vmcb.c revision 276403 
1/*-
2 * Copyright (c) 2013 Anish Gupta (akgupt3@gmail.com)
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 unmodified, this list of conditions, and the following
10 *    disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 *    notice, this list of conditions and the following disclaimer in the
13 *    documentation and/or other materials provided with the distribution.
14 *
15 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25 */
26
27#include <sys/cdefs.h>
28__FBSDID("$FreeBSD: stable/10/sys/amd64/vmm/amd/vmcb.c 276403 2014-12-30 08:24:14Z neel $");
29
30#include <sys/param.h>
31#include <sys/systm.h>
32#include <sys/cpuset.h>
33
34#include <machine/segments.h>
35#include <machine/specialreg.h>
36#include <machine/vmm.h>
37
38#include "vmm_ktr.h"
39
40#include "vmcb.h"
41#include "svm.h"
42#include "svm_softc.h"
43
44/*
45 * The VMCB aka Virtual Machine Control Block is a 4KB aligned page
46 * in memory that describes the virtual machine.
47 *
48 * The VMCB contains:
49 * - instructions or events in the guest to intercept
50 * - control bits that modify execution environment of the guest
51 * - guest processor state (e.g. general purpose registers)
52 */
53
54/*
55 * Return VMCB segment area.
56 */
57static struct vmcb_segment *
58vmcb_segptr(struct vmcb *vmcb, int type)
59{
60	struct vmcb_state *state;
61	struct vmcb_segment *seg;
62
63	state = &vmcb->state;
64
65	switch (type) {
66	case VM_REG_GUEST_CS:
67		seg = &state->cs;
68		break;
69
70	case VM_REG_GUEST_DS:
71		seg = &state->ds;
72		break;
73
74	case VM_REG_GUEST_ES:
75		seg = &state->es;
76		break;
77
78	case VM_REG_GUEST_FS:
79		seg = &state->fs;
80		break;
81
82	case VM_REG_GUEST_GS:
83		seg = &state->gs;
84		break;
85
86	case VM_REG_GUEST_SS:
87		seg = &state->ss;
88		break;
89
90	case VM_REG_GUEST_GDTR:
91		seg = &state->gdt;
92		break;
93
94	case VM_REG_GUEST_IDTR:
95		seg = &state->idt;
96		break;
97
98	case VM_REG_GUEST_LDTR:
99		seg = &state->ldt;
100		break;
101
102	case VM_REG_GUEST_TR:
103		seg = &state->tr;
104		break;
105
106	default:
107		seg = NULL;
108		break;
109	}
110
111	return (seg);
112}
113
114static int
115vmcb_access(struct svm_softc *softc, int vcpu, int write, int ident,
116	uint64_t *val)
117{
118	struct vmcb *vmcb;
119	int off, bytes;
120	char *ptr;
121
122	vmcb	= svm_get_vmcb(softc, vcpu);
123	off	= VMCB_ACCESS_OFFSET(ident);
124	bytes	= VMCB_ACCESS_BYTES(ident);
125
126	if ((off + bytes) >= sizeof (struct vmcb))
127		return (EINVAL);
128
129	ptr = (char *)vmcb;
130
131	if (!write)
132		*val = 0;
133
134	switch (bytes) {
135	case 8:
136	case 4:
137	case 2:
138		if (write)
139			memcpy(ptr + off, val, bytes);
140		else
141			memcpy(val, ptr + off, bytes);
142		break;
143	default:
144		VCPU_CTR1(softc->vm, vcpu,
145		    "Invalid size %d for VMCB access: %d", bytes);
146		return (EINVAL);
147	}
148
149	/* Invalidate all VMCB state cached by h/w. */
150	if (write)
151		svm_set_dirty(softc, vcpu, 0xffffffff);
152
153	return (0);
154}
155
156/*
157 * Read from segment selector, control and general purpose register of VMCB.
158 */
159int
160vmcb_read(struct svm_softc *sc, int vcpu, int ident, uint64_t *retval)
161{
162	struct vmcb *vmcb;
163	struct vmcb_state *state;
164	struct vmcb_segment *seg;
165	int err;
166
167	vmcb = svm_get_vmcb(sc, vcpu);
168	state = &vmcb->state;
169	err = 0;
170
171	if (VMCB_ACCESS_OK(ident))
172		return (vmcb_access(sc, vcpu, 0, ident, retval));
173
174	switch (ident) {
175	case VM_REG_GUEST_CR0:
176		*retval = state->cr0;
177		break;
178
179	case VM_REG_GUEST_CR2:
180		*retval = state->cr2;
181		break;
182
183	case VM_REG_GUEST_CR3:
184		*retval = state->cr3;
185		break;
186
187	case VM_REG_GUEST_CR4:
188		*retval = state->cr4;
189		break;
190
191	case VM_REG_GUEST_DR7:
192		*retval = state->dr7;
193		break;
194
195	case VM_REG_GUEST_EFER:
196		*retval = state->efer;
197		break;
198
199	case VM_REG_GUEST_RAX:
200		*retval = state->rax;
201		break;
202
203	case VM_REG_GUEST_RFLAGS:
204		*retval = state->rflags;
205		break;
206
207	case VM_REG_GUEST_RIP:
208		*retval = state->rip;
209		break;
210
211	case VM_REG_GUEST_RSP:
212		*retval = state->rsp;
213		break;
214
215	case VM_REG_GUEST_CS:
216	case VM_REG_GUEST_DS:
217	case VM_REG_GUEST_ES:
218	case VM_REG_GUEST_FS:
219	case VM_REG_GUEST_GS:
220	case VM_REG_GUEST_SS:
221	case VM_REG_GUEST_LDTR:
222	case VM_REG_GUEST_TR:
223		seg = vmcb_segptr(vmcb, ident);
224		KASSERT(seg != NULL, ("%s: unable to get segment %d from VMCB",
225		    __func__, ident));
226		*retval = seg->selector;
227		break;
228
229	case VM_REG_GUEST_GDTR:
230	case VM_REG_GUEST_IDTR:
231		/* GDTR and IDTR don't have segment selectors */
232		err = EINVAL;
233		break;
234	default:
235		err =  EINVAL;
236		break;
237	}
238
239	return (err);
240}
241
242/*
243 * Write to segment selector, control and general purpose register of VMCB.
244 */
245int
246vmcb_write(struct svm_softc *sc, int vcpu, int ident, uint64_t val)
247{
248	struct vmcb *vmcb;
249	struct vmcb_state *state;
250	struct vmcb_segment *seg;
251	int err, dirtyseg;
252
253	vmcb = svm_get_vmcb(sc, vcpu);
254	state = &vmcb->state;
255	dirtyseg = 0;
256	err = 0;
257
258	if (VMCB_ACCESS_OK(ident))
259		return (vmcb_access(sc, vcpu, 1, ident, &val));
260
261	switch (ident) {
262	case VM_REG_GUEST_CR0:
263		state->cr0 = val;
264		svm_set_dirty(sc, vcpu, VMCB_CACHE_CR);
265		break;
266
267	case VM_REG_GUEST_CR2:
268		state->cr2 = val;
269		svm_set_dirty(sc, vcpu, VMCB_CACHE_CR2);
270		break;
271
272	case VM_REG_GUEST_CR3:
273		state->cr3 = val;
274		svm_set_dirty(sc, vcpu, VMCB_CACHE_CR);
275		break;
276
277	case VM_REG_GUEST_CR4:
278		state->cr4 = val;
279		svm_set_dirty(sc, vcpu, VMCB_CACHE_CR);
280		break;
281
282	case VM_REG_GUEST_DR7:
283		state->dr7 = val;
284		break;
285
286	case VM_REG_GUEST_EFER:
287		/* EFER_SVM must always be set when the guest is executing */
288		state->efer = val | EFER_SVM;
289		svm_set_dirty(sc, vcpu, VMCB_CACHE_CR);
290		break;
291
292	case VM_REG_GUEST_RAX:
293		state->rax = val;
294		break;
295
296	case VM_REG_GUEST_RFLAGS:
297		state->rflags = val;
298		break;
299
300	case VM_REG_GUEST_RIP:
301		state->rip = val;
302		break;
303
304	case VM_REG_GUEST_RSP:
305		state->rsp = val;
306		break;
307
308	case VM_REG_GUEST_CS:
309	case VM_REG_GUEST_DS:
310	case VM_REG_GUEST_ES:
311	case VM_REG_GUEST_SS:
312		dirtyseg = 1;		/* FALLTHROUGH */
313	case VM_REG_GUEST_FS:
314	case VM_REG_GUEST_GS:
315	case VM_REG_GUEST_LDTR:
316	case VM_REG_GUEST_TR:
317		seg = vmcb_segptr(vmcb, ident);
318		KASSERT(seg != NULL, ("%s: unable to get segment %d from VMCB",
319		    __func__, ident));
320		seg->selector = val;
321		if (dirtyseg)
322			svm_set_dirty(sc, vcpu, VMCB_CACHE_SEG);
323		break;
324
325	case VM_REG_GUEST_GDTR:
326	case VM_REG_GUEST_IDTR:
327		/* GDTR and IDTR don't have segment selectors */
328		err = EINVAL;
329		break;
330	default:
331		err = EINVAL;
332		break;
333	}
334
335	return (err);
336}
337
338int
339vmcb_seg(struct vmcb *vmcb, int ident, struct vmcb_segment *seg2)
340{
341	struct vmcb_segment *seg;
342
343	seg = vmcb_segptr(vmcb, ident);
344	if (seg != NULL) {
345		bcopy(seg, seg2, sizeof(struct vmcb_segment));
346		return (0);
347	} else {
348		return (EINVAL);
349	}
350}
351
352int
353vmcb_setdesc(void *arg, int vcpu, int reg, struct seg_desc *desc)
354{
355	struct vmcb *vmcb;
356	struct svm_softc *sc;
357	struct vmcb_segment *seg;
358	uint16_t attrib;
359
360	sc = arg;
361	vmcb = svm_get_vmcb(sc, vcpu);
362
363	seg = vmcb_segptr(vmcb, reg);
364	KASSERT(seg != NULL, ("%s: invalid segment descriptor %d",
365	    __func__, reg));
366
367	seg->base = desc->base;
368	seg->limit = desc->limit;
369	if (reg != VM_REG_GUEST_GDTR && reg != VM_REG_GUEST_IDTR) {
370		/*
371		 * Map seg_desc access to VMCB attribute format.
372		 *
373		 * SVM uses the 'P' bit in the segment attributes to indicate a
374		 * NULL segment so clear it if the segment is marked unusable.
375		 */
376		attrib = ((desc->access & 0xF000) >> 4) | (desc->access & 0xFF);
377		if (SEG_DESC_UNUSABLE(desc->access)) {
378			attrib &= ~0x80;
379		}
380		seg->attrib = attrib;
381	}
382
383	VCPU_CTR4(sc->vm, vcpu, "Setting desc %d: base (%#lx), limit (%#x), "
384	    "attrib (%#x)", reg, seg->base, seg->limit, seg->attrib);
385
386	switch (reg) {
387	case VM_REG_GUEST_CS:
388	case VM_REG_GUEST_DS:
389	case VM_REG_GUEST_ES:
390	case VM_REG_GUEST_SS:
391		svm_set_dirty(sc, vcpu, VMCB_CACHE_SEG);
392		break;
393	case VM_REG_GUEST_GDTR:
394	case VM_REG_GUEST_IDTR:
395		svm_set_dirty(sc, vcpu, VMCB_CACHE_DT);
396		break;
397	default:
398		break;
399	}
400
401	return (0);
402}
403
404int
405vmcb_getdesc(void *arg, int vcpu, int reg, struct seg_desc *desc)
406{
407	struct vmcb *vmcb;
408	struct svm_softc *sc;
409	struct vmcb_segment *seg;
410
411	sc = arg;
412	vmcb = svm_get_vmcb(sc, vcpu);
413	seg = vmcb_segptr(vmcb, reg);
414	KASSERT(seg != NULL, ("%s: invalid segment descriptor %d",
415	    __func__, reg));
416
417	desc->base = seg->base;
418	desc->limit = seg->limit;
419	desc->access = 0;
420
421	if (reg != VM_REG_GUEST_GDTR && reg != VM_REG_GUEST_IDTR) {
422		/* Map seg_desc access to VMCB attribute format */
423		desc->access = ((seg->attrib & 0xF00) << 4) |
424		    (seg->attrib & 0xFF);
425
426		/*
427		 * VT-x uses bit 16 to indicate a segment that has been loaded
428		 * with a NULL selector (aka unusable). The 'desc->access'
429		 * field is interpreted in the VT-x format by the
430		 * processor-independent code.
431		 *
432		 * SVM uses the 'P' bit to convey the same information so
433		 * convert it into the VT-x format. For more details refer to
434		 * section "Segment State in the VMCB" in APMv2.
435		 */
436		if (reg != VM_REG_GUEST_CS && reg != VM_REG_GUEST_TR) {
437			if ((desc->access & 0x80) == 0)
438				desc->access |= 0x10000;  /* Unusable segment */
439		}
440	}
441
442	return (0);
443}
444