xref: /freebsd/sys/amd64/vmm/x86.c (revision c7a063741720ef81d4caa4613242579d12f1d605)
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 #if __FreeBSD_version < 1200060	/* Remove after 11 EOL helps MFCing */
65 extern u_int threads_per_core;
66 SYSCTL_UINT(_hw_vmm_topology, OID_AUTO, threads_per_core, CTLFLAG_RDTUN,
67     &threads_per_core, 0, NULL);
68 
69 extern u_int cores_per_package;
70 SYSCTL_UINT(_hw_vmm_topology, OID_AUTO, cores_per_package, CTLFLAG_RDTUN,
71     &cores_per_package, 0, NULL);
72 #endif
73 
74 static int cpuid_leaf_b = 1;
75 SYSCTL_INT(_hw_vmm_topology, OID_AUTO, cpuid_leaf_b, CTLFLAG_RDTUN,
76     &cpuid_leaf_b, 0, NULL);
77 
78 /*
79  * Round up to the next power of two, if necessary, and then take log2.
80  * Returns -1 if argument is zero.
81  */
82 static __inline int
83 log2(u_int x)
84 {
85 
86 	return (fls(x << (1 - powerof2(x))) - 1);
87 }
88 
89 int
90 x86_emulate_cpuid(struct vcpu *vcpu, uint64_t *rax, uint64_t *rbx,
91     uint64_t *rcx, uint64_t *rdx)
92 {
93 	struct vm *vm = vcpu_vm(vcpu);
94 	int vcpu_id = vcpu_vcpuid(vcpu);
95 	const struct xsave_limits *limits;
96 	uint64_t cr4;
97 	int error, enable_invpcid, enable_rdpid, enable_rdtscp, level,
98 	    width, x2apic_id;
99 	unsigned int func, regs[4], logical_cpus, param;
100 	enum x2apic_state x2apic_state;
101 	uint16_t cores, maxcpus, sockets, threads;
102 
103 	/*
104 	 * The function of CPUID is controlled through the provided value of
105 	 * %eax (and secondarily %ecx, for certain leaf data).
106 	 */
107 	func = (uint32_t)*rax;
108 	param = (uint32_t)*rcx;
109 
110 	VCPU_CTR2(vm, vcpu_id, "cpuid %#x,%#x", func, param);
111 
112 	/*
113 	 * Requests for invalid CPUID levels should map to the highest
114 	 * available level instead.
115 	 */
116 	if (cpu_exthigh != 0 && func >= 0x80000000) {
117 		if (func > cpu_exthigh)
118 			func = cpu_exthigh;
119 	} else if (func >= 0x40000000) {
120 		if (func > CPUID_VM_HIGH)
121 			func = CPUID_VM_HIGH;
122 	} else if (func > cpu_high) {
123 		func = cpu_high;
124 	}
125 
126 	/*
127 	 * In general the approach used for CPU topology is to
128 	 * advertise a flat topology where all CPUs are packages with
129 	 * no multi-core or SMT.
130 	 */
131 	switch (func) {
132 		/*
133 		 * Pass these through to the guest
134 		 */
135 		case CPUID_0000_0000:
136 		case CPUID_0000_0002:
137 		case CPUID_0000_0003:
138 		case CPUID_8000_0000:
139 		case CPUID_8000_0002:
140 		case CPUID_8000_0003:
141 		case CPUID_8000_0004:
142 		case CPUID_8000_0006:
143 			cpuid_count(func, param, regs);
144 			break;
145 		case CPUID_8000_0008:
146 			cpuid_count(func, param, regs);
147 			if (vmm_is_svm()) {
148 				/*
149 				 * As on Intel (0000_0007:0, EDX), mask out
150 				 * unsupported or unsafe AMD extended features
151 				 * (8000_0008 EBX).
152 				 */
153 				regs[1] &= (AMDFEID_CLZERO | AMDFEID_IRPERF |
154 				    AMDFEID_XSAVEERPTR);
155 
156 				vm_get_topology(vm, &sockets, &cores, &threads,
157 				    &maxcpus);
158 				/*
159 				 * Here, width is ApicIdCoreIdSize, present on
160 				 * at least Family 15h and newer.  It
161 				 * represents the "number of bits in the
162 				 * initial apicid that indicate thread id
163 				 * within a package."
164 				 *
165 				 * Our topo_probe_amd() uses it for
166 				 * pkg_id_shift and other OSes may rely on it.
167 				 */
168 				width = MIN(0xF, log2(threads * cores));
169 				if (width < 0x4)
170 					width = 0;
171 				logical_cpus = MIN(0xFF, threads * cores - 1);
172 				regs[2] = (width << AMDID_COREID_SIZE_SHIFT) | logical_cpus;
173 			}
174 			break;
175 
176 		case CPUID_8000_0001:
177 			cpuid_count(func, param, regs);
178 
179 			/*
180 			 * Hide SVM from guest.
181 			 */
182 			regs[2] &= ~AMDID2_SVM;
183 
184 			/*
185 			 * Don't advertise extended performance counter MSRs
186 			 * to the guest.
187 			 */
188 			regs[2] &= ~AMDID2_PCXC;
189 			regs[2] &= ~AMDID2_PNXC;
190 			regs[2] &= ~AMDID2_PTSCEL2I;
191 
192 			/*
193 			 * Don't advertise Instruction Based Sampling feature.
194 			 */
195 			regs[2] &= ~AMDID2_IBS;
196 
197 			/* NodeID MSR not available */
198 			regs[2] &= ~AMDID2_NODE_ID;
199 
200 			/* Don't advertise the OS visible workaround feature */
201 			regs[2] &= ~AMDID2_OSVW;
202 
203 			/* Hide mwaitx/monitorx capability from the guest */
204 			regs[2] &= ~AMDID2_MWAITX;
205 
206 			/* Advertise RDTSCP if it is enabled. */
207 			error = vm_get_capability(vcpu,
208 			    VM_CAP_RDTSCP, &enable_rdtscp);
209 			if (error == 0 && enable_rdtscp)
210 				regs[3] |= AMDID_RDTSCP;
211 			else
212 				regs[3] &= ~AMDID_RDTSCP;
213 			break;
214 
215 		case CPUID_8000_0007:
216 			/*
217 			 * AMD uses this leaf to advertise the processor's
218 			 * power monitoring and RAS capabilities. These
219 			 * features are hardware-specific and exposing
220 			 * them to a guest doesn't make a lot of sense.
221 			 *
222 			 * Intel uses this leaf only to advertise the
223 			 * "Invariant TSC" feature with all other bits
224 			 * being reserved (set to zero).
225 			 */
226 			regs[0] = 0;
227 			regs[1] = 0;
228 			regs[2] = 0;
229 			regs[3] = 0;
230 
231 			/*
232 			 * "Invariant TSC" can be advertised to the guest if:
233 			 * - host TSC frequency is invariant
234 			 * - host TSCs are synchronized across physical cpus
235 			 *
236 			 * XXX This still falls short because the vcpu
237 			 * can observe the TSC moving backwards as it
238 			 * migrates across physical cpus. But at least
239 			 * it should discourage the guest from using the
240 			 * TSC to keep track of time.
241 			 */
242 			if (tsc_is_invariant && smp_tsc)
243 				regs[3] |= AMDPM_TSC_INVARIANT;
244 			break;
245 
246 		case CPUID_8000_001D:
247 			/* AMD Cache topology, like 0000_0004 for Intel. */
248 			if (!vmm_is_svm())
249 				goto default_leaf;
250 
251 			/*
252 			 * Similar to Intel, generate a ficticious cache
253 			 * topology for the guest with L3 shared by the
254 			 * package, and L1 and L2 local to a core.
255 			 */
256 			vm_get_topology(vm, &sockets, &cores, &threads,
257 			    &maxcpus);
258 			switch (param) {
259 			case 0:
260 				logical_cpus = threads;
261 				level = 1;
262 				func = 1;	/* data cache */
263 				break;
264 			case 1:
265 				logical_cpus = threads;
266 				level = 2;
267 				func = 3;	/* unified cache */
268 				break;
269 			case 2:
270 				logical_cpus = threads * cores;
271 				level = 3;
272 				func = 3;	/* unified cache */
273 				break;
274 			default:
275 				logical_cpus = 0;
276 				level = 0;
277 				func = 0;
278 				break;
279 			}
280 
281 			logical_cpus = MIN(0xfff, logical_cpus - 1);
282 			regs[0] = (logical_cpus << 14) | (1 << 8) |
283 			    (level << 5) | func;
284 			regs[1] = (func > 0) ? (CACHE_LINE_SIZE - 1) : 0;
285 			regs[2] = 0;
286 			regs[3] = 0;
287 			break;
288 
289 		case CPUID_8000_001E:
290 			/*
291 			 * AMD Family 16h+ and Hygon Family 18h additional
292 			 * identifiers.
293 			 */
294 			if (!vmm_is_svm() || CPUID_TO_FAMILY(cpu_id) < 0x16)
295 				goto default_leaf;
296 
297 			vm_get_topology(vm, &sockets, &cores, &threads,
298 			    &maxcpus);
299 			regs[0] = vcpu_id;
300 			threads = MIN(0xFF, threads - 1);
301 			regs[1] = (threads << 8) |
302 			    (vcpu_id >> log2(threads + 1));
303 			/*
304 			 * XXX Bhyve topology cannot yet represent >1 node per
305 			 * processor.
306 			 */
307 			regs[2] = 0;
308 			regs[3] = 0;
309 			break;
310 
311 		case CPUID_0000_0001:
312 			do_cpuid(1, regs);
313 
314 			error = vm_get_x2apic_state(vcpu, &x2apic_state);
315 			if (error) {
316 				panic("x86_emulate_cpuid: error %d "
317 				      "fetching x2apic state", error);
318 			}
319 
320 			/*
321 			 * Override the APIC ID only in ebx
322 			 */
323 			regs[1] &= ~(CPUID_LOCAL_APIC_ID);
324 			regs[1] |= (vcpu_id << CPUID_0000_0001_APICID_SHIFT);
325 
326 			/*
327 			 * Don't expose VMX, SpeedStep, TME or SMX capability.
328 			 * Advertise x2APIC capability and Hypervisor guest.
329 			 */
330 			regs[2] &= ~(CPUID2_VMX | CPUID2_EST | CPUID2_TM2);
331 			regs[2] &= ~(CPUID2_SMX);
332 
333 			regs[2] |= CPUID2_HV;
334 
335 			if (x2apic_state != X2APIC_DISABLED)
336 				regs[2] |= CPUID2_X2APIC;
337 			else
338 				regs[2] &= ~CPUID2_X2APIC;
339 
340 			/*
341 			 * Only advertise CPUID2_XSAVE in the guest if
342 			 * the host is using XSAVE.
343 			 */
344 			if (!(regs[2] & CPUID2_OSXSAVE))
345 				regs[2] &= ~CPUID2_XSAVE;
346 
347 			/*
348 			 * If CPUID2_XSAVE is being advertised and the
349 			 * guest has set CR4_XSAVE, set
350 			 * CPUID2_OSXSAVE.
351 			 */
352 			regs[2] &= ~CPUID2_OSXSAVE;
353 			if (regs[2] & CPUID2_XSAVE) {
354 				error = vm_get_register(vcpu,
355 				    VM_REG_GUEST_CR4, &cr4);
356 				if (error)
357 					panic("x86_emulate_cpuid: error %d "
358 					      "fetching %%cr4", error);
359 				if (cr4 & CR4_XSAVE)
360 					regs[2] |= CPUID2_OSXSAVE;
361 			}
362 
363 			/*
364 			 * Hide monitor/mwait until we know how to deal with
365 			 * these instructions.
366 			 */
367 			regs[2] &= ~CPUID2_MON;
368 
369                         /*
370 			 * Hide the performance and debug features.
371 			 */
372 			regs[2] &= ~CPUID2_PDCM;
373 
374 			/*
375 			 * No TSC deadline support in the APIC yet
376 			 */
377 			regs[2] &= ~CPUID2_TSCDLT;
378 
379 			/*
380 			 * Hide thermal monitoring
381 			 */
382 			regs[3] &= ~(CPUID_ACPI | CPUID_TM);
383 
384 			/*
385 			 * Hide the debug store capability.
386 			 */
387 			regs[3] &= ~CPUID_DS;
388 
389 			/*
390 			 * Advertise the Machine Check and MTRR capability.
391 			 *
392 			 * Some guest OSes (e.g. Windows) will not boot if
393 			 * these features are absent.
394 			 */
395 			regs[3] |= (CPUID_MCA | CPUID_MCE | CPUID_MTRR);
396 
397 			vm_get_topology(vm, &sockets, &cores, &threads,
398 			    &maxcpus);
399 			logical_cpus = threads * cores;
400 			regs[1] &= ~CPUID_HTT_CORES;
401 			regs[1] |= (logical_cpus & 0xff) << 16;
402 			regs[3] |= CPUID_HTT;
403 			break;
404 
405 		case CPUID_0000_0004:
406 			cpuid_count(func, param, regs);
407 
408 			if (regs[0] || regs[1] || regs[2] || regs[3]) {
409 				vm_get_topology(vm, &sockets, &cores, &threads,
410 				    &maxcpus);
411 				regs[0] &= 0x3ff;
412 				regs[0] |= (cores - 1) << 26;
413 				/*
414 				 * Cache topology:
415 				 * - L1 and L2 are shared only by the logical
416 				 *   processors in a single core.
417 				 * - L3 and above are shared by all logical
418 				 *   processors in the package.
419 				 */
420 				logical_cpus = threads;
421 				level = (regs[0] >> 5) & 0x7;
422 				if (level >= 3)
423 					logical_cpus *= cores;
424 				regs[0] |= (logical_cpus - 1) << 14;
425 			}
426 			break;
427 
428 		case CPUID_0000_0007:
429 			regs[0] = 0;
430 			regs[1] = 0;
431 			regs[2] = 0;
432 			regs[3] = 0;
433 
434 			/* leaf 0 */
435 			if (param == 0) {
436 				cpuid_count(func, param, regs);
437 
438 				/* Only leaf 0 is supported */
439 				regs[0] = 0;
440 
441 				/*
442 				 * Expose known-safe features.
443 				 */
444 				regs[1] &= (CPUID_STDEXT_FSGSBASE |
445 				    CPUID_STDEXT_BMI1 | CPUID_STDEXT_HLE |
446 				    CPUID_STDEXT_AVX2 | CPUID_STDEXT_SMEP |
447 				    CPUID_STDEXT_BMI2 |
448 				    CPUID_STDEXT_ERMS | CPUID_STDEXT_RTM |
449 				    CPUID_STDEXT_AVX512F |
450 				    CPUID_STDEXT_RDSEED |
451 				    CPUID_STDEXT_SMAP |
452 				    CPUID_STDEXT_AVX512PF |
453 				    CPUID_STDEXT_AVX512ER |
454 				    CPUID_STDEXT_AVX512CD | CPUID_STDEXT_SHA);
455 				regs[2] = 0;
456 				regs[3] &= CPUID_STDEXT3_MD_CLEAR;
457 
458 				/* Advertise RDPID if it is enabled. */
459 				error = vm_get_capability(vcpu, VM_CAP_RDPID,
460 				    &enable_rdpid);
461 				if (error == 0 && enable_rdpid)
462 					regs[2] |= CPUID_STDEXT2_RDPID;
463 
464 				/* Advertise INVPCID if it is enabled. */
465 				error = vm_get_capability(vcpu,
466 				    VM_CAP_ENABLE_INVPCID, &enable_invpcid);
467 				if (error == 0 && enable_invpcid)
468 					regs[1] |= CPUID_STDEXT_INVPCID;
469 			}
470 			break;
471 
472 		case CPUID_0000_0006:
473 			regs[0] = CPUTPM1_ARAT;
474 			regs[1] = 0;
475 			regs[2] = 0;
476 			regs[3] = 0;
477 			break;
478 
479 		case CPUID_0000_000A:
480 			/*
481 			 * Handle the access, but report 0 for
482 			 * all options
483 			 */
484 			regs[0] = 0;
485 			regs[1] = 0;
486 			regs[2] = 0;
487 			regs[3] = 0;
488 			break;
489 
490 		case CPUID_0000_000B:
491 			/*
492 			 * Intel processor topology enumeration
493 			 */
494 			if (vmm_is_intel()) {
495 				vm_get_topology(vm, &sockets, &cores, &threads,
496 				    &maxcpus);
497 				if (param == 0) {
498 					logical_cpus = threads;
499 					width = log2(logical_cpus);
500 					level = CPUID_TYPE_SMT;
501 					x2apic_id = vcpu_id;
502 				}
503 
504 				if (param == 1) {
505 					logical_cpus = threads * cores;
506 					width = log2(logical_cpus);
507 					level = CPUID_TYPE_CORE;
508 					x2apic_id = vcpu_id;
509 				}
510 
511 				if (!cpuid_leaf_b || param >= 2) {
512 					width = 0;
513 					logical_cpus = 0;
514 					level = 0;
515 					x2apic_id = 0;
516 				}
517 
518 				regs[0] = width & 0x1f;
519 				regs[1] = logical_cpus & 0xffff;
520 				regs[2] = (level << 8) | (param & 0xff);
521 				regs[3] = x2apic_id;
522 			} else {
523 				regs[0] = 0;
524 				regs[1] = 0;
525 				regs[2] = 0;
526 				regs[3] = 0;
527 			}
528 			break;
529 
530 		case CPUID_0000_000D:
531 			limits = vmm_get_xsave_limits();
532 			if (!limits->xsave_enabled) {
533 				regs[0] = 0;
534 				regs[1] = 0;
535 				regs[2] = 0;
536 				regs[3] = 0;
537 				break;
538 			}
539 
540 			cpuid_count(func, param, regs);
541 			switch (param) {
542 			case 0:
543 				/*
544 				 * Only permit the guest to use bits
545 				 * that are active in the host in
546 				 * %xcr0.  Also, claim that the
547 				 * maximum save area size is
548 				 * equivalent to the host's current
549 				 * save area size.  Since this runs
550 				 * "inside" of vmrun(), it runs with
551 				 * the guest's xcr0, so the current
552 				 * save area size is correct as-is.
553 				 */
554 				regs[0] &= limits->xcr0_allowed;
555 				regs[2] = limits->xsave_max_size;
556 				regs[3] &= (limits->xcr0_allowed >> 32);
557 				break;
558 			case 1:
559 				/* Only permit XSAVEOPT. */
560 				regs[0] &= CPUID_EXTSTATE_XSAVEOPT;
561 				regs[1] = 0;
562 				regs[2] = 0;
563 				regs[3] = 0;
564 				break;
565 			default:
566 				/*
567 				 * If the leaf is for a permitted feature,
568 				 * pass through as-is, otherwise return
569 				 * all zeroes.
570 				 */
571 				if (!(limits->xcr0_allowed & (1ul << param))) {
572 					regs[0] = 0;
573 					regs[1] = 0;
574 					regs[2] = 0;
575 					regs[3] = 0;
576 				}
577 				break;
578 			}
579 			break;
580 
581 		case CPUID_0000_000F:
582 		case CPUID_0000_0010:
583 			/*
584 			 * Do not report any Resource Director Technology
585 			 * capabilities.  Exposing control of cache or memory
586 			 * controller resource partitioning to the guest is not
587 			 * at all sensible.
588 			 *
589 			 * This is already hidden at a high level by masking of
590 			 * leaf 0x7.  Even still, a guest may look here for
591 			 * detailed capability information.
592 			 */
593 			regs[0] = 0;
594 			regs[1] = 0;
595 			regs[2] = 0;
596 			regs[3] = 0;
597 			break;
598 
599 		case CPUID_0000_0015:
600 			/*
601 			 * Don't report CPU TSC/Crystal ratio and clock
602 			 * values since guests may use these to derive the
603 			 * local APIC frequency..
604 			 */
605 			regs[0] = 0;
606 			regs[1] = 0;
607 			regs[2] = 0;
608 			regs[3] = 0;
609 			break;
610 
611 		case 0x40000000:
612 			regs[0] = CPUID_VM_HIGH;
613 			bcopy(bhyve_id, &regs[1], 4);
614 			bcopy(bhyve_id + 4, &regs[2], 4);
615 			bcopy(bhyve_id + 8, &regs[3], 4);
616 			break;
617 
618 		default:
619 default_leaf:
620 			/*
621 			 * The leaf value has already been clamped so
622 			 * simply pass this through, keeping count of
623 			 * how many unhandled leaf values have been seen.
624 			 */
625 			atomic_add_long(&bhyve_xcpuids, 1);
626 			cpuid_count(func, param, regs);
627 			break;
628 	}
629 
630 	/*
631 	 * CPUID clears the upper 32-bits of the long-mode registers.
632 	 */
633 	*rax = regs[0];
634 	*rbx = regs[1];
635 	*rcx = regs[2];
636 	*rdx = regs[3];
637 
638 	return (1);
639 }
640 
641 bool
642 vm_cpuid_capability(struct vcpu *vcpu, enum vm_cpuid_capability cap)
643 {
644 	bool rv;
645 
646 	KASSERT(cap > 0 && cap < VCC_LAST, ("%s: invalid vm_cpu_capability %d",
647 	    __func__, cap));
648 
649 	/*
650 	 * Simply passthrough the capabilities of the host cpu for now.
651 	 */
652 	rv = false;
653 	switch (cap) {
654 	case VCC_NO_EXECUTE:
655 		if (amd_feature & AMDID_NX)
656 			rv = true;
657 		break;
658 	case VCC_FFXSR:
659 		if (amd_feature & AMDID_FFXSR)
660 			rv = true;
661 		break;
662 	case VCC_TCE:
663 		if (amd_feature2 & AMDID2_TCE)
664 			rv = true;
665 		break;
666 	default:
667 		panic("%s: unknown vm_cpu_capability %d", __func__, cap);
668 	}
669 	return (rv);
670 }
671 
672 int
673 vm_rdmtrr(struct vm_mtrr *mtrr, u_int num, uint64_t *val)
674 {
675 	switch (num) {
676 	case MSR_MTRRcap:
677 		*val = MTRR_CAP_WC | MTRR_CAP_FIXED | VMM_MTRR_VAR_MAX;
678 		break;
679 	case MSR_MTRRdefType:
680 		*val = mtrr->def_type;
681 		break;
682 	case MSR_MTRR4kBase ... MSR_MTRR4kBase + 7:
683 		*val = mtrr->fixed4k[num - MSR_MTRR4kBase];
684 		break;
685 	case MSR_MTRR16kBase ... MSR_MTRR16kBase + 1:
686 		*val = mtrr->fixed16k[num - MSR_MTRR16kBase];
687 		break;
688 	case MSR_MTRR64kBase:
689 		*val = mtrr->fixed64k;
690 		break;
691 	case MSR_MTRRVarBase ... MSR_MTRRVarBase + (VMM_MTRR_VAR_MAX * 2) - 1: {
692 		u_int offset = num - MSR_MTRRVarBase;
693 		if (offset % 2 == 0) {
694 			*val = mtrr->var[offset / 2].base;
695 		} else {
696 			*val = mtrr->var[offset / 2].mask;
697 		}
698 		break;
699 	}
700 	default:
701 		return (-1);
702 	}
703 
704 	return (0);
705 }
706 
707 int
708 vm_wrmtrr(struct vm_mtrr *mtrr, u_int num, uint64_t val)
709 {
710 	switch (num) {
711 	case MSR_MTRRcap:
712 		/* MTRRCAP is read only */
713 		return (-1);
714 	case MSR_MTRRdefType:
715 		if (val & ~VMM_MTRR_DEF_MASK) {
716 			/* generate #GP on writes to reserved fields */
717 			return (-1);
718 		}
719 		mtrr->def_type = val;
720 		break;
721 	case MSR_MTRR4kBase ... MSR_MTRR4kBase + 7:
722 		mtrr->fixed4k[num - MSR_MTRR4kBase] = val;
723 		break;
724 	case MSR_MTRR16kBase ... MSR_MTRR16kBase + 1:
725 		mtrr->fixed16k[num - MSR_MTRR16kBase] = val;
726 		break;
727 	case MSR_MTRR64kBase:
728 		mtrr->fixed64k = val;
729 		break;
730 	case MSR_MTRRVarBase ... MSR_MTRRVarBase + (VMM_MTRR_VAR_MAX * 2) - 1: {
731 		u_int offset = num - MSR_MTRRVarBase;
732 		if (offset % 2 == 0) {
733 			if (val & ~VMM_MTRR_PHYSBASE_MASK) {
734 				/* generate #GP on writes to reserved fields */
735 				return (-1);
736 			}
737 			mtrr->var[offset / 2].base = val;
738 		} else {
739 			if (val & ~VMM_MTRR_PHYSMASK_MASK) {
740 				/* generate #GP on writes to reserved fields */
741 				return (-1);
742 			}
743 			mtrr->var[offset / 2].mask = val;
744 		}
745 		break;
746 	}
747 	default:
748 		return (-1);
749 	}
750 
751 	return (0);
752 }
753