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/smp.h>
44#include <sys/sysctl.h>
45#include <sys/systm.h>
46#include <sys/taskqueue.h>
47
48#include <machine/bus.h>
49#include <machine/intr_machdep.h>
50#include <machine/metadata.h>
51#include <machine/md_var.h>
52#include <machine/resource.h>
53#include <x86/include/apicvar.h>
54
55#include <contrib/dev/acpica/include/acpi.h>
56#include <dev/acpica/acpivar.h>
57
58#include <dev/hyperv/include/hyperv.h>
59#include <dev/hyperv/include/vmbus_xact.h>
60#include <dev/hyperv/vmbus/hyperv_reg.h>
61#include <dev/hyperv/vmbus/hyperv_var.h>
62#include <dev/hyperv/vmbus/vmbus_reg.h>
63#include <dev/hyperv/vmbus/vmbus_var.h>
64#include <dev/hyperv/vmbus/vmbus_chanvar.h>
65
66#include "acpi_if.h"
67#include "pcib_if.h"
68#include "vmbus_if.h"
69
70#define VMBUS_GPADL_START		0xe1e10
71
72struct vmbus_msghc {
73	struct vmbus_xact		*mh_xact;
74	struct hypercall_postmsg_in	mh_inprm_save;
75};
76
77static void			vmbus_identify(driver_t *, device_t);
78static int			vmbus_probe(device_t);
79static int			vmbus_attach(device_t);
80static int			vmbus_detach(device_t);
81static int			vmbus_read_ivar(device_t, device_t, int,
82				    uintptr_t *);
83static int			vmbus_child_pnpinfo_str(device_t, device_t,
84				    char *, size_t);
85static struct resource		*vmbus_alloc_resource(device_t dev,
86				    device_t child, int type, int *rid,
87				    rman_res_t start, rman_res_t end,
88				    rman_res_t count, u_int flags);
89static int			vmbus_alloc_msi(device_t bus, device_t dev,
90				    int count, int maxcount, int *irqs);
91static int			vmbus_release_msi(device_t bus, device_t dev,
92				    int count, int *irqs);
93static int			vmbus_alloc_msix(device_t bus, device_t dev,
94				    int *irq);
95static int			vmbus_release_msix(device_t bus, device_t dev,
96				    int irq);
97static int			vmbus_map_msi(device_t bus, device_t dev,
98				    int irq, uint64_t *addr, uint32_t *data);
99static uint32_t			vmbus_get_version_method(device_t, device_t);
100static int			vmbus_probe_guid_method(device_t, device_t,
101				    const struct hyperv_guid *);
102static uint32_t			vmbus_get_vcpu_id_method(device_t bus,
103				    device_t dev, int cpu);
104static struct taskqueue		*vmbus_get_eventtq_method(device_t, device_t,
105				    int);
106#ifdef EARLY_AP_STARTUP
107static void			vmbus_intrhook(void *);
108#endif
109
110static int			vmbus_init(struct vmbus_softc *);
111static int			vmbus_connect(struct vmbus_softc *, uint32_t);
112static int			vmbus_req_channels(struct vmbus_softc *sc);
113static void			vmbus_disconnect(struct vmbus_softc *);
114static int			vmbus_scan(struct vmbus_softc *);
115static void			vmbus_scan_teardown(struct vmbus_softc *);
116static void			vmbus_scan_done(struct vmbus_softc *,
117				    const struct vmbus_message *);
118static void			vmbus_chanmsg_handle(struct vmbus_softc *,
119				    const struct vmbus_message *);
120static void			vmbus_msg_task(void *, int);
121static void			vmbus_synic_setup(void *);
122static void			vmbus_synic_teardown(void *);
123static int			vmbus_sysctl_version(SYSCTL_HANDLER_ARGS);
124static int			vmbus_dma_alloc(struct vmbus_softc *);
125static void			vmbus_dma_free(struct vmbus_softc *);
126static int			vmbus_intr_setup(struct vmbus_softc *);
127static void			vmbus_intr_teardown(struct vmbus_softc *);
128static int			vmbus_doattach(struct vmbus_softc *);
129static void			vmbus_event_proc_dummy(struct vmbus_softc *,
130				    int);
131
132static struct vmbus_softc	*vmbus_sc;
133
134SYSCTL_NODE(_hw, OID_AUTO, vmbus, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
135    "Hyper-V vmbus");
136
137static int			vmbus_pin_evttask = 1;
138SYSCTL_INT(_hw_vmbus, OID_AUTO, pin_evttask, CTLFLAG_RDTUN,
139    &vmbus_pin_evttask, 0, "Pin event tasks to their respective CPU");
140
141extern inthand_t IDTVEC(vmbus_isr), IDTVEC(vmbus_isr_pti);
142
143uint32_t			vmbus_current_version;
144
145static const uint32_t		vmbus_version[] = {
146	VMBUS_VERSION_WIN10,
147	VMBUS_VERSION_WIN8_1,
148	VMBUS_VERSION_WIN8,
149	VMBUS_VERSION_WIN7,
150	VMBUS_VERSION_WS2008
151};
152
153static const vmbus_chanmsg_proc_t
154vmbus_chanmsg_handlers[VMBUS_CHANMSG_TYPE_MAX] = {
155	VMBUS_CHANMSG_PROC(CHOFFER_DONE, vmbus_scan_done),
156	VMBUS_CHANMSG_PROC_WAKEUP(CONNECT_RESP)
157};
158
159static device_method_t vmbus_methods[] = {
160	/* Device interface */
161	DEVMETHOD(device_identify,		vmbus_identify),
162	DEVMETHOD(device_probe,			vmbus_probe),
163	DEVMETHOD(device_attach,		vmbus_attach),
164	DEVMETHOD(device_detach,		vmbus_detach),
165	DEVMETHOD(device_shutdown,		bus_generic_shutdown),
166	DEVMETHOD(device_suspend,		bus_generic_suspend),
167	DEVMETHOD(device_resume,		bus_generic_resume),
168
169	/* Bus interface */
170	DEVMETHOD(bus_add_child,		bus_generic_add_child),
171	DEVMETHOD(bus_print_child,		bus_generic_print_child),
172	DEVMETHOD(bus_read_ivar,		vmbus_read_ivar),
173	DEVMETHOD(bus_child_pnpinfo_str,	vmbus_child_pnpinfo_str),
174	DEVMETHOD(bus_alloc_resource,		vmbus_alloc_resource),
175	DEVMETHOD(bus_release_resource,		bus_generic_release_resource),
176	DEVMETHOD(bus_activate_resource,	bus_generic_activate_resource),
177	DEVMETHOD(bus_deactivate_resource,	bus_generic_deactivate_resource),
178	DEVMETHOD(bus_setup_intr,		bus_generic_setup_intr),
179	DEVMETHOD(bus_teardown_intr,		bus_generic_teardown_intr),
180#if __FreeBSD_version >= 1100000
181	DEVMETHOD(bus_get_cpus,			bus_generic_get_cpus),
182#endif
183
184	/* pcib interface */
185	DEVMETHOD(pcib_alloc_msi,		vmbus_alloc_msi),
186	DEVMETHOD(pcib_release_msi,		vmbus_release_msi),
187	DEVMETHOD(pcib_alloc_msix,		vmbus_alloc_msix),
188	DEVMETHOD(pcib_release_msix,		vmbus_release_msix),
189	DEVMETHOD(pcib_map_msi,			vmbus_map_msi),
190
191	/* Vmbus interface */
192	DEVMETHOD(vmbus_get_version,		vmbus_get_version_method),
193	DEVMETHOD(vmbus_probe_guid,		vmbus_probe_guid_method),
194	DEVMETHOD(vmbus_get_vcpu_id,		vmbus_get_vcpu_id_method),
195	DEVMETHOD(vmbus_get_event_taskq,	vmbus_get_eventtq_method),
196
197	DEVMETHOD_END
198};
199
200static driver_t vmbus_driver = {
201	"vmbus",
202	vmbus_methods,
203	sizeof(struct vmbus_softc)
204};
205
206static devclass_t vmbus_devclass;
207
208DRIVER_MODULE(vmbus, pcib, vmbus_driver, vmbus_devclass, NULL, NULL);
209DRIVER_MODULE(vmbus, acpi_syscontainer, vmbus_driver, vmbus_devclass,
210    NULL, NULL);
211
212MODULE_DEPEND(vmbus, acpi, 1, 1, 1);
213MODULE_DEPEND(vmbus, pci, 1, 1, 1);
214MODULE_VERSION(vmbus, 1);
215
216static __inline struct vmbus_softc *
217vmbus_get_softc(void)
218{
219	return vmbus_sc;
220}
221
222void
223vmbus_msghc_reset(struct vmbus_msghc *mh, size_t dsize)
224{
225	struct hypercall_postmsg_in *inprm;
226
227	if (dsize > HYPERCALL_POSTMSGIN_DSIZE_MAX)
228		panic("invalid data size %zu", dsize);
229
230	inprm = vmbus_xact_req_data(mh->mh_xact);
231	memset(inprm, 0, HYPERCALL_POSTMSGIN_SIZE);
232	inprm->hc_connid = VMBUS_CONNID_MESSAGE;
233	inprm->hc_msgtype = HYPERV_MSGTYPE_CHANNEL;
234	inprm->hc_dsize = dsize;
235}
236
237struct vmbus_msghc *
238vmbus_msghc_get(struct vmbus_softc *sc, size_t dsize)
239{
240	struct vmbus_msghc *mh;
241	struct vmbus_xact *xact;
242
243	if (dsize > HYPERCALL_POSTMSGIN_DSIZE_MAX)
244		panic("invalid data size %zu", dsize);
245
246	xact = vmbus_xact_get(sc->vmbus_xc,
247	    dsize + __offsetof(struct hypercall_postmsg_in, hc_data[0]));
248	if (xact == NULL)
249		return (NULL);
250
251	mh = vmbus_xact_priv(xact, sizeof(*mh));
252	mh->mh_xact = xact;
253
254	vmbus_msghc_reset(mh, dsize);
255	return (mh);
256}
257
258void
259vmbus_msghc_put(struct vmbus_softc *sc __unused, struct vmbus_msghc *mh)
260{
261
262	vmbus_xact_put(mh->mh_xact);
263}
264
265void *
266vmbus_msghc_dataptr(struct vmbus_msghc *mh)
267{
268	struct hypercall_postmsg_in *inprm;
269
270	inprm = vmbus_xact_req_data(mh->mh_xact);
271	return (inprm->hc_data);
272}
273
274int
275vmbus_msghc_exec_noresult(struct vmbus_msghc *mh)
276{
277	sbintime_t time = SBT_1MS;
278	struct hypercall_postmsg_in *inprm;
279	bus_addr_t inprm_paddr;
280	int i;
281
282	inprm = vmbus_xact_req_data(mh->mh_xact);
283	inprm_paddr = vmbus_xact_req_paddr(mh->mh_xact);
284
285	/*
286	 * Save the input parameter so that we could restore the input
287	 * parameter if the Hypercall failed.
288	 *
289	 * XXX
290	 * Is this really necessary?!  i.e. Will the Hypercall ever
291	 * overwrite the input parameter?
292	 */
293	memcpy(&mh->mh_inprm_save, inprm, HYPERCALL_POSTMSGIN_SIZE);
294
295	/*
296	 * In order to cope with transient failures, e.g. insufficient
297	 * resources on host side, we retry the post message Hypercall
298	 * several times.  20 retries seem sufficient.
299	 */
300#define HC_RETRY_MAX	20
301
302	for (i = 0; i < HC_RETRY_MAX; ++i) {
303		uint64_t status;
304
305		status = hypercall_post_message(inprm_paddr);
306		if (status == HYPERCALL_STATUS_SUCCESS)
307			return 0;
308
309		pause_sbt("hcpmsg", time, 0, C_HARDCLOCK);
310		if (time < SBT_1S * 2)
311			time *= 2;
312
313		/* Restore input parameter and try again */
314		memcpy(inprm, &mh->mh_inprm_save, HYPERCALL_POSTMSGIN_SIZE);
315	}
316
317#undef HC_RETRY_MAX
318
319	return EIO;
320}
321
322int
323vmbus_msghc_exec(struct vmbus_softc *sc __unused, struct vmbus_msghc *mh)
324{
325	int error;
326
327	vmbus_xact_activate(mh->mh_xact);
328	error = vmbus_msghc_exec_noresult(mh);
329	if (error)
330		vmbus_xact_deactivate(mh->mh_xact);
331	return error;
332}
333
334void
335vmbus_msghc_exec_cancel(struct vmbus_softc *sc __unused, struct vmbus_msghc *mh)
336{
337
338	vmbus_xact_deactivate(mh->mh_xact);
339}
340
341const struct vmbus_message *
342vmbus_msghc_wait_result(struct vmbus_softc *sc __unused, struct vmbus_msghc *mh)
343{
344	size_t resp_len;
345
346	return (vmbus_xact_wait(mh->mh_xact, &resp_len));
347}
348
349const struct vmbus_message *
350vmbus_msghc_poll_result(struct vmbus_softc *sc __unused, struct vmbus_msghc *mh)
351{
352	size_t resp_len;
353
354	return (vmbus_xact_poll(mh->mh_xact, &resp_len));
355}
356
357void
358vmbus_msghc_wakeup(struct vmbus_softc *sc, const struct vmbus_message *msg)
359{
360
361	vmbus_xact_ctx_wakeup(sc->vmbus_xc, msg, sizeof(*msg));
362}
363
364uint32_t
365vmbus_gpadl_alloc(struct vmbus_softc *sc)
366{
367	uint32_t gpadl;
368
369again:
370	gpadl = atomic_fetchadd_int(&sc->vmbus_gpadl, 1);
371	if (gpadl == 0)
372		goto again;
373	return (gpadl);
374}
375
376static int
377vmbus_connect(struct vmbus_softc *sc, uint32_t version)
378{
379	struct vmbus_chanmsg_connect *req;
380	const struct vmbus_message *msg;
381	struct vmbus_msghc *mh;
382	int error, done = 0;
383
384	mh = vmbus_msghc_get(sc, sizeof(*req));
385	if (mh == NULL)
386		return ENXIO;
387
388	req = vmbus_msghc_dataptr(mh);
389	req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_CONNECT;
390	req->chm_ver = version;
391	req->chm_evtflags = sc->vmbus_evtflags_dma.hv_paddr;
392	req->chm_mnf1 = sc->vmbus_mnf1_dma.hv_paddr;
393	req->chm_mnf2 = sc->vmbus_mnf2_dma.hv_paddr;
394
395	error = vmbus_msghc_exec(sc, mh);
396	if (error) {
397		vmbus_msghc_put(sc, mh);
398		return error;
399	}
400
401	msg = vmbus_msghc_wait_result(sc, mh);
402	done = ((const struct vmbus_chanmsg_connect_resp *)
403	    msg->msg_data)->chm_done;
404
405	vmbus_msghc_put(sc, mh);
406
407	return (done ? 0 : EOPNOTSUPP);
408}
409
410static int
411vmbus_init(struct vmbus_softc *sc)
412{
413	int i;
414
415	for (i = 0; i < nitems(vmbus_version); ++i) {
416		int error;
417
418		error = vmbus_connect(sc, vmbus_version[i]);
419		if (!error) {
420			vmbus_current_version = vmbus_version[i];
421			sc->vmbus_version = vmbus_version[i];
422			device_printf(sc->vmbus_dev, "version %u.%u\n",
423			    VMBUS_VERSION_MAJOR(sc->vmbus_version),
424			    VMBUS_VERSION_MINOR(sc->vmbus_version));
425			return 0;
426		}
427	}
428	return ENXIO;
429}
430
431static void
432vmbus_disconnect(struct vmbus_softc *sc)
433{
434	struct vmbus_chanmsg_disconnect *req;
435	struct vmbus_msghc *mh;
436	int error;
437
438	mh = vmbus_msghc_get(sc, sizeof(*req));
439	if (mh == NULL) {
440		device_printf(sc->vmbus_dev,
441		    "can not get msg hypercall for disconnect\n");
442		return;
443	}
444
445	req = vmbus_msghc_dataptr(mh);
446	req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_DISCONNECT;
447
448	error = vmbus_msghc_exec_noresult(mh);
449	vmbus_msghc_put(sc, mh);
450
451	if (error) {
452		device_printf(sc->vmbus_dev,
453		    "disconnect msg hypercall failed\n");
454	}
455}
456
457static int
458vmbus_req_channels(struct vmbus_softc *sc)
459{
460	struct vmbus_chanmsg_chrequest *req;
461	struct vmbus_msghc *mh;
462	int error;
463
464	mh = vmbus_msghc_get(sc, sizeof(*req));
465	if (mh == NULL)
466		return ENXIO;
467
468	req = vmbus_msghc_dataptr(mh);
469	req->chm_hdr.chm_type = VMBUS_CHANMSG_TYPE_CHREQUEST;
470
471	error = vmbus_msghc_exec_noresult(mh);
472	vmbus_msghc_put(sc, mh);
473
474	return error;
475}
476
477static void
478vmbus_scan_done_task(void *xsc, int pending __unused)
479{
480	struct vmbus_softc *sc = xsc;
481
482	mtx_lock(&Giant);
483	sc->vmbus_scandone = true;
484	mtx_unlock(&Giant);
485	wakeup(&sc->vmbus_scandone);
486}
487
488static void
489vmbus_scan_done(struct vmbus_softc *sc,
490    const struct vmbus_message *msg __unused)
491{
492
493	taskqueue_enqueue(sc->vmbus_devtq, &sc->vmbus_scandone_task);
494}
495
496static int
497vmbus_scan(struct vmbus_softc *sc)
498{
499	int error;
500
501	/*
502	 * Identify, probe and attach for non-channel devices.
503	 */
504	bus_generic_probe(sc->vmbus_dev);
505	bus_generic_attach(sc->vmbus_dev);
506
507	/*
508	 * This taskqueue serializes vmbus devices' attach and detach
509	 * for channel offer and rescind messages.
510	 */
511	sc->vmbus_devtq = taskqueue_create("vmbus dev", M_WAITOK,
512	    taskqueue_thread_enqueue, &sc->vmbus_devtq);
513	taskqueue_start_threads(&sc->vmbus_devtq, 1, PI_NET, "vmbusdev");
514	TASK_INIT(&sc->vmbus_scandone_task, 0, vmbus_scan_done_task, sc);
515
516	/*
517	 * This taskqueue handles sub-channel detach, so that vmbus
518	 * device's detach running in vmbus_devtq can drain its sub-
519	 * channels.
520	 */
521	sc->vmbus_subchtq = taskqueue_create("vmbus subch", M_WAITOK,
522	    taskqueue_thread_enqueue, &sc->vmbus_subchtq);
523	taskqueue_start_threads(&sc->vmbus_subchtq, 1, PI_NET, "vmbussch");
524
525	/*
526	 * Start vmbus scanning.
527	 */
528	error = vmbus_req_channels(sc);
529	if (error) {
530		device_printf(sc->vmbus_dev, "channel request failed: %d\n",
531		    error);
532		return (error);
533	}
534
535	/*
536	 * Wait for all vmbus devices from the initial channel offers to be
537	 * attached.
538	 */
539	GIANT_REQUIRED;
540	while (!sc->vmbus_scandone)
541		mtx_sleep(&sc->vmbus_scandone, &Giant, 0, "vmbusdev", 0);
542
543	if (bootverbose) {
544		device_printf(sc->vmbus_dev, "device scan, probe and attach "
545		    "done\n");
546	}
547	return (0);
548}
549
550static void
551vmbus_scan_teardown(struct vmbus_softc *sc)
552{
553
554	GIANT_REQUIRED;
555	if (sc->vmbus_devtq != NULL) {
556		mtx_unlock(&Giant);
557		taskqueue_free(sc->vmbus_devtq);
558		mtx_lock(&Giant);
559		sc->vmbus_devtq = NULL;
560	}
561	if (sc->vmbus_subchtq != NULL) {
562		mtx_unlock(&Giant);
563		taskqueue_free(sc->vmbus_subchtq);
564		mtx_lock(&Giant);
565		sc->vmbus_subchtq = NULL;
566	}
567}
568
569static void
570vmbus_chanmsg_handle(struct vmbus_softc *sc, const struct vmbus_message *msg)
571{
572	vmbus_chanmsg_proc_t msg_proc;
573	uint32_t msg_type;
574
575	msg_type = ((const struct vmbus_chanmsg_hdr *)msg->msg_data)->chm_type;
576	if (msg_type >= VMBUS_CHANMSG_TYPE_MAX) {
577		device_printf(sc->vmbus_dev, "unknown message type 0x%x\n",
578		    msg_type);
579		return;
580	}
581
582	msg_proc = vmbus_chanmsg_handlers[msg_type];
583	if (msg_proc != NULL)
584		msg_proc(sc, msg);
585
586	/* Channel specific processing */
587	vmbus_chan_msgproc(sc, msg);
588}
589
590static void
591vmbus_msg_task(void *xsc, int pending __unused)
592{
593	struct vmbus_softc *sc = xsc;
594	volatile struct vmbus_message *msg;
595
596	msg = VMBUS_PCPU_GET(sc, message, curcpu) + VMBUS_SINT_MESSAGE;
597	for (;;) {
598		if (msg->msg_type == HYPERV_MSGTYPE_NONE) {
599			/* No message */
600			break;
601		} else if (msg->msg_type == HYPERV_MSGTYPE_CHANNEL) {
602			/* Channel message */
603			vmbus_chanmsg_handle(sc,
604			    __DEVOLATILE(const struct vmbus_message *, msg));
605		}
606
607		msg->msg_type = HYPERV_MSGTYPE_NONE;
608		/*
609		 * Make sure the write to msg_type (i.e. set to
610		 * HYPERV_MSGTYPE_NONE) happens before we read the
611		 * msg_flags and EOMing. Otherwise, the EOMing will
612		 * not deliver any more messages since there is no
613		 * empty slot
614		 *
615		 * NOTE:
616		 * mb() is used here, since atomic_thread_fence_seq_cst()
617		 * will become compiler fence on UP kernel.
618		 */
619		mb();
620		if (msg->msg_flags & VMBUS_MSGFLAG_PENDING) {
621			/*
622			 * This will cause message queue rescan to possibly
623			 * deliver another msg from the hypervisor
624			 */
625			wrmsr(MSR_HV_EOM, 0);
626		}
627	}
628}
629
630static __inline int
631vmbus_handle_intr1(struct vmbus_softc *sc, struct trapframe *frame, int cpu)
632{
633	volatile struct vmbus_message *msg;
634	struct vmbus_message *msg_base;
635
636	msg_base = VMBUS_PCPU_GET(sc, message, cpu);
637
638	/*
639	 * Check event timer.
640	 *
641	 * TODO: move this to independent IDT vector.
642	 */
643	msg = msg_base + VMBUS_SINT_TIMER;
644	if (msg->msg_type == HYPERV_MSGTYPE_TIMER_EXPIRED) {
645		msg->msg_type = HYPERV_MSGTYPE_NONE;
646
647		vmbus_et_intr(frame);
648
649		/*
650		 * Make sure the write to msg_type (i.e. set to
651		 * HYPERV_MSGTYPE_NONE) happens before we read the
652		 * msg_flags and EOMing. Otherwise, the EOMing will
653		 * not deliver any more messages since there is no
654		 * empty slot
655		 *
656		 * NOTE:
657		 * mb() is used here, since atomic_thread_fence_seq_cst()
658		 * will become compiler fence on UP kernel.
659		 */
660		mb();
661		if (msg->msg_flags & VMBUS_MSGFLAG_PENDING) {
662			/*
663			 * This will cause message queue rescan to possibly
664			 * deliver another msg from the hypervisor
665			 */
666			wrmsr(MSR_HV_EOM, 0);
667		}
668	}
669
670	/*
671	 * Check events.  Hot path for network and storage I/O data; high rate.
672	 *
673	 * NOTE:
674	 * As recommended by the Windows guest fellows, we check events before
675	 * checking messages.
676	 */
677	sc->vmbus_event_proc(sc, cpu);
678
679	/*
680	 * Check messages.  Mainly management stuffs; ultra low rate.
681	 */
682	msg = msg_base + VMBUS_SINT_MESSAGE;
683	if (__predict_false(msg->msg_type != HYPERV_MSGTYPE_NONE)) {
684		taskqueue_enqueue(VMBUS_PCPU_GET(sc, message_tq, cpu),
685		    VMBUS_PCPU_PTR(sc, message_task, cpu));
686	}
687
688	return (FILTER_HANDLED);
689}
690
691void
692vmbus_handle_intr(struct trapframe *trap_frame)
693{
694	struct vmbus_softc *sc = vmbus_get_softc();
695	int cpu = curcpu;
696
697	/*
698	 * Disable preemption.
699	 */
700	critical_enter();
701
702	/*
703	 * Do a little interrupt counting.
704	 */
705	(*VMBUS_PCPU_GET(sc, intr_cnt, cpu))++;
706
707	vmbus_handle_intr1(sc, trap_frame, cpu);
708
709	/*
710	 * Enable preemption.
711	 */
712	critical_exit();
713}
714
715static void
716vmbus_synic_setup(void *xsc)
717{
718	struct vmbus_softc *sc = xsc;
719	int cpu = curcpu;
720	uint64_t val, orig;
721	uint32_t sint;
722
723	if (hyperv_features & CPUID_HV_MSR_VP_INDEX) {
724		/* Save virtual processor id. */
725		VMBUS_PCPU_GET(sc, vcpuid, cpu) = rdmsr(MSR_HV_VP_INDEX);
726	} else {
727		/* Set virtual processor id to 0 for compatibility. */
728		VMBUS_PCPU_GET(sc, vcpuid, cpu) = 0;
729	}
730
731	/*
732	 * Setup the SynIC message.
733	 */
734	orig = rdmsr(MSR_HV_SIMP);
735	val = MSR_HV_SIMP_ENABLE | (orig & MSR_HV_SIMP_RSVD_MASK) |
736	    ((VMBUS_PCPU_GET(sc, message_dma.hv_paddr, cpu) >> PAGE_SHIFT) <<
737	     MSR_HV_SIMP_PGSHIFT);
738	wrmsr(MSR_HV_SIMP, val);
739
740	/*
741	 * Setup the SynIC event flags.
742	 */
743	orig = rdmsr(MSR_HV_SIEFP);
744	val = MSR_HV_SIEFP_ENABLE | (orig & MSR_HV_SIEFP_RSVD_MASK) |
745	    ((VMBUS_PCPU_GET(sc, event_flags_dma.hv_paddr, cpu)
746	      >> PAGE_SHIFT) << MSR_HV_SIEFP_PGSHIFT);
747	wrmsr(MSR_HV_SIEFP, val);
748
749
750	/*
751	 * Configure and unmask SINT for message and event flags.
752	 */
753	sint = MSR_HV_SINT0 + VMBUS_SINT_MESSAGE;
754	orig = rdmsr(sint);
755	val = sc->vmbus_idtvec | MSR_HV_SINT_AUTOEOI |
756	    (orig & MSR_HV_SINT_RSVD_MASK);
757	wrmsr(sint, val);
758
759	/*
760	 * Configure and unmask SINT for timer.
761	 */
762	sint = MSR_HV_SINT0 + VMBUS_SINT_TIMER;
763	orig = rdmsr(sint);
764	val = sc->vmbus_idtvec | MSR_HV_SINT_AUTOEOI |
765	    (orig & MSR_HV_SINT_RSVD_MASK);
766	wrmsr(sint, val);
767
768	/*
769	 * All done; enable SynIC.
770	 */
771	orig = rdmsr(MSR_HV_SCONTROL);
772	val = MSR_HV_SCTRL_ENABLE | (orig & MSR_HV_SCTRL_RSVD_MASK);
773	wrmsr(MSR_HV_SCONTROL, val);
774}
775
776static void
777vmbus_synic_teardown(void *arg)
778{
779	uint64_t orig;
780	uint32_t sint;
781
782	/*
783	 * Disable SynIC.
784	 */
785	orig = rdmsr(MSR_HV_SCONTROL);
786	wrmsr(MSR_HV_SCONTROL, (orig & MSR_HV_SCTRL_RSVD_MASK));
787
788	/*
789	 * Mask message and event flags SINT.
790	 */
791	sint = MSR_HV_SINT0 + VMBUS_SINT_MESSAGE;
792	orig = rdmsr(sint);
793	wrmsr(sint, orig | MSR_HV_SINT_MASKED);
794
795	/*
796	 * Mask timer SINT.
797	 */
798	sint = MSR_HV_SINT0 + VMBUS_SINT_TIMER;
799	orig = rdmsr(sint);
800	wrmsr(sint, orig | MSR_HV_SINT_MASKED);
801
802	/*
803	 * Teardown SynIC message.
804	 */
805	orig = rdmsr(MSR_HV_SIMP);
806	wrmsr(MSR_HV_SIMP, (orig & MSR_HV_SIMP_RSVD_MASK));
807
808	/*
809	 * Teardown SynIC event flags.
810	 */
811	orig = rdmsr(MSR_HV_SIEFP);
812	wrmsr(MSR_HV_SIEFP, (orig & MSR_HV_SIEFP_RSVD_MASK));
813}
814
815static int
816vmbus_dma_alloc(struct vmbus_softc *sc)
817{
818	bus_dma_tag_t parent_dtag;
819	uint8_t *evtflags;
820	int cpu;
821
822	parent_dtag = bus_get_dma_tag(sc->vmbus_dev);
823	CPU_FOREACH(cpu) {
824		void *ptr;
825
826		/*
827		 * Per-cpu messages and event flags.
828		 */
829		ptr = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0,
830		    PAGE_SIZE, VMBUS_PCPU_PTR(sc, message_dma, cpu),
831		    BUS_DMA_WAITOK | BUS_DMA_ZERO);
832		if (ptr == NULL)
833			return ENOMEM;
834		VMBUS_PCPU_GET(sc, message, cpu) = ptr;
835
836		ptr = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0,
837		    PAGE_SIZE, VMBUS_PCPU_PTR(sc, event_flags_dma, cpu),
838		    BUS_DMA_WAITOK | BUS_DMA_ZERO);
839		if (ptr == NULL)
840			return ENOMEM;
841		VMBUS_PCPU_GET(sc, event_flags, cpu) = ptr;
842	}
843
844	evtflags = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0,
845	    PAGE_SIZE, &sc->vmbus_evtflags_dma, BUS_DMA_WAITOK | BUS_DMA_ZERO);
846	if (evtflags == NULL)
847		return ENOMEM;
848	sc->vmbus_rx_evtflags = (u_long *)evtflags;
849	sc->vmbus_tx_evtflags = (u_long *)(evtflags + (PAGE_SIZE / 2));
850	sc->vmbus_evtflags = evtflags;
851
852	sc->vmbus_mnf1 = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0,
853	    PAGE_SIZE, &sc->vmbus_mnf1_dma, BUS_DMA_WAITOK | BUS_DMA_ZERO);
854	if (sc->vmbus_mnf1 == NULL)
855		return ENOMEM;
856
857	sc->vmbus_mnf2 = hyperv_dmamem_alloc(parent_dtag, PAGE_SIZE, 0,
858	    sizeof(struct vmbus_mnf), &sc->vmbus_mnf2_dma,
859	    BUS_DMA_WAITOK | BUS_DMA_ZERO);
860	if (sc->vmbus_mnf2 == NULL)
861		return ENOMEM;
862
863	return 0;
864}
865
866static void
867vmbus_dma_free(struct vmbus_softc *sc)
868{
869	int cpu;
870
871	if (sc->vmbus_evtflags != NULL) {
872		hyperv_dmamem_free(&sc->vmbus_evtflags_dma, sc->vmbus_evtflags);
873		sc->vmbus_evtflags = NULL;
874		sc->vmbus_rx_evtflags = NULL;
875		sc->vmbus_tx_evtflags = NULL;
876	}
877	if (sc->vmbus_mnf1 != NULL) {
878		hyperv_dmamem_free(&sc->vmbus_mnf1_dma, sc->vmbus_mnf1);
879		sc->vmbus_mnf1 = NULL;
880	}
881	if (sc->vmbus_mnf2 != NULL) {
882		hyperv_dmamem_free(&sc->vmbus_mnf2_dma, sc->vmbus_mnf2);
883		sc->vmbus_mnf2 = NULL;
884	}
885
886	CPU_FOREACH(cpu) {
887		if (VMBUS_PCPU_GET(sc, message, cpu) != NULL) {
888			hyperv_dmamem_free(
889			    VMBUS_PCPU_PTR(sc, message_dma, cpu),
890			    VMBUS_PCPU_GET(sc, message, cpu));
891			VMBUS_PCPU_GET(sc, message, cpu) = NULL;
892		}
893		if (VMBUS_PCPU_GET(sc, event_flags, cpu) != NULL) {
894			hyperv_dmamem_free(
895			    VMBUS_PCPU_PTR(sc, event_flags_dma, cpu),
896			    VMBUS_PCPU_GET(sc, event_flags, cpu));
897			VMBUS_PCPU_GET(sc, event_flags, cpu) = NULL;
898		}
899	}
900}
901
902static int
903vmbus_intr_setup(struct vmbus_softc *sc)
904{
905	int cpu;
906
907	CPU_FOREACH(cpu) {
908		char buf[MAXCOMLEN + 1];
909		cpuset_t cpu_mask;
910
911		/* Allocate an interrupt counter for Hyper-V interrupt */
912		snprintf(buf, sizeof(buf), "cpu%d:hyperv", cpu);
913		intrcnt_add(buf, VMBUS_PCPU_PTR(sc, intr_cnt, cpu));
914
915		/*
916		 * Setup taskqueue to handle events.  Task will be per-
917		 * channel.
918		 */
919		VMBUS_PCPU_GET(sc, event_tq, cpu) = taskqueue_create_fast(
920		    "hyperv event", M_WAITOK, taskqueue_thread_enqueue,
921		    VMBUS_PCPU_PTR(sc, event_tq, cpu));
922		if (vmbus_pin_evttask) {
923			CPU_SETOF(cpu, &cpu_mask);
924			taskqueue_start_threads_cpuset(
925			    VMBUS_PCPU_PTR(sc, event_tq, cpu), 1, PI_NET,
926			    &cpu_mask, "hvevent%d", cpu);
927		} else {
928			taskqueue_start_threads(
929			    VMBUS_PCPU_PTR(sc, event_tq, cpu), 1, PI_NET,
930			    "hvevent%d", cpu);
931		}
932
933		/*
934		 * Setup tasks and taskqueues to handle messages.
935		 */
936		VMBUS_PCPU_GET(sc, message_tq, cpu) = taskqueue_create_fast(
937		    "hyperv msg", M_WAITOK, taskqueue_thread_enqueue,
938		    VMBUS_PCPU_PTR(sc, message_tq, cpu));
939		CPU_SETOF(cpu, &cpu_mask);
940		taskqueue_start_threads_cpuset(
941		    VMBUS_PCPU_PTR(sc, message_tq, cpu), 1, PI_NET, &cpu_mask,
942		    "hvmsg%d", cpu);
943		TASK_INIT(VMBUS_PCPU_PTR(sc, message_task, cpu), 0,
944		    vmbus_msg_task, sc);
945	}
946
947	/*
948	 * All Hyper-V ISR required resources are setup, now let's find a
949	 * free IDT vector for Hyper-V ISR and set it up.
950	 */
951	sc->vmbus_idtvec = lapic_ipi_alloc(pti ? IDTVEC(vmbus_isr_pti) :
952	    IDTVEC(vmbus_isr));
953	if (sc->vmbus_idtvec < 0) {
954		device_printf(sc->vmbus_dev, "cannot find free IDT vector\n");
955		return ENXIO;
956	}
957	if (bootverbose) {
958		device_printf(sc->vmbus_dev, "vmbus IDT vector %d\n",
959		    sc->vmbus_idtvec);
960	}
961	return 0;
962}
963
964static void
965vmbus_intr_teardown(struct vmbus_softc *sc)
966{
967	int cpu;
968
969	if (sc->vmbus_idtvec >= 0) {
970		lapic_ipi_free(sc->vmbus_idtvec);
971		sc->vmbus_idtvec = -1;
972	}
973
974	CPU_FOREACH(cpu) {
975		if (VMBUS_PCPU_GET(sc, event_tq, cpu) != NULL) {
976			taskqueue_free(VMBUS_PCPU_GET(sc, event_tq, cpu));
977			VMBUS_PCPU_GET(sc, event_tq, cpu) = NULL;
978		}
979		if (VMBUS_PCPU_GET(sc, message_tq, cpu) != NULL) {
980			taskqueue_drain(VMBUS_PCPU_GET(sc, message_tq, cpu),
981			    VMBUS_PCPU_PTR(sc, message_task, cpu));
982			taskqueue_free(VMBUS_PCPU_GET(sc, message_tq, cpu));
983			VMBUS_PCPU_GET(sc, message_tq, cpu) = NULL;
984		}
985	}
986}
987
988static int
989vmbus_read_ivar(device_t dev, device_t child, int index, uintptr_t *result)
990{
991	return (ENOENT);
992}
993
994static int
995vmbus_child_pnpinfo_str(device_t dev, device_t child, char *buf, size_t buflen)
996{
997	const struct vmbus_channel *chan;
998	char guidbuf[HYPERV_GUID_STRLEN];
999
1000	chan = vmbus_get_channel(child);
1001	if (chan == NULL) {
1002		/* Event timer device, which does not belong to a channel */
1003		return (0);
1004	}
1005
1006	strlcat(buf, "classid=", buflen);
1007	hyperv_guid2str(&chan->ch_guid_type, guidbuf, sizeof(guidbuf));
1008	strlcat(buf, guidbuf, buflen);
1009
1010	strlcat(buf, " deviceid=", buflen);
1011	hyperv_guid2str(&chan->ch_guid_inst, guidbuf, sizeof(guidbuf));
1012	strlcat(buf, guidbuf, buflen);
1013
1014	return (0);
1015}
1016
1017int
1018vmbus_add_child(struct vmbus_channel *chan)
1019{
1020	struct vmbus_softc *sc = chan->ch_vmbus;
1021	device_t parent = sc->vmbus_dev;
1022
1023	mtx_lock(&Giant);
1024
1025	chan->ch_dev = device_add_child(parent, NULL, -1);
1026	if (chan->ch_dev == NULL) {
1027		mtx_unlock(&Giant);
1028		device_printf(parent, "device_add_child for chan%u failed\n",
1029		    chan->ch_id);
1030		return (ENXIO);
1031	}
1032	device_set_ivars(chan->ch_dev, chan);
1033	device_probe_and_attach(chan->ch_dev);
1034
1035	mtx_unlock(&Giant);
1036	return (0);
1037}
1038
1039int
1040vmbus_delete_child(struct vmbus_channel *chan)
1041{
1042	int error = 0;
1043
1044	mtx_lock(&Giant);
1045	if (chan->ch_dev != NULL) {
1046		error = device_delete_child(chan->ch_vmbus->vmbus_dev,
1047		    chan->ch_dev);
1048		chan->ch_dev = NULL;
1049	}
1050	mtx_unlock(&Giant);
1051	return (error);
1052}
1053
1054static int
1055vmbus_sysctl_version(SYSCTL_HANDLER_ARGS)
1056{
1057	struct vmbus_softc *sc = arg1;
1058	char verstr[16];
1059
1060	snprintf(verstr, sizeof(verstr), "%u.%u",
1061	    VMBUS_VERSION_MAJOR(sc->vmbus_version),
1062	    VMBUS_VERSION_MINOR(sc->vmbus_version));
1063	return sysctl_handle_string(oidp, verstr, sizeof(verstr), req);
1064}
1065
1066/*
1067 * We need the function to make sure the MMIO resource is allocated from the
1068 * ranges found in _CRS.
1069 *
1070 * For the release function, we can use bus_generic_release_resource().
1071 */
1072static struct resource *
1073vmbus_alloc_resource(device_t dev, device_t child, int type, int *rid,
1074    rman_res_t start, rman_res_t end, rman_res_t count, u_int flags)
1075{
1076	device_t parent = device_get_parent(dev);
1077	struct resource *res;
1078
1079#ifdef NEW_PCIB
1080	if (type == SYS_RES_MEMORY) {
1081		struct vmbus_softc *sc = device_get_softc(dev);
1082
1083		res = pcib_host_res_alloc(&sc->vmbus_mmio_res, child, type,
1084		    rid, start, end, count, flags);
1085	} else
1086#endif
1087	{
1088		res = BUS_ALLOC_RESOURCE(parent, child, type, rid, start,
1089		    end, count, flags);
1090	}
1091
1092	return (res);
1093}
1094
1095static int
1096vmbus_alloc_msi(device_t bus, device_t dev, int count, int maxcount, int *irqs)
1097{
1098
1099	return (PCIB_ALLOC_MSI(device_get_parent(bus), dev, count, maxcount,
1100	    irqs));
1101}
1102
1103static int
1104vmbus_release_msi(device_t bus, device_t dev, int count, int *irqs)
1105{
1106
1107	return (PCIB_RELEASE_MSI(device_get_parent(bus), dev, count, irqs));
1108}
1109
1110static int
1111vmbus_alloc_msix(device_t bus, device_t dev, int *irq)
1112{
1113
1114	return (PCIB_ALLOC_MSIX(device_get_parent(bus), dev, irq));
1115}
1116
1117static int
1118vmbus_release_msix(device_t bus, device_t dev, int irq)
1119{
1120
1121	return (PCIB_RELEASE_MSIX(device_get_parent(bus), dev, irq));
1122}
1123
1124static int
1125vmbus_map_msi(device_t bus, device_t dev, int irq, uint64_t *addr,
1126	uint32_t *data)
1127{
1128
1129	return (PCIB_MAP_MSI(device_get_parent(bus), dev, irq, addr, data));
1130}
1131
1132static uint32_t
1133vmbus_get_version_method(device_t bus, device_t dev)
1134{
1135	struct vmbus_softc *sc = device_get_softc(bus);
1136
1137	return sc->vmbus_version;
1138}
1139
1140static int
1141vmbus_probe_guid_method(device_t bus, device_t dev,
1142    const struct hyperv_guid *guid)
1143{
1144	const struct vmbus_channel *chan = vmbus_get_channel(dev);
1145
1146	if (memcmp(&chan->ch_guid_type, guid, sizeof(struct hyperv_guid)) == 0)
1147		return 0;
1148	return ENXIO;
1149}
1150
1151static uint32_t
1152vmbus_get_vcpu_id_method(device_t bus, device_t dev, int cpu)
1153{
1154	const struct vmbus_softc *sc = device_get_softc(bus);
1155
1156	return (VMBUS_PCPU_GET(sc, vcpuid, cpu));
1157}
1158
1159static struct taskqueue *
1160vmbus_get_eventtq_method(device_t bus, device_t dev __unused, int cpu)
1161{
1162	const struct vmbus_softc *sc = device_get_softc(bus);
1163
1164	KASSERT(cpu >= 0 && cpu < mp_ncpus, ("invalid cpu%d", cpu));
1165	return (VMBUS_PCPU_GET(sc, event_tq, cpu));
1166}
1167
1168#ifdef NEW_PCIB
1169#define VTPM_BASE_ADDR 0xfed40000
1170#define FOUR_GB (1ULL << 32)
1171
1172enum parse_pass { parse_64, parse_32 };
1173
1174struct parse_context {
1175	device_t vmbus_dev;
1176	enum parse_pass pass;
1177};
1178
1179static ACPI_STATUS
1180parse_crs(ACPI_RESOURCE *res, void *ctx)
1181{
1182	const struct parse_context *pc = ctx;
1183	device_t vmbus_dev = pc->vmbus_dev;
1184
1185	struct vmbus_softc *sc = device_get_softc(vmbus_dev);
1186	UINT64 start, end;
1187
1188	switch (res->Type) {
1189	case ACPI_RESOURCE_TYPE_ADDRESS32:
1190		start = res->Data.Address32.Address.Minimum;
1191		end = res->Data.Address32.Address.Maximum;
1192		break;
1193
1194	case ACPI_RESOURCE_TYPE_ADDRESS64:
1195		start = res->Data.Address64.Address.Minimum;
1196		end = res->Data.Address64.Address.Maximum;
1197		break;
1198
1199	default:
1200		/* Unused types. */
1201		return (AE_OK);
1202	}
1203
1204	/*
1205	 * We don't use <1MB addresses.
1206	 */
1207	if (end < 0x100000)
1208		return (AE_OK);
1209
1210	/* Don't conflict with vTPM. */
1211	if (end >= VTPM_BASE_ADDR && start < VTPM_BASE_ADDR)
1212		end = VTPM_BASE_ADDR - 1;
1213
1214	if ((pc->pass == parse_32 && start < FOUR_GB) ||
1215	    (pc->pass == parse_64 && start >= FOUR_GB))
1216		pcib_host_res_decodes(&sc->vmbus_mmio_res, SYS_RES_MEMORY,
1217		    start, end, 0);
1218
1219	return (AE_OK);
1220}
1221
1222static void
1223vmbus_get_crs(device_t dev, device_t vmbus_dev, enum parse_pass pass)
1224{
1225	struct parse_context pc;
1226	ACPI_STATUS status;
1227
1228	if (bootverbose)
1229		device_printf(dev, "walking _CRS, pass=%d\n", pass);
1230
1231	pc.vmbus_dev = vmbus_dev;
1232	pc.pass = pass;
1233	status = AcpiWalkResources(acpi_get_handle(dev), "_CRS",
1234			parse_crs, &pc);
1235
1236	if (bootverbose && ACPI_FAILURE(status))
1237		device_printf(dev, "_CRS: not found, pass=%d\n", pass);
1238}
1239
1240static void
1241vmbus_get_mmio_res_pass(device_t dev, enum parse_pass pass)
1242{
1243	device_t acpi0, parent;
1244
1245	parent = device_get_parent(dev);
1246
1247	acpi0 = device_get_parent(parent);
1248	if (strcmp("acpi0", device_get_nameunit(acpi0)) == 0) {
1249		device_t *children;
1250		int count;
1251
1252		/*
1253		 * Try to locate VMBUS resources and find _CRS on them.
1254		 */
1255		if (device_get_children(acpi0, &children, &count) == 0) {
1256			int i;
1257
1258			for (i = 0; i < count; ++i) {
1259				if (!device_is_attached(children[i]))
1260					continue;
1261
1262				if (strcmp("vmbus_res",
1263				    device_get_name(children[i])) == 0)
1264					vmbus_get_crs(children[i], dev, pass);
1265			}
1266			free(children, M_TEMP);
1267		}
1268
1269		/*
1270		 * Try to find _CRS on acpi.
1271		 */
1272		vmbus_get_crs(acpi0, dev, pass);
1273	} else {
1274		device_printf(dev, "not grandchild of acpi\n");
1275	}
1276
1277	/*
1278	 * Try to find _CRS on parent.
1279	 */
1280	vmbus_get_crs(parent, dev, pass);
1281}
1282
1283static void
1284vmbus_get_mmio_res(device_t dev)
1285{
1286	struct vmbus_softc *sc = device_get_softc(dev);
1287	/*
1288	 * We walk the resources twice to make sure that: in the resource
1289	 * list, the 32-bit resources appear behind the 64-bit resources.
1290	 * NB: resource_list_add() uses INSERT_TAIL. This way, when we
1291	 * iterate through the list to find a range for a 64-bit BAR in
1292	 * vmbus_alloc_resource(), we can make sure we try to use >4GB
1293	 * ranges first.
1294	 */
1295	pcib_host_res_init(dev, &sc->vmbus_mmio_res);
1296
1297	vmbus_get_mmio_res_pass(dev, parse_64);
1298	vmbus_get_mmio_res_pass(dev, parse_32);
1299}
1300
1301/*
1302 * On Gen2 VMs, Hyper-V provides mmio space for framebuffer.
1303 * This mmio address range is not useable for other PCI devices.
1304 * Currently only efifb driver is using this range without reserving
1305 * it from system.
1306 * Therefore, vmbus driver reserves it before any other PCI device
1307 * drivers start to request mmio addresses.
1308 */
1309static struct resource *hv_fb_res;
1310
1311static void
1312vmbus_fb_mmio_res(device_t dev)
1313{
1314	struct efi_fb *efifb;
1315	caddr_t kmdp;
1316
1317	struct vmbus_softc *sc = device_get_softc(dev);
1318	int rid = 0;
1319
1320	kmdp = preload_search_by_type("elf kernel");
1321	if (kmdp == NULL)
1322		kmdp = preload_search_by_type("elf64 kernel");
1323	efifb = (struct efi_fb *)preload_search_info(kmdp,
1324	    MODINFO_METADATA | MODINFOMD_EFI_FB);
1325	if (efifb == NULL) {
1326		if (bootverbose)
1327			device_printf(dev,
1328			    "fb has no preloaded kernel efi information\n");
1329		/* We are on Gen1 VM, just return. */
1330		return;
1331	} else {
1332		if (bootverbose)
1333			device_printf(dev,
1334			    "efifb: fb_addr: %#jx, size: %#jx, "
1335			    "actual size needed: 0x%x\n",
1336			    efifb->fb_addr, efifb->fb_size,
1337			    (int) efifb->fb_height * efifb->fb_width);
1338	}
1339
1340	hv_fb_res = pcib_host_res_alloc(&sc->vmbus_mmio_res, dev,
1341	    SYS_RES_MEMORY, &rid,
1342	    efifb->fb_addr, efifb->fb_addr + efifb->fb_size, efifb->fb_size,
1343	    RF_ACTIVE | rman_make_alignment_flags(PAGE_SIZE));
1344
1345	if (hv_fb_res && bootverbose)
1346		device_printf(dev,
1347		    "successfully reserved memory for framebuffer "
1348		    "starting at %#jx, size %#jx\n",
1349		    efifb->fb_addr, efifb->fb_size);
1350}
1351
1352static void
1353vmbus_free_mmio_res(device_t dev)
1354{
1355	struct vmbus_softc *sc = device_get_softc(dev);
1356
1357	pcib_host_res_free(dev, &sc->vmbus_mmio_res);
1358
1359	if (hv_fb_res)
1360		hv_fb_res = NULL;
1361}
1362#endif	/* NEW_PCIB */
1363
1364static void
1365vmbus_identify(driver_t *driver, device_t parent)
1366{
1367
1368	if (device_get_unit(parent) != 0 || vm_guest != VM_GUEST_HV ||
1369	    (hyperv_features & CPUID_HV_MSR_SYNIC) == 0)
1370		return;
1371	device_add_child(parent, "vmbus", -1);
1372}
1373
1374static int
1375vmbus_probe(device_t dev)
1376{
1377
1378	if (device_get_unit(dev) != 0 || vm_guest != VM_GUEST_HV ||
1379	    (hyperv_features & CPUID_HV_MSR_SYNIC) == 0)
1380		return (ENXIO);
1381
1382	device_set_desc(dev, "Hyper-V Vmbus");
1383	return (BUS_PROBE_DEFAULT);
1384}
1385
1386/**
1387 * @brief Main vmbus driver initialization routine.
1388 *
1389 * Here, we
1390 * - initialize the vmbus driver context
1391 * - setup various driver entry points
1392 * - invoke the vmbus hv main init routine
1393 * - get the irq resource
1394 * - invoke the vmbus to add the vmbus root device
1395 * - setup the vmbus root device
1396 * - retrieve the channel offers
1397 */
1398static int
1399vmbus_doattach(struct vmbus_softc *sc)
1400{
1401	struct sysctl_oid_list *child;
1402	struct sysctl_ctx_list *ctx;
1403	int ret;
1404
1405	if (sc->vmbus_flags & VMBUS_FLAG_ATTACHED)
1406		return (0);
1407
1408#ifdef NEW_PCIB
1409	vmbus_get_mmio_res(sc->vmbus_dev);
1410	vmbus_fb_mmio_res(sc->vmbus_dev);
1411#endif
1412
1413	sc->vmbus_flags |= VMBUS_FLAG_ATTACHED;
1414
1415	sc->vmbus_gpadl = VMBUS_GPADL_START;
1416	mtx_init(&sc->vmbus_prichan_lock, "vmbus prichan", NULL, MTX_DEF);
1417	TAILQ_INIT(&sc->vmbus_prichans);
1418	mtx_init(&sc->vmbus_chan_lock, "vmbus channel", NULL, MTX_DEF);
1419	TAILQ_INIT(&sc->vmbus_chans);
1420	sc->vmbus_chmap = malloc(
1421	    sizeof(struct vmbus_channel *) * VMBUS_CHAN_MAX, M_DEVBUF,
1422	    M_WAITOK | M_ZERO);
1423
1424	/*
1425	 * Create context for "post message" Hypercalls
1426	 */
1427	sc->vmbus_xc = vmbus_xact_ctx_create(bus_get_dma_tag(sc->vmbus_dev),
1428	    HYPERCALL_POSTMSGIN_SIZE, VMBUS_MSG_SIZE,
1429	    sizeof(struct vmbus_msghc));
1430	if (sc->vmbus_xc == NULL) {
1431		ret = ENXIO;
1432		goto cleanup;
1433	}
1434
1435	/*
1436	 * Allocate DMA stuffs.
1437	 */
1438	ret = vmbus_dma_alloc(sc);
1439	if (ret != 0)
1440		goto cleanup;
1441
1442	/*
1443	 * Setup interrupt.
1444	 */
1445	ret = vmbus_intr_setup(sc);
1446	if (ret != 0)
1447		goto cleanup;
1448
1449	/*
1450	 * Setup SynIC.
1451	 */
1452	if (bootverbose)
1453		device_printf(sc->vmbus_dev, "smp_started = %d\n", smp_started);
1454	smp_rendezvous(NULL, vmbus_synic_setup, NULL, sc);
1455	sc->vmbus_flags |= VMBUS_FLAG_SYNIC;
1456
1457	/*
1458	 * Initialize vmbus, e.g. connect to Hypervisor.
1459	 */
1460	ret = vmbus_init(sc);
1461	if (ret != 0)
1462		goto cleanup;
1463
1464	if (sc->vmbus_version == VMBUS_VERSION_WS2008 ||
1465	    sc->vmbus_version == VMBUS_VERSION_WIN7)
1466		sc->vmbus_event_proc = vmbus_event_proc_compat;
1467	else
1468		sc->vmbus_event_proc = vmbus_event_proc;
1469
1470	ret = vmbus_scan(sc);
1471	if (ret != 0)
1472		goto cleanup;
1473
1474	ctx = device_get_sysctl_ctx(sc->vmbus_dev);
1475	child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->vmbus_dev));
1476	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "version",
1477	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
1478	    vmbus_sysctl_version, "A", "vmbus version");
1479
1480	return (ret);
1481
1482cleanup:
1483	vmbus_scan_teardown(sc);
1484	vmbus_intr_teardown(sc);
1485	vmbus_dma_free(sc);
1486	if (sc->vmbus_xc != NULL) {
1487		vmbus_xact_ctx_destroy(sc->vmbus_xc);
1488		sc->vmbus_xc = NULL;
1489	}
1490	free(__DEVOLATILE(void *, sc->vmbus_chmap), M_DEVBUF);
1491	mtx_destroy(&sc->vmbus_prichan_lock);
1492	mtx_destroy(&sc->vmbus_chan_lock);
1493
1494	return (ret);
1495}
1496
1497static void
1498vmbus_event_proc_dummy(struct vmbus_softc *sc __unused, int cpu __unused)
1499{
1500}
1501
1502#ifdef EARLY_AP_STARTUP
1503
1504static void
1505vmbus_intrhook(void *xsc)
1506{
1507	struct vmbus_softc *sc = xsc;
1508
1509	if (bootverbose)
1510		device_printf(sc->vmbus_dev, "intrhook\n");
1511	vmbus_doattach(sc);
1512	config_intrhook_disestablish(&sc->vmbus_intrhook);
1513}
1514
1515#endif	/* EARLY_AP_STARTUP */
1516
1517static int
1518vmbus_attach(device_t dev)
1519{
1520	vmbus_sc = device_get_softc(dev);
1521	vmbus_sc->vmbus_dev = dev;
1522	vmbus_sc->vmbus_idtvec = -1;
1523
1524	/*
1525	 * Event processing logic will be configured:
1526	 * - After the vmbus protocol version negotiation.
1527	 * - Before we request channel offers.
1528	 */
1529	vmbus_sc->vmbus_event_proc = vmbus_event_proc_dummy;
1530
1531#ifdef EARLY_AP_STARTUP
1532	/*
1533	 * Defer the real attach until the pause(9) works as expected.
1534	 */
1535	vmbus_sc->vmbus_intrhook.ich_func = vmbus_intrhook;
1536	vmbus_sc->vmbus_intrhook.ich_arg = vmbus_sc;
1537	config_intrhook_establish(&vmbus_sc->vmbus_intrhook);
1538#else	/* !EARLY_AP_STARTUP */
1539	/*
1540	 * If the system has already booted and thread
1541	 * scheduling is possible indicated by the global
1542	 * cold set to zero, we just call the driver
1543	 * initialization directly.
1544	 */
1545	if (!cold)
1546		vmbus_doattach(vmbus_sc);
1547#endif	/* EARLY_AP_STARTUP */
1548
1549	return (0);
1550}
1551
1552static int
1553vmbus_detach(device_t dev)
1554{
1555	struct vmbus_softc *sc = device_get_softc(dev);
1556
1557	bus_generic_detach(dev);
1558	vmbus_chan_destroy_all(sc);
1559
1560	vmbus_scan_teardown(sc);
1561
1562	vmbus_disconnect(sc);
1563
1564	if (sc->vmbus_flags & VMBUS_FLAG_SYNIC) {
1565		sc->vmbus_flags &= ~VMBUS_FLAG_SYNIC;
1566		smp_rendezvous(NULL, vmbus_synic_teardown, NULL, NULL);
1567	}
1568
1569	vmbus_intr_teardown(sc);
1570	vmbus_dma_free(sc);
1571
1572	if (sc->vmbus_xc != NULL) {
1573		vmbus_xact_ctx_destroy(sc->vmbus_xc);
1574		sc->vmbus_xc = NULL;
1575	}
1576
1577	free(__DEVOLATILE(void *, sc->vmbus_chmap), M_DEVBUF);
1578	mtx_destroy(&sc->vmbus_prichan_lock);
1579	mtx_destroy(&sc->vmbus_chan_lock);
1580
1581#ifdef NEW_PCIB
1582	vmbus_free_mmio_res(dev);
1583#endif
1584
1585	return (0);
1586}
1587
1588#ifndef EARLY_AP_STARTUP
1589
1590static void
1591vmbus_sysinit(void *arg __unused)
1592{
1593	struct vmbus_softc *sc = vmbus_get_softc();
1594
1595	if (vm_guest != VM_GUEST_HV || sc == NULL)
1596		return;
1597
1598	/*
1599	 * If the system has already booted and thread
1600	 * scheduling is possible, as indicated by the
1601	 * global cold set to zero, we just call the driver
1602	 * initialization directly.
1603	 */
1604	if (!cold)
1605		vmbus_doattach(sc);
1606}
1607/*
1608 * NOTE:
1609 * We have to start as the last step of SI_SUB_SMP, i.e. after SMP is
1610 * initialized.
1611 */
1612SYSINIT(vmbus_initialize, SI_SUB_SMP, SI_ORDER_ANY, vmbus_sysinit, NULL);
1613
1614#endif	/* !EARLY_AP_STARTUP */
1615