vmcs.c revision 242275
1/*-
2 * Copyright (c) 2011 NetApp, Inc.
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 *    notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 *    notice, this list of conditions and the following disclaimer in the
12 *    documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY NETAPP, INC ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED.  IN NO EVENT SHALL NETAPP, INC OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 *
26 * $FreeBSD$
27 */
28
29#include "opt_ddb.h"
30
31#include <sys/cdefs.h>
32__FBSDID("$FreeBSD$");
33
34#include <sys/param.h>
35#include <sys/systm.h>
36#include <sys/pcpu.h>
37
38#include <vm/vm.h>
39#include <vm/pmap.h>
40
41#include <machine/segments.h>
42#include <machine/pmap.h>
43
44#include <machine/vmm.h>
45#include "vmm_host.h"
46#include "vmcs.h"
47#include "vmx_cpufunc.h"
48#include "ept.h"
49#include "vmx.h"
50
51#ifdef DDB
52#include <ddb/ddb.h>
53#endif
54
55static uint64_t
56vmcs_fix_regval(uint32_t encoding, uint64_t val)
57{
58
59	switch (encoding) {
60	case VMCS_GUEST_CR0:
61		val = vmx_fix_cr0(val);
62		break;
63	case VMCS_GUEST_CR4:
64		val = vmx_fix_cr4(val);
65		break;
66	default:
67		break;
68	}
69	return (val);
70}
71
72static uint32_t
73vmcs_field_encoding(int ident)
74{
75	switch (ident) {
76	case VM_REG_GUEST_CR0:
77		return (VMCS_GUEST_CR0);
78	case VM_REG_GUEST_CR3:
79		return (VMCS_GUEST_CR3);
80	case VM_REG_GUEST_CR4:
81		return (VMCS_GUEST_CR4);
82	case VM_REG_GUEST_DR7:
83		return (VMCS_GUEST_DR7);
84	case VM_REG_GUEST_RSP:
85		return (VMCS_GUEST_RSP);
86	case VM_REG_GUEST_RIP:
87		return (VMCS_GUEST_RIP);
88	case VM_REG_GUEST_RFLAGS:
89		return (VMCS_GUEST_RFLAGS);
90	case VM_REG_GUEST_ES:
91		return (VMCS_GUEST_ES_SELECTOR);
92	case VM_REG_GUEST_CS:
93		return (VMCS_GUEST_CS_SELECTOR);
94	case VM_REG_GUEST_SS:
95		return (VMCS_GUEST_SS_SELECTOR);
96	case VM_REG_GUEST_DS:
97		return (VMCS_GUEST_DS_SELECTOR);
98	case VM_REG_GUEST_FS:
99		return (VMCS_GUEST_FS_SELECTOR);
100	case VM_REG_GUEST_GS:
101		return (VMCS_GUEST_GS_SELECTOR);
102	case VM_REG_GUEST_TR:
103		return (VMCS_GUEST_TR_SELECTOR);
104	case VM_REG_GUEST_LDTR:
105		return (VMCS_GUEST_LDTR_SELECTOR);
106	case VM_REG_GUEST_EFER:
107		return (VMCS_GUEST_IA32_EFER);
108	default:
109		return (-1);
110	}
111
112}
113
114static int
115vmcs_seg_desc_encoding(int seg, uint32_t *base, uint32_t *lim, uint32_t *acc)
116{
117
118	switch (seg) {
119	case VM_REG_GUEST_ES:
120		*base = VMCS_GUEST_ES_BASE;
121		*lim = VMCS_GUEST_ES_LIMIT;
122		*acc = VMCS_GUEST_ES_ACCESS_RIGHTS;
123		break;
124	case VM_REG_GUEST_CS:
125		*base = VMCS_GUEST_CS_BASE;
126		*lim = VMCS_GUEST_CS_LIMIT;
127		*acc = VMCS_GUEST_CS_ACCESS_RIGHTS;
128		break;
129	case VM_REG_GUEST_SS:
130		*base = VMCS_GUEST_SS_BASE;
131		*lim = VMCS_GUEST_SS_LIMIT;
132		*acc = VMCS_GUEST_SS_ACCESS_RIGHTS;
133		break;
134	case VM_REG_GUEST_DS:
135		*base = VMCS_GUEST_DS_BASE;
136		*lim = VMCS_GUEST_DS_LIMIT;
137		*acc = VMCS_GUEST_DS_ACCESS_RIGHTS;
138		break;
139	case VM_REG_GUEST_FS:
140		*base = VMCS_GUEST_FS_BASE;
141		*lim = VMCS_GUEST_FS_LIMIT;
142		*acc = VMCS_GUEST_FS_ACCESS_RIGHTS;
143		break;
144	case VM_REG_GUEST_GS:
145		*base = VMCS_GUEST_GS_BASE;
146		*lim = VMCS_GUEST_GS_LIMIT;
147		*acc = VMCS_GUEST_GS_ACCESS_RIGHTS;
148		break;
149	case VM_REG_GUEST_TR:
150		*base = VMCS_GUEST_TR_BASE;
151		*lim = VMCS_GUEST_TR_LIMIT;
152		*acc = VMCS_GUEST_TR_ACCESS_RIGHTS;
153		break;
154	case VM_REG_GUEST_LDTR:
155		*base = VMCS_GUEST_LDTR_BASE;
156		*lim = VMCS_GUEST_LDTR_LIMIT;
157		*acc = VMCS_GUEST_LDTR_ACCESS_RIGHTS;
158		break;
159	case VM_REG_GUEST_IDTR:
160		*base = VMCS_GUEST_IDTR_BASE;
161		*lim = VMCS_GUEST_IDTR_LIMIT;
162		*acc = VMCS_INVALID_ENCODING;
163		break;
164	case VM_REG_GUEST_GDTR:
165		*base = VMCS_GUEST_GDTR_BASE;
166		*lim = VMCS_GUEST_GDTR_LIMIT;
167		*acc = VMCS_INVALID_ENCODING;
168		break;
169	default:
170		return (EINVAL);
171	}
172
173	return (0);
174}
175
176int
177vmcs_getreg(struct vmcs *vmcs, int ident, uint64_t *retval)
178{
179	int error;
180	uint32_t encoding;
181
182	/*
183	 * If we need to get at vmx-specific state in the VMCS we can bypass
184	 * the translation of 'ident' to 'encoding' by simply setting the
185	 * sign bit. As it so happens the upper 16 bits are reserved (i.e
186	 * set to 0) in the encodings for the VMCS so we are free to use the
187	 * sign bit.
188	 */
189	if (ident < 0)
190		encoding = ident & 0x7fffffff;
191	else
192		encoding = vmcs_field_encoding(ident);
193
194	if (encoding == (uint32_t)-1)
195		return (EINVAL);
196
197	VMPTRLD(vmcs);
198	error = vmread(encoding, retval);
199	VMCLEAR(vmcs);
200	return (error);
201}
202
203int
204vmcs_setreg(struct vmcs *vmcs, int ident, uint64_t val)
205{
206	int error;
207	uint32_t encoding;
208
209	if (ident < 0)
210		encoding = ident & 0x7fffffff;
211	else
212		encoding = vmcs_field_encoding(ident);
213
214	if (encoding == (uint32_t)-1)
215		return (EINVAL);
216
217	val = vmcs_fix_regval(encoding, val);
218
219	VMPTRLD(vmcs);
220	error = vmwrite(encoding, val);
221	VMCLEAR(vmcs);
222	return (error);
223}
224
225int
226vmcs_setdesc(struct vmcs *vmcs, int seg, struct seg_desc *desc)
227{
228	int error;
229	uint32_t base, limit, access;
230
231	error = vmcs_seg_desc_encoding(seg, &base, &limit, &access);
232	if (error != 0)
233		panic("vmcs_setdesc: invalid segment register %d", seg);
234
235	VMPTRLD(vmcs);
236	if ((error = vmwrite(base, desc->base)) != 0)
237		goto done;
238
239	if ((error = vmwrite(limit, desc->limit)) != 0)
240		goto done;
241
242	if (access != VMCS_INVALID_ENCODING) {
243		if ((error = vmwrite(access, desc->access)) != 0)
244			goto done;
245	}
246done:
247	VMCLEAR(vmcs);
248	return (error);
249}
250
251int
252vmcs_getdesc(struct vmcs *vmcs, int seg, struct seg_desc *desc)
253{
254	int error;
255	uint32_t base, limit, access;
256	uint64_t u64;
257
258	error = vmcs_seg_desc_encoding(seg, &base, &limit, &access);
259	if (error != 0)
260		panic("vmcs_getdesc: invalid segment register %d", seg);
261
262	VMPTRLD(vmcs);
263	if ((error = vmread(base, &u64)) != 0)
264		goto done;
265	desc->base = u64;
266
267	if ((error = vmread(limit, &u64)) != 0)
268		goto done;
269	desc->limit = u64;
270
271	if (access != VMCS_INVALID_ENCODING) {
272		if ((error = vmread(access, &u64)) != 0)
273			goto done;
274		desc->access = u64;
275	}
276done:
277	VMCLEAR(vmcs);
278	return (error);
279}
280
281int
282vmcs_set_msr_save(struct vmcs *vmcs, u_long g_area, u_int g_count)
283{
284	int error;
285
286	VMPTRLD(vmcs);
287
288	/*
289	 * Guest MSRs are saved in the VM-exit MSR-store area.
290	 * Guest MSRs are loaded from the VM-entry MSR-load area.
291	 * Both areas point to the same location in memory.
292	 */
293	if ((error = vmwrite(VMCS_EXIT_MSR_STORE, g_area)) != 0)
294		goto done;
295	if ((error = vmwrite(VMCS_EXIT_MSR_STORE_COUNT, g_count)) != 0)
296		goto done;
297
298	if ((error = vmwrite(VMCS_ENTRY_MSR_LOAD, g_area)) != 0)
299		goto done;
300	if ((error = vmwrite(VMCS_ENTRY_MSR_LOAD_COUNT, g_count)) != 0)
301		goto done;
302
303	error = 0;
304done:
305	VMCLEAR(vmcs);
306	return (error);
307}
308
309int
310vmcs_set_defaults(struct vmcs *vmcs,
311		  u_long host_rip, u_long host_rsp, u_long ept_pml4,
312		  uint32_t pinbased_ctls, uint32_t procbased_ctls,
313		  uint32_t procbased_ctls2, uint32_t exit_ctls,
314		  uint32_t entry_ctls, u_long msr_bitmap, uint16_t vpid)
315{
316	int error, codesel, datasel, tsssel;
317	u_long cr0, cr4, efer;
318	uint64_t eptp, pat, fsbase, idtrbase;
319	uint32_t exc_bitmap;
320
321	codesel = vmm_get_host_codesel();
322	datasel = vmm_get_host_datasel();
323	tsssel = vmm_get_host_tsssel();
324
325	/*
326	 * Make sure we have a "current" VMCS to work with.
327	 */
328	VMPTRLD(vmcs);
329
330	/*
331	 * Load the VMX controls
332	 */
333	if ((error = vmwrite(VMCS_PIN_BASED_CTLS, pinbased_ctls)) != 0)
334		goto done;
335	if ((error = vmwrite(VMCS_PRI_PROC_BASED_CTLS, procbased_ctls)) != 0)
336		goto done;
337	if ((error = vmwrite(VMCS_SEC_PROC_BASED_CTLS, procbased_ctls2)) != 0)
338		goto done;
339	if ((error = vmwrite(VMCS_EXIT_CTLS, exit_ctls)) != 0)
340		goto done;
341	if ((error = vmwrite(VMCS_ENTRY_CTLS, entry_ctls)) != 0)
342		goto done;
343
344	/* Guest state */
345
346	/* Initialize guest IA32_PAT MSR with the default value */
347	pat = PAT_VALUE(0, PAT_WRITE_BACK)	|
348	      PAT_VALUE(1, PAT_WRITE_THROUGH)	|
349	      PAT_VALUE(2, PAT_UNCACHED)	|
350	      PAT_VALUE(3, PAT_UNCACHEABLE)	|
351	      PAT_VALUE(4, PAT_WRITE_BACK)	|
352	      PAT_VALUE(5, PAT_WRITE_THROUGH)	|
353	      PAT_VALUE(6, PAT_UNCACHED)	|
354	      PAT_VALUE(7, PAT_UNCACHEABLE);
355	if ((error = vmwrite(VMCS_GUEST_IA32_PAT, pat)) != 0)
356		goto done;
357
358	/* Host state */
359
360	/* Initialize host IA32_PAT MSR */
361	pat = vmm_get_host_pat();
362	if ((error = vmwrite(VMCS_HOST_IA32_PAT, pat)) != 0)
363		goto done;
364
365	/* Load the IA32_EFER MSR */
366	efer = vmm_get_host_efer();
367	if ((error = vmwrite(VMCS_HOST_IA32_EFER, efer)) != 0)
368		goto done;
369
370	/* Load the control registers */
371
372	cr0 = vmm_get_host_cr0();
373	if ((error = vmwrite(VMCS_HOST_CR0, cr0)) != 0)
374		goto done;
375
376	cr4 = vmm_get_host_cr4() | CR4_VMXE;
377	if ((error = vmwrite(VMCS_HOST_CR4, cr4)) != 0)
378		goto done;
379
380	/* Load the segment selectors */
381	if ((error = vmwrite(VMCS_HOST_ES_SELECTOR, datasel)) != 0)
382		goto done;
383
384	if ((error = vmwrite(VMCS_HOST_CS_SELECTOR, codesel)) != 0)
385		goto done;
386
387	if ((error = vmwrite(VMCS_HOST_SS_SELECTOR, datasel)) != 0)
388		goto done;
389
390	if ((error = vmwrite(VMCS_HOST_DS_SELECTOR, datasel)) != 0)
391		goto done;
392
393	if ((error = vmwrite(VMCS_HOST_FS_SELECTOR, datasel)) != 0)
394		goto done;
395
396	if ((error = vmwrite(VMCS_HOST_GS_SELECTOR, datasel)) != 0)
397		goto done;
398
399	if ((error = vmwrite(VMCS_HOST_TR_SELECTOR, tsssel)) != 0)
400		goto done;
401
402	/*
403	 * Load the Base-Address for %fs and idtr.
404	 *
405	 * Note that we exclude %gs, tss and gdtr here because their base
406	 * address is pcpu specific.
407	 */
408	fsbase = vmm_get_host_fsbase();
409	if ((error = vmwrite(VMCS_HOST_FS_BASE, fsbase)) != 0)
410		goto done;
411
412	idtrbase = vmm_get_host_idtrbase();
413	if ((error = vmwrite(VMCS_HOST_IDTR_BASE, idtrbase)) != 0)
414		goto done;
415
416	/* instruction pointer */
417	if ((error = vmwrite(VMCS_HOST_RIP, host_rip)) != 0)
418		goto done;
419
420	/* stack pointer */
421	if ((error = vmwrite(VMCS_HOST_RSP, host_rsp)) != 0)
422		goto done;
423
424	/* eptp */
425	eptp = EPTP(ept_pml4);
426	if ((error = vmwrite(VMCS_EPTP, eptp)) != 0)
427		goto done;
428
429	/* vpid */
430	if ((error = vmwrite(VMCS_VPID, vpid)) != 0)
431		goto done;
432
433	/* msr bitmap */
434	if ((error = vmwrite(VMCS_MSR_BITMAP, msr_bitmap)) != 0)
435		goto done;
436
437	/* exception bitmap */
438	exc_bitmap = 1 << IDT_MC;
439	if ((error = vmwrite(VMCS_EXCEPTION_BITMAP, exc_bitmap)) != 0)
440		goto done;
441
442	/* link pointer */
443	if ((error = vmwrite(VMCS_LINK_POINTER, ~0)) != 0)
444		goto done;
445done:
446	VMCLEAR(vmcs);
447	return (error);
448}
449
450uint64_t
451vmcs_read(uint32_t encoding)
452{
453	int error;
454	uint64_t val;
455
456	error = vmread(encoding, &val);
457	if (error != 0)
458		panic("vmcs_read(%u) error %d", encoding, error);
459
460	return (val);
461}
462
463#ifdef DDB
464extern int vmxon_enabled[];
465
466DB_SHOW_COMMAND(vmcs, db_show_vmcs)
467{
468	uint64_t cur_vmcs, val;
469	uint32_t exit;
470
471	if (!vmxon_enabled[curcpu]) {
472		db_printf("VMX not enabled\n");
473		return;
474	}
475
476	if (have_addr) {
477		db_printf("Only current VMCS supported\n");
478		return;
479	}
480
481	vmptrst(&cur_vmcs);
482	if (cur_vmcs == VMCS_INITIAL) {
483		db_printf("No current VM context\n");
484		return;
485	}
486	db_printf("VMCS: %jx\n", cur_vmcs);
487	db_printf("VPID: %lu\n", vmcs_read(VMCS_VPID));
488	db_printf("Activity: ");
489	val = vmcs_read(VMCS_GUEST_ACTIVITY);
490	switch (val) {
491	case 0:
492		db_printf("Active");
493		break;
494	case 1:
495		db_printf("HLT");
496		break;
497	case 2:
498		db_printf("Shutdown");
499		break;
500	case 3:
501		db_printf("Wait for SIPI");
502		break;
503	default:
504		db_printf("Unknown: %#lx", val);
505	}
506	db_printf("\n");
507	exit = vmcs_read(VMCS_EXIT_REASON);
508	if (exit & 0x80000000)
509		db_printf("Entry Failure Reason: %u\n", exit & 0xffff);
510	else
511		db_printf("Exit Reason: %u\n", exit & 0xffff);
512	db_printf("Qualification: %#lx\n", vmcs_exit_qualification());
513	db_printf("Guest Linear Address: %#lx\n",
514	    vmcs_read(VMCS_GUEST_LINEAR_ADDRESS));
515	switch (exit & 0x8000ffff) {
516	case EXIT_REASON_EXCEPTION:
517	case EXIT_REASON_EXT_INTR:
518		val = vmcs_read(VMCS_EXIT_INTERRUPTION_INFO);
519		db_printf("Interrupt Type: ");
520		switch (val >> 8 & 0x7) {
521		case 0:
522			db_printf("external");
523			break;
524		case 2:
525			db_printf("NMI");
526			break;
527		case 3:
528			db_printf("HW exception");
529			break;
530		case 4:
531			db_printf("SW exception");
532			break;
533		default:
534			db_printf("?? %lu", val >> 8 & 0x7);
535			break;
536		}
537		db_printf("  Vector: %lu", val & 0xff);
538		if (val & 0x800)
539			db_printf("  Error Code: %lx",
540			    vmcs_read(VMCS_EXIT_INTERRUPTION_ERROR));
541		db_printf("\n");
542		break;
543	case EXIT_REASON_EPT_FAULT:
544	case EXIT_REASON_EPT_MISCONFIG:
545		db_printf("Guest Physical Address: %#lx\n",
546		    vmcs_read(VMCS_GUEST_PHYSICAL_ADDRESS));
547		break;
548	}
549	db_printf("VM-instruction error: %#lx\n", vmcs_instruction_error());
550}
551#endif
552