?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
KVM: Add ioctl to tss address from userspace,
[linux-2.6/linux-2.6-openrd.git] / drivers / kvm / vmx.c
blob46b29184a4d8ee352239af36b102d6844ddbbe56
1 /*
2 * Kernel-based Virtual Machine driver for Linux
3 *
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
6 *
7 * Copyright (C) 2006 Qumranet, Inc.
8 *
9 * Authors:
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
12 *
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
15 *
16 */
17
18 #include "kvm.h"
19 #include "x86.h"
20 #include "x86_emulate.h"
21 #include "irq.h"
22 #include "vmx.h"
23 #include "segment_descriptor.h"
24
25 #include <linux/module.h>
26 #include <linux/kernel.h>
27 #include <linux/mm.h>
28 #include <linux/highmem.h>
29 #include <linux/sched.h>
30 #include <linux/moduleparam.h>
31
32 #include <asm/io.h>
33 #include <asm/desc.h>
34
35 MODULE_AUTHOR("Qumranet");
36 MODULE_LICENSE("GPL");
37
38 static int bypass_guest_pf = 1;
39 module_param(bypass_guest_pf, bool, 0);
40
41 struct vmcs {
42 u32 revision_id;
43 u32 abort;
44 char data[0];
45 };
46
47 struct vcpu_vmx {
48 struct kvm_vcpu vcpu;
49 int launched;
50 u8 fail;
51 struct kvm_msr_entry *guest_msrs;
52 struct kvm_msr_entry *host_msrs;
53 int nmsrs;
54 int save_nmsrs;
55 int msr_offset_efer;
56 #ifdef CONFIG_X86_64
57 int msr_offset_kernel_gs_base;
58 #endif
59 struct vmcs *vmcs;
60 struct {
61 int loaded;
62 u16 fs_sel, gs_sel, ldt_sel;
63 int gs_ldt_reload_needed;
64 int fs_reload_needed;
65 int guest_efer_loaded;
66 } host_state;
67
68 };
69
70 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
71 {
72 return container_of(vcpu, struct vcpu_vmx, vcpu);
73 }
74
75 static int init_rmode_tss(struct kvm *kvm);
76
77 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
78 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
79
80 static struct page *vmx_io_bitmap_a;
81 static struct page *vmx_io_bitmap_b;
82
83 static struct vmcs_config {
84 int size;
85 int order;
86 u32 revision_id;
87 u32 pin_based_exec_ctrl;
88 u32 cpu_based_exec_ctrl;
89 u32 vmexit_ctrl;
90 u32 vmentry_ctrl;
91 } vmcs_config;
92
93 #define VMX_SEGMENT_FIELD(seg) \
94 [VCPU_SREG_##seg] = { \
95 .selector = GUEST_##seg##_SELECTOR, \
96 .base = GUEST_##seg##_BASE, \
97 .limit = GUEST_##seg##_LIMIT, \
98 .ar_bytes = GUEST_##seg##_AR_BYTES, \
99 }
100
101 static struct kvm_vmx_segment_field {
102 unsigned selector;
103 unsigned base;
104 unsigned limit;
105 unsigned ar_bytes;
106 } kvm_vmx_segment_fields[] = {
107 VMX_SEGMENT_FIELD(CS),
108 VMX_SEGMENT_FIELD(DS),
109 VMX_SEGMENT_FIELD(ES),
110 VMX_SEGMENT_FIELD(FS),
111 VMX_SEGMENT_FIELD(GS),
112 VMX_SEGMENT_FIELD(SS),
113 VMX_SEGMENT_FIELD(TR),
114 VMX_SEGMENT_FIELD(LDTR),
115 };
116
117 /*
118 * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
119 * away by decrementing the array size.
120 */
121 static const u32 vmx_msr_index[] = {
122 #ifdef CONFIG_X86_64
123 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
124 #endif
125 MSR_EFER, MSR_K6_STAR,
126 };
127 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
128
129 static void load_msrs(struct kvm_msr_entry *e, int n)
130 {
131 int i;
132
133 for (i = 0; i < n; ++i)
134 wrmsrl(e[i].index, e[i].data);
135 }
136
137 static void save_msrs(struct kvm_msr_entry *e, int n)
138 {
139 int i;
140
141 for (i = 0; i < n; ++i)
142 rdmsrl(e[i].index, e[i].data);
143 }
144
145 static inline int is_page_fault(u32 intr_info)
146 {
147 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
148 INTR_INFO_VALID_MASK)) ==
149 (INTR_TYPE_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
150 }
151
152 static inline int is_no_device(u32 intr_info)
153 {
154 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
155 INTR_INFO_VALID_MASK)) ==
156 (INTR_TYPE_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
157 }
158
159 static inline int is_invalid_opcode(u32 intr_info)
160 {
161 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
162 INTR_INFO_VALID_MASK)) ==
163 (INTR_TYPE_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
164 }
165
166 static inline int is_external_interrupt(u32 intr_info)
167 {
168 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
169 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
170 }
171
172 static inline int cpu_has_vmx_tpr_shadow(void)
173 {
174 return (vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW);
175 }
176
177 static inline int vm_need_tpr_shadow(struct kvm *kvm)
178 {
179 return ((cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm)));
180 }
181
182 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
183 {
184 int i;
185
186 for (i = 0; i < vmx->nmsrs; ++i)
187 if (vmx->guest_msrs[i].index == msr)
188 return i;
189 return -1;
190 }
191
192 static struct kvm_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
193 {
194 int i;
195
196 i = __find_msr_index(vmx, msr);
197 if (i >= 0)
198 return &vmx->guest_msrs[i];
199 return NULL;
200 }
201
202 static void vmcs_clear(struct vmcs *vmcs)
203 {
204 u64 phys_addr = __pa(vmcs);
205 u8 error;
206
207 asm volatile (ASM_VMX_VMCLEAR_RAX "; setna %0"
208 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
209 : "cc", "memory");
210 if (error)
211 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
212 vmcs, phys_addr);
213 }
214
215 static void __vcpu_clear(void *arg)
216 {
217 struct vcpu_vmx *vmx = arg;
218 int cpu = raw_smp_processor_id();
219
220 if (vmx->vcpu.cpu == cpu)
221 vmcs_clear(vmx->vmcs);
222 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
223 per_cpu(current_vmcs, cpu) = NULL;
224 rdtscll(vmx->vcpu.host_tsc);
225 }
226
227 static void vcpu_clear(struct vcpu_vmx *vmx)
228 {
229 if (vmx->vcpu.cpu == -1)
230 return;
231 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 0, 1);
232 vmx->launched = 0;
233 }
234
235 static unsigned long vmcs_readl(unsigned long field)
236 {
237 unsigned long value;
238
239 asm volatile (ASM_VMX_VMREAD_RDX_RAX
240 : "=a"(value) : "d"(field) : "cc");
241 return value;
242 }
243
244 static u16 vmcs_read16(unsigned long field)
245 {
246 return vmcs_readl(field);
247 }
248
249 static u32 vmcs_read32(unsigned long field)
250 {
251 return vmcs_readl(field);
252 }
253
254 static u64 vmcs_read64(unsigned long field)
255 {
256 #ifdef CONFIG_X86_64
257 return vmcs_readl(field);
258 #else
259 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
260 #endif
261 }
262
263 static noinline void vmwrite_error(unsigned long field, unsigned long value)
264 {
265 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
266 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
267 dump_stack();
268 }
269
270 static void vmcs_writel(unsigned long field, unsigned long value)
271 {
272 u8 error;
273
274 asm volatile (ASM_VMX_VMWRITE_RAX_RDX "; setna %0"
275 : "=q"(error) : "a"(value), "d"(field) : "cc");
276 if (unlikely(error))
277 vmwrite_error(field, value);
278 }
279
280 static void vmcs_write16(unsigned long field, u16 value)
281 {
282 vmcs_writel(field, value);
283 }
284
285 static void vmcs_write32(unsigned long field, u32 value)
286 {
287 vmcs_writel(field, value);
288 }
289
290 static void vmcs_write64(unsigned long field, u64 value)
291 {
292 #ifdef CONFIG_X86_64
293 vmcs_writel(field, value);
294 #else
295 vmcs_writel(field, value);
296 asm volatile ("");
297 vmcs_writel(field+1, value >> 32);
298 #endif
299 }
300
301 static void vmcs_clear_bits(unsigned long field, u32 mask)
302 {
303 vmcs_writel(field, vmcs_readl(field) & ~mask);
304 }
305
306 static void vmcs_set_bits(unsigned long field, u32 mask)
307 {
308 vmcs_writel(field, vmcs_readl(field) | mask);
309 }
310
311 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
312 {
313 u32 eb;
314
315 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR);
316 if (!vcpu->fpu_active)
317 eb |= 1u << NM_VECTOR;
318 if (vcpu->guest_debug.enabled)
319 eb |= 1u << 1;
320 if (vcpu->rmode.active)
321 eb = ~0;
322 vmcs_write32(EXCEPTION_BITMAP, eb);
323 }
324
325 static void reload_tss(void)
326 {
327 #ifndef CONFIG_X86_64
328
329 /*
330 * VT restores TR but not its size. Useless.
331 */
332 struct descriptor_table gdt;
333 struct segment_descriptor *descs;
334
335 get_gdt(&gdt);
336 descs = (void *)gdt.base;
337 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
338 load_TR_desc();
339 #endif
340 }
341
342 static void load_transition_efer(struct vcpu_vmx *vmx)
343 {
344 int efer_offset = vmx->msr_offset_efer;
345 u64 host_efer = vmx->host_msrs[efer_offset].data;
346 u64 guest_efer = vmx->guest_msrs[efer_offset].data;
347 u64 ignore_bits;
348
349 if (efer_offset < 0)
350 return;
351 /*
352 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
353 * outside long mode
354 */
355 ignore_bits = EFER_NX | EFER_SCE;
356 #ifdef CONFIG_X86_64
357 ignore_bits |= EFER_LMA | EFER_LME;
358 /* SCE is meaningful only in long mode on Intel */
359 if (guest_efer & EFER_LMA)
360 ignore_bits &= ~(u64)EFER_SCE;
361 #endif
362 if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
363 return;
364
365 vmx->host_state.guest_efer_loaded = 1;
366 guest_efer &= ~ignore_bits;
367 guest_efer |= host_efer & ignore_bits;
368 wrmsrl(MSR_EFER, guest_efer);
369 vmx->vcpu.stat.efer_reload++;
370 }
371
372 static void reload_host_efer(struct vcpu_vmx *vmx)
373 {
374 if (vmx->host_state.guest_efer_loaded) {
375 vmx->host_state.guest_efer_loaded = 0;
376 load_msrs(vmx->host_msrs + vmx->msr_offset_efer, 1);
377 }
378 }
379
380 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
381 {
382 struct vcpu_vmx *vmx = to_vmx(vcpu);
383
384 if (vmx->host_state.loaded)
385 return;
386
387 vmx->host_state.loaded = 1;
388 /*
389 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
390 * allow segment selectors with cpl > 0 or ti == 1.
391 */
392 vmx->host_state.ldt_sel = read_ldt();
393 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
394 vmx->host_state.fs_sel = read_fs();
395 if (!(vmx->host_state.fs_sel & 7)) {
396 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
397 vmx->host_state.fs_reload_needed = 0;
398 } else {
399 vmcs_write16(HOST_FS_SELECTOR, 0);
400 vmx->host_state.fs_reload_needed = 1;
401 }
402 vmx->host_state.gs_sel = read_gs();
403 if (!(vmx->host_state.gs_sel & 7))
404 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
405 else {
406 vmcs_write16(HOST_GS_SELECTOR, 0);
407 vmx->host_state.gs_ldt_reload_needed = 1;
408 }
409
410 #ifdef CONFIG_X86_64
411 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
412 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
413 #else
414 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
415 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
416 #endif
417
418 #ifdef CONFIG_X86_64
419 if (is_long_mode(&vmx->vcpu))
420 save_msrs(vmx->host_msrs +
421 vmx->msr_offset_kernel_gs_base, 1);
422
423 #endif
424 load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
425 load_transition_efer(vmx);
426 }
427
428 static void vmx_load_host_state(struct vcpu_vmx *vmx)
429 {
430 unsigned long flags;
431
432 if (!vmx->host_state.loaded)
433 return;
434
435 vmx->host_state.loaded = 0;
436 if (vmx->host_state.fs_reload_needed)
437 load_fs(vmx->host_state.fs_sel);
438 if (vmx->host_state.gs_ldt_reload_needed) {
439 load_ldt(vmx->host_state.ldt_sel);
440 /*
441 * If we have to reload gs, we must take care to
442 * preserve our gs base.
443 */
444 local_irq_save(flags);
445 load_gs(vmx->host_state.gs_sel);
446 #ifdef CONFIG_X86_64
447 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
448 #endif
449 local_irq_restore(flags);
450 }
451 reload_tss();
452 save_msrs(vmx->guest_msrs, vmx->save_nmsrs);
453 load_msrs(vmx->host_msrs, vmx->save_nmsrs);
454 reload_host_efer(vmx);
455 }
456
457 /*
458 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
459 * vcpu mutex is already taken.
460 */
461 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
462 {
463 struct vcpu_vmx *vmx = to_vmx(vcpu);
464 u64 phys_addr = __pa(vmx->vmcs);
465 u64 tsc_this, delta;
466
467 if (vcpu->cpu != cpu) {
468 vcpu_clear(vmx);
469 kvm_migrate_apic_timer(vcpu);
470 }
471
472 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
473 u8 error;
474
475 per_cpu(current_vmcs, cpu) = vmx->vmcs;
476 asm volatile (ASM_VMX_VMPTRLD_RAX "; setna %0"
477 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
478 : "cc");
479 if (error)
480 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
481 vmx->vmcs, phys_addr);
482 }
483
484 if (vcpu->cpu != cpu) {
485 struct descriptor_table dt;
486 unsigned long sysenter_esp;
487
488 vcpu->cpu = cpu;
489 /*
490 * Linux uses per-cpu TSS and GDT, so set these when switching
491 * processors.
492 */
493 vmcs_writel(HOST_TR_BASE, read_tr_base()); /* 22.2.4 */
494 get_gdt(&dt);
495 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
496
497 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
498 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
499
500 /*
501 * Make sure the time stamp counter is monotonous.
502 */
503 rdtscll(tsc_this);
504 delta = vcpu->host_tsc - tsc_this;
505 vmcs_write64(TSC_OFFSET, vmcs_read64(TSC_OFFSET) + delta);
506 }
507 }
508
509 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
510 {
511 vmx_load_host_state(to_vmx(vcpu));
512 kvm_put_guest_fpu(vcpu);
513 }
514
515 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
516 {
517 if (vcpu->fpu_active)
518 return;
519 vcpu->fpu_active = 1;
520 vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
521 if (vcpu->cr0 & X86_CR0_TS)
522 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
523 update_exception_bitmap(vcpu);
524 }
525
526 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
527 {
528 if (!vcpu->fpu_active)
529 return;
530 vcpu->fpu_active = 0;
531 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
532 update_exception_bitmap(vcpu);
533 }
534
535 static void vmx_vcpu_decache(struct kvm_vcpu *vcpu)
536 {
537 vcpu_clear(to_vmx(vcpu));
538 }
539
540 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
541 {
542 return vmcs_readl(GUEST_RFLAGS);
543 }
544
545 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
546 {
547 if (vcpu->rmode.active)
548 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
549 vmcs_writel(GUEST_RFLAGS, rflags);
550 }
551
552 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
553 {
554 unsigned long rip;
555 u32 interruptibility;
556
557 rip = vmcs_readl(GUEST_RIP);
558 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
559 vmcs_writel(GUEST_RIP, rip);
560
561 /*
562 * We emulated an instruction, so temporary interrupt blocking
563 * should be removed, if set.
564 */
565 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
566 if (interruptibility & 3)
567 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
568 interruptibility & ~3);
569 vcpu->interrupt_window_open = 1;
570 }
571
572 static void vmx_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
573 {
574 printk(KERN_DEBUG "inject_general_protection: rip 0x%lx\n",
575 vmcs_readl(GUEST_RIP));
576 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
577 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
578 GP_VECTOR |
579 INTR_TYPE_EXCEPTION |
580 INTR_INFO_DELIEVER_CODE_MASK |
581 INTR_INFO_VALID_MASK);
582 }
583
584 static void vmx_inject_ud(struct kvm_vcpu *vcpu)
585 {
586 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
587 UD_VECTOR |
588 INTR_TYPE_EXCEPTION |
589 INTR_INFO_VALID_MASK);
590 }
591
592 /*
593 * Swap MSR entry in host/guest MSR entry array.
594 */
595 #ifdef CONFIG_X86_64
596 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
597 {
598 struct kvm_msr_entry tmp;
599
600 tmp = vmx->guest_msrs[to];
601 vmx->guest_msrs[to] = vmx->guest_msrs[from];
602 vmx->guest_msrs[from] = tmp;
603 tmp = vmx->host_msrs[to];
604 vmx->host_msrs[to] = vmx->host_msrs[from];
605 vmx->host_msrs[from] = tmp;
606 }
607 #endif
608
609 /*
610 * Set up the vmcs to automatically save and restore system
611 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
612 * mode, as fiddling with msrs is very expensive.
613 */
614 static void setup_msrs(struct vcpu_vmx *vmx)
615 {
616 int save_nmsrs;
617
618 save_nmsrs = 0;
619 #ifdef CONFIG_X86_64
620 if (is_long_mode(&vmx->vcpu)) {
621 int index;
622
623 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
624 if (index >= 0)
625 move_msr_up(vmx, index, save_nmsrs++);
626 index = __find_msr_index(vmx, MSR_LSTAR);
627 if (index >= 0)
628 move_msr_up(vmx, index, save_nmsrs++);
629 index = __find_msr_index(vmx, MSR_CSTAR);
630 if (index >= 0)
631 move_msr_up(vmx, index, save_nmsrs++);
632 index = __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
633 if (index >= 0)
634 move_msr_up(vmx, index, save_nmsrs++);
635 /*
636 * MSR_K6_STAR is only needed on long mode guests, and only
637 * if efer.sce is enabled.
638 */
639 index = __find_msr_index(vmx, MSR_K6_STAR);
640 if ((index >= 0) && (vmx->vcpu.shadow_efer & EFER_SCE))
641 move_msr_up(vmx, index, save_nmsrs++);
642 }
643 #endif
644 vmx->save_nmsrs = save_nmsrs;
645
646 #ifdef CONFIG_X86_64
647 vmx->msr_offset_kernel_gs_base =
648 __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
649 #endif
650 vmx->msr_offset_efer = __find_msr_index(vmx, MSR_EFER);
651 }
652
653 /*
654 * reads and returns guest's timestamp counter "register"
655 * guest_tsc = host_tsc + tsc_offset -- 21.3
656 */
657 static u64 guest_read_tsc(void)
658 {
659 u64 host_tsc, tsc_offset;
660
661 rdtscll(host_tsc);
662 tsc_offset = vmcs_read64(TSC_OFFSET);
663 return host_tsc + tsc_offset;
664 }
665
666 /*
667 * writes 'guest_tsc' into guest's timestamp counter "register"
668 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
669 */
670 static void guest_write_tsc(u64 guest_tsc)
671 {
672 u64 host_tsc;
673
674 rdtscll(host_tsc);
675 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
676 }
677
678 /*
679 * Reads an msr value (of 'msr_index') into 'pdata'.
680 * Returns 0 on success, non-0 otherwise.
681 * Assumes vcpu_load() was already called.
682 */
683 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
684 {
685 u64 data;
686 struct kvm_msr_entry *msr;
687
688 if (!pdata) {
689 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
690 return -EINVAL;
691 }
692
693 switch (msr_index) {
694 #ifdef CONFIG_X86_64
695 case MSR_FS_BASE:
696 data = vmcs_readl(GUEST_FS_BASE);
697 break;
698 case MSR_GS_BASE:
699 data = vmcs_readl(GUEST_GS_BASE);
700 break;
701 case MSR_EFER:
702 return kvm_get_msr_common(vcpu, msr_index, pdata);
703 #endif
704 case MSR_IA32_TIME_STAMP_COUNTER:
705 data = guest_read_tsc();
706 break;
707 case MSR_IA32_SYSENTER_CS:
708 data = vmcs_read32(GUEST_SYSENTER_CS);
709 break;
710 case MSR_IA32_SYSENTER_EIP:
711 data = vmcs_readl(GUEST_SYSENTER_EIP);
712 break;
713 case MSR_IA32_SYSENTER_ESP:
714 data = vmcs_readl(GUEST_SYSENTER_ESP);
715 break;
716 default:
717 msr = find_msr_entry(to_vmx(vcpu), msr_index);
718 if (msr) {
719 data = msr->data;
720 break;
721 }
722 return kvm_get_msr_common(vcpu, msr_index, pdata);
723 }
724
725 *pdata = data;
726 return 0;
727 }
728
729 /*
730 * Writes msr value into into the appropriate "register".
731 * Returns 0 on success, non-0 otherwise.
732 * Assumes vcpu_load() was already called.
733 */
734 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
735 {
736 struct vcpu_vmx *vmx = to_vmx(vcpu);
737 struct kvm_msr_entry *msr;
738 int ret = 0;
739
740 switch (msr_index) {
741 #ifdef CONFIG_X86_64
742 case MSR_EFER:
743 ret = kvm_set_msr_common(vcpu, msr_index, data);
744 if (vmx->host_state.loaded) {
745 reload_host_efer(vmx);
746 load_transition_efer(vmx);
747 }
748 break;
749 case MSR_FS_BASE:
750 vmcs_writel(GUEST_FS_BASE, data);
751 break;
752 case MSR_GS_BASE:
753 vmcs_writel(GUEST_GS_BASE, data);
754 break;
755 #endif
756 case MSR_IA32_SYSENTER_CS:
757 vmcs_write32(GUEST_SYSENTER_CS, data);
758 break;
759 case MSR_IA32_SYSENTER_EIP:
760 vmcs_writel(GUEST_SYSENTER_EIP, data);
761 break;
762 case MSR_IA32_SYSENTER_ESP:
763 vmcs_writel(GUEST_SYSENTER_ESP, data);
764 break;
765 case MSR_IA32_TIME_STAMP_COUNTER:
766 guest_write_tsc(data);
767 break;
768 default:
769 msr = find_msr_entry(vmx, msr_index);
770 if (msr) {
771 msr->data = data;
772 if (vmx->host_state.loaded)
773 load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
774 break;
775 }
776 ret = kvm_set_msr_common(vcpu, msr_index, data);
777 }
778
779 return ret;
780 }
781
782 /*
783 * Sync the rsp and rip registers into the vcpu structure. This allows
784 * registers to be accessed by indexing vcpu->regs.
785 */
786 static void vcpu_load_rsp_rip(struct kvm_vcpu *vcpu)
787 {
788 vcpu->regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
789 vcpu->rip = vmcs_readl(GUEST_RIP);
790 }
791
792 /*
793 * Syncs rsp and rip back into the vmcs. Should be called after possible
794 * modification.
795 */
796 static void vcpu_put_rsp_rip(struct kvm_vcpu *vcpu)
797 {
798 vmcs_writel(GUEST_RSP, vcpu->regs[VCPU_REGS_RSP]);
799 vmcs_writel(GUEST_RIP, vcpu->rip);
800 }
801
802 static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
803 {
804 unsigned long dr7 = 0x400;
805 int old_singlestep;
806
807 old_singlestep = vcpu->guest_debug.singlestep;
808
809 vcpu->guest_debug.enabled = dbg->enabled;
810 if (vcpu->guest_debug.enabled) {
811 int i;
812
813 dr7 |= 0x200; /* exact */
814 for (i = 0; i < 4; ++i) {
815 if (!dbg->breakpoints[i].enabled)
816 continue;
817 vcpu->guest_debug.bp[i] = dbg->breakpoints[i].address;
818 dr7 |= 2 << (i*2); /* global enable */
819 dr7 |= 0 << (i*4+16); /* execution breakpoint */
820 }
821
822 vcpu->guest_debug.singlestep = dbg->singlestep;
823 } else
824 vcpu->guest_debug.singlestep = 0;
825
826 if (old_singlestep && !vcpu->guest_debug.singlestep) {
827 unsigned long flags;
828
829 flags = vmcs_readl(GUEST_RFLAGS);
830 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
831 vmcs_writel(GUEST_RFLAGS, flags);
832 }
833
834 update_exception_bitmap(vcpu);
835 vmcs_writel(GUEST_DR7, dr7);
836
837 return 0;
838 }
839
840 static int vmx_get_irq(struct kvm_vcpu *vcpu)
841 {
842 u32 idtv_info_field;
843
844 idtv_info_field = vmcs_read32(IDT_VECTORING_INFO_FIELD);
845 if (idtv_info_field & INTR_INFO_VALID_MASK) {
846 if (is_external_interrupt(idtv_info_field))
847 return idtv_info_field & VECTORING_INFO_VECTOR_MASK;
848 else
849 printk(KERN_DEBUG "pending exception: not handled yet\n");
850 }
851 return -1;
852 }
853
854 static __init int cpu_has_kvm_support(void)
855 {
856 unsigned long ecx = cpuid_ecx(1);
857 return test_bit(5, &ecx); /* CPUID.1:ECX.VMX[bit 5] -> VT */
858 }
859
860 static __init int vmx_disabled_by_bios(void)
861 {
862 u64 msr;
863
864 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
865 return (msr & (MSR_IA32_FEATURE_CONTROL_LOCKED |
866 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
867 == MSR_IA32_FEATURE_CONTROL_LOCKED;
868 /* locked but not enabled */
869 }
870
871 static void hardware_enable(void *garbage)
872 {
873 int cpu = raw_smp_processor_id();
874 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
875 u64 old;
876
877 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
878 if ((old & (MSR_IA32_FEATURE_CONTROL_LOCKED |
879 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
880 != (MSR_IA32_FEATURE_CONTROL_LOCKED |
881 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
882 /* enable and lock */
883 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
884 MSR_IA32_FEATURE_CONTROL_LOCKED |
885 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED);
886 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
887 asm volatile (ASM_VMX_VMXON_RAX : : "a"(&phys_addr), "m"(phys_addr)
888 : "memory", "cc");
889 }
890
891 static void hardware_disable(void *garbage)
892 {
893 asm volatile (ASM_VMX_VMXOFF : : : "cc");
894 }
895
896 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
897 u32 msr, u32 *result)
898 {
899 u32 vmx_msr_low, vmx_msr_high;
900 u32 ctl = ctl_min | ctl_opt;
901
902 rdmsr(msr, vmx_msr_low, vmx_msr_high);
903
904 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
905 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
906
907 /* Ensure minimum (required) set of control bits are supported. */
908 if (ctl_min & ~ctl)
909 return -EIO;
910
911 *result = ctl;
912 return 0;
913 }
914
915 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
916 {
917 u32 vmx_msr_low, vmx_msr_high;
918 u32 min, opt;
919 u32 _pin_based_exec_control = 0;
920 u32 _cpu_based_exec_control = 0;
921 u32 _vmexit_control = 0;
922 u32 _vmentry_control = 0;
923
924 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
925 opt = 0;
926 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
927 &_pin_based_exec_control) < 0)
928 return -EIO;
929
930 min = CPU_BASED_HLT_EXITING |
931 #ifdef CONFIG_X86_64
932 CPU_BASED_CR8_LOAD_EXITING |
933 CPU_BASED_CR8_STORE_EXITING |
934 #endif
935 CPU_BASED_USE_IO_BITMAPS |
936 CPU_BASED_MOV_DR_EXITING |
937 CPU_BASED_USE_TSC_OFFSETING;
938 #ifdef CONFIG_X86_64
939 opt = CPU_BASED_TPR_SHADOW;
940 #else
941 opt = 0;
942 #endif
943 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
944 &_cpu_based_exec_control) < 0)
945 return -EIO;
946 #ifdef CONFIG_X86_64
947 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
948 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
949 ~CPU_BASED_CR8_STORE_EXITING;
950 #endif
951
952 min = 0;
953 #ifdef CONFIG_X86_64
954 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
955 #endif
956 opt = 0;
957 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
958 &_vmexit_control) < 0)
959 return -EIO;
960
961 min = opt = 0;
962 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
963 &_vmentry_control) < 0)
964 return -EIO;
965
966 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
967
968 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
969 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
970 return -EIO;
971
972 #ifdef CONFIG_X86_64
973 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
974 if (vmx_msr_high & (1u<<16))
975 return -EIO;
976 #endif
977
978 /* Require Write-Back (WB) memory type for VMCS accesses. */
979 if (((vmx_msr_high >> 18) & 15) != 6)
980 return -EIO;
981
982 vmcs_conf->size = vmx_msr_high & 0x1fff;
983 vmcs_conf->order = get_order(vmcs_config.size);
984 vmcs_conf->revision_id = vmx_msr_low;
985
986 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
987 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
988 vmcs_conf->vmexit_ctrl = _vmexit_control;
989 vmcs_conf->vmentry_ctrl = _vmentry_control;
990
991 return 0;
992 }
993
994 static struct vmcs *alloc_vmcs_cpu(int cpu)
995 {
996 int node = cpu_to_node(cpu);
997 struct page *pages;
998 struct vmcs *vmcs;
999
1000 pages = alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
1001 if (!pages)
1002 return NULL;
1003 vmcs = page_address(pages);
1004 memset(vmcs, 0, vmcs_config.size);
1005 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1006 return vmcs;
1007 }
1008
1009 static struct vmcs *alloc_vmcs(void)
1010 {
1011 return alloc_vmcs_cpu(raw_smp_processor_id());
1012 }
1013
1014 static void free_vmcs(struct vmcs *vmcs)
1015 {
1016 free_pages((unsigned long)vmcs, vmcs_config.order);
1017 }
1018
1019 static void free_kvm_area(void)
1020 {
1021 int cpu;
1022
1023 for_each_online_cpu(cpu)
1024 free_vmcs(per_cpu(vmxarea, cpu));
1025 }
1026
1027 static __init int alloc_kvm_area(void)
1028 {
1029 int cpu;
1030
1031 for_each_online_cpu(cpu) {
1032 struct vmcs *vmcs;
1033
1034 vmcs = alloc_vmcs_cpu(cpu);
1035 if (!vmcs) {
1036 free_kvm_area();
1037 return -ENOMEM;
1038 }
1039
1040 per_cpu(vmxarea, cpu) = vmcs;
1041 }
1042 return 0;
1043 }
1044
1045 static __init int hardware_setup(void)
1046 {
1047 if (setup_vmcs_config(&vmcs_config) < 0)
1048 return -EIO;
1049 return alloc_kvm_area();
1050 }
1051
1052 static __exit void hardware_unsetup(void)
1053 {
1054 free_kvm_area();
1055 }
1056
1057 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1058 {
1059 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1060
1061 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1062 vmcs_write16(sf->selector, save->selector);
1063 vmcs_writel(sf->base, save->base);
1064 vmcs_write32(sf->limit, save->limit);
1065 vmcs_write32(sf->ar_bytes, save->ar);
1066 } else {
1067 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1068 << AR_DPL_SHIFT;
1069 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1070 }
1071 }
1072
1073 static void enter_pmode(struct kvm_vcpu *vcpu)
1074 {
1075 unsigned long flags;
1076
1077 vcpu->rmode.active = 0;
1078
1079 vmcs_writel(GUEST_TR_BASE, vcpu->rmode.tr.base);
1080 vmcs_write32(GUEST_TR_LIMIT, vcpu->rmode.tr.limit);
1081 vmcs_write32(GUEST_TR_AR_BYTES, vcpu->rmode.tr.ar);
1082
1083 flags = vmcs_readl(GUEST_RFLAGS);
1084 flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
1085 flags |= (vcpu->rmode.save_iopl << IOPL_SHIFT);
1086 vmcs_writel(GUEST_RFLAGS, flags);
1087
1088 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1089 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1090
1091 update_exception_bitmap(vcpu);
1092
1093 fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->rmode.es);
1094 fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->rmode.ds);
1095 fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->rmode.gs);
1096 fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->rmode.fs);
1097
1098 vmcs_write16(GUEST_SS_SELECTOR, 0);
1099 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1100
1101 vmcs_write16(GUEST_CS_SELECTOR,
1102 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1103 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1104 }
1105
1106 static gva_t rmode_tss_base(struct kvm *kvm)
1107 {
1108 if (!kvm->tss_addr) {
1109 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1110 kvm->memslots[0].npages - 3;
1111 return base_gfn << PAGE_SHIFT;
1112 }
1113 return kvm->tss_addr;
1114 }
1115
1116 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1117 {
1118 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1119
1120 save->selector = vmcs_read16(sf->selector);
1121 save->base = vmcs_readl(sf->base);
1122 save->limit = vmcs_read32(sf->limit);
1123 save->ar = vmcs_read32(sf->ar_bytes);
1124 vmcs_write16(sf->selector, vmcs_readl(sf->base) >> 4);
1125 vmcs_write32(sf->limit, 0xffff);
1126 vmcs_write32(sf->ar_bytes, 0xf3);
1127 }
1128
1129 static void enter_rmode(struct kvm_vcpu *vcpu)
1130 {
1131 unsigned long flags;
1132
1133 vcpu->rmode.active = 1;
1134
1135 vcpu->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1136 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1137
1138 vcpu->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1139 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1140
1141 vcpu->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1142 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1143
1144 flags = vmcs_readl(GUEST_RFLAGS);
1145 vcpu->rmode.save_iopl = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
1146
1147 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1148
1149 vmcs_writel(GUEST_RFLAGS, flags);
1150 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1151 update_exception_bitmap(vcpu);
1152
1153 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1154 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1155 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1156
1157 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1158 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1159 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1160 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1161 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1162
1163 fix_rmode_seg(VCPU_SREG_ES, &vcpu->rmode.es);
1164 fix_rmode_seg(VCPU_SREG_DS, &vcpu->rmode.ds);
1165 fix_rmode_seg(VCPU_SREG_GS, &vcpu->rmode.gs);
1166 fix_rmode_seg(VCPU_SREG_FS, &vcpu->rmode.fs);
1167
1168 kvm_mmu_reset_context(vcpu);
1169 init_rmode_tss(vcpu->kvm);
1170 }
1171
1172 #ifdef CONFIG_X86_64
1173
1174 static void enter_lmode(struct kvm_vcpu *vcpu)
1175 {
1176 u32 guest_tr_ar;
1177
1178 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1179 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1180 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1181 __FUNCTION__);
1182 vmcs_write32(GUEST_TR_AR_BYTES,
1183 (guest_tr_ar & ~AR_TYPE_MASK)
1184 | AR_TYPE_BUSY_64_TSS);
1185 }
1186
1187 vcpu->shadow_efer |= EFER_LMA;
1188
1189 find_msr_entry(to_vmx(vcpu), MSR_EFER)->data |= EFER_LMA | EFER_LME;
1190 vmcs_write32(VM_ENTRY_CONTROLS,
1191 vmcs_read32(VM_ENTRY_CONTROLS)
1192 | VM_ENTRY_IA32E_MODE);
1193 }
1194
1195 static void exit_lmode(struct kvm_vcpu *vcpu)
1196 {
1197 vcpu->shadow_efer &= ~EFER_LMA;
1198
1199 vmcs_write32(VM_ENTRY_CONTROLS,
1200 vmcs_read32(VM_ENTRY_CONTROLS)
1201 & ~VM_ENTRY_IA32E_MODE);
1202 }
1203
1204 #endif
1205
1206 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1207 {
1208 vcpu->cr4 &= KVM_GUEST_CR4_MASK;
1209 vcpu->cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
1210 }
1211
1212 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1213 {
1214 vmx_fpu_deactivate(vcpu);
1215
1216 if (vcpu->rmode.active && (cr0 & X86_CR0_PE))
1217 enter_pmode(vcpu);
1218
1219 if (!vcpu->rmode.active && !(cr0 & X86_CR0_PE))
1220 enter_rmode(vcpu);
1221
1222 #ifdef CONFIG_X86_64
1223 if (vcpu->shadow_efer & EFER_LME) {
1224 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1225 enter_lmode(vcpu);
1226 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1227 exit_lmode(vcpu);
1228 }
1229 #endif
1230
1231 vmcs_writel(CR0_READ_SHADOW, cr0);
1232 vmcs_writel(GUEST_CR0,
1233 (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON);
1234 vcpu->cr0 = cr0;
1235
1236 if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
1237 vmx_fpu_activate(vcpu);
1238 }
1239
1240 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1241 {
1242 vmcs_writel(GUEST_CR3, cr3);
1243 if (vcpu->cr0 & X86_CR0_PE)
1244 vmx_fpu_deactivate(vcpu);
1245 }
1246
1247 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1248 {
1249 vmcs_writel(CR4_READ_SHADOW, cr4);
1250 vmcs_writel(GUEST_CR4, cr4 | (vcpu->rmode.active ?
1251 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON));
1252 vcpu->cr4 = cr4;
1253 }
1254
1255 #ifdef CONFIG_X86_64
1256
1257 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1258 {
1259 struct vcpu_vmx *vmx = to_vmx(vcpu);
1260 struct kvm_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1261
1262 vcpu->shadow_efer = efer;
1263 if (efer & EFER_LMA) {
1264 vmcs_write32(VM_ENTRY_CONTROLS,
1265 vmcs_read32(VM_ENTRY_CONTROLS) |
1266 VM_ENTRY_IA32E_MODE);
1267 msr->data = efer;
1268
1269 } else {
1270 vmcs_write32(VM_ENTRY_CONTROLS,
1271 vmcs_read32(VM_ENTRY_CONTROLS) &
1272 ~VM_ENTRY_IA32E_MODE);
1273
1274 msr->data = efer & ~EFER_LME;
1275 }
1276 setup_msrs(vmx);
1277 }
1278
1279 #endif
1280
1281 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1282 {
1283 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1284
1285 return vmcs_readl(sf->base);
1286 }
1287
1288 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1289 struct kvm_segment *var, int seg)
1290 {
1291 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1292 u32 ar;
1293
1294 var->base = vmcs_readl(sf->base);
1295 var->limit = vmcs_read32(sf->limit);
1296 var->selector = vmcs_read16(sf->selector);
1297 ar = vmcs_read32(sf->ar_bytes);
1298 if (ar & AR_UNUSABLE_MASK)
1299 ar = 0;
1300 var->type = ar & 15;
1301 var->s = (ar >> 4) & 1;
1302 var->dpl = (ar >> 5) & 3;
1303 var->present = (ar >> 7) & 1;
1304 var->avl = (ar >> 12) & 1;
1305 var->l = (ar >> 13) & 1;
1306 var->db = (ar >> 14) & 1;
1307 var->g = (ar >> 15) & 1;
1308 var->unusable = (ar >> 16) & 1;
1309 }
1310
1311 static u32 vmx_segment_access_rights(struct kvm_segment *var)
1312 {
1313 u32 ar;
1314
1315 if (var->unusable)
1316 ar = 1 << 16;
1317 else {
1318 ar = var->type & 15;
1319 ar |= (var->s & 1) << 4;
1320 ar |= (var->dpl & 3) << 5;
1321 ar |= (var->present & 1) << 7;
1322 ar |= (var->avl & 1) << 12;
1323 ar |= (var->l & 1) << 13;
1324 ar |= (var->db & 1) << 14;
1325 ar |= (var->g & 1) << 15;
1326 }
1327 if (ar == 0) /* a 0 value means unusable */
1328 ar = AR_UNUSABLE_MASK;
1329
1330 return ar;
1331 }
1332
1333 static void vmx_set_segment(struct kvm_vcpu *vcpu,
1334 struct kvm_segment *var, int seg)
1335 {
1336 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1337 u32 ar;
1338
1339 if (vcpu->rmode.active && seg == VCPU_SREG_TR) {
1340 vcpu->rmode.tr.selector = var->selector;
1341 vcpu->rmode.tr.base = var->base;
1342 vcpu->rmode.tr.limit = var->limit;
1343 vcpu->rmode.tr.ar = vmx_segment_access_rights(var);
1344 return;
1345 }
1346 vmcs_writel(sf->base, var->base);
1347 vmcs_write32(sf->limit, var->limit);
1348 vmcs_write16(sf->selector, var->selector);
1349 if (vcpu->rmode.active && var->s) {
1350 /*
1351 * Hack real-mode segments into vm86 compatibility.
1352 */
1353 if (var->base == 0xffff0000 && var->selector == 0xf000)
1354 vmcs_writel(sf->base, 0xf0000);
1355 ar = 0xf3;
1356 } else
1357 ar = vmx_segment_access_rights(var);
1358 vmcs_write32(sf->ar_bytes, ar);
1359 }
1360
1361 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1362 {
1363 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1364
1365 *db = (ar >> 14) & 1;
1366 *l = (ar >> 13) & 1;
1367 }
1368
1369 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1370 {
1371 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1372 dt->base = vmcs_readl(GUEST_IDTR_BASE);
1373 }
1374
1375 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1376 {
1377 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1378 vmcs_writel(GUEST_IDTR_BASE, dt->base);
1379 }
1380
1381 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1382 {
1383 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1384 dt->base = vmcs_readl(GUEST_GDTR_BASE);
1385 }
1386
1387 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1388 {
1389 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1390 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1391 }
1392
1393 static int init_rmode_tss(struct kvm *kvm)
1394 {
1395 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
1396 u16 data = 0;
1397 int r;
1398
1399 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1400 if (r < 0)
1401 return 0;
1402 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
1403 r = kvm_write_guest_page(kvm, fn++, &data, 0x66, sizeof(u16));
1404 if (r < 0)
1405 return 0;
1406 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
1407 if (r < 0)
1408 return 0;
1409 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1410 if (r < 0)
1411 return 0;
1412 data = ~0;
1413 r = kvm_write_guest_page(kvm, fn, &data, RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
1414 sizeof(u8));
1415 if (r < 0)
1416 return 0;
1417 return 1;
1418 }
1419
1420 static void seg_setup(int seg)
1421 {
1422 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1423
1424 vmcs_write16(sf->selector, 0);
1425 vmcs_writel(sf->base, 0);
1426 vmcs_write32(sf->limit, 0xffff);
1427 vmcs_write32(sf->ar_bytes, 0x93);
1428 }
1429
1430 /*
1431 * Sets up the vmcs for emulated real mode.
1432 */
1433 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
1434 {
1435 u32 host_sysenter_cs;
1436 u32 junk;
1437 unsigned long a;
1438 struct descriptor_table dt;
1439 int i;
1440 unsigned long kvm_vmx_return;
1441 u32 exec_control;
1442
1443 /* I/O */
1444 vmcs_write64(IO_BITMAP_A, page_to_phys(vmx_io_bitmap_a));
1445 vmcs_write64(IO_BITMAP_B, page_to_phys(vmx_io_bitmap_b));
1446
1447 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
1448
1449 /* Control */
1450 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
1451 vmcs_config.pin_based_exec_ctrl);
1452
1453 exec_control = vmcs_config.cpu_based_exec_ctrl;
1454 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
1455 exec_control &= ~CPU_BASED_TPR_SHADOW;
1456 #ifdef CONFIG_X86_64
1457 exec_control |= CPU_BASED_CR8_STORE_EXITING |
1458 CPU_BASED_CR8_LOAD_EXITING;
1459 #endif
1460 }
1461 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
1462
1463 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
1464 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
1465 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
1466
1467 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
1468 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
1469 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
1470
1471 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
1472 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1473 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1474 vmcs_write16(HOST_FS_SELECTOR, read_fs()); /* 22.2.4 */
1475 vmcs_write16(HOST_GS_SELECTOR, read_gs()); /* 22.2.4 */
1476 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1477 #ifdef CONFIG_X86_64
1478 rdmsrl(MSR_FS_BASE, a);
1479 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
1480 rdmsrl(MSR_GS_BASE, a);
1481 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
1482 #else
1483 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
1484 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
1485 #endif
1486
1487 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
1488
1489 get_idt(&dt);
1490 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
1491
1492 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
1493 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
1494 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
1495 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
1496 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
1497
1498 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
1499 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
1500 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
1501 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
1502 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
1503 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
1504
1505 for (i = 0; i < NR_VMX_MSR; ++i) {
1506 u32 index = vmx_msr_index[i];
1507 u32 data_low, data_high;
1508 u64 data;
1509 int j = vmx->nmsrs;
1510
1511 if (rdmsr_safe(index, &data_low, &data_high) < 0)
1512 continue;
1513 if (wrmsr_safe(index, data_low, data_high) < 0)
1514 continue;
1515 data = data_low | ((u64)data_high << 32);
1516 vmx->host_msrs[j].index = index;
1517 vmx->host_msrs[j].reserved = 0;
1518 vmx->host_msrs[j].data = data;
1519 vmx->guest_msrs[j] = vmx->host_msrs[j];
1520 ++vmx->nmsrs;
1521 }
1522
1523 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
1524
1525 /* 22.2.1, 20.8.1 */
1526 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
1527
1528 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
1529 vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
1530
1531 return 0;
1532 }
1533
1534 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
1535 {
1536 struct vcpu_vmx *vmx = to_vmx(vcpu);
1537 u64 msr;
1538 int ret;
1539
1540 if (!init_rmode_tss(vmx->vcpu.kvm)) {
1541 ret = -ENOMEM;
1542 goto out;
1543 }
1544
1545 vmx->vcpu.rmode.active = 0;
1546
1547 vmx->vcpu.regs[VCPU_REGS_RDX] = get_rdx_init_val();
1548 set_cr8(&vmx->vcpu, 0);
1549 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
1550 if (vmx->vcpu.vcpu_id == 0)
1551 msr |= MSR_IA32_APICBASE_BSP;
1552 kvm_set_apic_base(&vmx->vcpu, msr);
1553
1554 fx_init(&vmx->vcpu);
1555
1556 /*
1557 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
1558 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
1559 */
1560 if (vmx->vcpu.vcpu_id == 0) {
1561 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
1562 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
1563 } else {
1564 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.sipi_vector << 8);
1565 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.sipi_vector << 12);
1566 }
1567 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1568 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1569
1570 seg_setup(VCPU_SREG_DS);
1571 seg_setup(VCPU_SREG_ES);
1572 seg_setup(VCPU_SREG_FS);
1573 seg_setup(VCPU_SREG_GS);
1574 seg_setup(VCPU_SREG_SS);
1575
1576 vmcs_write16(GUEST_TR_SELECTOR, 0);
1577 vmcs_writel(GUEST_TR_BASE, 0);
1578 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
1579 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1580
1581 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
1582 vmcs_writel(GUEST_LDTR_BASE, 0);
1583 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
1584 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
1585
1586 vmcs_write32(GUEST_SYSENTER_CS, 0);
1587 vmcs_writel(GUEST_SYSENTER_ESP, 0);
1588 vmcs_writel(GUEST_SYSENTER_EIP, 0);
1589
1590 vmcs_writel(GUEST_RFLAGS, 0x02);
1591 if (vmx->vcpu.vcpu_id == 0)
1592 vmcs_writel(GUEST_RIP, 0xfff0);
1593 else
1594 vmcs_writel(GUEST_RIP, 0);
1595 vmcs_writel(GUEST_RSP, 0);
1596
1597 /* todo: dr0 = dr1 = dr2 = dr3 = 0; dr6 = 0xffff0ff0 */
1598 vmcs_writel(GUEST_DR7, 0x400);
1599
1600 vmcs_writel(GUEST_GDTR_BASE, 0);
1601 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
1602
1603 vmcs_writel(GUEST_IDTR_BASE, 0);
1604 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
1605
1606 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
1607 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
1608 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
1609
1610 guest_write_tsc(0);
1611
1612 /* Special registers */
1613 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
1614
1615 setup_msrs(vmx);
1616
1617 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
1618
1619 #ifdef CONFIG_X86_64
1620 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
1621 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
1622 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
1623 page_to_phys(vmx->vcpu.apic->regs_page));
1624 vmcs_write32(TPR_THRESHOLD, 0);
1625 #endif
1626
1627 vmx->vcpu.cr0 = 0x60000010;
1628 vmx_set_cr0(&vmx->vcpu, vmx->vcpu.cr0); /* enter rmode */
1629 vmx_set_cr4(&vmx->vcpu, 0);
1630 #ifdef CONFIG_X86_64
1631 vmx_set_efer(&vmx->vcpu, 0);
1632 #endif
1633 vmx_fpu_activate(&vmx->vcpu);
1634 update_exception_bitmap(&vmx->vcpu);
1635
1636 return 0;
1637
1638 out:
1639 return ret;
1640 }
1641
1642 static void inject_rmode_irq(struct kvm_vcpu *vcpu, int irq)
1643 {
1644 u16 ent[2];
1645 u16 cs;
1646 u16 ip;
1647 unsigned long flags;
1648 unsigned long ss_base = vmcs_readl(GUEST_SS_BASE);
1649 u16 sp = vmcs_readl(GUEST_RSP);
1650 u32 ss_limit = vmcs_read32(GUEST_SS_LIMIT);
1651
1652 if (sp > ss_limit || sp < 6) {
1653 vcpu_printf(vcpu, "%s: #SS, rsp 0x%lx ss 0x%lx limit 0x%x\n",
1654 __FUNCTION__,
1655 vmcs_readl(GUEST_RSP),
1656 vmcs_readl(GUEST_SS_BASE),
1657 vmcs_read32(GUEST_SS_LIMIT));
1658 return;
1659 }
1660
1661 if (emulator_read_std(irq * sizeof(ent), &ent, sizeof(ent), vcpu) !=
1662 X86EMUL_CONTINUE) {
1663 vcpu_printf(vcpu, "%s: read guest err\n", __FUNCTION__);
1664 return;
1665 }
1666
1667 flags = vmcs_readl(GUEST_RFLAGS);
1668 cs = vmcs_readl(GUEST_CS_BASE) >> 4;
1669 ip = vmcs_readl(GUEST_RIP);
1670
1671
1672 if (emulator_write_emulated(
1673 ss_base + sp - 2, &flags, 2, vcpu) != X86EMUL_CONTINUE ||
1674 emulator_write_emulated(
1675 ss_base + sp - 4, &cs, 2, vcpu) != X86EMUL_CONTINUE ||
1676 emulator_write_emulated(
1677 ss_base + sp - 6, &ip, 2, vcpu) != X86EMUL_CONTINUE) {
1678 vcpu_printf(vcpu, "%s: write guest err\n", __FUNCTION__);
1679 return;
1680 }
1681
1682 vmcs_writel(GUEST_RFLAGS, flags &
1683 ~(X86_EFLAGS_IF | X86_EFLAGS_AC | X86_EFLAGS_TF));
1684 vmcs_write16(GUEST_CS_SELECTOR, ent[1]) ;
1685 vmcs_writel(GUEST_CS_BASE, ent[1] << 4);
1686 vmcs_writel(GUEST_RIP, ent[0]);
1687 vmcs_writel(GUEST_RSP, (vmcs_readl(GUEST_RSP) & ~0xffff) | (sp - 6));
1688 }
1689
1690 static void vmx_inject_irq(struct kvm_vcpu *vcpu, int irq)
1691 {
1692 if (vcpu->rmode.active) {
1693 inject_rmode_irq(vcpu, irq);
1694 return;
1695 }
1696 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1697 irq | INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1698 }
1699
1700 static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1701 {
1702 int word_index = __ffs(vcpu->irq_summary);
1703 int bit_index = __ffs(vcpu->irq_pending[word_index]);
1704 int irq = word_index * BITS_PER_LONG + bit_index;
1705
1706 clear_bit(bit_index, &vcpu->irq_pending[word_index]);
1707 if (!vcpu->irq_pending[word_index])
1708 clear_bit(word_index, &vcpu->irq_summary);
1709 vmx_inject_irq(vcpu, irq);
1710 }
1711
1712
1713 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1714 struct kvm_run *kvm_run)
1715 {
1716 u32 cpu_based_vm_exec_control;
1717
1718 vcpu->interrupt_window_open =
1719 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
1720 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
1721
1722 if (vcpu->interrupt_window_open &&
1723 vcpu->irq_summary &&
1724 !(vmcs_read32(VM_ENTRY_INTR_INFO_FIELD) & INTR_INFO_VALID_MASK))
1725 /*
1726 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1727 */
1728 kvm_do_inject_irq(vcpu);
1729
1730 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
1731 if (!vcpu->interrupt_window_open &&
1732 (vcpu->irq_summary || kvm_run->request_interrupt_window))
1733 /*
1734 * Interrupts blocked. Wait for unblock.
1735 */
1736 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
1737 else
1738 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
1739 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
1740 }
1741
1742 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
1743 {
1744 int ret;
1745 struct kvm_userspace_memory_region tss_mem = {
1746 .slot = 8,
1747 .guest_phys_addr = addr,
1748 .memory_size = PAGE_SIZE * 3,
1749 .flags = 0,
1750 };
1751
1752 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
1753 if (ret)
1754 return ret;
1755 kvm->tss_addr = addr;
1756 return 0;
1757 }
1758
1759 static void kvm_guest_debug_pre(struct kvm_vcpu *vcpu)
1760 {
1761 struct kvm_guest_debug *dbg = &vcpu->guest_debug;
1762
1763 set_debugreg(dbg->bp[0], 0);
1764 set_debugreg(dbg->bp[1], 1);
1765 set_debugreg(dbg->bp[2], 2);
1766 set_debugreg(dbg->bp[3], 3);
1767
1768 if (dbg->singlestep) {
1769 unsigned long flags;
1770
1771 flags = vmcs_readl(GUEST_RFLAGS);
1772 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1773 vmcs_writel(GUEST_RFLAGS, flags);
1774 }
1775 }
1776
1777 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
1778 int vec, u32 err_code)
1779 {
1780 if (!vcpu->rmode.active)
1781 return 0;
1782
1783 /*
1784 * Instruction with address size override prefix opcode 0x67
1785 * Cause the #SS fault with 0 error code in VM86 mode.
1786 */
1787 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
1788 if (emulate_instruction(vcpu, NULL, 0, 0, 0) == EMULATE_DONE)
1789 return 1;
1790 return 0;
1791 }
1792
1793 static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1794 {
1795 u32 intr_info, error_code;
1796 unsigned long cr2, rip;
1797 u32 vect_info;
1798 enum emulation_result er;
1799 int r;
1800
1801 vect_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
1802 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
1803
1804 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
1805 !is_page_fault(intr_info))
1806 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
1807 "intr info 0x%x\n", __FUNCTION__, vect_info, intr_info);
1808
1809 if (!irqchip_in_kernel(vcpu->kvm) && is_external_interrupt(vect_info)) {
1810 int irq = vect_info & VECTORING_INFO_VECTOR_MASK;
1811 set_bit(irq, vcpu->irq_pending);
1812 set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
1813 }
1814
1815 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) /* nmi */
1816 return 1; /* already handled by vmx_vcpu_run() */
1817
1818 if (is_no_device(intr_info)) {
1819 vmx_fpu_activate(vcpu);
1820 return 1;
1821 }
1822
1823 if (is_invalid_opcode(intr_info)) {
1824 er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
1825 if (er != EMULATE_DONE)
1826 vmx_inject_ud(vcpu);
1827
1828 return 1;
1829 }
1830
1831 error_code = 0;
1832 rip = vmcs_readl(GUEST_RIP);
1833 if (intr_info & INTR_INFO_DELIEVER_CODE_MASK)
1834 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
1835 if (is_page_fault(intr_info)) {
1836 cr2 = vmcs_readl(EXIT_QUALIFICATION);
1837
1838 mutex_lock(&vcpu->kvm->lock);
1839 r = kvm_mmu_page_fault(vcpu, cr2, error_code);
1840 if (r < 0) {
1841 mutex_unlock(&vcpu->kvm->lock);
1842 return r;
1843 }
1844 if (!r) {
1845 mutex_unlock(&vcpu->kvm->lock);
1846 return 1;
1847 }
1848
1849 er = emulate_instruction(vcpu, kvm_run, cr2, error_code, 0);
1850 mutex_unlock(&vcpu->kvm->lock);
1851
1852 switch (er) {
1853 case EMULATE_DONE:
1854 return 1;
1855 case EMULATE_DO_MMIO:
1856 ++vcpu->stat.mmio_exits;
1857 return 0;
1858 case EMULATE_FAIL:
1859 kvm_report_emulation_failure(vcpu, "pagetable");
1860 break;
1861 default:
1862 BUG();
1863 }
1864 }
1865
1866 if (vcpu->rmode.active &&
1867 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
1868 error_code)) {
1869 if (vcpu->halt_request) {
1870 vcpu->halt_request = 0;
1871 return kvm_emulate_halt(vcpu);
1872 }
1873 return 1;
1874 }
1875
1876 if ((intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK)) ==
1877 (INTR_TYPE_EXCEPTION | 1)) {
1878 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1879 return 0;
1880 }
1881 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
1882 kvm_run->ex.exception = intr_info & INTR_INFO_VECTOR_MASK;
1883 kvm_run->ex.error_code = error_code;
1884 return 0;
1885 }
1886
1887 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
1888 struct kvm_run *kvm_run)
1889 {
1890 ++vcpu->stat.irq_exits;
1891 return 1;
1892 }
1893
1894 static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1895 {
1896 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1897 return 0;
1898 }
1899
1900 static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1901 {
1902 unsigned long exit_qualification;
1903 int size, down, in, string, rep;
1904 unsigned port;
1905
1906 ++vcpu->stat.io_exits;
1907 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
1908 string = (exit_qualification & 16) != 0;
1909
1910 if (string) {
1911 if (emulate_instruction(vcpu,
1912 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
1913 return 0;
1914 return 1;
1915 }
1916
1917 size = (exit_qualification & 7) + 1;
1918 in = (exit_qualification & 8) != 0;
1919 down = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_DF) != 0;
1920 rep = (exit_qualification & 32) != 0;
1921 port = exit_qualification >> 16;
1922
1923 return kvm_emulate_pio(vcpu, kvm_run, in, size, port);
1924 }
1925
1926 static void
1927 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1928 {
1929 /*
1930 * Patch in the VMCALL instruction:
1931 */
1932 hypercall[0] = 0x0f;
1933 hypercall[1] = 0x01;
1934 hypercall[2] = 0xc1;
1935 }
1936
1937 static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1938 {
1939 unsigned long exit_qualification;
1940 int cr;
1941 int reg;
1942
1943 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
1944 cr = exit_qualification & 15;
1945 reg = (exit_qualification >> 8) & 15;
1946 switch ((exit_qualification >> 4) & 3) {
1947 case 0: /* mov to cr */
1948 switch (cr) {
1949 case 0:
1950 vcpu_load_rsp_rip(vcpu);
1951 set_cr0(vcpu, vcpu->regs[reg]);
1952 skip_emulated_instruction(vcpu);
1953 return 1;
1954 case 3:
1955 vcpu_load_rsp_rip(vcpu);
1956 set_cr3(vcpu, vcpu->regs[reg]);
1957 skip_emulated_instruction(vcpu);
1958 return 1;
1959 case 4:
1960 vcpu_load_rsp_rip(vcpu);
1961 set_cr4(vcpu, vcpu->regs[reg]);
1962 skip_emulated_instruction(vcpu);
1963 return 1;
1964 case 8:
1965 vcpu_load_rsp_rip(vcpu);
1966 set_cr8(vcpu, vcpu->regs[reg]);
1967 skip_emulated_instruction(vcpu);
1968 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
1969 return 0;
1970 };
1971 break;
1972 case 2: /* clts */
1973 vcpu_load_rsp_rip(vcpu);
1974 vmx_fpu_deactivate(vcpu);
1975 vcpu->cr0 &= ~X86_CR0_TS;
1976 vmcs_writel(CR0_READ_SHADOW, vcpu->cr0);
1977 vmx_fpu_activate(vcpu);
1978 skip_emulated_instruction(vcpu);
1979 return 1;
1980 case 1: /*mov from cr*/
1981 switch (cr) {
1982 case 3:
1983 vcpu_load_rsp_rip(vcpu);
1984 vcpu->regs[reg] = vcpu->cr3;
1985 vcpu_put_rsp_rip(vcpu);
1986 skip_emulated_instruction(vcpu);
1987 return 1;
1988 case 8:
1989 vcpu_load_rsp_rip(vcpu);
1990 vcpu->regs[reg] = get_cr8(vcpu);
1991 vcpu_put_rsp_rip(vcpu);
1992 skip_emulated_instruction(vcpu);
1993 return 1;
1994 }
1995 break;
1996 case 3: /* lmsw */
1997 lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
1998
1999 skip_emulated_instruction(vcpu);
2000 return 1;
2001 default:
2002 break;
2003 }
2004 kvm_run->exit_reason = 0;
2005 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
2006 (int)(exit_qualification >> 4) & 3, cr);
2007 return 0;
2008 }
2009
2010 static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2011 {
2012 unsigned long exit_qualification;
2013 unsigned long val;
2014 int dr, reg;
2015
2016 /*
2017 * FIXME: this code assumes the host is debugging the guest.
2018 * need to deal with guest debugging itself too.
2019 */
2020 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2021 dr = exit_qualification & 7;
2022 reg = (exit_qualification >> 8) & 15;
2023 vcpu_load_rsp_rip(vcpu);
2024 if (exit_qualification & 16) {
2025 /* mov from dr */
2026 switch (dr) {
2027 case 6:
2028 val = 0xffff0ff0;
2029 break;
2030 case 7:
2031 val = 0x400;
2032 break;
2033 default:
2034 val = 0;
2035 }
2036 vcpu->regs[reg] = val;
2037 } else {
2038 /* mov to dr */
2039 }
2040 vcpu_put_rsp_rip(vcpu);
2041 skip_emulated_instruction(vcpu);
2042 return 1;
2043 }
2044
2045 static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2046 {
2047 kvm_emulate_cpuid(vcpu);
2048 return 1;
2049 }
2050
2051 static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2052 {
2053 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
2054 u64 data;
2055
2056 if (vmx_get_msr(vcpu, ecx, &data)) {
2057 vmx_inject_gp(vcpu, 0);
2058 return 1;
2059 }
2060
2061 /* FIXME: handling of bits 32:63 of rax, rdx */
2062 vcpu->regs[VCPU_REGS_RAX] = data & -1u;
2063 vcpu->regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
2064 skip_emulated_instruction(vcpu);
2065 return 1;
2066 }
2067
2068 static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2069 {
2070 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
2071 u64 data = (vcpu->regs[VCPU_REGS_RAX] & -1u)
2072 | ((u64)(vcpu->regs[VCPU_REGS_RDX] & -1u) << 32);
2073
2074 if (vmx_set_msr(vcpu, ecx, data) != 0) {
2075 vmx_inject_gp(vcpu, 0);
2076 return 1;
2077 }
2078
2079 skip_emulated_instruction(vcpu);
2080 return 1;
2081 }
2082
2083 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu,
2084 struct kvm_run *kvm_run)
2085 {
2086 return 1;
2087 }
2088
2089 static int handle_interrupt_window(struct kvm_vcpu *vcpu,
2090 struct kvm_run *kvm_run)
2091 {
2092 u32 cpu_based_vm_exec_control;
2093
2094 /* clear pending irq */
2095 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2096 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2097 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2098 /*
2099 * If the user space waits to inject interrupts, exit as soon as
2100 * possible
2101 */
2102 if (kvm_run->request_interrupt_window &&
2103 !vcpu->irq_summary) {
2104 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2105 ++vcpu->stat.irq_window_exits;
2106 return 0;
2107 }
2108 return 1;
2109 }
2110
2111 static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2112 {
2113 skip_emulated_instruction(vcpu);
2114 return kvm_emulate_halt(vcpu);
2115 }
2116
2117 static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2118 {
2119 skip_emulated_instruction(vcpu);
2120 kvm_emulate_hypercall(vcpu);
2121 return 1;
2122 }
2123
2124 /*
2125 * The exit handlers return 1 if the exit was handled fully and guest execution
2126 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
2127 * to be done to userspace and return 0.
2128 */
2129 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
2130 struct kvm_run *kvm_run) = {
2131 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
2132 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
2133 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
2134 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
2135 [EXIT_REASON_CR_ACCESS] = handle_cr,
2136 [EXIT_REASON_DR_ACCESS] = handle_dr,
2137 [EXIT_REASON_CPUID] = handle_cpuid,
2138 [EXIT_REASON_MSR_READ] = handle_rdmsr,
2139 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
2140 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
2141 [EXIT_REASON_HLT] = handle_halt,
2142 [EXIT_REASON_VMCALL] = handle_vmcall,
2143 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold
2144 };
2145
2146 static const int kvm_vmx_max_exit_handlers =
2147 ARRAY_SIZE(kvm_vmx_exit_handlers);
2148
2149 /*
2150 * The guest has exited. See if we can fix it or if we need userspace
2151 * assistance.
2152 */
2153 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2154 {
2155 u32 vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
2156 u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
2157 struct vcpu_vmx *vmx = to_vmx(vcpu);
2158
2159 if (unlikely(vmx->fail)) {
2160 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2161 kvm_run->fail_entry.hardware_entry_failure_reason
2162 = vmcs_read32(VM_INSTRUCTION_ERROR);
2163 return 0;
2164 }
2165
2166 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
2167 exit_reason != EXIT_REASON_EXCEPTION_NMI)
2168 printk(KERN_WARNING "%s: unexpected, valid vectoring info and "
2169 "exit reason is 0x%x\n", __FUNCTION__, exit_reason);
2170 if (exit_reason < kvm_vmx_max_exit_handlers
2171 && kvm_vmx_exit_handlers[exit_reason])
2172 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
2173 else {
2174 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2175 kvm_run->hw.hardware_exit_reason = exit_reason;
2176 }
2177 return 0;
2178 }
2179
2180 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
2181 {
2182 }
2183
2184 static void update_tpr_threshold(struct kvm_vcpu *vcpu)
2185 {
2186 int max_irr, tpr;
2187
2188 if (!vm_need_tpr_shadow(vcpu->kvm))
2189 return;
2190
2191 if (!kvm_lapic_enabled(vcpu) ||
2192 ((max_irr = kvm_lapic_find_highest_irr(vcpu)) == -1)) {
2193 vmcs_write32(TPR_THRESHOLD, 0);
2194 return;
2195 }
2196
2197 tpr = (kvm_lapic_get_cr8(vcpu) & 0x0f) << 4;
2198 vmcs_write32(TPR_THRESHOLD, (max_irr > tpr) ? tpr >> 4 : max_irr >> 4);
2199 }
2200
2201 static void enable_irq_window(struct kvm_vcpu *vcpu)
2202 {
2203 u32 cpu_based_vm_exec_control;
2204
2205 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2206 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2207 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2208 }
2209
2210 static void vmx_intr_assist(struct kvm_vcpu *vcpu)
2211 {
2212 u32 idtv_info_field, intr_info_field;
2213 int has_ext_irq, interrupt_window_open;
2214 int vector;
2215
2216 update_tpr_threshold(vcpu);
2217
2218 has_ext_irq = kvm_cpu_has_interrupt(vcpu);
2219 intr_info_field = vmcs_read32(VM_ENTRY_INTR_INFO_FIELD);
2220 idtv_info_field = vmcs_read32(IDT_VECTORING_INFO_FIELD);
2221 if (intr_info_field & INTR_INFO_VALID_MASK) {
2222 if (idtv_info_field & INTR_INFO_VALID_MASK) {
2223 /* TODO: fault when IDT_Vectoring */
2224 printk(KERN_ERR "Fault when IDT_Vectoring\n");
2225 }
2226 if (has_ext_irq)
2227 enable_irq_window(vcpu);
2228 return;
2229 }
2230 if (unlikely(idtv_info_field & INTR_INFO_VALID_MASK)) {
2231 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, idtv_info_field);
2232 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2233 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
2234
2235 if (unlikely(idtv_info_field & INTR_INFO_DELIEVER_CODE_MASK))
2236 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2237 vmcs_read32(IDT_VECTORING_ERROR_CODE));
2238 if (unlikely(has_ext_irq))
2239 enable_irq_window(vcpu);
2240 return;
2241 }
2242 if (!has_ext_irq)
2243 return;
2244 interrupt_window_open =
2245 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2246 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
2247 if (interrupt_window_open) {
2248 vector = kvm_cpu_get_interrupt(vcpu);
2249 vmx_inject_irq(vcpu, vector);
2250 kvm_timer_intr_post(vcpu, vector);
2251 } else
2252 enable_irq_window(vcpu);
2253 }
2254
2255 static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2256 {
2257 struct vcpu_vmx *vmx = to_vmx(vcpu);
2258 u32 intr_info;
2259
2260 /*
2261 * Loading guest fpu may have cleared host cr0.ts
2262 */
2263 vmcs_writel(HOST_CR0, read_cr0());
2264
2265 asm(
2266 /* Store host registers */
2267 #ifdef CONFIG_X86_64
2268 "push %%rax; push %%rbx; push %%rdx;"
2269 "push %%rsi; push %%rdi; push %%rbp;"
2270 "push %%r8; push %%r9; push %%r10; push %%r11;"
2271 "push %%r12; push %%r13; push %%r14; push %%r15;"
2272 "push %%rcx \n\t"
2273 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
2274 #else
2275 "pusha; push %%ecx \n\t"
2276 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
2277 #endif
2278 /* Check if vmlaunch of vmresume is needed */
2279 "cmp $0, %1 \n\t"
2280 /* Load guest registers. Don't clobber flags. */
2281 #ifdef CONFIG_X86_64
2282 "mov %c[cr2](%3), %%rax \n\t"
2283 "mov %%rax, %%cr2 \n\t"
2284 "mov %c[rax](%3), %%rax \n\t"
2285 "mov %c[rbx](%3), %%rbx \n\t"
2286 "mov %c[rdx](%3), %%rdx \n\t"
2287 "mov %c[rsi](%3), %%rsi \n\t"
2288 "mov %c[rdi](%3), %%rdi \n\t"
2289 "mov %c[rbp](%3), %%rbp \n\t"
2290 "mov %c[r8](%3), %%r8 \n\t"
2291 "mov %c[r9](%3), %%r9 \n\t"
2292 "mov %c[r10](%3), %%r10 \n\t"
2293 "mov %c[r11](%3), %%r11 \n\t"
2294 "mov %c[r12](%3), %%r12 \n\t"
2295 "mov %c[r13](%3), %%r13 \n\t"
2296 "mov %c[r14](%3), %%r14 \n\t"
2297 "mov %c[r15](%3), %%r15 \n\t"
2298 "mov %c[rcx](%3), %%rcx \n\t" /* kills %3 (rcx) */
2299 #else
2300 "mov %c[cr2](%3), %%eax \n\t"
2301 "mov %%eax, %%cr2 \n\t"
2302 "mov %c[rax](%3), %%eax \n\t"
2303 "mov %c[rbx](%3), %%ebx \n\t"
2304 "mov %c[rdx](%3), %%edx \n\t"
2305 "mov %c[rsi](%3), %%esi \n\t"
2306 "mov %c[rdi](%3), %%edi \n\t"
2307 "mov %c[rbp](%3), %%ebp \n\t"
2308 "mov %c[rcx](%3), %%ecx \n\t" /* kills %3 (ecx) */
2309 #endif
2310 /* Enter guest mode */
2311 "jne .Llaunched \n\t"
2312 ASM_VMX_VMLAUNCH "\n\t"
2313 "jmp .Lkvm_vmx_return \n\t"
2314 ".Llaunched: " ASM_VMX_VMRESUME "\n\t"
2315 ".Lkvm_vmx_return: "
2316 /* Save guest registers, load host registers, keep flags */
2317 #ifdef CONFIG_X86_64
2318 "xchg %3, (%%rsp) \n\t"
2319 "mov %%rax, %c[rax](%3) \n\t"
2320 "mov %%rbx, %c[rbx](%3) \n\t"
2321 "pushq (%%rsp); popq %c[rcx](%3) \n\t"
2322 "mov %%rdx, %c[rdx](%3) \n\t"
2323 "mov %%rsi, %c[rsi](%3) \n\t"
2324 "mov %%rdi, %c[rdi](%3) \n\t"
2325 "mov %%rbp, %c[rbp](%3) \n\t"
2326 "mov %%r8, %c[r8](%3) \n\t"
2327 "mov %%r9, %c[r9](%3) \n\t"
2328 "mov %%r10, %c[r10](%3) \n\t"
2329 "mov %%r11, %c[r11](%3) \n\t"
2330 "mov %%r12, %c[r12](%3) \n\t"
2331 "mov %%r13, %c[r13](%3) \n\t"
2332 "mov %%r14, %c[r14](%3) \n\t"
2333 "mov %%r15, %c[r15](%3) \n\t"
2334 "mov %%cr2, %%rax \n\t"
2335 "mov %%rax, %c[cr2](%3) \n\t"
2336 "mov (%%rsp), %3 \n\t"
2337
2338 "pop %%rcx; pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
2339 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
2340 "pop %%rbp; pop %%rdi; pop %%rsi;"
2341 "pop %%rdx; pop %%rbx; pop %%rax \n\t"
2342 #else
2343 "xchg %3, (%%esp) \n\t"
2344 "mov %%eax, %c[rax](%3) \n\t"
2345 "mov %%ebx, %c[rbx](%3) \n\t"
2346 "pushl (%%esp); popl %c[rcx](%3) \n\t"
2347 "mov %%edx, %c[rdx](%3) \n\t"
2348 "mov %%esi, %c[rsi](%3) \n\t"
2349 "mov %%edi, %c[rdi](%3) \n\t"
2350 "mov %%ebp, %c[rbp](%3) \n\t"
2351 "mov %%cr2, %%eax \n\t"
2352 "mov %%eax, %c[cr2](%3) \n\t"
2353 "mov (%%esp), %3 \n\t"
2354
2355 "pop %%ecx; popa \n\t"
2356 #endif
2357 "setbe %0 \n\t"
2358 : "=q" (vmx->fail)
2359 : "r"(vmx->launched), "d"((unsigned long)HOST_RSP),
2360 "c"(vcpu),
2361 [rax]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RAX])),
2362 [rbx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBX])),
2363 [rcx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RCX])),
2364 [rdx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDX])),
2365 [rsi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RSI])),
2366 [rdi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDI])),
2367 [rbp]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBP])),
2368 #ifdef CONFIG_X86_64
2369 [r8]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R8])),
2370 [r9]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R9])),
2371 [r10]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R10])),
2372 [r11]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R11])),
2373 [r12]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R12])),
2374 [r13]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R13])),
2375 [r14]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R14])),
2376 [r15]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R15])),
2377 #endif
2378 [cr2]"i"(offsetof(struct kvm_vcpu, cr2))
2379 : "cc", "memory");
2380
2381 vcpu->interrupt_window_open =
2382 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0;
2383
2384 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
2385 vmx->launched = 1;
2386
2387 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2388
2389 /* We need to handle NMIs before interrupts are enabled */
2390 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) /* nmi */
2391 asm("int $2");
2392 }
2393
2394 static void vmx_inject_page_fault(struct kvm_vcpu *vcpu,
2395 unsigned long addr,
2396 u32 err_code)
2397 {
2398 u32 vect_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
2399
2400 ++vcpu->stat.pf_guest;
2401
2402 if (is_page_fault(vect_info)) {
2403 printk(KERN_DEBUG "inject_page_fault: "
2404 "double fault 0x%lx @ 0x%lx\n",
2405 addr, vmcs_readl(GUEST_RIP));
2406 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, 0);
2407 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2408 DF_VECTOR |
2409 INTR_TYPE_EXCEPTION |
2410 INTR_INFO_DELIEVER_CODE_MASK |
2411 INTR_INFO_VALID_MASK);
2412 return;
2413 }
2414 vcpu->cr2 = addr;
2415 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, err_code);
2416 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2417 PF_VECTOR |
2418 INTR_TYPE_EXCEPTION |
2419 INTR_INFO_DELIEVER_CODE_MASK |
2420 INTR_INFO_VALID_MASK);
2421
2422 }
2423
2424 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
2425 {
2426 struct vcpu_vmx *vmx = to_vmx(vcpu);
2427
2428 if (vmx->vmcs) {
2429 on_each_cpu(__vcpu_clear, vmx, 0, 1);
2430 free_vmcs(vmx->vmcs);
2431 vmx->vmcs = NULL;
2432 }
2433 }
2434
2435 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
2436 {
2437 struct vcpu_vmx *vmx = to_vmx(vcpu);
2438
2439 vmx_free_vmcs(vcpu);
2440 kfree(vmx->host_msrs);
2441 kfree(vmx->guest_msrs);
2442 kvm_vcpu_uninit(vcpu);
2443 kmem_cache_free(kvm_vcpu_cache, vmx);
2444 }
2445
2446 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
2447 {
2448 int err;
2449 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2450 int cpu;
2451
2452 if (!vmx)
2453 return ERR_PTR(-ENOMEM);
2454
2455 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
2456 if (err)
2457 goto free_vcpu;
2458
2459 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2460 if (!vmx->guest_msrs) {
2461 err = -ENOMEM;
2462 goto uninit_vcpu;
2463 }
2464
2465 vmx->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2466 if (!vmx->host_msrs)
2467 goto free_guest_msrs;
2468
2469 vmx->vmcs = alloc_vmcs();
2470 if (!vmx->vmcs)
2471 goto free_msrs;
2472
2473 vmcs_clear(vmx->vmcs);
2474
2475 cpu = get_cpu();
2476 vmx_vcpu_load(&vmx->vcpu, cpu);
2477 err = vmx_vcpu_setup(vmx);
2478 vmx_vcpu_put(&vmx->vcpu);
2479 put_cpu();
2480 if (err)
2481 goto free_vmcs;
2482
2483 return &vmx->vcpu;
2484
2485 free_vmcs:
2486 free_vmcs(vmx->vmcs);
2487 free_msrs:
2488 kfree(vmx->host_msrs);
2489 free_guest_msrs:
2490 kfree(vmx->guest_msrs);
2491 uninit_vcpu:
2492 kvm_vcpu_uninit(&vmx->vcpu);
2493 free_vcpu:
2494 kmem_cache_free(kvm_vcpu_cache, vmx);
2495 return ERR_PTR(err);
2496 }
2497
2498 static void __init vmx_check_processor_compat(void *rtn)
2499 {
2500 struct vmcs_config vmcs_conf;
2501
2502 *(int *)rtn = 0;
2503 if (setup_vmcs_config(&vmcs_conf) < 0)
2504 *(int *)rtn = -EIO;
2505 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
2506 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
2507 smp_processor_id());
2508 *(int *)rtn = -EIO;
2509 }
2510 }
2511
2512 static struct kvm_x86_ops vmx_x86_ops = {
2513 .cpu_has_kvm_support = cpu_has_kvm_support,
2514 .disabled_by_bios = vmx_disabled_by_bios,
2515 .hardware_setup = hardware_setup,
2516 .hardware_unsetup = hardware_unsetup,
2517 .check_processor_compatibility = vmx_check_processor_compat,
2518 .hardware_enable = hardware_enable,
2519 .hardware_disable = hardware_disable,
2520
2521 .vcpu_create = vmx_create_vcpu,
2522 .vcpu_free = vmx_free_vcpu,
2523 .vcpu_reset = vmx_vcpu_reset,
2524
2525 .prepare_guest_switch = vmx_save_host_state,
2526 .vcpu_load = vmx_vcpu_load,
2527 .vcpu_put = vmx_vcpu_put,
2528 .vcpu_decache = vmx_vcpu_decache,
2529
2530 .set_guest_debug = set_guest_debug,
2531 .guest_debug_pre = kvm_guest_debug_pre,
2532 .get_msr = vmx_get_msr,
2533 .set_msr = vmx_set_msr,
2534 .get_segment_base = vmx_get_segment_base,
2535 .get_segment = vmx_get_segment,
2536 .set_segment = vmx_set_segment,
2537 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
2538 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
2539 .set_cr0 = vmx_set_cr0,
2540 .set_cr3 = vmx_set_cr3,
2541 .set_cr4 = vmx_set_cr4,
2542 #ifdef CONFIG_X86_64
2543 .set_efer = vmx_set_efer,
2544 #endif
2545 .get_idt = vmx_get_idt,
2546 .set_idt = vmx_set_idt,
2547 .get_gdt = vmx_get_gdt,
2548 .set_gdt = vmx_set_gdt,
2549 .cache_regs = vcpu_load_rsp_rip,
2550 .decache_regs = vcpu_put_rsp_rip,
2551 .get_rflags = vmx_get_rflags,
2552 .set_rflags = vmx_set_rflags,
2553
2554 .tlb_flush = vmx_flush_tlb,
2555 .inject_page_fault = vmx_inject_page_fault,
2556
2557 .inject_gp = vmx_inject_gp,
2558
2559 .run = vmx_vcpu_run,
2560 .handle_exit = kvm_handle_exit,
2561 .skip_emulated_instruction = skip_emulated_instruction,
2562 .patch_hypercall = vmx_patch_hypercall,
2563 .get_irq = vmx_get_irq,
2564 .set_irq = vmx_inject_irq,
2565 .inject_pending_irq = vmx_intr_assist,
2566 .inject_pending_vectors = do_interrupt_requests,
2567
2568 .set_tss_addr = vmx_set_tss_addr,
2569 };
2570
2571 static int __init vmx_init(void)
2572 {
2573 void *iova;
2574 int r;
2575
2576 vmx_io_bitmap_a = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2577 if (!vmx_io_bitmap_a)
2578 return -ENOMEM;
2579
2580 vmx_io_bitmap_b = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2581 if (!vmx_io_bitmap_b) {
2582 r = -ENOMEM;
2583 goto out;
2584 }
2585
2586 /*
2587 * Allow direct access to the PC debug port (it is often used for I/O
2588 * delays, but the vmexits simply slow things down).
2589 */
2590 iova = kmap(vmx_io_bitmap_a);
2591 memset(iova, 0xff, PAGE_SIZE);
2592 clear_bit(0x80, iova);
2593 kunmap(vmx_io_bitmap_a);
2594
2595 iova = kmap(vmx_io_bitmap_b);
2596 memset(iova, 0xff, PAGE_SIZE);
2597 kunmap(vmx_io_bitmap_b);
2598
2599 r = kvm_init_x86(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
2600 if (r)
2601 goto out1;
2602
2603 if (bypass_guest_pf)
2604 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
2605
2606 return 0;
2607
2608 out1:
2609 __free_page(vmx_io_bitmap_b);
2610 out:
2611 __free_page(vmx_io_bitmap_a);
2612 return r;
2613 }
2614
2615 static void __exit vmx_exit(void)
2616 {
2617 __free_page(vmx_io_bitmap_b);
2618 __free_page(vmx_io_bitmap_a);
2619
2620 kvm_exit_x86();
2621 }
2622
2623 module_init(vmx_init)
2624 module_exit(vmx_exit)
linux 2.6 git repository for openrd