xref: /freebsd/sys/amd64/vmm/x86.c (revision f540a43052c12c76d3453ead881248d5467a1ab0)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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  * $FreeBSD$
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include <sys/param.h>
35 #include <sys/pcpu.h>
36 #include <sys/systm.h>
37 #include <sys/sysctl.h>
38 
39 #include <machine/clock.h>
40 #include <machine/cpufunc.h>
41 #include <machine/md_var.h>
42 #include <machine/segments.h>
43 #include <machine/specialreg.h>
44 
45 #include <machine/vmm.h>
46 
47 #include "vmm_host.h"
48 #include "vmm_ktr.h"
49 #include "vmm_util.h"
50 #include "x86.h"
51 
52 SYSCTL_DECL(_hw_vmm);
53 static SYSCTL_NODE(_hw_vmm, OID_AUTO, topology, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
54     NULL);
55 
56 #define	CPUID_VM_HIGH		0x40000000
57 
58 static const char bhyve_id[12] = "bhyve bhyve ";
59 
60 static uint64_t bhyve_xcpuids;
61 SYSCTL_ULONG(_hw_vmm, OID_AUTO, bhyve_xcpuids, CTLFLAG_RW, &bhyve_xcpuids, 0,
62     "Number of times an unknown cpuid leaf was accessed");
63 
64 static int cpuid_leaf_b = 1;
65 SYSCTL_INT(_hw_vmm_topology, OID_AUTO, cpuid_leaf_b, CTLFLAG_RDTUN,
66     &cpuid_leaf_b, 0, NULL);
67 
68 /*
69  * Round up to the next power of two, if necessary, and then take log2.
70  * Returns -1 if argument is zero.
71  */
72 static __inline int
73 log2(u_int x)
74 {
75 
76 	return (fls(x << (1 - powerof2(x))) - 1);
77 }
78 
79 int
80 x86_emulate_cpuid(struct vcpu *vcpu, uint64_t *rax, uint64_t *rbx,
81     uint64_t *rcx, uint64_t *rdx)
82 {
83 	struct vm *vm = vcpu_vm(vcpu);
84 	int vcpu_id = vcpu_vcpuid(vcpu);
85 	const struct xsave_limits *limits;
86 	uint64_t cr4;
87 	int error, enable_invpcid, enable_rdpid, enable_rdtscp, level,
88 	    width, x2apic_id;
89 	unsigned int func, regs[4], logical_cpus, param;
90 	enum x2apic_state x2apic_state;
91 	uint16_t cores, maxcpus, sockets, threads;
92 
93 	/*
94 	 * The function of CPUID is controlled through the provided value of
95 	 * %eax (and secondarily %ecx, for certain leaf data).
96 	 */
97 	func = (uint32_t)*rax;
98 	param = (uint32_t)*rcx;
99 
100 	VCPU_CTR2(vm, vcpu_id, "cpuid %#x,%#x", func, param);
101 
102 	/*
103 	 * Requests for invalid CPUID levels should map to the highest
104 	 * available level instead.
105 	 */
106 	if (cpu_exthigh != 0 && func >= 0x80000000) {
107 		if (func > cpu_exthigh)
108 			func = cpu_exthigh;
109 	} else if (func >= 0x40000000) {
110 		if (func > CPUID_VM_HIGH)
111 			func = CPUID_VM_HIGH;
112 	} else if (func > cpu_high) {
113 		func = cpu_high;
114 	}
115 
116 	/*
117 	 * In general the approach used for CPU topology is to
118 	 * advertise a flat topology where all CPUs are packages with
119 	 * no multi-core or SMT.
120 	 */
121 	switch (func) {
122 		/*
123 		 * Pass these through to the guest
124 		 */
125 		case CPUID_0000_0000:
126 		case CPUID_0000_0002:
127 		case CPUID_0000_0003:
128 		case CPUID_8000_0000:
129 		case CPUID_8000_0002:
130 		case CPUID_8000_0003:
131 		case CPUID_8000_0004:
132 		case CPUID_8000_0006:
133 			cpuid_count(func, param, regs);
134 			break;
135 		case CPUID_8000_0008:
136 			cpuid_count(func, param, regs);
137 			if (vmm_is_svm()) {
138 				/*
139 				 * As on Intel (0000_0007:0, EDX), mask out
140 				 * unsupported or unsafe AMD extended features
141 				 * (8000_0008 EBX).
142 				 */
143 				regs[1] &= (AMDFEID_CLZERO | AMDFEID_IRPERF |
144 				    AMDFEID_XSAVEERPTR);
145 
146 				vm_get_topology(vm, &sockets, &cores, &threads,
147 				    &maxcpus);
148 				/*
149 				 * Here, width is ApicIdCoreIdSize, present on
150 				 * at least Family 15h and newer.  It
151 				 * represents the "number of bits in the
152 				 * initial apicid that indicate thread id
153 				 * within a package."
154 				 *
155 				 * Our topo_probe_amd() uses it for
156 				 * pkg_id_shift and other OSes may rely on it.
157 				 */
158 				width = MIN(0xF, log2(threads * cores));
159 				if (width < 0x4)
160 					width = 0;
161 				logical_cpus = MIN(0xFF, threads * cores - 1);
162 				regs[2] = (width << AMDID_COREID_SIZE_SHIFT) | logical_cpus;
163 			}
164 			break;
165 
166 		case CPUID_8000_0001:
167 			cpuid_count(func, param, regs);
168 
169 			/*
170 			 * Hide SVM from guest.
171 			 */
172 			regs[2] &= ~AMDID2_SVM;
173 
174 			/*
175 			 * Don't advertise extended performance counter MSRs
176 			 * to the guest.
177 			 */
178 			regs[2] &= ~AMDID2_PCXC;
179 			regs[2] &= ~AMDID2_PNXC;
180 			regs[2] &= ~AMDID2_PTSCEL2I;
181 
182 			/*
183 			 * Don't advertise Instruction Based Sampling feature.
184 			 */
185 			regs[2] &= ~AMDID2_IBS;
186 
187 			/* NodeID MSR not available */
188 			regs[2] &= ~AMDID2_NODE_ID;
189 
190 			/* Don't advertise the OS visible workaround feature */
191 			regs[2] &= ~AMDID2_OSVW;
192 
193 			/* Hide mwaitx/monitorx capability from the guest */
194 			regs[2] &= ~AMDID2_MWAITX;
195 
196 			/* Advertise RDTSCP if it is enabled. */
197 			error = vm_get_capability(vcpu,
198 			    VM_CAP_RDTSCP, &enable_rdtscp);
199 			if (error == 0 && enable_rdtscp)
200 				regs[3] |= AMDID_RDTSCP;
201 			else
202 				regs[3] &= ~AMDID_RDTSCP;
203 			break;
204 
205 		case CPUID_8000_0007:
206 			/*
207 			 * AMD uses this leaf to advertise the processor's
208 			 * power monitoring and RAS capabilities. These
209 			 * features are hardware-specific and exposing
210 			 * them to a guest doesn't make a lot of sense.
211 			 *
212 			 * Intel uses this leaf only to advertise the
213 			 * "Invariant TSC" feature with all other bits
214 			 * being reserved (set to zero).
215 			 */
216 			regs[0] = 0;
217 			regs[1] = 0;
218 			regs[2] = 0;
219 			regs[3] = 0;
220 
221 			/*
222 			 * "Invariant TSC" can be advertised to the guest if:
223 			 * - host TSC frequency is invariant
224 			 * - host TSCs are synchronized across physical cpus
225 			 *
226 			 * XXX This still falls short because the vcpu
227 			 * can observe the TSC moving backwards as it
228 			 * migrates across physical cpus. But at least
229 			 * it should discourage the guest from using the
230 			 * TSC to keep track of time.
231 			 */
232 			if (tsc_is_invariant && smp_tsc)
233 				regs[3] |= AMDPM_TSC_INVARIANT;
234 			break;
235 
236 		case CPUID_8000_001D:
237 			/* AMD Cache topology, like 0000_0004 for Intel. */
238 			if (!vmm_is_svm())
239 				goto default_leaf;
240 
241 			/*
242 			 * Similar to Intel, generate a ficticious cache
243 			 * topology for the guest with L3 shared by the
244 			 * package, and L1 and L2 local to a core.
245 			 */
246 			vm_get_topology(vm, &sockets, &cores, &threads,
247 			    &maxcpus);
248 			switch (param) {
249 			case 0:
250 				logical_cpus = threads;
251 				level = 1;
252 				func = 1;	/* data cache */
253 				break;
254 			case 1:
255 				logical_cpus = threads;
256 				level = 2;
257 				func = 3;	/* unified cache */
258 				break;
259 			case 2:
260 				logical_cpus = threads * cores;
261 				level = 3;
262 				func = 3;	/* unified cache */
263 				break;
264 			default:
265 				logical_cpus = 0;
266 				level = 0;
267 				func = 0;
268 				break;
269 			}
270 
271 			logical_cpus = MIN(0xfff, logical_cpus - 1);
272 			regs[0] = (logical_cpus << 14) | (1 << 8) |
273 			    (level << 5) | func;
274 			regs[1] = (func > 0) ? (CACHE_LINE_SIZE - 1) : 0;
275 			regs[2] = 0;
276 			regs[3] = 0;
277 			break;
278 
279 		case CPUID_8000_001E:
280 			/*
281 			 * AMD Family 16h+ and Hygon Family 18h additional
282 			 * identifiers.
283 			 */
284 			if (!vmm_is_svm() || CPUID_TO_FAMILY(cpu_id) < 0x16)
285 				goto default_leaf;
286 
287 			vm_get_topology(vm, &sockets, &cores, &threads,
288 			    &maxcpus);
289 			regs[0] = vcpu_id;
290 			threads = MIN(0xFF, threads - 1);
291 			regs[1] = (threads << 8) |
292 			    (vcpu_id >> log2(threads + 1));
293 			/*
294 			 * XXX Bhyve topology cannot yet represent >1 node per
295 			 * processor.
296 			 */
297 			regs[2] = 0;
298 			regs[3] = 0;
299 			break;
300 
301 		case CPUID_0000_0001:
302 			do_cpuid(1, regs);
303 
304 			error = vm_get_x2apic_state(vcpu, &x2apic_state);
305 			if (error) {
306 				panic("x86_emulate_cpuid: error %d "
307 				      "fetching x2apic state", error);
308 			}
309 
310 			/*
311 			 * Override the APIC ID only in ebx
312 			 */
313 			regs[1] &= ~(CPUID_LOCAL_APIC_ID);
314 			regs[1] |= (vcpu_id << CPUID_0000_0001_APICID_SHIFT);
315 
316 			/*
317 			 * Don't expose VMX, SpeedStep, TME or SMX capability.
318 			 * Advertise x2APIC capability and Hypervisor guest.
319 			 */
320 			regs[2] &= ~(CPUID2_VMX | CPUID2_EST | CPUID2_TM2);
321 			regs[2] &= ~(CPUID2_SMX);
322 
323 			regs[2] |= CPUID2_HV;
324 
325 			if (x2apic_state != X2APIC_DISABLED)
326 				regs[2] |= CPUID2_X2APIC;
327 			else
328 				regs[2] &= ~CPUID2_X2APIC;
329 
330 			/*
331 			 * Only advertise CPUID2_XSAVE in the guest if
332 			 * the host is using XSAVE.
333 			 */
334 			if (!(regs[2] & CPUID2_OSXSAVE))
335 				regs[2] &= ~CPUID2_XSAVE;
336 
337 			/*
338 			 * If CPUID2_XSAVE is being advertised and the
339 			 * guest has set CR4_XSAVE, set
340 			 * CPUID2_OSXSAVE.
341 			 */
342 			regs[2] &= ~CPUID2_OSXSAVE;
343 			if (regs[2] & CPUID2_XSAVE) {
344 				error = vm_get_register(vcpu,
345 				    VM_REG_GUEST_CR4, &cr4);
346 				if (error)
347 					panic("x86_emulate_cpuid: error %d "
348 					      "fetching %%cr4", error);
349 				if (cr4 & CR4_XSAVE)
350 					regs[2] |= CPUID2_OSXSAVE;
351 			}
352 
353 			/*
354 			 * Hide monitor/mwait until we know how to deal with
355 			 * these instructions.
356 			 */
357 			regs[2] &= ~CPUID2_MON;
358 
359                         /*
360 			 * Hide the performance and debug features.
361 			 */
362 			regs[2] &= ~CPUID2_PDCM;
363 
364 			/*
365 			 * No TSC deadline support in the APIC yet
366 			 */
367 			regs[2] &= ~CPUID2_TSCDLT;
368 
369 			/*
370 			 * Hide thermal monitoring
371 			 */
372 			regs[3] &= ~(CPUID_ACPI | CPUID_TM);
373 
374 			/*
375 			 * Hide the debug store capability.
376 			 */
377 			regs[3] &= ~CPUID_DS;
378 
379 			/*
380 			 * Advertise the Machine Check and MTRR capability.
381 			 *
382 			 * Some guest OSes (e.g. Windows) will not boot if
383 			 * these features are absent.
384 			 */
385 			regs[3] |= (CPUID_MCA | CPUID_MCE | CPUID_MTRR);
386 
387 			vm_get_topology(vm, &sockets, &cores, &threads,
388 			    &maxcpus);
389 			logical_cpus = threads * cores;
390 			regs[1] &= ~CPUID_HTT_CORES;
391 			regs[1] |= (logical_cpus & 0xff) << 16;
392 			regs[3] |= CPUID_HTT;
393 			break;
394 
395 		case CPUID_0000_0004:
396 			cpuid_count(func, param, regs);
397 
398 			if (regs[0] || regs[1] || regs[2] || regs[3]) {
399 				vm_get_topology(vm, &sockets, &cores, &threads,
400 				    &maxcpus);
401 				regs[0] &= 0x3ff;
402 				regs[0] |= (cores - 1) << 26;
403 				/*
404 				 * Cache topology:
405 				 * - L1 and L2 are shared only by the logical
406 				 *   processors in a single core.
407 				 * - L3 and above are shared by all logical
408 				 *   processors in the package.
409 				 */
410 				logical_cpus = threads;
411 				level = (regs[0] >> 5) & 0x7;
412 				if (level >= 3)
413 					logical_cpus *= cores;
414 				regs[0] |= (logical_cpus - 1) << 14;
415 			}
416 			break;
417 
418 		case CPUID_0000_0007:
419 			regs[0] = 0;
420 			regs[1] = 0;
421 			regs[2] = 0;
422 			regs[3] = 0;
423 
424 			/* leaf 0 */
425 			if (param == 0) {
426 				cpuid_count(func, param, regs);
427 
428 				/* Only leaf 0 is supported */
429 				regs[0] = 0;
430 
431 				/*
432 				 * Expose known-safe features.
433 				 */
434 				regs[1] &= CPUID_STDEXT_FSGSBASE |
435 				    CPUID_STDEXT_BMI1 | CPUID_STDEXT_HLE |
436 				    CPUID_STDEXT_AVX2 | CPUID_STDEXT_SMEP |
437 				    CPUID_STDEXT_BMI2 |
438 				    CPUID_STDEXT_ERMS | CPUID_STDEXT_RTM |
439 				    CPUID_STDEXT_AVX512F |
440 				    CPUID_STDEXT_AVX512DQ |
441 				    CPUID_STDEXT_RDSEED |
442 				    CPUID_STDEXT_SMAP |
443 				    CPUID_STDEXT_AVX512PF |
444 				    CPUID_STDEXT_AVX512ER |
445 				    CPUID_STDEXT_AVX512CD | CPUID_STDEXT_SHA |
446 				    CPUID_STDEXT_AVX512BW |
447 				    CPUID_STDEXT_AVX512VL;
448 				regs[2] &= CPUID_STDEXT2_VAES |
449 				    CPUID_STDEXT2_VPCLMULQDQ;
450 				regs[3] &= CPUID_STDEXT3_MD_CLEAR;
451 
452 				/* Advertise RDPID if it is enabled. */
453 				error = vm_get_capability(vcpu, VM_CAP_RDPID,
454 				    &enable_rdpid);
455 				if (error == 0 && enable_rdpid)
456 					regs[2] |= CPUID_STDEXT2_RDPID;
457 
458 				/* Advertise INVPCID if it is enabled. */
459 				error = vm_get_capability(vcpu,
460 				    VM_CAP_ENABLE_INVPCID, &enable_invpcid);
461 				if (error == 0 && enable_invpcid)
462 					regs[1] |= CPUID_STDEXT_INVPCID;
463 			}
464 			break;
465 
466 		case CPUID_0000_0006:
467 			regs[0] = CPUTPM1_ARAT;
468 			regs[1] = 0;
469 			regs[2] = 0;
470 			regs[3] = 0;
471 			break;
472 
473 		case CPUID_0000_000A:
474 			/*
475 			 * Handle the access, but report 0 for
476 			 * all options
477 			 */
478 			regs[0] = 0;
479 			regs[1] = 0;
480 			regs[2] = 0;
481 			regs[3] = 0;
482 			break;
483 
484 		case CPUID_0000_000B:
485 			/*
486 			 * Intel processor topology enumeration
487 			 */
488 			if (vmm_is_intel()) {
489 				vm_get_topology(vm, &sockets, &cores, &threads,
490 				    &maxcpus);
491 				if (param == 0) {
492 					logical_cpus = threads;
493 					width = log2(logical_cpus);
494 					level = CPUID_TYPE_SMT;
495 					x2apic_id = vcpu_id;
496 				}
497 
498 				if (param == 1) {
499 					logical_cpus = threads * cores;
500 					width = log2(logical_cpus);
501 					level = CPUID_TYPE_CORE;
502 					x2apic_id = vcpu_id;
503 				}
504 
505 				if (!cpuid_leaf_b || param >= 2) {
506 					width = 0;
507 					logical_cpus = 0;
508 					level = 0;
509 					x2apic_id = 0;
510 				}
511 
512 				regs[0] = width & 0x1f;
513 				regs[1] = logical_cpus & 0xffff;
514 				regs[2] = (level << 8) | (param & 0xff);
515 				regs[3] = x2apic_id;
516 			} else {
517 				regs[0] = 0;
518 				regs[1] = 0;
519 				regs[2] = 0;
520 				regs[3] = 0;
521 			}
522 			break;
523 
524 		case CPUID_0000_000D:
525 			limits = vmm_get_xsave_limits();
526 			if (!limits->xsave_enabled) {
527 				regs[0] = 0;
528 				regs[1] = 0;
529 				regs[2] = 0;
530 				regs[3] = 0;
531 				break;
532 			}
533 
534 			cpuid_count(func, param, regs);
535 			switch (param) {
536 			case 0:
537 				/*
538 				 * Only permit the guest to use bits
539 				 * that are active in the host in
540 				 * %xcr0.  Also, claim that the
541 				 * maximum save area size is
542 				 * equivalent to the host's current
543 				 * save area size.  Since this runs
544 				 * "inside" of vmrun(), it runs with
545 				 * the guest's xcr0, so the current
546 				 * save area size is correct as-is.
547 				 */
548 				regs[0] &= limits->xcr0_allowed;
549 				regs[2] = limits->xsave_max_size;
550 				regs[3] &= (limits->xcr0_allowed >> 32);
551 				break;
552 			case 1:
553 				/* Only permit XSAVEOPT. */
554 				regs[0] &= CPUID_EXTSTATE_XSAVEOPT;
555 				regs[1] = 0;
556 				regs[2] = 0;
557 				regs[3] = 0;
558 				break;
559 			default:
560 				/*
561 				 * If the leaf is for a permitted feature,
562 				 * pass through as-is, otherwise return
563 				 * all zeroes.
564 				 */
565 				if (!(limits->xcr0_allowed & (1ul << param))) {
566 					regs[0] = 0;
567 					regs[1] = 0;
568 					regs[2] = 0;
569 					regs[3] = 0;
570 				}
571 				break;
572 			}
573 			break;
574 
575 		case CPUID_0000_000F:
576 		case CPUID_0000_0010:
577 			/*
578 			 * Do not report any Resource Director Technology
579 			 * capabilities.  Exposing control of cache or memory
580 			 * controller resource partitioning to the guest is not
581 			 * at all sensible.
582 			 *
583 			 * This is already hidden at a high level by masking of
584 			 * leaf 0x7.  Even still, a guest may look here for
585 			 * detailed capability information.
586 			 */
587 			regs[0] = 0;
588 			regs[1] = 0;
589 			regs[2] = 0;
590 			regs[3] = 0;
591 			break;
592 
593 		case CPUID_0000_0015:
594 			/*
595 			 * Don't report CPU TSC/Crystal ratio and clock
596 			 * values since guests may use these to derive the
597 			 * local APIC frequency..
598 			 */
599 			regs[0] = 0;
600 			regs[1] = 0;
601 			regs[2] = 0;
602 			regs[3] = 0;
603 			break;
604 
605 		case 0x40000000:
606 			regs[0] = CPUID_VM_HIGH;
607 			bcopy(bhyve_id, &regs[1], 4);
608 			bcopy(bhyve_id + 4, &regs[2], 4);
609 			bcopy(bhyve_id + 8, &regs[3], 4);
610 			break;
611 
612 		default:
613 default_leaf:
614 			/*
615 			 * The leaf value has already been clamped so
616 			 * simply pass this through, keeping count of
617 			 * how many unhandled leaf values have been seen.
618 			 */
619 			atomic_add_long(&bhyve_xcpuids, 1);
620 			cpuid_count(func, param, regs);
621 			break;
622 	}
623 
624 	/*
625 	 * CPUID clears the upper 32-bits of the long-mode registers.
626 	 */
627 	*rax = regs[0];
628 	*rbx = regs[1];
629 	*rcx = regs[2];
630 	*rdx = regs[3];
631 
632 	return (1);
633 }
634 
635 bool
636 vm_cpuid_capability(struct vcpu *vcpu, enum vm_cpuid_capability cap)
637 {
638 	bool rv;
639 
640 	KASSERT(cap > 0 && cap < VCC_LAST, ("%s: invalid vm_cpu_capability %d",
641 	    __func__, cap));
642 
643 	/*
644 	 * Simply passthrough the capabilities of the host cpu for now.
645 	 */
646 	rv = false;
647 	switch (cap) {
648 	case VCC_NO_EXECUTE:
649 		if (amd_feature & AMDID_NX)
650 			rv = true;
651 		break;
652 	case VCC_FFXSR:
653 		if (amd_feature & AMDID_FFXSR)
654 			rv = true;
655 		break;
656 	case VCC_TCE:
657 		if (amd_feature2 & AMDID2_TCE)
658 			rv = true;
659 		break;
660 	default:
661 		panic("%s: unknown vm_cpu_capability %d", __func__, cap);
662 	}
663 	return (rv);
664 }
665 
666 int
667 vm_rdmtrr(struct vm_mtrr *mtrr, u_int num, uint64_t *val)
668 {
669 	switch (num) {
670 	case MSR_MTRRcap:
671 		*val = MTRR_CAP_WC | MTRR_CAP_FIXED | VMM_MTRR_VAR_MAX;
672 		break;
673 	case MSR_MTRRdefType:
674 		*val = mtrr->def_type;
675 		break;
676 	case MSR_MTRR4kBase ... MSR_MTRR4kBase + 7:
677 		*val = mtrr->fixed4k[num - MSR_MTRR4kBase];
678 		break;
679 	case MSR_MTRR16kBase ... MSR_MTRR16kBase + 1:
680 		*val = mtrr->fixed16k[num - MSR_MTRR16kBase];
681 		break;
682 	case MSR_MTRR64kBase:
683 		*val = mtrr->fixed64k;
684 		break;
685 	case MSR_MTRRVarBase ... MSR_MTRRVarBase + (VMM_MTRR_VAR_MAX * 2) - 1: {
686 		u_int offset = num - MSR_MTRRVarBase;
687 		if (offset % 2 == 0) {
688 			*val = mtrr->var[offset / 2].base;
689 		} else {
690 			*val = mtrr->var[offset / 2].mask;
691 		}
692 		break;
693 	}
694 	default:
695 		return (-1);
696 	}
697 
698 	return (0);
699 }
700 
701 int
702 vm_wrmtrr(struct vm_mtrr *mtrr, u_int num, uint64_t val)
703 {
704 	switch (num) {
705 	case MSR_MTRRcap:
706 		/* MTRRCAP is read only */
707 		return (-1);
708 	case MSR_MTRRdefType:
709 		if (val & ~VMM_MTRR_DEF_MASK) {
710 			/* generate #GP on writes to reserved fields */
711 			return (-1);
712 		}
713 		mtrr->def_type = val;
714 		break;
715 	case MSR_MTRR4kBase ... MSR_MTRR4kBase + 7:
716 		mtrr->fixed4k[num - MSR_MTRR4kBase] = val;
717 		break;
718 	case MSR_MTRR16kBase ... MSR_MTRR16kBase + 1:
719 		mtrr->fixed16k[num - MSR_MTRR16kBase] = val;
720 		break;
721 	case MSR_MTRR64kBase:
722 		mtrr->fixed64k = val;
723 		break;
724 	case MSR_MTRRVarBase ... MSR_MTRRVarBase + (VMM_MTRR_VAR_MAX * 2) - 1: {
725 		u_int offset = num - MSR_MTRRVarBase;
726 		if (offset % 2 == 0) {
727 			if (val & ~VMM_MTRR_PHYSBASE_MASK) {
728 				/* generate #GP on writes to reserved fields */
729 				return (-1);
730 			}
731 			mtrr->var[offset / 2].base = val;
732 		} else {
733 			if (val & ~VMM_MTRR_PHYSMASK_MASK) {
734 				/* generate #GP on writes to reserved fields */
735 				return (-1);
736 			}
737 			mtrr->var[offset / 2].mask = val;
738 		}
739 		break;
740 	}
741 	default:
742 		return (-1);
743 	}
744 
745 	return (0);
746 }
747