xref: /freebsd/sys/amd64/vmm/x86.c (revision edf8578117e8844e02c0121147f45e4609b30680)
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 <sys/cdefs.h>
30 #include <sys/param.h>
31 #include <sys/pcpu.h>
32 #include <sys/systm.h>
33 #include <sys/sysctl.h>
34 
35 #include <machine/clock.h>
36 #include <machine/cpufunc.h>
37 #include <machine/md_var.h>
38 #include <machine/segments.h>
39 #include <machine/specialreg.h>
40 
41 #include <machine/vmm.h>
42 
43 #include "vmm_host.h"
44 #include "vmm_ktr.h"
45 #include "vmm_util.h"
46 #include "x86.h"
47 
48 SYSCTL_DECL(_hw_vmm);
49 static SYSCTL_NODE(_hw_vmm, OID_AUTO, topology, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
50     NULL);
51 
52 #define	CPUID_VM_HIGH		0x40000000
53 
54 static const char bhyve_id[12] = "bhyve bhyve ";
55 
56 static uint64_t bhyve_xcpuids;
57 SYSCTL_ULONG(_hw_vmm, OID_AUTO, bhyve_xcpuids, CTLFLAG_RW, &bhyve_xcpuids, 0,
58     "Number of times an unknown cpuid leaf was accessed");
59 
60 static int cpuid_leaf_b = 1;
61 SYSCTL_INT(_hw_vmm_topology, OID_AUTO, cpuid_leaf_b, CTLFLAG_RDTUN,
62     &cpuid_leaf_b, 0, NULL);
63 
64 /*
65  * Round up to the next power of two, if necessary, and then take log2.
66  * Returns -1 if argument is zero.
67  */
68 static __inline int
69 log2(u_int x)
70 {
71 
72 	return (fls(x << (1 - powerof2(x))) - 1);
73 }
74 
75 int
76 x86_emulate_cpuid(struct vcpu *vcpu, uint64_t *rax, uint64_t *rbx,
77     uint64_t *rcx, uint64_t *rdx)
78 {
79 	struct vm *vm = vcpu_vm(vcpu);
80 	int vcpu_id = vcpu_vcpuid(vcpu);
81 	const struct xsave_limits *limits;
82 	uint64_t cr4;
83 	int error, enable_invpcid, enable_rdpid, enable_rdtscp, level,
84 	    width, x2apic_id;
85 	unsigned int func, regs[4], logical_cpus, param;
86 	enum x2apic_state x2apic_state;
87 	uint16_t cores, maxcpus, sockets, threads;
88 
89 	/*
90 	 * The function of CPUID is controlled through the provided value of
91 	 * %eax (and secondarily %ecx, for certain leaf data).
92 	 */
93 	func = (uint32_t)*rax;
94 	param = (uint32_t)*rcx;
95 
96 	VCPU_CTR2(vm, vcpu_id, "cpuid %#x,%#x", func, param);
97 
98 	/*
99 	 * Requests for invalid CPUID levels should map to the highest
100 	 * available level instead.
101 	 */
102 	if (cpu_exthigh != 0 && func >= 0x80000000) {
103 		if (func > cpu_exthigh)
104 			func = cpu_exthigh;
105 	} else if (func >= 0x40000000) {
106 		if (func > CPUID_VM_HIGH)
107 			func = CPUID_VM_HIGH;
108 	} else if (func > cpu_high) {
109 		func = cpu_high;
110 	}
111 
112 	/*
113 	 * In general the approach used for CPU topology is to
114 	 * advertise a flat topology where all CPUs are packages with
115 	 * no multi-core or SMT.
116 	 */
117 	switch (func) {
118 		/*
119 		 * Pass these through to the guest
120 		 */
121 		case CPUID_0000_0000:
122 		case CPUID_0000_0002:
123 		case CPUID_0000_0003:
124 		case CPUID_8000_0000:
125 		case CPUID_8000_0002:
126 		case CPUID_8000_0003:
127 		case CPUID_8000_0004:
128 		case CPUID_8000_0006:
129 			cpuid_count(func, param, regs);
130 			break;
131 		case CPUID_8000_0008:
132 			cpuid_count(func, param, regs);
133 			if (vmm_is_svm()) {
134 				/*
135 				 * As on Intel (0000_0007:0, EDX), mask out
136 				 * unsupported or unsafe AMD extended features
137 				 * (8000_0008 EBX).
138 				 */
139 				regs[1] &= (AMDFEID_CLZERO | AMDFEID_IRPERF |
140 				    AMDFEID_XSAVEERPTR);
141 
142 				vm_get_topology(vm, &sockets, &cores, &threads,
143 				    &maxcpus);
144 				/*
145 				 * Here, width is ApicIdCoreIdSize, present on
146 				 * at least Family 15h and newer.  It
147 				 * represents the "number of bits in the
148 				 * initial apicid that indicate thread id
149 				 * within a package."
150 				 *
151 				 * Our topo_probe_amd() uses it for
152 				 * pkg_id_shift and other OSes may rely on it.
153 				 */
154 				width = MIN(0xF, log2(threads * cores));
155 				if (width < 0x4)
156 					width = 0;
157 				logical_cpus = MIN(0xFF, threads * cores - 1);
158 				regs[2] = (width << AMDID_COREID_SIZE_SHIFT) | logical_cpus;
159 			}
160 			break;
161 
162 		case CPUID_8000_0001:
163 			cpuid_count(func, param, regs);
164 
165 			/*
166 			 * Hide SVM from guest.
167 			 */
168 			regs[2] &= ~AMDID2_SVM;
169 
170 			/*
171 			 * Don't advertise extended performance counter MSRs
172 			 * to the guest.
173 			 */
174 			regs[2] &= ~AMDID2_PCXC;
175 			regs[2] &= ~AMDID2_PNXC;
176 			regs[2] &= ~AMDID2_PTSCEL2I;
177 
178 			/*
179 			 * Don't advertise Instruction Based Sampling feature.
180 			 */
181 			regs[2] &= ~AMDID2_IBS;
182 
183 			/* NodeID MSR not available */
184 			regs[2] &= ~AMDID2_NODE_ID;
185 
186 			/* Don't advertise the OS visible workaround feature */
187 			regs[2] &= ~AMDID2_OSVW;
188 
189 			/* Hide mwaitx/monitorx capability from the guest */
190 			regs[2] &= ~AMDID2_MWAITX;
191 
192 			/* Advertise RDTSCP if it is enabled. */
193 			error = vm_get_capability(vcpu,
194 			    VM_CAP_RDTSCP, &enable_rdtscp);
195 			if (error == 0 && enable_rdtscp)
196 				regs[3] |= AMDID_RDTSCP;
197 			else
198 				regs[3] &= ~AMDID_RDTSCP;
199 			break;
200 
201 		case CPUID_8000_0007:
202 			/*
203 			 * AMD uses this leaf to advertise the processor's
204 			 * power monitoring and RAS capabilities. These
205 			 * features are hardware-specific and exposing
206 			 * them to a guest doesn't make a lot of sense.
207 			 *
208 			 * Intel uses this leaf only to advertise the
209 			 * "Invariant TSC" feature with all other bits
210 			 * being reserved (set to zero).
211 			 */
212 			regs[0] = 0;
213 			regs[1] = 0;
214 			regs[2] = 0;
215 			regs[3] = 0;
216 
217 			/*
218 			 * "Invariant TSC" can be advertised to the guest if:
219 			 * - host TSC frequency is invariant
220 			 * - host TSCs are synchronized across physical cpus
221 			 *
222 			 * XXX This still falls short because the vcpu
223 			 * can observe the TSC moving backwards as it
224 			 * migrates across physical cpus. But at least
225 			 * it should discourage the guest from using the
226 			 * TSC to keep track of time.
227 			 */
228 			if (tsc_is_invariant && smp_tsc)
229 				regs[3] |= AMDPM_TSC_INVARIANT;
230 			break;
231 
232 		case CPUID_8000_001D:
233 			/* AMD Cache topology, like 0000_0004 for Intel. */
234 			if (!vmm_is_svm())
235 				goto default_leaf;
236 
237 			/*
238 			 * Similar to Intel, generate a ficticious cache
239 			 * topology for the guest with L3 shared by the
240 			 * package, and L1 and L2 local to a core.
241 			 */
242 			vm_get_topology(vm, &sockets, &cores, &threads,
243 			    &maxcpus);
244 			switch (param) {
245 			case 0:
246 				logical_cpus = threads;
247 				level = 1;
248 				func = 1;	/* data cache */
249 				break;
250 			case 1:
251 				logical_cpus = threads;
252 				level = 2;
253 				func = 3;	/* unified cache */
254 				break;
255 			case 2:
256 				logical_cpus = threads * cores;
257 				level = 3;
258 				func = 3;	/* unified cache */
259 				break;
260 			default:
261 				logical_cpus = 0;
262 				level = 0;
263 				func = 0;
264 				break;
265 			}
266 
267 			logical_cpus = MIN(0xfff, logical_cpus - 1);
268 			regs[0] = (logical_cpus << 14) | (1 << 8) |
269 			    (level << 5) | func;
270 			regs[1] = (func > 0) ? (CACHE_LINE_SIZE - 1) : 0;
271 			regs[2] = 0;
272 			regs[3] = 0;
273 			break;
274 
275 		case CPUID_8000_001E:
276 			/*
277 			 * AMD Family 16h+ and Hygon Family 18h additional
278 			 * identifiers.
279 			 */
280 			if (!vmm_is_svm() || CPUID_TO_FAMILY(cpu_id) < 0x16)
281 				goto default_leaf;
282 
283 			vm_get_topology(vm, &sockets, &cores, &threads,
284 			    &maxcpus);
285 			regs[0] = vcpu_id;
286 			threads = MIN(0xFF, threads - 1);
287 			regs[1] = (threads << 8) |
288 			    (vcpu_id >> log2(threads + 1));
289 			/*
290 			 * XXX Bhyve topology cannot yet represent >1 node per
291 			 * processor.
292 			 */
293 			regs[2] = 0;
294 			regs[3] = 0;
295 			break;
296 
297 		case CPUID_0000_0001:
298 			do_cpuid(1, regs);
299 
300 			error = vm_get_x2apic_state(vcpu, &x2apic_state);
301 			if (error) {
302 				panic("x86_emulate_cpuid: error %d "
303 				      "fetching x2apic state", error);
304 			}
305 
306 			/*
307 			 * Override the APIC ID only in ebx
308 			 */
309 			regs[1] &= ~(CPUID_LOCAL_APIC_ID);
310 			regs[1] |= (vcpu_id << CPUID_0000_0001_APICID_SHIFT);
311 
312 			/*
313 			 * Don't expose VMX, SpeedStep, TME or SMX capability.
314 			 * Advertise x2APIC capability and Hypervisor guest.
315 			 */
316 			regs[2] &= ~(CPUID2_VMX | CPUID2_EST | CPUID2_TM2);
317 			regs[2] &= ~(CPUID2_SMX);
318 
319 			regs[2] |= CPUID2_HV;
320 
321 			if (x2apic_state != X2APIC_DISABLED)
322 				regs[2] |= CPUID2_X2APIC;
323 			else
324 				regs[2] &= ~CPUID2_X2APIC;
325 
326 			/*
327 			 * Only advertise CPUID2_XSAVE in the guest if
328 			 * the host is using XSAVE.
329 			 */
330 			if (!(regs[2] & CPUID2_OSXSAVE))
331 				regs[2] &= ~CPUID2_XSAVE;
332 
333 			/*
334 			 * If CPUID2_XSAVE is being advertised and the
335 			 * guest has set CR4_XSAVE, set
336 			 * CPUID2_OSXSAVE.
337 			 */
338 			regs[2] &= ~CPUID2_OSXSAVE;
339 			if (regs[2] & CPUID2_XSAVE) {
340 				error = vm_get_register(vcpu,
341 				    VM_REG_GUEST_CR4, &cr4);
342 				if (error)
343 					panic("x86_emulate_cpuid: error %d "
344 					      "fetching %%cr4", error);
345 				if (cr4 & CR4_XSAVE)
346 					regs[2] |= CPUID2_OSXSAVE;
347 			}
348 
349 			/*
350 			 * Hide monitor/mwait until we know how to deal with
351 			 * these instructions.
352 			 */
353 			regs[2] &= ~CPUID2_MON;
354 
355                         /*
356 			 * Hide the performance and debug features.
357 			 */
358 			regs[2] &= ~CPUID2_PDCM;
359 
360 			/*
361 			 * No TSC deadline support in the APIC yet
362 			 */
363 			regs[2] &= ~CPUID2_TSCDLT;
364 
365 			/*
366 			 * Hide thermal monitoring
367 			 */
368 			regs[3] &= ~(CPUID_ACPI | CPUID_TM);
369 
370 			/*
371 			 * Hide the debug store capability.
372 			 */
373 			regs[3] &= ~CPUID_DS;
374 
375 			/*
376 			 * Advertise the Machine Check and MTRR capability.
377 			 *
378 			 * Some guest OSes (e.g. Windows) will not boot if
379 			 * these features are absent.
380 			 */
381 			regs[3] |= (CPUID_MCA | CPUID_MCE | CPUID_MTRR);
382 
383 			vm_get_topology(vm, &sockets, &cores, &threads,
384 			    &maxcpus);
385 			logical_cpus = threads * cores;
386 			regs[1] &= ~CPUID_HTT_CORES;
387 			regs[1] |= (logical_cpus & 0xff) << 16;
388 			regs[3] |= CPUID_HTT;
389 			break;
390 
391 		case CPUID_0000_0004:
392 			cpuid_count(func, param, regs);
393 
394 			if (regs[0] || regs[1] || regs[2] || regs[3]) {
395 				vm_get_topology(vm, &sockets, &cores, &threads,
396 				    &maxcpus);
397 				regs[0] &= 0x3ff;
398 				regs[0] |= (cores - 1) << 26;
399 				/*
400 				 * Cache topology:
401 				 * - L1 and L2 are shared only by the logical
402 				 *   processors in a single core.
403 				 * - L3 and above are shared by all logical
404 				 *   processors in the package.
405 				 */
406 				logical_cpus = threads;
407 				level = (regs[0] >> 5) & 0x7;
408 				if (level >= 3)
409 					logical_cpus *= cores;
410 				regs[0] |= (logical_cpus - 1) << 14;
411 			}
412 			break;
413 
414 		case CPUID_0000_0007:
415 			regs[0] = 0;
416 			regs[1] = 0;
417 			regs[2] = 0;
418 			regs[3] = 0;
419 
420 			/* leaf 0 */
421 			if (param == 0) {
422 				cpuid_count(func, param, regs);
423 
424 				/* Only leaf 0 is supported */
425 				regs[0] = 0;
426 
427 				/*
428 				 * Expose known-safe features.
429 				 */
430 				regs[1] &= CPUID_STDEXT_FSGSBASE |
431 				    CPUID_STDEXT_BMI1 | CPUID_STDEXT_HLE |
432 				    CPUID_STDEXT_AVX2 | CPUID_STDEXT_SMEP |
433 				    CPUID_STDEXT_BMI2 |
434 				    CPUID_STDEXT_ERMS | CPUID_STDEXT_RTM |
435 				    CPUID_STDEXT_AVX512F |
436 				    CPUID_STDEXT_AVX512DQ |
437 				    CPUID_STDEXT_RDSEED |
438 				    CPUID_STDEXT_SMAP |
439 				    CPUID_STDEXT_AVX512PF |
440 				    CPUID_STDEXT_AVX512ER |
441 				    CPUID_STDEXT_AVX512CD | CPUID_STDEXT_SHA |
442 				    CPUID_STDEXT_AVX512BW |
443 				    CPUID_STDEXT_AVX512VL;
444 				regs[2] &= CPUID_STDEXT2_VAES |
445 				    CPUID_STDEXT2_VPCLMULQDQ;
446 				regs[3] &= CPUID_STDEXT3_MD_CLEAR;
447 
448 				/* Advertise RDPID if it is enabled. */
449 				error = vm_get_capability(vcpu, VM_CAP_RDPID,
450 				    &enable_rdpid);
451 				if (error == 0 && enable_rdpid)
452 					regs[2] |= CPUID_STDEXT2_RDPID;
453 
454 				/* Advertise INVPCID if it is enabled. */
455 				error = vm_get_capability(vcpu,
456 				    VM_CAP_ENABLE_INVPCID, &enable_invpcid);
457 				if (error == 0 && enable_invpcid)
458 					regs[1] |= CPUID_STDEXT_INVPCID;
459 			}
460 			break;
461 
462 		case CPUID_0000_0006:
463 			regs[0] = CPUTPM1_ARAT;
464 			regs[1] = 0;
465 			regs[2] = 0;
466 			regs[3] = 0;
467 			break;
468 
469 		case CPUID_0000_000A:
470 			/*
471 			 * Handle the access, but report 0 for
472 			 * all options
473 			 */
474 			regs[0] = 0;
475 			regs[1] = 0;
476 			regs[2] = 0;
477 			regs[3] = 0;
478 			break;
479 
480 		case CPUID_0000_000B:
481 			/*
482 			 * Intel processor topology enumeration
483 			 */
484 			if (vmm_is_intel()) {
485 				vm_get_topology(vm, &sockets, &cores, &threads,
486 				    &maxcpus);
487 				if (param == 0) {
488 					logical_cpus = threads;
489 					width = log2(logical_cpus);
490 					level = CPUID_TYPE_SMT;
491 					x2apic_id = vcpu_id;
492 				}
493 
494 				if (param == 1) {
495 					logical_cpus = threads * cores;
496 					width = log2(logical_cpus);
497 					level = CPUID_TYPE_CORE;
498 					x2apic_id = vcpu_id;
499 				}
500 
501 				if (!cpuid_leaf_b || param >= 2) {
502 					width = 0;
503 					logical_cpus = 0;
504 					level = 0;
505 					x2apic_id = 0;
506 				}
507 
508 				regs[0] = width & 0x1f;
509 				regs[1] = logical_cpus & 0xffff;
510 				regs[2] = (level << 8) | (param & 0xff);
511 				regs[3] = x2apic_id;
512 			} else {
513 				regs[0] = 0;
514 				regs[1] = 0;
515 				regs[2] = 0;
516 				regs[3] = 0;
517 			}
518 			break;
519 
520 		case CPUID_0000_000D:
521 			limits = vmm_get_xsave_limits();
522 			if (!limits->xsave_enabled) {
523 				regs[0] = 0;
524 				regs[1] = 0;
525 				regs[2] = 0;
526 				regs[3] = 0;
527 				break;
528 			}
529 
530 			cpuid_count(func, param, regs);
531 			switch (param) {
532 			case 0:
533 				/*
534 				 * Only permit the guest to use bits
535 				 * that are active in the host in
536 				 * %xcr0.  Also, claim that the
537 				 * maximum save area size is
538 				 * equivalent to the host's current
539 				 * save area size.  Since this runs
540 				 * "inside" of vmrun(), it runs with
541 				 * the guest's xcr0, so the current
542 				 * save area size is correct as-is.
543 				 */
544 				regs[0] &= limits->xcr0_allowed;
545 				regs[2] = limits->xsave_max_size;
546 				regs[3] &= (limits->xcr0_allowed >> 32);
547 				break;
548 			case 1:
549 				/* Only permit XSAVEOPT. */
550 				regs[0] &= CPUID_EXTSTATE_XSAVEOPT;
551 				regs[1] = 0;
552 				regs[2] = 0;
553 				regs[3] = 0;
554 				break;
555 			default:
556 				/*
557 				 * If the leaf is for a permitted feature,
558 				 * pass through as-is, otherwise return
559 				 * all zeroes.
560 				 */
561 				if (!(limits->xcr0_allowed & (1ul << param))) {
562 					regs[0] = 0;
563 					regs[1] = 0;
564 					regs[2] = 0;
565 					regs[3] = 0;
566 				}
567 				break;
568 			}
569 			break;
570 
571 		case CPUID_0000_000F:
572 		case CPUID_0000_0010:
573 			/*
574 			 * Do not report any Resource Director Technology
575 			 * capabilities.  Exposing control of cache or memory
576 			 * controller resource partitioning to the guest is not
577 			 * at all sensible.
578 			 *
579 			 * This is already hidden at a high level by masking of
580 			 * leaf 0x7.  Even still, a guest may look here for
581 			 * detailed capability information.
582 			 */
583 			regs[0] = 0;
584 			regs[1] = 0;
585 			regs[2] = 0;
586 			regs[3] = 0;
587 			break;
588 
589 		case CPUID_0000_0015:
590 			/*
591 			 * Don't report CPU TSC/Crystal ratio and clock
592 			 * values since guests may use these to derive the
593 			 * local APIC frequency..
594 			 */
595 			regs[0] = 0;
596 			regs[1] = 0;
597 			regs[2] = 0;
598 			regs[3] = 0;
599 			break;
600 
601 		case 0x40000000:
602 			regs[0] = CPUID_VM_HIGH;
603 			bcopy(bhyve_id, &regs[1], 4);
604 			bcopy(bhyve_id + 4, &regs[2], 4);
605 			bcopy(bhyve_id + 8, &regs[3], 4);
606 			break;
607 
608 		default:
609 default_leaf:
610 			/*
611 			 * The leaf value has already been clamped so
612 			 * simply pass this through, keeping count of
613 			 * how many unhandled leaf values have been seen.
614 			 */
615 			atomic_add_long(&bhyve_xcpuids, 1);
616 			cpuid_count(func, param, regs);
617 			break;
618 	}
619 
620 	/*
621 	 * CPUID clears the upper 32-bits of the long-mode registers.
622 	 */
623 	*rax = regs[0];
624 	*rbx = regs[1];
625 	*rcx = regs[2];
626 	*rdx = regs[3];
627 
628 	return (1);
629 }
630 
631 bool
632 vm_cpuid_capability(struct vcpu *vcpu, enum vm_cpuid_capability cap)
633 {
634 	bool rv;
635 
636 	KASSERT(cap > 0 && cap < VCC_LAST, ("%s: invalid vm_cpu_capability %d",
637 	    __func__, cap));
638 
639 	/*
640 	 * Simply passthrough the capabilities of the host cpu for now.
641 	 */
642 	rv = false;
643 	switch (cap) {
644 	case VCC_NO_EXECUTE:
645 		if (amd_feature & AMDID_NX)
646 			rv = true;
647 		break;
648 	case VCC_FFXSR:
649 		if (amd_feature & AMDID_FFXSR)
650 			rv = true;
651 		break;
652 	case VCC_TCE:
653 		if (amd_feature2 & AMDID2_TCE)
654 			rv = true;
655 		break;
656 	default:
657 		panic("%s: unknown vm_cpu_capability %d", __func__, cap);
658 	}
659 	return (rv);
660 }
661 
662 int
663 vm_rdmtrr(struct vm_mtrr *mtrr, u_int num, uint64_t *val)
664 {
665 	switch (num) {
666 	case MSR_MTRRcap:
667 		*val = MTRR_CAP_WC | MTRR_CAP_FIXED | VMM_MTRR_VAR_MAX;
668 		break;
669 	case MSR_MTRRdefType:
670 		*val = mtrr->def_type;
671 		break;
672 	case MSR_MTRR4kBase ... MSR_MTRR4kBase + 7:
673 		*val = mtrr->fixed4k[num - MSR_MTRR4kBase];
674 		break;
675 	case MSR_MTRR16kBase ... MSR_MTRR16kBase + 1:
676 		*val = mtrr->fixed16k[num - MSR_MTRR16kBase];
677 		break;
678 	case MSR_MTRR64kBase:
679 		*val = mtrr->fixed64k;
680 		break;
681 	case MSR_MTRRVarBase ... MSR_MTRRVarBase + (VMM_MTRR_VAR_MAX * 2) - 1: {
682 		u_int offset = num - MSR_MTRRVarBase;
683 		if (offset % 2 == 0) {
684 			*val = mtrr->var[offset / 2].base;
685 		} else {
686 			*val = mtrr->var[offset / 2].mask;
687 		}
688 		break;
689 	}
690 	default:
691 		return (-1);
692 	}
693 
694 	return (0);
695 }
696 
697 int
698 vm_wrmtrr(struct vm_mtrr *mtrr, u_int num, uint64_t val)
699 {
700 	switch (num) {
701 	case MSR_MTRRcap:
702 		/* MTRRCAP is read only */
703 		return (-1);
704 	case MSR_MTRRdefType:
705 		if (val & ~VMM_MTRR_DEF_MASK) {
706 			/* generate #GP on writes to reserved fields */
707 			return (-1);
708 		}
709 		mtrr->def_type = val;
710 		break;
711 	case MSR_MTRR4kBase ... MSR_MTRR4kBase + 7:
712 		mtrr->fixed4k[num - MSR_MTRR4kBase] = val;
713 		break;
714 	case MSR_MTRR16kBase ... MSR_MTRR16kBase + 1:
715 		mtrr->fixed16k[num - MSR_MTRR16kBase] = val;
716 		break;
717 	case MSR_MTRR64kBase:
718 		mtrr->fixed64k = val;
719 		break;
720 	case MSR_MTRRVarBase ... MSR_MTRRVarBase + (VMM_MTRR_VAR_MAX * 2) - 1: {
721 		u_int offset = num - MSR_MTRRVarBase;
722 		if (offset % 2 == 0) {
723 			if (val & ~VMM_MTRR_PHYSBASE_MASK) {
724 				/* generate #GP on writes to reserved fields */
725 				return (-1);
726 			}
727 			mtrr->var[offset / 2].base = val;
728 		} else {
729 			if (val & ~VMM_MTRR_PHYSMASK_MASK) {
730 				/* generate #GP on writes to reserved fields */
731 				return (-1);
732 			}
733 			mtrr->var[offset / 2].mask = val;
734 		}
735 		break;
736 	}
737 	default:
738 		return (-1);
739 	}
740 
741 	return (0);
742 }
743