		device_t *children;
1216 		int count;
1217 
1218 		/*
1219 		 * Try to locate VMBUS resources and find _CRS on them.
1220 		 */
1221 		if (device_get_children(acpi0, &children, &count) == 0) {
1222 			int i;
1223 
1224 			for (i = 0; i < count; ++i) {
1225 				if (!device_is_attached(children[i]))
1226 					continue;
1227 
1228 				if (strcmp("vmbus_res",
1229 				    device_get_name(children[i])) == 0)
1230 					vmbus_get_crs(children[i], dev, pass);
1231 			}
1232 			free(children, M_TEMP);
1233 		}
1234 
1235 		/*
1236 		 * Try to find _CRS on acpi.
1237 		 */
1238 		vmbus_get_crs(acpi0, dev, pass);
1239 	} else {
1240 		device_printf(dev, "not grandchild of acpi\n");
1241 	}
1242 
1243 	/*
1244 	 * Try to find _CRS on parent.
1245 	 */
1246 	vmbus_get_crs(parent, dev, pass);
1247 }
1248 
1249 static void
1250 vmbus_get_mmio_res(device_t dev)
1251 {
1252 	struct vmbus_softc *sc = device_get_softc(dev);
1253 	/*
1254 	 * We walk the resources twice to make sure that: in the resource
1255 	 * list, the 32-bit resources appear behind the 64-bit resources.
1256 	 * NB: resource_list_add() uses INSERT_TAIL. This way, when we
1257 	 * iterate through the list to find a range for a 64-bit BAR in
1258 	 * vmbus_alloc_resource(), we can make sure we try to use >4GB
1259 	 * ranges first.
1260 	 */
1261 	pcib_host_res_init(dev, &sc->vmbus_mmio_res);
1262 
1263 	vmbus_get_mmio_res_pass(dev, parse_64);
1264 	vmbus_get_mmio_res_pass(dev, parse_32);
1265 }
1266 
1267 /*
1268  * On Gen2 VMs, Hyper-V provides mmio space for framebuffer.
1269  * This mmio address range is not useable for other PCI devices.
1270  * Currently only efifb and vbefb drivers are using this range without
1271  * reserving it from system.
1272  * Therefore, vmbus driver reserves it before any other PCI device
1273  * drivers start to request mmio addresses.
1274  */
1275 static struct resource *hv_fb_res;
1276 
1277 static void
1278 vmbus_fb_mmio_res(device_t dev)
1279 {
1280 	struct efi_fb *efifb;
1281 #if !defined(__aarch64__)
1282 	struct vbe_fb *vbefb;
1283 #endif /* aarch64 */
1284 	rman_res_t fb_start, fb_end, fb_count;
1285 	int fb_height, fb_width;
1286 	caddr_t kmdp;
1287 
1288 	struct vmbus_softc *sc = device_get_softc(dev);
1289 	int rid = 0;
1290 
1291 	kmdp = preload_search_by_type("elf kernel");
1292 	if (kmdp == NULL)
1293 		kmdp = preload_search_by_type("elf64 kernel");
1294 	efifb = (struct efi_fb *)preload_search_info(kmdp,
1295 	    MODINFO_METADATA | MODINFOMD_EFI_FB);
1296 #if !defined(__aarch64__)
1297 	vbefb = (struct vbe_fb *)preload_search_info(kmdp,
1298 	    MODINFO_METADATA | MODINFOMD_VBE_FB);
1299 #endif /* aarch64 */
1300 	if (efifb != NULL) {
1301 		fb_start = efifb->fb_addr;
1302 		fb_end = efifb->fb_addr + efifb->fb_size;
1303 		fb_count = efifb->fb_size;
1304 		fb_height = efifb->fb_height;
1305 		fb_width = efifb->fb_width;
1306 	}
1307 #if !defined(__aarch64__)
1308 	else if (vbefb != NULL) {
1309 		fb_start = vbefb->fb_addr;
1310 		fb_end = vbefb->fb_addr + vbefb->fb_size;
1311 		fb_count = vbefb->fb_size;
1312 		fb_height = vbefb->fb_height;
1313 		fb_width = vbefb->fb_width;
1314 	}
1315 #endif /* aarch64 */
1316 	else {
1317 		if (bootverbose)
1318 			device_printf(dev,
1319 			    "no preloaded kernel fb information\n");
1320 		/* We are on Gen1 VM, just return. */
1321 		return;
1322 	}
1323 
1324 	if (bootverbose)
1325 		device_printf(dev,
1326 		    "fb: fb_addr: %#jx, size: %#jx, "
1327 		    "actual size needed: 0x%x\n",
1328 		    fb_start, fb_count, fb_height * fb_width);
1329 
1330 	hv_fb_res = pcib_host_res_alloc(&sc->vmbus_mmio_res, dev,
1331 	    SYS_RES_MEMORY, &rid, fb_start, fb_end, fb_count,
1332 	    RF_ACTIVE | rman_make_alignment_flags(PAGE_SIZE));
1333 
1334 	if (hv_fb_res && bootverbose)
1335 		device_printf(dev,
1336 		    "successfully reserved memory for framebuffer "
1337 		    "starting at %#jx, size %#jx\n",
1338 		    fb_start, fb_count);
1339 }
1340 
1341 static void
1342 vmbus_free_mmio_res(device_t dev)
1343 {
1344 	struct vmbus_softc *sc = device_get_softc(dev);
1345 
1346 	pcib_host_res_free(dev, &sc->vmbus_mmio_res);
1347 
1348 	if (hv_fb_res)
1349 		hv_fb_res = NULL;
1350 }
1351 #endif	/* NEW_PCIB */
1352 
1353 static void
1354 vmbus_identify(driver_t *driver, device_t parent)
1355 {
1356 
1357 	if (device_get_unit(parent) != 0 || vm_guest != VM_GUEST_HV ||
1358 	    (hyperv_features & CPUID_HV_MSR_SYNIC) == 0)
1359 		return;
1360 	device_add_child(parent, "vmbus", -1);
1361 }
1362 
1363 static int
1364 vmbus_probe(device_t dev)
1365 {
1366 
1367 	if (device_get_unit(dev) != 0 || vm_guest != VM_GUEST_HV ||
1368 	    (hyperv_features & CPUID_HV_MSR_SYNIC) == 0)
1369 		return (ENXIO);
1370 
1371 	device_set_desc(dev, "Hyper-V Vmbus");
1372 	return (BUS_PROBE_DEFAULT);
1373 }
1374 
1375 /**
1376  * @brief Main vmbus driver initialization routine.
1377  *
1378  * Here, we
1379  * - initialize the vmbus driver context
1380  * - setup various driver entry points
1381  * - invoke the vmbus hv main init routine
1382  * - get the irq resource
1383  * - invoke the vmbus to add the vmbus root device
1384  * - setup the vmbus root device
1385  * - retrieve the channel offers
1386  */
1387 static int
1388 vmbus_doattach(struct vmbus_softc *sc)
1389 {
1390 	struct sysctl_oid_list *child;
1391 	struct sysctl_ctx_list *ctx;
1392 	int ret;
1393 
1394 	if (sc->vmbus_flags & VMBUS_FLAG_ATTACHED)
1395 		return (0);
1396 
1397 #ifdef NEW_PCIB
1398 	vmbus_get_mmio_res(sc->vmbus_dev);
1399 	vmbus_fb_mmio_res(sc->vmbus_dev);
1400 #endif
1401 
1402 	sc->vmbus_flags |= VMBUS_FLAG_ATTACHED;
1403 
1404 	sc->vmbus_gpadl = VMBUS_GPADL_START;
1405 	mtx_init(&sc->vmbus_prichan_lock, "vmbus prichan", NULL, MTX_DEF);
1406 	TAILQ_INIT(&sc->vmbus_prichans);
1407 	mtx_init(&sc->vmbus_chan_lock, "vmbus channel", NULL, MTX_DEF);
1408 	TAILQ_INIT(&sc->vmbus_chans);
1409 	sc->vmbus_chmap = malloc(
1410 	    sizeof(struct vmbus_channel *) * VMBUS_CHAN_MAX, M_DEVBUF,
1411 	    M_WAITOK | M_ZERO);
1412 
1413 	/*
1414 	 * Create context for "post message" Hypercalls
1415 	 */
1416 	sc->vmbus_xc = vmbus_xact_ctx_create(bus_get_dma_tag(sc->vmbus_dev),
1417 	    HYPERCALL_POSTMSGIN_SIZE, VMBUS_MSG_SIZE,
1418 	    sizeof(struct vmbus_msghc));
1419 	if (sc->vmbus_xc == NULL) {
1420 		ret = ENXIO;
1421 		goto cleanup;
1422 	}
1423 
1424 	/*
1425 	 * Allocate DMA stuffs.
1426 	 */
1427 	ret = vmbus_dma_alloc(sc);
1428 	if (ret != 0)
1429 		goto cleanup;
1430 
1431 	/*
1432 	 * Setup interrupt.
1433 	 */
1434 	ret = vmbus_intr_setup(sc);
1435 	if (ret != 0)
1436 		goto cleanup;
1437 
1438 	/*
1439 	 * Setup SynIC.
1440 	 */
1441 	if (bootverbose)
1442 		device_printf(sc->vmbus_dev, "smp_started = %d\n", smp_started);
1443 	smp_rendezvous(NULL, vmbus_synic_setup, NULL, sc);
1444 	sc->vmbus_flags |= VMBUS_FLAG_SYNIC;
1445 
1446 	/*
1447 	 * Initialize vmbus, e.g. connect to Hypervisor.
1448 	 */
1449 	ret = vmbus_init(sc);
1450 	if (ret != 0)
1451 		goto cleanup;
1452 
1453 	if (sc->vmbus_version == VMBUS_VERSION_WS2008 ||
1454 	    sc->vmbus_version == VMBUS_VERSION_WIN7)
1455 		sc->vmbus_event_proc = vmbus_event_proc_compat;
1456 	else
1457 		sc->vmbus_event_proc = vmbus_event_proc;
1458 
1459 	ret = vmbus_scan(sc);
1460 	if (ret != 0)
1461 		goto cleanup;
1462 
1463 	ctx = device_get_sysctl_ctx(sc->vmbus_dev);
1464 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->vmbus_dev));
1465 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "version",
1466 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
1467 	    vmbus_sysctl_version, "A", "vmbus version");
1468 
1469 	return (ret);
1470 
1471 cleanup:
1472 	vmbus_scan_teardown(sc);
1473 	vmbus_intr_teardown(sc);
1474 	vmbus_dma_free(sc);
1475 	if (sc->vmbus_xc != NULL) {
1476 		vmbus_xact_ctx_destroy(sc->vmbus_xc);
1477 		sc->vmbus_xc = NULL;
1478 	}
1479 	free(__DEVOLATILE(void *, sc->vmbus_chmap), M_DEVBUF);
1480 	mtx_destroy(&sc->vmbus_prichan_lock);
1481 	mtx_destroy(&sc->vmbus_chan_lock);
1482 
1483 	return (ret);
1484 }
1485 
1486 static void
1487 vmbus_event_proc_dummy(struct vmbus_softc *sc __unused, int cpu __unused)
1488 {
1489 }
1490 
1491 #if defined(EARLY_AP_STARTUP) || defined(__aarch64__)
1492 
1493 static void
1494 vmbus_intrhook(void *xsc)
1495 {
1496 	struct vmbus_softc *sc = xsc;
1497 
1498 	if (bootverbose)
1499 		device_printf(sc->vmbus_dev, "intrhook\n");
1500 	vmbus_doattach(sc);
1501 	config_intrhook_disestablish(&sc->vmbus_intrhook);
1502 }
1503 
1504 #endif /* EARLY_AP_STARTUP  aarch64 */
1505 
1506 static int
1507 vmbus_attach(device_t dev)
1508 {
1509 	vmbus_sc = device_get_softc(dev);
1510 	vmbus_sc->vmbus_dev = dev;
1511 	vmbus_sc->vmbus_idtvec = -1;
1512 
1513 	/*
1514 	 * Event processing logic will be configured:
1515 	 * - After the vmbus protocol version negotiation.
1516 	 * - Before we request channel offers.
1517 	 */
1518 	vmbus_sc->vmbus_event_proc = vmbus_event_proc_dummy;
1519 
1520 #if defined(EARLY_AP_STARTUP) || defined(__aarch64__)
1521 	/*
1522 	 * Defer the real attach until the pause(9) works as expected.
1523 	 */
1524 	vmbus_sc->vmbus_intrhook.ich_func = vmbus_intrhook;
1525 	vmbus_sc->vmbus_intrhook.ich_arg = vmbus_sc;
1526 	config_intrhook_establish(&vmbus_sc->vmbus_intrhook);
1527 #else	/* !EARLY_AP_STARTUP */
1528 	/*
1529 	 * If the system has already booted and thread
1530 	 * scheduling is possible indicated by the global
1531 	 * cold set to zero, we just call the driver
1532 	 * initialization directly.
1533 	 */
1534 	if (!cold)
1535 		vmbus_doattach(vmbus_sc);
1536 #endif /* EARLY_AP_STARTUP  and aarch64 */
1537 
1538 	return (0);
1539 }
1540 
1541 static int
1542 vmbus_detach(device_t dev)
1543 {
1544 	struct vmbus_softc *sc = device_get_softc(dev);
1545 
1546 	bus_generic_detach(dev);
1547 	vmbus_chan_destroy_all(sc);
1548 
1549 	vmbus_scan_teardown(sc);
1550 
1551 	vmbus_disconnect(sc);
1552 
1553 	if (sc->vmbus_flags & VMBUS_FLAG_SYNIC) {
1554 		sc->vmbus_flags &= ~VMBUS_FLAG_SYNIC;
1555 		smp_rendezvous(NULL, vmbus_synic_teardown, NULL, NULL);
1556 	}
1557 
1558 	vmbus_intr_teardown(sc);
1559 	vmbus_dma_free(sc);
1560 
1561 	if (sc->vmbus_xc != NULL) {
1562 		vmbus_xact_ctx_destroy(sc->vmbus_xc);
1563 		sc->vmbus_xc = NULL;
1564 	}
1565 
1566 	free(__DEVOLATILE(void *, sc->vmbus_chmap), M_DEVBUF);
1567 	mtx_destroy(&sc->vmbus_prichan_lock);
1568 	mtx_destroy(&sc->vmbus_chan_lock);
1569 
1570 #ifdef NEW_PCIB
1571 	vmbus_free_mmio_res(dev);
1572 #endif
1573 
1574 #if defined(__aarch64__)
1575 	bus_release_resource(device_get_parent(dev), SYS_RES_IRQ, sc->vector,
1576 	    sc->ires);
1577 #endif
1578 	return (0);
1579 }
1580 
1581 #if !defined(EARLY_AP_STARTUP) && !defined(__aarch64__)
1582 
1583 static void
1584 vmbus_sysinit(void *arg __unused)
1585 {
1586 	struct vmbus_softc *sc = vmbus_get_softc();
1587 
1588 	if (vm_guest != VM_GUEST_HV || sc == NULL)
1589 		return;
1590 
1591 	/*
1592 	 * If the system has already booted and thread
1593 	 * scheduling is possible, as indicated by the
1594 	 * global cold set to zero, we just call the driver
1595 	 * initialization directly.
1596 	 */
1597 	if (!cold)
1598 		vmbus_doattach(sc);
1599 }
1600 /*
1601  * NOTE:
1602  * We have to start as the last step of SI_SUB_SMP, i.e. after SMP is
1603  * initialized.
1604  */
1605 SYSINIT(vmbus_initialize, SI_SUB_SMP, SI_ORDER_ANY, vmbus_sysinit, NULL);
1606 #endif	/* !EARLY_AP_STARTUP */
1607