xref: /src/sys/amd64/vmm/x86.c (revision ed85203fb7a0334041db6da07e45ddda4caef13d)
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/param.h>
30 #include <sys/pcpu.h>
31 #include <sys/systm.h>
32 #include <sys/sysctl.h>
33 
34 #include <machine/clock.h>
35 #include <machine/cpufunc.h>
36 #include <machine/md_var.h>
37 #include <machine/segments.h>
38 #include <machine/specialreg.h>
39 #include <machine/vmm.h>
40 
41 #include <dev/vmm/vmm_ktr.h>
42 #include <dev/vmm/vmm_vm.h>
43 
44 #include "vmm_host.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_SIGNATURE	0x40000000
53 #define CPUID_BHYVE_FEATURES	0x40000001
54 #define	CPUID_VM_HIGH		CPUID_BHYVE_FEATURES
55 
56 /* Features advertised in CPUID_BHYVE_FEATURES %eax */
57 #define CPUID_BHYVE_FEAT_EXT_DEST_ID	(1UL << 0) /* MSI Extended Dest ID */
58 
59 static const char bhyve_id[12] = "bhyve bhyve ";
60 
61 static uint64_t bhyve_xcpuids;
62 SYSCTL_ULONG(_hw_vmm, OID_AUTO, bhyve_xcpuids, CTLFLAG_RW, &bhyve_xcpuids, 0,
63     "Number of times an unknown cpuid leaf was accessed");
64 
65 static int cpuid_leaf_b = 1;
66 SYSCTL_INT(_hw_vmm_topology, OID_AUTO, cpuid_leaf_b, CTLFLAG_RDTUN,
67     &cpuid_leaf_b, 0, NULL);
68 
69 /*
70  * Compute ceil(log2(x)).  Returns -1 if x is zero.
71  */
72 static __inline int
log2(u_int x)73 log2(u_int x)
74 {
75 
76 	return (x == 0 ? -1 : order_base_2(x));
77 }
78 
79 int
x86_emulate_cpuid(struct vcpu * vcpu,uint64_t * rax,uint64_t * rbx,uint64_t * rcx,uint64_t * rdx)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 >= CPUID_VM_SIGNATURE) {
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 				logical_cpus = MIN(0xFF, threads * cores - 1);
160 				regs[2] = (width << AMDID_COREID_SIZE_SHIFT) | logical_cpus;
161 			}
162 			break;
163 
164 		case CPUID_8000_0001:
165 			cpuid_count(func, param, regs);
166 
167 			/*
168 			 * Hide SVM from guest.
169 			 */
170 			regs[2] &= ~AMDID2_SVM;
171 
172 			/*
173 			 * Don't advertise extended performance counter MSRs
174 			 * to the guest.
175 			 */
176 			regs[2] &= ~AMDID2_PCXC;
177 			regs[2] &= ~AMDID2_PNXC;
178 			regs[2] &= ~AMDID2_PTSCEL2I;
179 
180 			/*
181 			 * Don't advertise Instruction Based Sampling feature.
182 			 */
183 			regs[2] &= ~AMDID2_IBS;
184 
185 			/* NodeID MSR not available */
186 			regs[2] &= ~AMDID2_NODE_ID;
187 
188 			/* Don't advertise the OS visible workaround feature */
189 			regs[2] &= ~AMDID2_OSVW;
190 
191 			/* Hide mwaitx/monitorx capability from the guest */
192 			regs[2] &= ~AMDID2_MWAITX;
193 
194 			/* Advertise RDTSCP if it is enabled. */
195 			error = vm_get_capability(vcpu,
196 			    VM_CAP_RDTSCP, &enable_rdtscp);
197 			if (error == 0 && enable_rdtscp)
198 				regs[3] |= AMDID_RDTSCP;
199 			else
200 				regs[3] &= ~AMDID_RDTSCP;
201 			break;
202 
203 		case CPUID_8000_0007:
204 			/*
205 			 * AMD uses this leaf to advertise the processor's
206 			 * power monitoring and RAS capabilities. These
207 			 * features are hardware-specific and exposing
208 			 * them to a guest doesn't make a lot of sense.
209 			 *
210 			 * Intel uses this leaf only to advertise the
211 			 * "Invariant TSC" feature with all other bits
212 			 * being reserved (set to zero).
213 			 */
214 			regs[0] = 0;
215 			regs[1] = 0;
216 			regs[2] = 0;
217 			regs[3] = 0;
218 
219 			/*
220 			 * "Invariant TSC" can be advertised to the guest if:
221 			 * - host TSC frequency is invariant
222 			 * - host TSCs are synchronized across physical cpus
223 			 *
224 			 * XXX This still falls short because the vcpu
225 			 * can observe the TSC moving backwards as it
226 			 * migrates across physical cpus. But at least
227 			 * it should discourage the guest from using the
228 			 * TSC to keep track of time.
229 			 */
230 			if (tsc_is_invariant && smp_tsc)
231 				regs[3] |= AMDPM_TSC_INVARIANT;
232 			break;
233 
234 		case CPUID_8000_001D:
235 			/* AMD Cache topology, like 0000_0004 for Intel. */
236 			if (!vmm_is_svm())
237 				goto default_leaf;
238 
239 			/*
240 			 * Similar to Intel, generate a fictitious cache
241 			 * topology for the guest with L3 shared by the
242 			 * package, and L1 and L2 local to a core.
243 			 */
244 			vm_get_topology(vm, &sockets, &cores, &threads,
245 			    &maxcpus);
246 			switch (param) {
247 			case 0:
248 				logical_cpus = threads;
249 				level = 1;
250 				func = 1;	/* data cache */
251 				break;
252 			case 1:
253 				logical_cpus = threads;
254 				level = 2;
255 				func = 3;	/* unified cache */
256 				break;
257 			case 2:
258 				logical_cpus = threads * cores;
259 				level = 3;
260 				func = 3;	/* unified cache */
261 				break;
262 			default:
263 				logical_cpus = sockets * threads * cores;
264 				level = 0;
265 				func = 0;
266 				break;
267 			}
268 
269 			logical_cpus = MIN(0xfff, logical_cpus - 1);
270 			regs[0] = (logical_cpus << 14) | (1 << 8) |
271 			    (level << 5) | func;
272 			regs[1] = (func > 0) ? (CACHE_LINE_SIZE - 1) : 0;
273 
274 			/*
275 			 * ecx: Number of cache ways for non-fully
276 			 * associative cache, minus 1.  Reported value
277 			 * of zero means there is one way.
278 			 */
279 			regs[2] = 0;
280 
281 			regs[3] = 0;
282 			break;
283 
284 		case CPUID_8000_001E:
285 			/*
286 			 * AMD Family 16h+ and Hygon Family 18h additional
287 			 * identifiers.
288 			 */
289 			if (!vmm_is_svm() || CPUID_TO_FAMILY(cpu_id) < 0x16)
290 				goto default_leaf;
291 
292 			vm_get_topology(vm, &sockets, &cores, &threads,
293 			    &maxcpus);
294 			regs[0] = vcpu_id;
295 			threads = MIN(0xFF, threads - 1);
296 			regs[1] = (threads << 8) |
297 			    (vcpu_id >> log2(threads + 1));
298 			/*
299 			 * XXX Bhyve topology cannot yet represent >1 node per
300 			 * processor.
301 			 */
302 			regs[2] = 0;
303 			regs[3] = 0;
304 			break;
305 
306 		case CPUID_0000_0001:
307 			do_cpuid(1, regs);
308 
309 			error = vm_get_x2apic_state(vcpu, &x2apic_state);
310 			if (error) {
311 				panic("x86_emulate_cpuid: error %d "
312 				      "fetching x2apic state", error);
313 			}
314 
315 			/*
316 			 * Override the APIC ID only in ebx
317 			 */
318 			regs[1] &= ~(CPUID_LOCAL_APIC_ID);
319 			regs[1] |= (vcpu_id << CPUID_0000_0001_APICID_SHIFT);
320 
321 			/*
322 			 * Don't expose VMX, SpeedStep, TME or SMX capability.
323 			 * Advertise x2APIC capability and Hypervisor guest.
324 			 */
325 			regs[2] &= ~(CPUID2_VMX | CPUID2_EST | CPUID2_TM2);
326 			regs[2] &= ~(CPUID2_SMX);
327 
328 			regs[2] |= CPUID2_HV;
329 
330 			if (x2apic_state != X2APIC_DISABLED)
331 				regs[2] |= CPUID2_X2APIC;
332 			else
333 				regs[2] &= ~CPUID2_X2APIC;
334 
335 			/*
336 			 * Only advertise CPUID2_XSAVE in the guest if
337 			 * the host is using XSAVE.
338 			 */
339 			if (!(regs[2] & CPUID2_OSXSAVE))
340 				regs[2] &= ~CPUID2_XSAVE;
341 
342 			/*
343 			 * If CPUID2_XSAVE is being advertised and the
344 			 * guest has set CR4_XSAVE, set
345 			 * CPUID2_OSXSAVE.
346 			 */
347 			regs[2] &= ~CPUID2_OSXSAVE;
348 			if (regs[2] & CPUID2_XSAVE) {
349 				error = vm_get_register(vcpu,
350 				    VM_REG_GUEST_CR4, &cr4);
351 				if (error)
352 					panic("x86_emulate_cpuid: error %d "
353 					      "fetching %%cr4", error);
354 				if (cr4 & CR4_XSAVE)
355 					regs[2] |= CPUID2_OSXSAVE;
356 			}
357 
358 			/*
359 			 * Hide monitor/mwait until we know how to deal with
360 			 * these instructions.
361 			 */
362 			regs[2] &= ~CPUID2_MON;
363 
364                         /*
365 			 * Hide the performance and debug features.
366 			 */
367 			regs[2] &= ~CPUID2_PDCM;
368 
369 			/*
370 			 * No TSC deadline support in the APIC yet
371 			 */
372 			regs[2] &= ~CPUID2_TSCDLT;
373 
374 			/*
375 			 * Hide thermal monitoring
376 			 */
377 			regs[3] &= ~(CPUID_ACPI | CPUID_TM);
378 
379 			/*
380 			 * Hide the debug store capability.
381 			 */
382 			regs[3] &= ~CPUID_DS;
383 
384 			/*
385 			 * Advertise the Machine Check and MTRR capability.
386 			 *
387 			 * Some guest OSes (e.g. Windows) will not boot if
388 			 * these features are absent.
389 			 */
390 			regs[3] |= (CPUID_MCA | CPUID_MCE | CPUID_MTRR);
391 
392 			vm_get_topology(vm, &sockets, &cores, &threads,
393 			    &maxcpus);
394 			logical_cpus = threads * cores;
395 			regs[1] &= ~CPUID_HTT_CORES;
396 			regs[1] |= (logical_cpus & 0xff) << 16;
397 			regs[3] |= CPUID_HTT;
398 			break;
399 
400 		case CPUID_0000_0004:
401 			cpuid_count(func, param, regs);
402 
403 			if (regs[0] || regs[1] || regs[2] || regs[3]) {
404 				vm_get_topology(vm, &sockets, &cores, &threads,
405 				    &maxcpus);
406 				regs[0] &= 0x3ff;
407 				regs[0] |= (cores - 1) << 26;
408 				/*
409 				 * Cache topology:
410 				 * - L1 and L2 are shared only by the logical
411 				 *   processors in a single core.
412 				 * - L3 and above are shared by all logical
413 				 *   processors in the package.
414 				 */
415 				logical_cpus = threads;
416 				level = (regs[0] >> 5) & 0x7;
417 				if (level >= 3)
418 					logical_cpus *= cores;
419 				regs[0] |= (logical_cpus - 1) << 14;
420 			}
421 			break;
422 
423 		case CPUID_0000_0007:
424 			regs[0] = 0;
425 			regs[1] = 0;
426 			regs[2] = 0;
427 			regs[3] = 0;
428 
429 			/* leaf 0 */
430 			if (param == 0) {
431 				cpuid_count(func, param, regs);
432 
433 				/* Only leaf 0 is supported */
434 				regs[0] = 0;
435 
436 				/*
437 				 * Expose known-safe features.
438 				 */
439 				regs[1] &= CPUID_STDEXT_FSGSBASE |
440 				    CPUID_STDEXT_BMI1 | CPUID_STDEXT_HLE |
441 				    CPUID_STDEXT_AVX2 | CPUID_STDEXT_SMEP |
442 				    CPUID_STDEXT_BMI2 |
443 				    CPUID_STDEXT_ERMS | CPUID_STDEXT_RTM |
444 				    CPUID_STDEXT_AVX512F |
445 				    CPUID_STDEXT_AVX512DQ |
446 				    CPUID_STDEXT_RDSEED |
447 				    CPUID_STDEXT_SMAP |
448 				    CPUID_STDEXT_AVX512PF |
449 				    CPUID_STDEXT_AVX512ER |
450 				    CPUID_STDEXT_AVX512CD | CPUID_STDEXT_SHA |
451 				    CPUID_STDEXT_AVX512BW |
452 				    CPUID_STDEXT_AVX512VL;
453 				regs[2] &= CPUID_STDEXT2_VAES |
454 				    CPUID_STDEXT2_VPCLMULQDQ;
455 				regs[3] &= CPUID_STDEXT3_MD_CLEAR;
456 
457 				/* Advertise RDPID if it is enabled. */
458 				error = vm_get_capability(vcpu, VM_CAP_RDPID,
459 				    &enable_rdpid);
460 				if (error == 0 && enable_rdpid)
461 					regs[2] |= CPUID_STDEXT2_RDPID;
462 
463 				/* Advertise INVPCID if it is enabled. */
464 				error = vm_get_capability(vcpu,
465 				    VM_CAP_ENABLE_INVPCID, &enable_invpcid);
466 				if (error == 0 && enable_invpcid)
467 					regs[1] |= CPUID_STDEXT_INVPCID;
468 			}
469 			break;
470 
471 		case CPUID_0000_0006:
472 			regs[0] = CPUTPM1_ARAT;
473 			regs[1] = 0;
474 			regs[2] = 0;
475 			regs[3] = 0;
476 			break;
477 
478 		case CPUID_0000_000A:
479 			/*
480 			 * Handle the access, but report 0 for
481 			 * all options
482 			 */
483 			regs[0] = 0;
484 			regs[1] = 0;
485 			regs[2] = 0;
486 			regs[3] = 0;
487 			break;
488 
489 		case CPUID_0000_000B:
490 			/*
491 			 * Intel processor topology enumeration
492 			 */
493 			if (vmm_is_intel()) {
494 				vm_get_topology(vm, &sockets, &cores, &threads,
495 				    &maxcpus);
496 				if (param == 0) {
497 					logical_cpus = threads;
498 					width = log2(logical_cpus);
499 					level = CPUID_TYPE_SMT;
500 					x2apic_id = vcpu_id;
501 				}
502 
503 				if (param == 1) {
504 					logical_cpus = threads * cores;
505 					width = log2(logical_cpus);
506 					level = CPUID_TYPE_CORE;
507 					x2apic_id = vcpu_id;
508 				}
509 
510 				if (!cpuid_leaf_b || param >= 2) {
511 					width = 0;
512 					logical_cpus = 0;
513 					level = 0;
514 					x2apic_id = 0;
515 				}
516 
517 				regs[0] = width & 0x1f;
518 				regs[1] = logical_cpus & 0xffff;
519 				regs[2] = (level << 8) | (param & 0xff);
520 				regs[3] = x2apic_id;
521 			} else {
522 				regs[0] = 0;
523 				regs[1] = 0;
524 				regs[2] = 0;
525 				regs[3] = 0;
526 			}
527 			break;
528 
529 		case CPUID_0000_000D:
530 			limits = vmm_get_xsave_limits();
531 			if (!limits->xsave_enabled) {
532 				regs[0] = 0;
533 				regs[1] = 0;
534 				regs[2] = 0;
535 				regs[3] = 0;
536 				break;
537 			}
538 
539 			cpuid_count(func, param, regs);
540 			switch (param) {
541 			case 0:
542 				/*
543 				 * Only permit the guest to use bits
544 				 * that are active in the host in
545 				 * %xcr0.  Also, claim that the
546 				 * maximum save area size is
547 				 * equivalent to the host's current
548 				 * save area size.  Since this runs
549 				 * "inside" of vmrun(), it runs with
550 				 * the guest's xcr0, so the current
551 				 * save area size is correct as-is.
552 				 */
553 				regs[0] &= limits->xcr0_allowed;
554 				regs[2] = limits->xsave_max_size;
555 				regs[3] &= (limits->xcr0_allowed >> 32);
556 				break;
557 			case 1:
558 				/* Only permit XSAVEOPT. */
559 				regs[0] &= CPUID_EXTSTATE_XSAVEOPT;
560 				regs[1] = 0;
561 				regs[2] = 0;
562 				regs[3] = 0;
563 				break;
564 			default:
565 				/*
566 				 * If the leaf is for a permitted feature,
567 				 * pass through as-is, otherwise return
568 				 * all zeroes.
569 				 */
570 				if (!(limits->xcr0_allowed & (1ul << param))) {
571 					regs[0] = 0;
572 					regs[1] = 0;
573 					regs[2] = 0;
574 					regs[3] = 0;
575 				}
576 				break;
577 			}
578 			break;
579 
580 		case CPUID_0000_000F:
581 		case CPUID_0000_0010:
582 			/*
583 			 * Do not report any Resource Director Technology
584 			 * capabilities.  Exposing control of cache or memory
585 			 * controller resource partitioning to the guest is not
586 			 * at all sensible.
587 			 *
588 			 * This is already hidden at a high level by masking of
589 			 * leaf 0x7.  Even still, a guest may look here for
590 			 * detailed capability information.
591 			 */
592 			regs[0] = 0;
593 			regs[1] = 0;
594 			regs[2] = 0;
595 			regs[3] = 0;
596 			break;
597 
598 		case CPUID_0000_0015:
599 			/*
600 			 * Don't report CPU TSC/Crystal ratio and clock
601 			 * values since guests may use these to derive the
602 			 * local APIC frequency..
603 			 */
604 			regs[0] = 0;
605 			regs[1] = 0;
606 			regs[2] = 0;
607 			regs[3] = 0;
608 			break;
609 
610 		case CPUID_VM_SIGNATURE:
611 			regs[0] = CPUID_VM_HIGH;
612 			bcopy(bhyve_id, &regs[1], 4);
613 			bcopy(bhyve_id + 4, &regs[2], 4);
614 			bcopy(bhyve_id + 8, &regs[3], 4);
615 			break;
616 
617 		case CPUID_BHYVE_FEATURES:
618 			regs[0] = CPUID_BHYVE_FEAT_EXT_DEST_ID;
619 			regs[1] = 0;
620 			regs[2] = 0;
621 			regs[3] = 0;
622 			break;
623 
624 		default:
625 default_leaf:
626 			/*
627 			 * The leaf value has already been clamped so
628 			 * simply pass this through, keeping count of
629 			 * how many unhandled leaf values have been seen.
630 			 */
631 			atomic_add_long(&bhyve_xcpuids, 1);
632 			cpuid_count(func, param, regs);
633 			break;
634 	}
635 
636 	/*
637 	 * CPUID clears the upper 32-bits of the long-mode registers.
638 	 */
639 	*rax = regs[0];
640 	*rbx = regs[1];
641 	*rcx = regs[2];
642 	*rdx = regs[3];
643 
644 	return (1);
645 }
646 
647 bool
vm_cpuid_capability(struct vcpu * vcpu,enum vm_cpuid_capability cap)648 vm_cpuid_capability(struct vcpu *vcpu, enum vm_cpuid_capability cap)
649 {
650 	bool rv;
651 
652 	KASSERT(cap > 0 && cap < VCC_LAST, ("%s: invalid vm_cpu_capability %d",
653 	    __func__, cap));
654 
655 	/*
656 	 * Simply passthrough the capabilities of the host cpu for now.
657 	 */
658 	rv = false;
659 	switch (cap) {
660 	case VCC_NO_EXECUTE:
661 		if (amd_feature & AMDID_NX)
662 			rv = true;
663 		break;
664 	case VCC_FFXSR:
665 		if (amd_feature & AMDID_FFXSR)
666 			rv = true;
667 		break;
668 	case VCC_TCE:
669 		if (amd_feature2 & AMDID2_TCE)
670 			rv = true;
671 		break;
672 	default:
673 		panic("%s: unknown vm_cpu_capability %d", __func__, cap);
674 	}
675 	return (rv);
676 }
677 
678 int
vm_rdmtrr(struct vm_mtrr * mtrr,u_int num,uint64_t * val)679 vm_rdmtrr(struct vm_mtrr *mtrr, u_int num, uint64_t *val)
680 {
681 	switch (num) {
682 	case MSR_MTRRcap:
683 		*val = MTRR_CAP_WC | MTRR_CAP_FIXED | VMM_MTRR_VAR_MAX;
684 		break;
685 	case MSR_MTRRdefType:
686 		*val = mtrr->def_type;
687 		break;
688 	case MSR_MTRR4kBase ... MSR_MTRR4kBase + 7:
689 		*val = mtrr->fixed4k[num - MSR_MTRR4kBase];
690 		break;
691 	case MSR_MTRR16kBase ... MSR_MTRR16kBase + 1:
692 		*val = mtrr->fixed16k[num - MSR_MTRR16kBase];
693 		break;
694 	case MSR_MTRR64kBase:
695 		*val = mtrr->fixed64k;
696 		break;
697 	case MSR_MTRRVarBase ... MSR_MTRRVarBase + (VMM_MTRR_VAR_MAX * 2) - 1: {
698 		u_int offset = num - MSR_MTRRVarBase;
699 		if (offset % 2 == 0) {
700 			*val = mtrr->var[offset / 2].base;
701 		} else {
702 			*val = mtrr->var[offset / 2].mask;
703 		}
704 		break;
705 	}
706 	default:
707 		return (-1);
708 	}
709 
710 	return (0);
711 }
712 
713 int
vm_wrmtrr(struct vm_mtrr * mtrr,u_int num,uint64_t val)714 vm_wrmtrr(struct vm_mtrr *mtrr, u_int num, uint64_t val)
715 {
716 	switch (num) {
717 	case MSR_MTRRcap:
718 		/* MTRRCAP is read only */
719 		return (-1);
720 	case MSR_MTRRdefType:
721 		if (val & ~VMM_MTRR_DEF_MASK) {
722 			/* generate #GP on writes to reserved fields */
723 			return (-1);
724 		}
725 		mtrr->def_type = val;
726 		break;
727 	case MSR_MTRR4kBase ... MSR_MTRR4kBase + 7:
728 		mtrr->fixed4k[num - MSR_MTRR4kBase] = val;
729 		break;
730 	case MSR_MTRR16kBase ... MSR_MTRR16kBase + 1:
731 		mtrr->fixed16k[num - MSR_MTRR16kBase] = val;
732 		break;
733 	case MSR_MTRR64kBase:
734 		mtrr->fixed64k = val;
735 		break;
736 	case MSR_MTRRVarBase ... MSR_MTRRVarBase + (VMM_MTRR_VAR_MAX * 2) - 1: {
737 		u_int offset = num - MSR_MTRRVarBase;
738 		if (offset % 2 == 0) {
739 			if (val & ~VMM_MTRR_PHYSBASE_MASK) {
740 				/* generate #GP on writes to reserved fields */
741 				return (-1);
742 			}
743 			mtrr->var[offset / 2].base = val;
744 		} else {
745 			if (val & ~VMM_MTRR_PHYSMASK_MASK) {
746 				/* generate #GP on writes to reserved fields */
747 				return (-1);
748 			}
749 			mtrr->var[offset / 2].mask = val;
750 		}
751 		break;
752 	}
753 	default:
754 		return (-1);
755 	}
756 
757 	return (0);
758 }
759