xref: /freebsd/usr.sbin/bhyve/amd64/bhyverun_machdep.c (revision a90b9d0159070121c221b966469c3e36d912bf82)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2011 NetApp, 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, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <assert.h>
30 #include <err.h>
31 #include <stdbool.h>
32 #include <stdlib.h>
33 #include <sysexits.h>
34 
35 #include <vmmapi.h>
36 
37 #include "acpi.h"
38 #include "atkbdc.h"
39 #include "bhyverun.h"
40 #include "config.h"
41 #include "debug.h"
42 #include "e820.h"
43 #include "fwctl.h"
44 #include "ioapic.h"
45 #include "inout.h"
46 #include "kernemu_dev.h"
47 #include "mptbl.h"
48 #include "pci_emul.h"
49 #include "pci_irq.h"
50 #include "pci_lpc.h"
51 #include "rtc.h"
52 #include "smbiostbl.h"
53 #include "xmsr.h"
54 
55 void
56 bhyve_init_config(void)
57 {
58 	init_config();
59 
60 	/* Set default values prior to option parsing. */
61 	set_config_bool("acpi_tables", true);
62 	set_config_bool("acpi_tables_in_memory", true);
63 	set_config_value("memory.size", "256M");
64 	set_config_bool("x86.strictmsr", true);
65 	set_config_value("lpc.fwcfg", "bhyve");
66 }
67 
68 void
69 bhyve_usage(int code)
70 {
71 	const char *progname;
72 
73 	progname = getprogname();
74 
75 	fprintf(stderr,
76 	    "Usage: %s [-aCDeHhPSuWwxY]\n"
77 	    "       %*s [-c [[cpus=]numcpus][,sockets=n][,cores=n][,threads=n]]\n"
78 	    "       %*s [-G port] [-k config_file] [-l lpc] [-m mem] [-o var=value]\n"
79 	    "       %*s [-p vcpu:hostcpu] [-r file] [-s pci] [-U uuid] vmname\n"
80 	    "       -a: local apic is in xAPIC mode (deprecated)\n"
81 	    "       -C: include guest memory in core file\n"
82 	    "       -c: number of CPUs and/or topology specification\n"
83 	    "       -D: destroy on power-off\n"
84 	    "       -e: exit on unhandled I/O access\n"
85 	    "       -G: start a debug server\n"
86 	    "       -H: vmexit from the guest on HLT\n"
87 	    "       -h: help\n"
88 	    "       -k: key=value flat config file\n"
89 	    "       -K: PS2 keyboard layout\n"
90 	    "       -l: LPC device configuration\n"
91 	    "       -m: memory size\n"
92 	    "       -o: set config 'var' to 'value'\n"
93 	    "       -P: vmexit from the guest on pause\n"
94 	    "       -p: pin 'vcpu' to 'hostcpu'\n"
95 #ifdef BHYVE_SNAPSHOT
96 	    "       -r: path to checkpoint file\n"
97 #endif
98 	    "       -S: guest memory cannot be swapped\n"
99 	    "       -s: <slot,driver,configinfo> PCI slot config\n"
100 	    "       -U: UUID\n"
101 	    "       -u: RTC keeps UTC time\n"
102 	    "       -W: force virtio to use single-vector MSI\n"
103 	    "       -w: ignore unimplemented MSRs\n"
104 	    "       -x: local APIC is in x2APIC mode\n"
105 	    "       -Y: disable MPtable generation\n",
106 	    progname, (int)strlen(progname), "", (int)strlen(progname), "",
107 	    (int)strlen(progname), "");
108 	exit(code);
109 }
110 
111 void
112 bhyve_optparse(int argc, char **argv)
113 {
114 	const char *optstr;
115 	int c;
116 
117 #ifdef BHYVE_SNAPSHOT
118 	optstr = "aehuwxACDHIPSWYk:f:o:p:G:c:s:m:l:K:U:r:";
119 #else
120 	optstr = "aehuwxACDHIPSWYk:f:o:p:G:c:s:m:l:K:U:";
121 #endif
122 	while ((c = getopt(argc, argv, optstr)) != -1) {
123 		switch (c) {
124 		case 'a':
125 			set_config_bool("x86.x2apic", false);
126 			break;
127 		case 'A':
128 			/*
129 			 * NOP. For backward compatibility. Most systems don't
130 			 * work properly without sane ACPI tables. Therefore,
131 			 * we're always generating them.
132 			 */
133 			break;
134 		case 'D':
135 			set_config_bool("destroy_on_poweroff", true);
136 			break;
137 		case 'p':
138 			if (bhyve_pincpu_parse(optarg) != 0) {
139 				errx(EX_USAGE, "invalid vcpu pinning "
140 				    "configuration '%s'", optarg);
141 			}
142 			break;
143 		case 'c':
144 			if (bhyve_topology_parse(optarg) != 0) {
145 			    errx(EX_USAGE, "invalid cpu topology "
146 				"'%s'", optarg);
147 			}
148 			break;
149 		case 'C':
150 			set_config_bool("memory.guest_in_core", true);
151 			break;
152 		case 'f':
153 			if (qemu_fwcfg_parse_cmdline_arg(optarg) != 0) {
154 				errx(EX_USAGE, "invalid fwcfg item '%s'",
155 				    optarg);
156 			}
157 			break;
158 		case 'G':
159 			bhyve_parse_gdb_options(optarg);
160 			break;
161 		case 'k':
162 			bhyve_parse_simple_config_file(optarg);
163 			break;
164 		case 'K':
165 			set_config_value("keyboard.layout", optarg);
166 			break;
167 		case 'l':
168 			if (strncmp(optarg, "help", strlen(optarg)) == 0) {
169 				lpc_print_supported_devices();
170 				exit(0);
171 			} else if (lpc_device_parse(optarg) != 0) {
172 				errx(EX_USAGE, "invalid lpc device "
173 				    "configuration '%s'", optarg);
174 			}
175 			break;
176 #ifdef BHYVE_SNAPSHOT
177 		case 'r':
178 			restore_file = optarg;
179 			break;
180 #endif
181 		case 's':
182 			if (strncmp(optarg, "help", strlen(optarg)) == 0) {
183 				pci_print_supported_devices();
184 				exit(0);
185 			} else if (pci_parse_slot(optarg) != 0)
186 				exit(4);
187 			else
188 				break;
189 		case 'S':
190 			set_config_bool("memory.wired", true);
191 			break;
192 		case 'm':
193 			set_config_value("memory.size", optarg);
194 			break;
195 		case 'o':
196 			if (!bhyve_parse_config_option(optarg)) {
197 				errx(EX_USAGE,
198 				    "invalid configuration option '%s'",
199 				    optarg);
200 			}
201 			break;
202 		case 'H':
203 			set_config_bool("x86.vmexit_on_hlt", true);
204 			break;
205 		case 'I':
206 			/*
207 			 * The "-I" option was used to add an ioapic to the
208 			 * virtual machine.
209 			 *
210 			 * An ioapic is now provided unconditionally for each
211 			 * virtual machine and this option is now deprecated.
212 			 */
213 			break;
214 		case 'P':
215 			set_config_bool("x86.vmexit_on_pause", true);
216 			break;
217 		case 'e':
218 			set_config_bool("x86.strictio", true);
219 			break;
220 		case 'u':
221 			set_config_bool("rtc.use_localtime", false);
222 			break;
223 		case 'U':
224 			set_config_value("uuid", optarg);
225 			break;
226 		case 'w':
227 			set_config_bool("x86.strictmsr", false);
228 			break;
229 		case 'W':
230 			set_config_bool("virtio_msix", false);
231 			break;
232 		case 'x':
233 			set_config_bool("x86.x2apic", true);
234 			break;
235 		case 'Y':
236 			set_config_bool("x86.mptable", false);
237 			break;
238 		case 'h':
239 			bhyve_usage(0);
240 		default:
241 			bhyve_usage(1);
242 		}
243 	}
244 }
245 
246 void
247 bhyve_init_vcpu(struct vcpu *vcpu)
248 {
249 	int err, tmp;
250 
251 	if (get_config_bool_default("x86.vmexit_on_hlt", false)) {
252 		err = vm_get_capability(vcpu, VM_CAP_HALT_EXIT, &tmp);
253 		if (err < 0) {
254 			EPRINTLN("VM exit on HLT not supported");
255 			exit(4);
256 		}
257 		vm_set_capability(vcpu, VM_CAP_HALT_EXIT, 1);
258 	}
259 
260 	if (get_config_bool_default("x86.vmexit_on_pause", false)) {
261 		/*
262 		 * pause exit support required for this mode
263 		 */
264 		err = vm_get_capability(vcpu, VM_CAP_PAUSE_EXIT, &tmp);
265 		if (err < 0) {
266 			EPRINTLN("SMP mux requested, no pause support");
267 			exit(4);
268 		}
269 		vm_set_capability(vcpu, VM_CAP_PAUSE_EXIT, 1);
270 	}
271 
272 	if (get_config_bool_default("x86.x2apic", false))
273 		err = vm_set_x2apic_state(vcpu, X2APIC_ENABLED);
274 	else
275 		err = vm_set_x2apic_state(vcpu, X2APIC_DISABLED);
276 
277 	if (err) {
278 		EPRINTLN("Unable to set x2apic state (%d)", err);
279 		exit(4);
280 	}
281 
282 	vm_set_capability(vcpu, VM_CAP_ENABLE_INVPCID, 1);
283 
284 	err = vm_set_capability(vcpu, VM_CAP_IPI_EXIT, 1);
285 	assert(err == 0);
286 }
287 
288 void
289 bhyve_start_vcpu(struct vcpu *vcpu, bool bsp)
290 {
291 	int error;
292 
293 	if (bsp) {
294 		if (lpc_bootrom()) {
295 			error = vm_set_capability(vcpu,
296 			    VM_CAP_UNRESTRICTED_GUEST, 1);
297 			if (error != 0) {
298 				err(4, "ROM boot failed: unrestricted guest "
299 				    "capability not available");
300 			}
301 			error = vcpu_reset(vcpu);
302 			assert(error == 0);
303 		}
304 	} else {
305 		bhyve_init_vcpu(vcpu);
306 
307 		/*
308 		 * Enable the 'unrestricted guest' mode for APs.
309 		 *
310 		 * APs startup in power-on 16-bit mode.
311 		 */
312 		error = vm_set_capability(vcpu, VM_CAP_UNRESTRICTED_GUEST, 1);
313 		assert(error == 0);
314 	}
315 
316 	fbsdrun_addcpu(vcpu_id(vcpu));
317 }
318 
319 int
320 bhyve_init_platform(struct vmctx *ctx, struct vcpu *bsp __unused)
321 {
322 	int error;
323 
324 	error = init_msr();
325 	if (error != 0)
326 		return (error);
327 	init_inout();
328 	kernemu_dev_init();
329 	atkbdc_init(ctx);
330 	pci_irq_init(ctx);
331 	ioapic_init(ctx);
332 	rtc_init(ctx);
333 	sci_init(ctx);
334 	error = e820_init(ctx);
335 	if (error != 0)
336 		return (error);
337 
338 	return (0);
339 }
340 
341 int
342 bhyve_init_platform_late(struct vmctx *ctx, struct vcpu *bsp __unused)
343 {
344 	int error;
345 
346 	if (get_config_bool_default("x86.mptable", true)) {
347 		error = mptable_build(ctx, guest_ncpus);
348 		if (error != 0)
349 			return (error);
350 	}
351 	error = smbios_build(ctx);
352 	if (error != 0)
353 		return (error);
354 	error = e820_finalize();
355 	if (error != 0)
356 		return (error);
357 
358 	if (lpc_bootrom() && strcmp(lpc_fwcfg(), "bhyve") == 0)
359 		fwctl_init();
360 
361 	if (get_config_bool("acpi_tables")) {
362 		error = acpi_build(ctx, guest_ncpus);
363 		assert(error == 0);
364 	}
365 
366 	return (0);
367 }
368