hyperv.c revision 302170 
1/*-
2 * Copyright (c) 2009-2012,2016 Microsoft Corp.
3 * Copyright (c) 2012 NetApp Inc.
4 * Copyright (c) 2012 Citrix Inc.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice unmodified, this list of conditions, and the following
12 *    disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29/**
30 * Implements low-level interactions with Hypver-V/Azure
31 */
32#include <sys/cdefs.h>
33__FBSDID("$FreeBSD: stable/10/sys/dev/hyperv/vmbus/hyperv.c 302170 2016-06-24 02:30:14Z sephe $");
34
35#include <sys/param.h>
36#include <sys/kernel.h>
37#include <sys/malloc.h>
38#include <sys/pcpu.h>
39#include <sys/timetc.h>
40#include <machine/bus.h>
41#include <machine/md_var.h>
42#include <vm/vm.h>
43#include <vm/vm_param.h>
44#include <vm/pmap.h>
45
46#include <dev/hyperv/include/hyperv_busdma.h>
47#include <dev/hyperv/vmbus/hv_vmbus_priv.h>
48#include <dev/hyperv/vmbus/hyperv_machdep.h>
49#include <dev/hyperv/vmbus/hyperv_reg.h>
50#include <dev/hyperv/vmbus/hyperv_var.h>
51#include <dev/hyperv/vmbus/vmbus_var.h>
52
53#define HYPERV_FREEBSD_BUILD		0ULL
54#define HYPERV_FREEBSD_VERSION		((uint64_t)__FreeBSD_version)
55#define HYPERV_FREEBSD_OSID		0ULL
56
57#define MSR_HV_GUESTID_BUILD_FREEBSD	\
58	(HYPERV_FREEBSD_BUILD & MSR_HV_GUESTID_BUILD_MASK)
59#define MSR_HV_GUESTID_VERSION_FREEBSD	\
60	((HYPERV_FREEBSD_VERSION << MSR_HV_GUESTID_VERSION_SHIFT) & \
61	 MSR_HV_GUESTID_VERSION_MASK)
62#define MSR_HV_GUESTID_OSID_FREEBSD	\
63	((HYPERV_FREEBSD_OSID << MSR_HV_GUESTID_OSID_SHIFT) & \
64	 MSR_HV_GUESTID_OSID_MASK)
65
66#define MSR_HV_GUESTID_FREEBSD		\
67	(MSR_HV_GUESTID_BUILD_FREEBSD |	\
68	 MSR_HV_GUESTID_VERSION_FREEBSD | \
69	 MSR_HV_GUESTID_OSID_FREEBSD |	\
70	 MSR_HV_GUESTID_OSTYPE_FREEBSD)
71
72struct hypercall_ctx {
73	void			*hc_addr;
74	struct hyperv_dma	hc_dma;
75};
76
77static u_int	hyperv_get_timecount(struct timecounter *tc);
78
79u_int		hyperv_features;
80u_int		hyperv_recommends;
81
82static u_int	hyperv_pm_features;
83static u_int	hyperv_features3;
84
85static struct timecounter	hyperv_timecounter = {
86	.tc_get_timecount	= hyperv_get_timecount,
87	.tc_poll_pps		= NULL,
88	.tc_counter_mask	= 0xffffffff,
89	.tc_frequency		= HYPERV_TIMER_FREQ,
90	.tc_name		= "Hyper-V",
91	.tc_quality		= 2000,
92	.tc_flags		= 0,
93	.tc_priv		= NULL
94};
95
96static struct hypercall_ctx	hypercall_context;
97
98static u_int
99hyperv_get_timecount(struct timecounter *tc __unused)
100{
101	return rdmsr(MSR_HV_TIME_REF_COUNT);
102}
103
104/**
105 * @brief Invoke the specified hypercall
106 */
107static uint64_t
108hv_vmbus_do_hypercall(uint64_t value, void *input, void *output)
109{
110	uint64_t in_paddr = 0, out_paddr = 0;
111
112	if (input != NULL)
113		in_paddr = hv_get_phys_addr(input);
114	if (output != NULL)
115		out_paddr = hv_get_phys_addr(output);
116
117	return hypercall_md(hypercall_context.hc_addr, value,
118	    in_paddr, out_paddr);
119}
120
121/**
122 * @brief Post a message using the hypervisor message IPC.
123 * (This involves a hypercall.)
124 */
125hv_vmbus_status
126hv_vmbus_post_msg_via_msg_ipc(
127	hv_vmbus_connection_id	connection_id,
128	hv_vmbus_msg_type	message_type,
129	void*			payload,
130	size_t			payload_size)
131{
132	struct alignedinput {
133	    uint64_t alignment8;
134	    hv_vmbus_input_post_message msg;
135	};
136
137	hv_vmbus_input_post_message*	aligned_msg;
138	hv_vmbus_status 		status;
139	size_t				addr;
140
141	if (payload_size > HV_MESSAGE_PAYLOAD_BYTE_COUNT)
142	    return (EMSGSIZE);
143
144	addr = (size_t) malloc(sizeof(struct alignedinput), M_DEVBUF,
145			    M_ZERO | M_NOWAIT);
146	KASSERT(addr != 0,
147	    ("Error VMBUS: malloc failed to allocate message buffer!"));
148	if (addr == 0)
149	    return (ENOMEM);
150
151	aligned_msg = (hv_vmbus_input_post_message*)
152	    (HV_ALIGN_UP(addr, HV_HYPERCALL_PARAM_ALIGN));
153
154	aligned_msg->connection_id = connection_id;
155	aligned_msg->message_type = message_type;
156	aligned_msg->payload_size = payload_size;
157	memcpy((void*) aligned_msg->payload, payload, payload_size);
158
159	status = hv_vmbus_do_hypercall(
160		    HV_CALL_POST_MESSAGE, aligned_msg, 0) & 0xFFFF;
161
162	free((void *) addr, M_DEVBUF);
163	return (status);
164}
165
166/**
167 * @brief Signal an event on the specified connection using the hypervisor
168 * event IPC. (This involves a hypercall.)
169 */
170hv_vmbus_status
171hv_vmbus_signal_event(void *con_id)
172{
173	hv_vmbus_status status;
174
175	status = hv_vmbus_do_hypercall(
176		    HV_CALL_SIGNAL_EVENT,
177		    con_id,
178		    0) & 0xFFFF;
179
180	return (status);
181}
182
183int
184hyperv_guid2str(const struct hv_guid *guid, char *buf, size_t sz)
185{
186	const uint8_t *d = guid->data;
187
188	return snprintf(buf, sz, "%02x%02x%02x%02x-"
189	    "%02x%02x-%02x%02x-%02x%02x-"
190	    "%02x%02x%02x%02x%02x%02x",
191	    d[3], d[2], d[1], d[0],
192	    d[5], d[4], d[7], d[6], d[8], d[9],
193	    d[10], d[11], d[12], d[13], d[14], d[15]);
194}
195
196static bool
197hyperv_identify(void)
198{
199	u_int regs[4];
200	unsigned int maxleaf;
201
202	if (vm_guest != VM_GUEST_HV)
203		return (false);
204
205	do_cpuid(CPUID_LEAF_HV_MAXLEAF, regs);
206	maxleaf = regs[0];
207	if (maxleaf < CPUID_LEAF_HV_LIMITS)
208		return (false);
209
210	do_cpuid(CPUID_LEAF_HV_INTERFACE, regs);
211	if (regs[0] != CPUID_HV_IFACE_HYPERV)
212		return (false);
213
214	do_cpuid(CPUID_LEAF_HV_FEATURES, regs);
215	if ((regs[0] & CPUID_HV_MSR_HYPERCALL) == 0) {
216		/*
217		 * Hyper-V w/o Hypercall is impossible; someone
218		 * is faking Hyper-V.
219		 */
220		return (false);
221	}
222	hyperv_features = regs[0];
223	hyperv_pm_features = regs[2];
224	hyperv_features3 = regs[3];
225
226	do_cpuid(CPUID_LEAF_HV_IDENTITY, regs);
227	printf("Hyper-V Version: %d.%d.%d [SP%d]\n",
228	    regs[1] >> 16, regs[1] & 0xffff, regs[0], regs[2]);
229
230	printf("  Features=0x%b\n", hyperv_features,
231	    "\020"
232	    "\001VPRUNTIME"	/* MSR_HV_VP_RUNTIME */
233	    "\002TMREFCNT"	/* MSR_HV_TIME_REF_COUNT */
234	    "\003SYNIC"		/* MSRs for SynIC */
235	    "\004SYNTM"		/* MSRs for SynTimer */
236	    "\005APIC"		/* MSR_HV_{EOI,ICR,TPR} */
237	    "\006HYPERCALL"	/* MSR_HV_{GUEST_OS_ID,HYPERCALL} */
238	    "\007VPINDEX"	/* MSR_HV_VP_INDEX */
239	    "\010RESET"		/* MSR_HV_RESET */
240	    "\011STATS"		/* MSR_HV_STATS_ */
241	    "\012REFTSC"	/* MSR_HV_REFERENCE_TSC */
242	    "\013IDLE"		/* MSR_HV_GUEST_IDLE */
243	    "\014TMFREQ"	/* MSR_HV_{TSC,APIC}_FREQUENCY */
244	    "\015DEBUG");	/* MSR_HV_SYNTH_DEBUG_ */
245	printf("  PM Features=0x%b [C%u]\n",
246	    (hyperv_pm_features & ~CPUPM_HV_CSTATE_MASK),
247	    "\020"
248	    "\005C3HPET",	/* HPET is required for C3 state */
249	    CPUPM_HV_CSTATE(hyperv_pm_features));
250	printf("  Features3=0x%b\n", hyperv_features3,
251	    "\020"
252	    "\001MWAIT"		/* MWAIT */
253	    "\002DEBUG"		/* guest debug support */
254	    "\003PERFMON"	/* performance monitor */
255	    "\004PCPUDPE"	/* physical CPU dynamic partition event */
256	    "\005XMMHC"		/* hypercall input through XMM regs */
257	    "\006IDLE"		/* guest idle support */
258	    "\007SLEEP"		/* hypervisor sleep support */
259	    "\010NUMA"		/* NUMA distance query support */
260	    "\011TMFREQ"	/* timer frequency query (TSC, LAPIC) */
261	    "\012SYNCMC"	/* inject synthetic machine checks */
262	    "\013CRASH"		/* MSRs for guest crash */
263	    "\014DEBUGMSR"	/* MSRs for guest debug */
264	    "\015NPIEP"		/* NPIEP */
265	    "\016HVDIS");	/* disabling hypervisor */
266
267	do_cpuid(CPUID_LEAF_HV_RECOMMENDS, regs);
268	hyperv_recommends = regs[0];
269	if (bootverbose)
270		printf("  Recommends: %08x %08x\n", regs[0], regs[1]);
271
272	do_cpuid(CPUID_LEAF_HV_LIMITS, regs);
273	if (bootverbose) {
274		printf("  Limits: Vcpu:%d Lcpu:%d Int:%d\n",
275		    regs[0], regs[1], regs[2]);
276	}
277
278	if (maxleaf >= CPUID_LEAF_HV_HWFEATURES) {
279		do_cpuid(CPUID_LEAF_HV_HWFEATURES, regs);
280		if (bootverbose) {
281			printf("  HW Features: %08x, AMD: %08x\n",
282			    regs[0], regs[3]);
283		}
284	}
285
286	return (true);
287}
288
289static void
290hyperv_init(void *dummy __unused)
291{
292	if (!hyperv_identify()) {
293		/* Not Hyper-V; reset guest id to the generic one. */
294		if (vm_guest == VM_GUEST_HV)
295			vm_guest = VM_GUEST_VM;
296		return;
297	}
298
299	/* Set guest id */
300	wrmsr(MSR_HV_GUEST_OS_ID, MSR_HV_GUESTID_FREEBSD);
301
302	if (hyperv_features & CPUID_HV_MSR_TIME_REFCNT) {
303		/* Register Hyper-V timecounter */
304		tc_init(&hyperv_timecounter);
305	}
306}
307SYSINIT(hyperv_initialize, SI_SUB_HYPERVISOR, SI_ORDER_FIRST, hyperv_init,
308    NULL);
309
310static void
311hypercall_memfree(void)
312{
313	hyperv_dmamem_free(&hypercall_context.hc_dma,
314	    hypercall_context.hc_addr);
315	hypercall_context.hc_addr = NULL;
316}
317
318static void
319hypercall_create(void *arg __unused)
320{
321	uint64_t hc, hc_orig;
322
323	if (vm_guest != VM_GUEST_HV)
324		return;
325
326	hypercall_context.hc_addr = hyperv_dmamem_alloc(NULL, PAGE_SIZE, 0,
327	    PAGE_SIZE, &hypercall_context.hc_dma, BUS_DMA_WAITOK);
328	if (hypercall_context.hc_addr == NULL) {
329		printf("hyperv: Hypercall page allocation failed\n");
330		/* Can't perform any Hyper-V specific actions */
331		vm_guest = VM_GUEST_VM;
332		return;
333	}
334
335	/* Get the 'reserved' bits, which requires preservation. */
336	hc_orig = rdmsr(MSR_HV_HYPERCALL);
337
338	/*
339	 * Setup the Hypercall page.
340	 *
341	 * NOTE: 'reserved' bits MUST be preserved.
342	 */
343	hc = ((hypercall_context.hc_dma.hv_paddr >> PAGE_SHIFT) <<
344	    MSR_HV_HYPERCALL_PGSHIFT) |
345	    (hc_orig & MSR_HV_HYPERCALL_RSVD_MASK) |
346	    MSR_HV_HYPERCALL_ENABLE;
347	wrmsr(MSR_HV_HYPERCALL, hc);
348
349	/*
350	 * Confirm that Hypercall page did get setup.
351	 */
352	hc = rdmsr(MSR_HV_HYPERCALL);
353	if ((hc & MSR_HV_HYPERCALL_ENABLE) == 0) {
354		printf("hyperv: Hypercall setup failed\n");
355		hypercall_memfree();
356		/* Can't perform any Hyper-V specific actions */
357		vm_guest = VM_GUEST_VM;
358		return;
359	}
360	if (bootverbose)
361		printf("hyperv: Hypercall created\n");
362}
363SYSINIT(hypercall_ctor, SI_SUB_DRIVERS, SI_ORDER_FIRST, hypercall_create, NULL);
364
365static void
366hypercall_destroy(void *arg __unused)
367{
368	uint64_t hc;
369
370	if (hypercall_context.hc_addr == NULL)
371		return;
372
373	/* Disable Hypercall */
374	hc = rdmsr(MSR_HV_HYPERCALL);
375	wrmsr(MSR_HV_HYPERCALL, (hc & MSR_HV_HYPERCALL_RSVD_MASK));
376	hypercall_memfree();
377
378	if (bootverbose)
379		printf("hyperv: Hypercall destroyed\n");
380}
381SYSUNINIT(hypercall_dtor, SI_SUB_DRIVERS, SI_ORDER_FIRST, hypercall_destroy,
382    NULL);
383