aboutsummaryrefslogtreecommitdiffstats
path: root/arch/x86/kvm/vmx.c
diff options
context:
space:
mode:
Diffstat (limited to 'arch/x86/kvm/vmx.c')
-rw-r--r--arch/x86/kvm/vmx.c2671
1 files changed, 2671 insertions, 0 deletions
diff --git a/arch/x86/kvm/vmx.c b/arch/x86/kvm/vmx.c
new file mode 100644
index 000000000000..fc494aff5d8b
--- /dev/null
+++ b/arch/x86/kvm/vmx.c
@@ -0,0 +1,2671 @@
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 "irq.h"
19#include "vmx.h"
20#include "segment_descriptor.h"
21#include "mmu.h"
22
23#include <linux/kvm_host.h>
24#include <linux/module.h>
25#include <linux/kernel.h>
26#include <linux/mm.h>
27#include <linux/highmem.h>
28#include <linux/sched.h>
29#include <linux/moduleparam.h>
30
31#include <asm/io.h>
32#include <asm/desc.h>
33
34MODULE_AUTHOR("Qumranet");
35MODULE_LICENSE("GPL");
36
37static int bypass_guest_pf = 1;
38module_param(bypass_guest_pf, bool, 0);
39
40struct vmcs {
41 u32 revision_id;
42 u32 abort;
43 char data[0];
44};
45
46struct vcpu_vmx {
47 struct kvm_vcpu vcpu;
48 int launched;
49 u8 fail;
50 u32 idt_vectoring_info;
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 struct {
68 struct {
69 bool pending;
70 u8 vector;
71 unsigned rip;
72 } irq;
73 } rmode;
74};
75
76static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
77{
78 return container_of(vcpu, struct vcpu_vmx, vcpu);
79}
80
81static int init_rmode_tss(struct kvm *kvm);
82
83static DEFINE_PER_CPU(struct vmcs *, vmxarea);
84static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
85
86static struct page *vmx_io_bitmap_a;
87static struct page *vmx_io_bitmap_b;
88
89static struct vmcs_config {
90 int size;
91 int order;
92 u32 revision_id;
93 u32 pin_based_exec_ctrl;
94 u32 cpu_based_exec_ctrl;
95 u32 cpu_based_2nd_exec_ctrl;
96 u32 vmexit_ctrl;
97 u32 vmentry_ctrl;
98} vmcs_config;
99
100#define VMX_SEGMENT_FIELD(seg) \
101 [VCPU_SREG_##seg] = { \
102 .selector = GUEST_##seg##_SELECTOR, \
103 .base = GUEST_##seg##_BASE, \
104 .limit = GUEST_##seg##_LIMIT, \
105 .ar_bytes = GUEST_##seg##_AR_BYTES, \
106 }
107
108static struct kvm_vmx_segment_field {
109 unsigned selector;
110 unsigned base;
111 unsigned limit;
112 unsigned ar_bytes;
113} kvm_vmx_segment_fields[] = {
114 VMX_SEGMENT_FIELD(CS),
115 VMX_SEGMENT_FIELD(DS),
116 VMX_SEGMENT_FIELD(ES),
117 VMX_SEGMENT_FIELD(FS),
118 VMX_SEGMENT_FIELD(GS),
119 VMX_SEGMENT_FIELD(SS),
120 VMX_SEGMENT_FIELD(TR),
121 VMX_SEGMENT_FIELD(LDTR),
122};
123
124/*
125 * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
126 * away by decrementing the array size.
127 */
128static const u32 vmx_msr_index[] = {
129#ifdef CONFIG_X86_64
130 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
131#endif
132 MSR_EFER, MSR_K6_STAR,
133};
134#define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
135
136static void load_msrs(struct kvm_msr_entry *e, int n)
137{
138 int i;
139
140 for (i = 0; i < n; ++i)
141 wrmsrl(e[i].index, e[i].data);
142}
143
144static void save_msrs(struct kvm_msr_entry *e, int n)
145{
146 int i;
147
148 for (i = 0; i < n; ++i)
149 rdmsrl(e[i].index, e[i].data);
150}
151
152static inline int is_page_fault(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 | PF_VECTOR | INTR_INFO_VALID_MASK);
157}
158
159static inline int is_no_device(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 | NM_VECTOR | INTR_INFO_VALID_MASK);
164}
165
166static inline int is_invalid_opcode(u32 intr_info)
167{
168 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
169 INTR_INFO_VALID_MASK)) ==
170 (INTR_TYPE_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
171}
172
173static inline int is_external_interrupt(u32 intr_info)
174{
175 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
176 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
177}
178
179static inline int cpu_has_vmx_tpr_shadow(void)
180{
181 return (vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW);
182}
183
184static inline int vm_need_tpr_shadow(struct kvm *kvm)
185{
186 return ((cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm)));
187}
188
189static inline int cpu_has_secondary_exec_ctrls(void)
190{
191 return (vmcs_config.cpu_based_exec_ctrl &
192 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS);
193}
194
195static inline int cpu_has_vmx_virtualize_apic_accesses(void)
196{
197 return (vmcs_config.cpu_based_2nd_exec_ctrl &
198 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
199}
200
201static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
202{
203 return ((cpu_has_vmx_virtualize_apic_accesses()) &&
204 (irqchip_in_kernel(kvm)));
205}
206
207static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
208{
209 int i;
210
211 for (i = 0; i < vmx->nmsrs; ++i)
212 if (vmx->guest_msrs[i].index == msr)
213 return i;
214 return -1;
215}
216
217static struct kvm_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
218{
219 int i;
220
221 i = __find_msr_index(vmx, msr);
222 if (i >= 0)
223 return &vmx->guest_msrs[i];
224 return NULL;
225}
226
227static void vmcs_clear(struct vmcs *vmcs)
228{
229 u64 phys_addr = __pa(vmcs);
230 u8 error;
231
232 asm volatile (ASM_VMX_VMCLEAR_RAX "; setna %0"
233 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
234 : "cc", "memory");
235 if (error)
236 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
237 vmcs, phys_addr);
238}
239
240static void __vcpu_clear(void *arg)
241{
242 struct vcpu_vmx *vmx = arg;
243 int cpu = raw_smp_processor_id();
244
245 if (vmx->vcpu.cpu == cpu)
246 vmcs_clear(vmx->vmcs);
247 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
248 per_cpu(current_vmcs, cpu) = NULL;
249 rdtscll(vmx->vcpu.arch.host_tsc);
250}
251
252static void vcpu_clear(struct vcpu_vmx *vmx)
253{
254 if (vmx->vcpu.cpu == -1)
255 return;
256 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 0, 1);
257 vmx->launched = 0;
258}
259
260static unsigned long vmcs_readl(unsigned long field)
261{
262 unsigned long value;
263
264 asm volatile (ASM_VMX_VMREAD_RDX_RAX
265 : "=a"(value) : "d"(field) : "cc");
266 return value;
267}
268
269static u16 vmcs_read16(unsigned long field)
270{
271 return vmcs_readl(field);
272}
273
274static u32 vmcs_read32(unsigned long field)
275{
276 return vmcs_readl(field);
277}
278
279static u64 vmcs_read64(unsigned long field)
280{
281#ifdef CONFIG_X86_64
282 return vmcs_readl(field);
283#else
284 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
285#endif
286}
287
288static noinline void vmwrite_error(unsigned long field, unsigned long value)
289{
290 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
291 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
292 dump_stack();
293}
294
295static void vmcs_writel(unsigned long field, unsigned long value)
296{
297 u8 error;
298
299 asm volatile (ASM_VMX_VMWRITE_RAX_RDX "; setna %0"
300 : "=q"(error) : "a"(value), "d"(field) : "cc");
301 if (unlikely(error))
302 vmwrite_error(field, value);
303}
304
305static void vmcs_write16(unsigned long field, u16 value)
306{
307 vmcs_writel(field, value);
308}
309
310static void vmcs_write32(unsigned long field, u32 value)
311{
312 vmcs_writel(field, value);
313}
314
315static void vmcs_write64(unsigned long field, u64 value)
316{
317#ifdef CONFIG_X86_64
318 vmcs_writel(field, value);
319#else
320 vmcs_writel(field, value);
321 asm volatile ("");
322 vmcs_writel(field+1, value >> 32);
323#endif
324}
325
326static void vmcs_clear_bits(unsigned long field, u32 mask)
327{
328 vmcs_writel(field, vmcs_readl(field) & ~mask);
329}
330
331static void vmcs_set_bits(unsigned long field, u32 mask)
332{
333 vmcs_writel(field, vmcs_readl(field) | mask);
334}
335
336static void update_exception_bitmap(struct kvm_vcpu *vcpu)
337{
338 u32 eb;
339
340 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR);
341 if (!vcpu->fpu_active)
342 eb |= 1u << NM_VECTOR;
343 if (vcpu->guest_debug.enabled)
344 eb |= 1u << 1;
345 if (vcpu->arch.rmode.active)
346 eb = ~0;
347 vmcs_write32(EXCEPTION_BITMAP, eb);
348}
349
350static void reload_tss(void)
351{
352#ifndef CONFIG_X86_64
353
354 /*
355 * VT restores TR but not its size. Useless.
356 */
357 struct descriptor_table gdt;
358 struct segment_descriptor *descs;
359
360 get_gdt(&gdt);
361 descs = (void *)gdt.base;
362 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
363 load_TR_desc();
364#endif
365}
366
367static void load_transition_efer(struct vcpu_vmx *vmx)
368{
369 int efer_offset = vmx->msr_offset_efer;
370 u64 host_efer = vmx->host_msrs[efer_offset].data;
371 u64 guest_efer = vmx->guest_msrs[efer_offset].data;
372 u64 ignore_bits;
373
374 if (efer_offset < 0)
375 return;
376 /*
377 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
378 * outside long mode
379 */
380 ignore_bits = EFER_NX | EFER_SCE;
381#ifdef CONFIG_X86_64
382 ignore_bits |= EFER_LMA | EFER_LME;
383 /* SCE is meaningful only in long mode on Intel */
384 if (guest_efer & EFER_LMA)
385 ignore_bits &= ~(u64)EFER_SCE;
386#endif
387 if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
388 return;
389
390 vmx->host_state.guest_efer_loaded = 1;
391 guest_efer &= ~ignore_bits;
392 guest_efer |= host_efer & ignore_bits;
393 wrmsrl(MSR_EFER, guest_efer);
394 vmx->vcpu.stat.efer_reload++;
395}
396
397static void reload_host_efer(struct vcpu_vmx *vmx)
398{
399 if (vmx->host_state.guest_efer_loaded) {
400 vmx->host_state.guest_efer_loaded = 0;
401 load_msrs(vmx->host_msrs + vmx->msr_offset_efer, 1);
402 }
403}
404
405static void vmx_save_host_state(struct kvm_vcpu *vcpu)
406{
407 struct vcpu_vmx *vmx = to_vmx(vcpu);
408
409 if (vmx->host_state.loaded)
410 return;
411
412 vmx->host_state.loaded = 1;
413 /*
414 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
415 * allow segment selectors with cpl > 0 or ti == 1.
416 */
417 vmx->host_state.ldt_sel = read_ldt();
418 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
419 vmx->host_state.fs_sel = read_fs();
420 if (!(vmx->host_state.fs_sel & 7)) {
421 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
422 vmx->host_state.fs_reload_needed = 0;
423 } else {
424 vmcs_write16(HOST_FS_SELECTOR, 0);
425 vmx->host_state.fs_reload_needed = 1;
426 }
427 vmx->host_state.gs_sel = read_gs();
428 if (!(vmx->host_state.gs_sel & 7))
429 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
430 else {
431 vmcs_write16(HOST_GS_SELECTOR, 0);
432 vmx->host_state.gs_ldt_reload_needed = 1;
433 }
434
435#ifdef CONFIG_X86_64
436 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
437 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
438#else
439 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
440 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
441#endif
442
443#ifdef CONFIG_X86_64
444 if (is_long_mode(&vmx->vcpu))
445 save_msrs(vmx->host_msrs +
446 vmx->msr_offset_kernel_gs_base, 1);
447
448#endif
449 load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
450 load_transition_efer(vmx);
451}
452
453static void vmx_load_host_state(struct vcpu_vmx *vmx)
454{
455 unsigned long flags;
456
457 if (!vmx->host_state.loaded)
458 return;
459
460 ++vmx->vcpu.stat.host_state_reload;
461 vmx->host_state.loaded = 0;
462 if (vmx->host_state.fs_reload_needed)
463 load_fs(vmx->host_state.fs_sel);
464 if (vmx->host_state.gs_ldt_reload_needed) {
465 load_ldt(vmx->host_state.ldt_sel);
466 /*
467 * If we have to reload gs, we must take care to
468 * preserve our gs base.
469 */
470 local_irq_save(flags);
471 load_gs(vmx->host_state.gs_sel);
472#ifdef CONFIG_X86_64
473 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
474#endif
475 local_irq_restore(flags);
476 }
477 reload_tss();
478 save_msrs(vmx->guest_msrs, vmx->save_nmsrs);
479 load_msrs(vmx->host_msrs, vmx->save_nmsrs);
480 reload_host_efer(vmx);
481}
482
483/*
484 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
485 * vcpu mutex is already taken.
486 */
487static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
488{
489 struct vcpu_vmx *vmx = to_vmx(vcpu);
490 u64 phys_addr = __pa(vmx->vmcs);
491 u64 tsc_this, delta;
492
493 if (vcpu->cpu != cpu) {
494 vcpu_clear(vmx);
495 kvm_migrate_apic_timer(vcpu);
496 }
497
498 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
499 u8 error;
500
501 per_cpu(current_vmcs, cpu) = vmx->vmcs;
502 asm volatile (ASM_VMX_VMPTRLD_RAX "; setna %0"
503 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
504 : "cc");
505 if (error)
506 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
507 vmx->vmcs, phys_addr);
508 }
509
510 if (vcpu->cpu != cpu) {
511 struct descriptor_table dt;
512 unsigned long sysenter_esp;
513
514 vcpu->cpu = cpu;
515 /*
516 * Linux uses per-cpu TSS and GDT, so set these when switching
517 * processors.
518 */
519 vmcs_writel(HOST_TR_BASE, read_tr_base()); /* 22.2.4 */
520 get_gdt(&dt);
521 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
522
523 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
524 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
525
526 /*
527 * Make sure the time stamp counter is monotonous.
528 */
529 rdtscll(tsc_this);
530 delta = vcpu->arch.host_tsc - tsc_this;
531 vmcs_write64(TSC_OFFSET, vmcs_read64(TSC_OFFSET) + delta);
532 }
533}
534
535static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
536{
537 vmx_load_host_state(to_vmx(vcpu));
538}
539
540static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
541{
542 if (vcpu->fpu_active)
543 return;
544 vcpu->fpu_active = 1;
545 vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
546 if (vcpu->arch.cr0 & X86_CR0_TS)
547 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
548 update_exception_bitmap(vcpu);
549}
550
551static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
552{
553 if (!vcpu->fpu_active)
554 return;
555 vcpu->fpu_active = 0;
556 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
557 update_exception_bitmap(vcpu);
558}
559
560static void vmx_vcpu_decache(struct kvm_vcpu *vcpu)
561{
562 vcpu_clear(to_vmx(vcpu));
563}
564
565static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
566{
567 return vmcs_readl(GUEST_RFLAGS);
568}
569
570static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
571{
572 if (vcpu->arch.rmode.active)
573 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
574 vmcs_writel(GUEST_RFLAGS, rflags);
575}
576
577static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
578{
579 unsigned long rip;
580 u32 interruptibility;
581
582 rip = vmcs_readl(GUEST_RIP);
583 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
584 vmcs_writel(GUEST_RIP, rip);
585
586 /*
587 * We emulated an instruction, so temporary interrupt blocking
588 * should be removed, if set.
589 */
590 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
591 if (interruptibility & 3)
592 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
593 interruptibility & ~3);
594 vcpu->arch.interrupt_window_open = 1;
595}
596
597static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
598 bool has_error_code, u32 error_code)
599{
600 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
601 nr | INTR_TYPE_EXCEPTION
602 | (has_error_code ? INTR_INFO_DELIEVER_CODE_MASK : 0)
603 | INTR_INFO_VALID_MASK);
604 if (has_error_code)
605 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
606}
607
608static bool vmx_exception_injected(struct kvm_vcpu *vcpu)
609{
610 struct vcpu_vmx *vmx = to_vmx(vcpu);
611
612 return !(vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
613}
614
615/*
616 * Swap MSR entry in host/guest MSR entry array.
617 */
618#ifdef CONFIG_X86_64
619static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
620{
621 struct kvm_msr_entry tmp;
622
623 tmp = vmx->guest_msrs[to];
624 vmx->guest_msrs[to] = vmx->guest_msrs[from];
625 vmx->guest_msrs[from] = tmp;
626 tmp = vmx->host_msrs[to];
627 vmx->host_msrs[to] = vmx->host_msrs[from];
628 vmx->host_msrs[from] = tmp;
629}
630#endif
631
632/*
633 * Set up the vmcs to automatically save and restore system
634 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
635 * mode, as fiddling with msrs is very expensive.
636 */
637static void setup_msrs(struct vcpu_vmx *vmx)
638{
639 int save_nmsrs;
640
641 save_nmsrs = 0;
642#ifdef CONFIG_X86_64
643 if (is_long_mode(&vmx->vcpu)) {
644 int index;
645
646 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
647 if (index >= 0)
648 move_msr_up(vmx, index, save_nmsrs++);
649 index = __find_msr_index(vmx, MSR_LSTAR);
650 if (index >= 0)
651 move_msr_up(vmx, index, save_nmsrs++);
652 index = __find_msr_index(vmx, MSR_CSTAR);
653 if (index >= 0)
654 move_msr_up(vmx, index, save_nmsrs++);
655 index = __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
656 if (index >= 0)
657 move_msr_up(vmx, index, save_nmsrs++);
658 /*
659 * MSR_K6_STAR is only needed on long mode guests, and only
660 * if efer.sce is enabled.
661 */
662 index = __find_msr_index(vmx, MSR_K6_STAR);
663 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
664 move_msr_up(vmx, index, save_nmsrs++);
665 }
666#endif
667 vmx->save_nmsrs = save_nmsrs;
668
669#ifdef CONFIG_X86_64
670 vmx->msr_offset_kernel_gs_base =
671 __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
672#endif
673 vmx->msr_offset_efer = __find_msr_index(vmx, MSR_EFER);
674}
675
676/*
677 * reads and returns guest's timestamp counter "register"
678 * guest_tsc = host_tsc + tsc_offset -- 21.3
679 */
680static u64 guest_read_tsc(void)
681{
682 u64 host_tsc, tsc_offset;
683
684 rdtscll(host_tsc);
685 tsc_offset = vmcs_read64(TSC_OFFSET);
686 return host_tsc + tsc_offset;
687}
688
689/*
690 * writes 'guest_tsc' into guest's timestamp counter "register"
691 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
692 */
693static void guest_write_tsc(u64 guest_tsc)
694{
695 u64 host_tsc;
696
697 rdtscll(host_tsc);
698 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
699}
700
701/*
702 * Reads an msr value (of 'msr_index') into 'pdata'.
703 * Returns 0 on success, non-0 otherwise.
704 * Assumes vcpu_load() was already called.
705 */
706static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
707{
708 u64 data;
709 struct kvm_msr_entry *msr;
710
711 if (!pdata) {
712 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
713 return -EINVAL;
714 }
715
716 switch (msr_index) {
717#ifdef CONFIG_X86_64
718 case MSR_FS_BASE:
719 data = vmcs_readl(GUEST_FS_BASE);
720 break;
721 case MSR_GS_BASE:
722 data = vmcs_readl(GUEST_GS_BASE);
723 break;
724 case MSR_EFER:
725 return kvm_get_msr_common(vcpu, msr_index, pdata);
726#endif
727 case MSR_IA32_TIME_STAMP_COUNTER:
728 data = guest_read_tsc();
729 break;
730 case MSR_IA32_SYSENTER_CS:
731 data = vmcs_read32(GUEST_SYSENTER_CS);
732 break;
733 case MSR_IA32_SYSENTER_EIP:
734 data = vmcs_readl(GUEST_SYSENTER_EIP);
735 break;
736 case MSR_IA32_SYSENTER_ESP:
737 data = vmcs_readl(GUEST_SYSENTER_ESP);
738 break;
739 default:
740 msr = find_msr_entry(to_vmx(vcpu), msr_index);
741 if (msr) {
742 data = msr->data;
743 break;
744 }
745 return kvm_get_msr_common(vcpu, msr_index, pdata);
746 }
747
748 *pdata = data;
749 return 0;
750}
751
752/*
753 * Writes msr value into into the appropriate "register".
754 * Returns 0 on success, non-0 otherwise.
755 * Assumes vcpu_load() was already called.
756 */
757static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
758{
759 struct vcpu_vmx *vmx = to_vmx(vcpu);
760 struct kvm_msr_entry *msr;
761 int ret = 0;
762
763 switch (msr_index) {
764#ifdef CONFIG_X86_64
765 case MSR_EFER:
766 ret = kvm_set_msr_common(vcpu, msr_index, data);
767 if (vmx->host_state.loaded) {
768 reload_host_efer(vmx);
769 load_transition_efer(vmx);
770 }
771 break;
772 case MSR_FS_BASE:
773 vmcs_writel(GUEST_FS_BASE, data);
774 break;
775 case MSR_GS_BASE:
776 vmcs_writel(GUEST_GS_BASE, data);
777 break;
778#endif
779 case MSR_IA32_SYSENTER_CS:
780 vmcs_write32(GUEST_SYSENTER_CS, data);
781 break;
782 case MSR_IA32_SYSENTER_EIP:
783 vmcs_writel(GUEST_SYSENTER_EIP, data);
784 break;
785 case MSR_IA32_SYSENTER_ESP:
786 vmcs_writel(GUEST_SYSENTER_ESP, data);
787 break;
788 case MSR_IA32_TIME_STAMP_COUNTER:
789 guest_write_tsc(data);
790 break;
791 default:
792 msr = find_msr_entry(vmx, msr_index);
793 if (msr) {
794 msr->data = data;
795 if (vmx->host_state.loaded)
796 load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
797 break;
798 }
799 ret = kvm_set_msr_common(vcpu, msr_index, data);
800 }
801
802 return ret;
803}
804
805/*
806 * Sync the rsp and rip registers into the vcpu structure. This allows
807 * registers to be accessed by indexing vcpu->arch.regs.
808 */
809static void vcpu_load_rsp_rip(struct kvm_vcpu *vcpu)
810{
811 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
812 vcpu->arch.rip = vmcs_readl(GUEST_RIP);
813}
814
815/*
816 * Syncs rsp and rip back into the vmcs. Should be called after possible
817 * modification.
818 */
819static void vcpu_put_rsp_rip(struct kvm_vcpu *vcpu)
820{
821 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
822 vmcs_writel(GUEST_RIP, vcpu->arch.rip);
823}
824
825static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
826{
827 unsigned long dr7 = 0x400;
828 int old_singlestep;
829
830 old_singlestep = vcpu->guest_debug.singlestep;
831
832 vcpu->guest_debug.enabled = dbg->enabled;
833 if (vcpu->guest_debug.enabled) {
834 int i;
835
836 dr7 |= 0x200; /* exact */
837 for (i = 0; i < 4; ++i) {
838 if (!dbg->breakpoints[i].enabled)
839 continue;
840 vcpu->guest_debug.bp[i] = dbg->breakpoints[i].address;
841 dr7 |= 2 << (i*2); /* global enable */
842 dr7 |= 0 << (i*4+16); /* execution breakpoint */
843 }
844
845 vcpu->guest_debug.singlestep = dbg->singlestep;
846 } else
847 vcpu->guest_debug.singlestep = 0;
848
849 if (old_singlestep && !vcpu->guest_debug.singlestep) {
850 unsigned long flags;
851
852 flags = vmcs_readl(GUEST_RFLAGS);
853 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
854 vmcs_writel(GUEST_RFLAGS, flags);
855 }
856
857 update_exception_bitmap(vcpu);
858 vmcs_writel(GUEST_DR7, dr7);
859
860 return 0;
861}
862
863static int vmx_get_irq(struct kvm_vcpu *vcpu)
864{
865 struct vcpu_vmx *vmx = to_vmx(vcpu);
866 u32 idtv_info_field;
867
868 idtv_info_field = vmx->idt_vectoring_info;
869 if (idtv_info_field & INTR_INFO_VALID_MASK) {
870 if (is_external_interrupt(idtv_info_field))
871 return idtv_info_field & VECTORING_INFO_VECTOR_MASK;
872 else
873 printk(KERN_DEBUG "pending exception: not handled yet\n");
874 }
875 return -1;
876}
877
878static __init int cpu_has_kvm_support(void)
879{
880 unsigned long ecx = cpuid_ecx(1);
881 return test_bit(5, &ecx); /* CPUID.1:ECX.VMX[bit 5] -> VT */
882}
883
884static __init int vmx_disabled_by_bios(void)
885{
886 u64 msr;
887
888 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
889 return (msr & (MSR_IA32_FEATURE_CONTROL_LOCKED |
890 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
891 == MSR_IA32_FEATURE_CONTROL_LOCKED;
892 /* locked but not enabled */
893}
894
895static void hardware_enable(void *garbage)
896{
897 int cpu = raw_smp_processor_id();
898 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
899 u64 old;
900
901 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
902 if ((old & (MSR_IA32_FEATURE_CONTROL_LOCKED |
903 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
904 != (MSR_IA32_FEATURE_CONTROL_LOCKED |
905 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
906 /* enable and lock */
907 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
908 MSR_IA32_FEATURE_CONTROL_LOCKED |
909 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED);
910 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
911 asm volatile (ASM_VMX_VMXON_RAX : : "a"(&phys_addr), "m"(phys_addr)
912 : "memory", "cc");
913}
914
915static void hardware_disable(void *garbage)
916{
917 asm volatile (ASM_VMX_VMXOFF : : : "cc");
918}
919
920static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
921 u32 msr, u32 *result)
922{
923 u32 vmx_msr_low, vmx_msr_high;
924 u32 ctl = ctl_min | ctl_opt;
925
926 rdmsr(msr, vmx_msr_low, vmx_msr_high);
927
928 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
929 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
930
931 /* Ensure minimum (required) set of control bits are supported. */
932 if (ctl_min & ~ctl)
933 return -EIO;
934
935 *result = ctl;
936 return 0;
937}
938
939static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
940{
941 u32 vmx_msr_low, vmx_msr_high;
942 u32 min, opt;
943 u32 _pin_based_exec_control = 0;
944 u32 _cpu_based_exec_control = 0;
945 u32 _cpu_based_2nd_exec_control = 0;
946 u32 _vmexit_control = 0;
947 u32 _vmentry_control = 0;
948
949 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
950 opt = 0;
951 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
952 &_pin_based_exec_control) < 0)
953 return -EIO;
954
955 min = CPU_BASED_HLT_EXITING |
956#ifdef CONFIG_X86_64
957 CPU_BASED_CR8_LOAD_EXITING |
958 CPU_BASED_CR8_STORE_EXITING |
959#endif
960 CPU_BASED_USE_IO_BITMAPS |
961 CPU_BASED_MOV_DR_EXITING |
962 CPU_BASED_USE_TSC_OFFSETING;
963 opt = CPU_BASED_TPR_SHADOW |
964 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
965 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
966 &_cpu_based_exec_control) < 0)
967 return -EIO;
968#ifdef CONFIG_X86_64
969 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
970 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
971 ~CPU_BASED_CR8_STORE_EXITING;
972#endif
973 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
974 min = 0;
975 opt = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
976 SECONDARY_EXEC_WBINVD_EXITING;
977 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS2,
978 &_cpu_based_2nd_exec_control) < 0)
979 return -EIO;
980 }
981#ifndef CONFIG_X86_64
982 if (!(_cpu_based_2nd_exec_control &
983 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
984 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
985#endif
986
987 min = 0;
988#ifdef CONFIG_X86_64
989 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
990#endif
991 opt = 0;
992 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
993 &_vmexit_control) < 0)
994 return -EIO;
995
996 min = opt = 0;
997 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
998 &_vmentry_control) < 0)
999 return -EIO;
1000
1001 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1002
1003 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1004 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1005 return -EIO;
1006
1007#ifdef CONFIG_X86_64
1008 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1009 if (vmx_msr_high & (1u<<16))
1010 return -EIO;
1011#endif
1012
1013 /* Require Write-Back (WB) memory type for VMCS accesses. */
1014 if (((vmx_msr_high >> 18) & 15) != 6)
1015 return -EIO;
1016
1017 vmcs_conf->size = vmx_msr_high & 0x1fff;
1018 vmcs_conf->order = get_order(vmcs_config.size);
1019 vmcs_conf->revision_id = vmx_msr_low;
1020
1021 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1022 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1023 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1024 vmcs_conf->vmexit_ctrl = _vmexit_control;
1025 vmcs_conf->vmentry_ctrl = _vmentry_control;
1026
1027 return 0;
1028}
1029
1030static struct vmcs *alloc_vmcs_cpu(int cpu)
1031{
1032 int node = cpu_to_node(cpu);
1033 struct page *pages;
1034 struct vmcs *vmcs;
1035
1036 pages = alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
1037 if (!pages)
1038 return NULL;
1039 vmcs = page_address(pages);
1040 memset(vmcs, 0, vmcs_config.size);
1041 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1042 return vmcs;
1043}
1044
1045static struct vmcs *alloc_vmcs(void)
1046{
1047 return alloc_vmcs_cpu(raw_smp_processor_id());
1048}
1049
1050static void free_vmcs(struct vmcs *vmcs)
1051{
1052 free_pages((unsigned long)vmcs, vmcs_config.order);
1053}
1054
1055static void free_kvm_area(void)
1056{
1057 int cpu;
1058
1059 for_each_online_cpu(cpu)
1060 free_vmcs(per_cpu(vmxarea, cpu));
1061}
1062
1063static __init int alloc_kvm_area(void)
1064{
1065 int cpu;
1066
1067 for_each_online_cpu(cpu) {
1068 struct vmcs *vmcs;
1069
1070 vmcs = alloc_vmcs_cpu(cpu);
1071 if (!vmcs) {
1072 free_kvm_area();
1073 return -ENOMEM;
1074 }
1075
1076 per_cpu(vmxarea, cpu) = vmcs;
1077 }
1078 return 0;
1079}
1080
1081static __init int hardware_setup(void)
1082{
1083 if (setup_vmcs_config(&vmcs_config) < 0)
1084 return -EIO;
1085 return alloc_kvm_area();
1086}
1087
1088static __exit void hardware_unsetup(void)
1089{
1090 free_kvm_area();
1091}
1092
1093static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1094{
1095 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1096
1097 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1098 vmcs_write16(sf->selector, save->selector);
1099 vmcs_writel(sf->base, save->base);
1100 vmcs_write32(sf->limit, save->limit);
1101 vmcs_write32(sf->ar_bytes, save->ar);
1102 } else {
1103 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1104 << AR_DPL_SHIFT;
1105 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1106 }
1107}
1108
1109static void enter_pmode(struct kvm_vcpu *vcpu)
1110{
1111 unsigned long flags;
1112
1113 vcpu->arch.rmode.active = 0;
1114
1115 vmcs_writel(GUEST_TR_BASE, vcpu->arch.rmode.tr.base);
1116 vmcs_write32(GUEST_TR_LIMIT, vcpu->arch.rmode.tr.limit);
1117 vmcs_write32(GUEST_TR_AR_BYTES, vcpu->arch.rmode.tr.ar);
1118
1119 flags = vmcs_readl(GUEST_RFLAGS);
1120 flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
1121 flags |= (vcpu->arch.rmode.save_iopl << IOPL_SHIFT);
1122 vmcs_writel(GUEST_RFLAGS, flags);
1123
1124 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1125 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1126
1127 update_exception_bitmap(vcpu);
1128
1129 fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1130 fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1131 fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1132 fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1133
1134 vmcs_write16(GUEST_SS_SELECTOR, 0);
1135 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1136
1137 vmcs_write16(GUEST_CS_SELECTOR,
1138 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1139 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1140}
1141
1142static gva_t rmode_tss_base(struct kvm *kvm)
1143{
1144 if (!kvm->arch.tss_addr) {
1145 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1146 kvm->memslots[0].npages - 3;
1147 return base_gfn << PAGE_SHIFT;
1148 }
1149 return kvm->arch.tss_addr;
1150}
1151
1152static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1153{
1154 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1155
1156 save->selector = vmcs_read16(sf->selector);
1157 save->base = vmcs_readl(sf->base);
1158 save->limit = vmcs_read32(sf->limit);
1159 save->ar = vmcs_read32(sf->ar_bytes);
1160 vmcs_write16(sf->selector, save->base >> 4);
1161 vmcs_write32(sf->base, save->base & 0xfffff);
1162 vmcs_write32(sf->limit, 0xffff);
1163 vmcs_write32(sf->ar_bytes, 0xf3);
1164}
1165
1166static void enter_rmode(struct kvm_vcpu *vcpu)
1167{
1168 unsigned long flags;
1169
1170 vcpu->arch.rmode.active = 1;
1171
1172 vcpu->arch.rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1173 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1174
1175 vcpu->arch.rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1176 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1177
1178 vcpu->arch.rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1179 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1180
1181 flags = vmcs_readl(GUEST_RFLAGS);
1182 vcpu->arch.rmode.save_iopl
1183 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
1184
1185 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1186
1187 vmcs_writel(GUEST_RFLAGS, flags);
1188 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1189 update_exception_bitmap(vcpu);
1190
1191 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1192 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1193 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1194
1195 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1196 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1197 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1198 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1199 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1200
1201 fix_rmode_seg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1202 fix_rmode_seg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1203 fix_rmode_seg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1204 fix_rmode_seg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1205
1206 kvm_mmu_reset_context(vcpu);
1207 init_rmode_tss(vcpu->kvm);
1208}
1209
1210#ifdef CONFIG_X86_64
1211
1212static void enter_lmode(struct kvm_vcpu *vcpu)
1213{
1214 u32 guest_tr_ar;
1215
1216 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1217 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1218 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1219 __FUNCTION__);
1220 vmcs_write32(GUEST_TR_AR_BYTES,
1221 (guest_tr_ar & ~AR_TYPE_MASK)
1222 | AR_TYPE_BUSY_64_TSS);
1223 }
1224
1225 vcpu->arch.shadow_efer |= EFER_LMA;
1226
1227 find_msr_entry(to_vmx(vcpu), MSR_EFER)->data |= EFER_LMA | EFER_LME;
1228 vmcs_write32(VM_ENTRY_CONTROLS,
1229 vmcs_read32(VM_ENTRY_CONTROLS)
1230 | VM_ENTRY_IA32E_MODE);
1231}
1232
1233static void exit_lmode(struct kvm_vcpu *vcpu)
1234{
1235 vcpu->arch.shadow_efer &= ~EFER_LMA;
1236
1237 vmcs_write32(VM_ENTRY_CONTROLS,
1238 vmcs_read32(VM_ENTRY_CONTROLS)
1239 & ~VM_ENTRY_IA32E_MODE);
1240}
1241
1242#endif
1243
1244static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1245{
1246 vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK;
1247 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
1248}
1249
1250static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1251{
1252 vmx_fpu_deactivate(vcpu);
1253
1254 if (vcpu->arch.rmode.active && (cr0 & X86_CR0_PE))
1255 enter_pmode(vcpu);
1256
1257 if (!vcpu->arch.rmode.active && !(cr0 & X86_CR0_PE))
1258 enter_rmode(vcpu);
1259
1260#ifdef CONFIG_X86_64
1261 if (vcpu->arch.shadow_efer & EFER_LME) {
1262 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1263 enter_lmode(vcpu);
1264 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1265 exit_lmode(vcpu);
1266 }
1267#endif
1268
1269 vmcs_writel(CR0_READ_SHADOW, cr0);
1270 vmcs_writel(GUEST_CR0,
1271 (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON);
1272 vcpu->arch.cr0 = cr0;
1273
1274 if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
1275 vmx_fpu_activate(vcpu);
1276}
1277
1278static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1279{
1280 vmcs_writel(GUEST_CR3, cr3);
1281 if (vcpu->arch.cr0 & X86_CR0_PE)
1282 vmx_fpu_deactivate(vcpu);
1283}
1284
1285static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1286{
1287 vmcs_writel(CR4_READ_SHADOW, cr4);
1288 vmcs_writel(GUEST_CR4, cr4 | (vcpu->arch.rmode.active ?
1289 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON));
1290 vcpu->arch.cr4 = cr4;
1291}
1292
1293#ifdef CONFIG_X86_64
1294
1295static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1296{
1297 struct vcpu_vmx *vmx = to_vmx(vcpu);
1298 struct kvm_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1299
1300 vcpu->arch.shadow_efer = efer;
1301 if (efer & EFER_LMA) {
1302 vmcs_write32(VM_ENTRY_CONTROLS,
1303 vmcs_read32(VM_ENTRY_CONTROLS) |
1304 VM_ENTRY_IA32E_MODE);
1305 msr->data = efer;
1306
1307 } else {
1308 vmcs_write32(VM_ENTRY_CONTROLS,
1309 vmcs_read32(VM_ENTRY_CONTROLS) &
1310 ~VM_ENTRY_IA32E_MODE);
1311
1312 msr->data = efer & ~EFER_LME;
1313 }
1314 setup_msrs(vmx);
1315}
1316
1317#endif
1318
1319static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1320{
1321 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1322
1323 return vmcs_readl(sf->base);
1324}
1325
1326static void vmx_get_segment(struct kvm_vcpu *vcpu,
1327 struct kvm_segment *var, int seg)
1328{
1329 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1330 u32 ar;
1331
1332 var->base = vmcs_readl(sf->base);
1333 var->limit = vmcs_read32(sf->limit);
1334 var->selector = vmcs_read16(sf->selector);
1335 ar = vmcs_read32(sf->ar_bytes);
1336 if (ar & AR_UNUSABLE_MASK)
1337 ar = 0;
1338 var->type = ar & 15;
1339 var->s = (ar >> 4) & 1;
1340 var->dpl = (ar >> 5) & 3;
1341 var->present = (ar >> 7) & 1;
1342 var->avl = (ar >> 12) & 1;
1343 var->l = (ar >> 13) & 1;
1344 var->db = (ar >> 14) & 1;
1345 var->g = (ar >> 15) & 1;
1346 var->unusable = (ar >> 16) & 1;
1347}
1348
1349static u32 vmx_segment_access_rights(struct kvm_segment *var)
1350{
1351 u32 ar;
1352
1353 if (var->unusable)
1354 ar = 1 << 16;
1355 else {
1356 ar = var->type & 15;
1357 ar |= (var->s & 1) << 4;
1358 ar |= (var->dpl & 3) << 5;
1359 ar |= (var->present & 1) << 7;
1360 ar |= (var->avl & 1) << 12;
1361 ar |= (var->l & 1) << 13;
1362 ar |= (var->db & 1) << 14;
1363 ar |= (var->g & 1) << 15;
1364 }
1365 if (ar == 0) /* a 0 value means unusable */
1366 ar = AR_UNUSABLE_MASK;
1367
1368 return ar;
1369}
1370
1371static void vmx_set_segment(struct kvm_vcpu *vcpu,
1372 struct kvm_segment *var, int seg)
1373{
1374 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1375 u32 ar;
1376
1377 if (vcpu->arch.rmode.active && seg == VCPU_SREG_TR) {
1378 vcpu->arch.rmode.tr.selector = var->selector;
1379 vcpu->arch.rmode.tr.base = var->base;
1380 vcpu->arch.rmode.tr.limit = var->limit;
1381 vcpu->arch.rmode.tr.ar = vmx_segment_access_rights(var);
1382 return;
1383 }
1384 vmcs_writel(sf->base, var->base);
1385 vmcs_write32(sf->limit, var->limit);
1386 vmcs_write16(sf->selector, var->selector);
1387 if (vcpu->arch.rmode.active && var->s) {
1388 /*
1389 * Hack real-mode segments into vm86 compatibility.
1390 */
1391 if (var->base == 0xffff0000 && var->selector == 0xf000)
1392 vmcs_writel(sf->base, 0xf0000);
1393 ar = 0xf3;
1394 } else
1395 ar = vmx_segment_access_rights(var);
1396 vmcs_write32(sf->ar_bytes, ar);
1397}
1398
1399static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1400{
1401 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1402
1403 *db = (ar >> 14) & 1;
1404 *l = (ar >> 13) & 1;
1405}
1406
1407static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1408{
1409 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1410 dt->base = vmcs_readl(GUEST_IDTR_BASE);
1411}
1412
1413static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1414{
1415 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1416 vmcs_writel(GUEST_IDTR_BASE, dt->base);
1417}
1418
1419static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1420{
1421 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1422 dt->base = vmcs_readl(GUEST_GDTR_BASE);
1423}
1424
1425static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1426{
1427 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1428 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1429}
1430
1431static int init_rmode_tss(struct kvm *kvm)
1432{
1433 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
1434 u16 data = 0;
1435 int r;
1436
1437 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1438 if (r < 0)
1439 return 0;
1440 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
1441 r = kvm_write_guest_page(kvm, fn++, &data, 0x66, sizeof(u16));
1442 if (r < 0)
1443 return 0;
1444 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
1445 if (r < 0)
1446 return 0;
1447 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1448 if (r < 0)
1449 return 0;
1450 data = ~0;
1451 r = kvm_write_guest_page(kvm, fn, &data, RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
1452 sizeof(u8));
1453 if (r < 0)
1454 return 0;
1455 return 1;
1456}
1457
1458static void seg_setup(int seg)
1459{
1460 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1461
1462 vmcs_write16(sf->selector, 0);
1463 vmcs_writel(sf->base, 0);
1464 vmcs_write32(sf->limit, 0xffff);
1465 vmcs_write32(sf->ar_bytes, 0x93);
1466}
1467
1468static int alloc_apic_access_page(struct kvm *kvm)
1469{
1470 struct kvm_userspace_memory_region kvm_userspace_mem;
1471 int r = 0;
1472
1473 mutex_lock(&kvm->lock);
1474 if (kvm->arch.apic_access_page)
1475 goto out;
1476 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
1477 kvm_userspace_mem.flags = 0;
1478 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
1479 kvm_userspace_mem.memory_size = PAGE_SIZE;
1480 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
1481 if (r)
1482 goto out;
1483 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
1484out:
1485 mutex_unlock(&kvm->lock);
1486 return r;
1487}
1488
1489/*
1490 * Sets up the vmcs for emulated real mode.
1491 */
1492static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
1493{
1494 u32 host_sysenter_cs;
1495 u32 junk;
1496 unsigned long a;
1497 struct descriptor_table dt;
1498 int i;
1499 unsigned long kvm_vmx_return;
1500 u32 exec_control;
1501
1502 /* I/O */
1503 vmcs_write64(IO_BITMAP_A, page_to_phys(vmx_io_bitmap_a));
1504 vmcs_write64(IO_BITMAP_B, page_to_phys(vmx_io_bitmap_b));
1505
1506 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
1507
1508 /* Control */
1509 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
1510 vmcs_config.pin_based_exec_ctrl);
1511
1512 exec_control = vmcs_config.cpu_based_exec_ctrl;
1513 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
1514 exec_control &= ~CPU_BASED_TPR_SHADOW;
1515#ifdef CONFIG_X86_64
1516 exec_control |= CPU_BASED_CR8_STORE_EXITING |
1517 CPU_BASED_CR8_LOAD_EXITING;
1518#endif
1519 }
1520 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
1521
1522 if (cpu_has_secondary_exec_ctrls()) {
1523 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
1524 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
1525 exec_control &=
1526 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1527 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
1528 }
1529
1530 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
1531 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
1532 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
1533
1534 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
1535 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
1536 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
1537
1538 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
1539 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1540 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1541 vmcs_write16(HOST_FS_SELECTOR, read_fs()); /* 22.2.4 */
1542 vmcs_write16(HOST_GS_SELECTOR, read_gs()); /* 22.2.4 */
1543 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1544#ifdef CONFIG_X86_64
1545 rdmsrl(MSR_FS_BASE, a);
1546 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
1547 rdmsrl(MSR_GS_BASE, a);
1548 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
1549#else
1550 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
1551 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
1552#endif
1553
1554 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
1555
1556 get_idt(&dt);
1557 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
1558
1559 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
1560 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
1561 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
1562 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
1563 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
1564
1565 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
1566 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
1567 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
1568 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
1569 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
1570 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
1571
1572 for (i = 0; i < NR_VMX_MSR; ++i) {
1573 u32 index = vmx_msr_index[i];
1574 u32 data_low, data_high;
1575 u64 data;
1576 int j = vmx->nmsrs;
1577
1578 if (rdmsr_safe(index, &data_low, &data_high) < 0)
1579 continue;
1580 if (wrmsr_safe(index, data_low, data_high) < 0)
1581 continue;
1582 data = data_low | ((u64)data_high << 32);
1583 vmx->host_msrs[j].index = index;
1584 vmx->host_msrs[j].reserved = 0;
1585 vmx->host_msrs[j].data = data;
1586 vmx->guest_msrs[j] = vmx->host_msrs[j];
1587 ++vmx->nmsrs;
1588 }
1589
1590 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
1591
1592 /* 22.2.1, 20.8.1 */
1593 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
1594
1595 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
1596 vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
1597
1598 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
1599 if (alloc_apic_access_page(vmx->vcpu.kvm) != 0)
1600 return -ENOMEM;
1601
1602 return 0;
1603}
1604
1605static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
1606{
1607 struct vcpu_vmx *vmx = to_vmx(vcpu);
1608 u64 msr;
1609 int ret;
1610
1611 if (!init_rmode_tss(vmx->vcpu.kvm)) {
1612 ret = -ENOMEM;
1613 goto out;
1614 }
1615
1616 vmx->vcpu.arch.rmode.active = 0;
1617
1618 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
1619 set_cr8(&vmx->vcpu, 0);
1620 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
1621 if (vmx->vcpu.vcpu_id == 0)
1622 msr |= MSR_IA32_APICBASE_BSP;
1623 kvm_set_apic_base(&vmx->vcpu, msr);
1624
1625 fx_init(&vmx->vcpu);
1626
1627 /*
1628 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
1629 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
1630 */
1631 if (vmx->vcpu.vcpu_id == 0) {
1632 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
1633 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
1634 } else {
1635 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
1636 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
1637 }
1638 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1639 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1640
1641 seg_setup(VCPU_SREG_DS);
1642 seg_setup(VCPU_SREG_ES);
1643 seg_setup(VCPU_SREG_FS);
1644 seg_setup(VCPU_SREG_GS);
1645 seg_setup(VCPU_SREG_SS);
1646
1647 vmcs_write16(GUEST_TR_SELECTOR, 0);
1648 vmcs_writel(GUEST_TR_BASE, 0);
1649 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
1650 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1651
1652 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
1653 vmcs_writel(GUEST_LDTR_BASE, 0);
1654 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
1655 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
1656
1657 vmcs_write32(GUEST_SYSENTER_CS, 0);
1658 vmcs_writel(GUEST_SYSENTER_ESP, 0);
1659 vmcs_writel(GUEST_SYSENTER_EIP, 0);
1660
1661 vmcs_writel(GUEST_RFLAGS, 0x02);
1662 if (vmx->vcpu.vcpu_id == 0)
1663 vmcs_writel(GUEST_RIP, 0xfff0);
1664 else
1665 vmcs_writel(GUEST_RIP, 0);
1666 vmcs_writel(GUEST_RSP, 0);
1667
1668 /* todo: dr0 = dr1 = dr2 = dr3 = 0; dr6 = 0xffff0ff0 */
1669 vmcs_writel(GUEST_DR7, 0x400);
1670
1671 vmcs_writel(GUEST_GDTR_BASE, 0);
1672 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
1673
1674 vmcs_writel(GUEST_IDTR_BASE, 0);
1675 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
1676
1677 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
1678 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
1679 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
1680
1681 guest_write_tsc(0);
1682
1683 /* Special registers */
1684 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
1685
1686 setup_msrs(vmx);
1687
1688 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
1689
1690 if (cpu_has_vmx_tpr_shadow()) {
1691 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
1692 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
1693 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
1694 page_to_phys(vmx->vcpu.arch.apic->regs_page));
1695 vmcs_write32(TPR_THRESHOLD, 0);
1696 }
1697
1698 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
1699 vmcs_write64(APIC_ACCESS_ADDR,
1700 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
1701
1702 vmx->vcpu.arch.cr0 = 0x60000010;
1703 vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
1704 vmx_set_cr4(&vmx->vcpu, 0);
1705#ifdef CONFIG_X86_64
1706 vmx_set_efer(&vmx->vcpu, 0);
1707#endif
1708 vmx_fpu_activate(&vmx->vcpu);
1709 update_exception_bitmap(&vmx->vcpu);
1710
1711 return 0;
1712
1713out:
1714 return ret;
1715}
1716
1717static void vmx_inject_irq(struct kvm_vcpu *vcpu, int irq)
1718{
1719 struct vcpu_vmx *vmx = to_vmx(vcpu);
1720
1721 if (vcpu->arch.rmode.active) {
1722 vmx->rmode.irq.pending = true;
1723 vmx->rmode.irq.vector = irq;
1724 vmx->rmode.irq.rip = vmcs_readl(GUEST_RIP);
1725 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1726 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
1727 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
1728 vmcs_writel(GUEST_RIP, vmx->rmode.irq.rip - 1);
1729 return;
1730 }
1731 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1732 irq | INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1733}
1734
1735static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1736{
1737 int word_index = __ffs(vcpu->arch.irq_summary);
1738 int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
1739 int irq = word_index * BITS_PER_LONG + bit_index;
1740
1741 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
1742 if (!vcpu->arch.irq_pending[word_index])
1743 clear_bit(word_index, &vcpu->arch.irq_summary);
1744 vmx_inject_irq(vcpu, irq);
1745}
1746
1747
1748static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1749 struct kvm_run *kvm_run)
1750{
1751 u32 cpu_based_vm_exec_control;
1752
1753 vcpu->arch.interrupt_window_open =
1754 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
1755 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
1756
1757 if (vcpu->arch.interrupt_window_open &&
1758 vcpu->arch.irq_summary &&
1759 !(vmcs_read32(VM_ENTRY_INTR_INFO_FIELD) & INTR_INFO_VALID_MASK))
1760 /*
1761 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1762 */
1763 kvm_do_inject_irq(vcpu);
1764
1765 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
1766 if (!vcpu->arch.interrupt_window_open &&
1767 (vcpu->arch.irq_summary || kvm_run->request_interrupt_window))
1768 /*
1769 * Interrupts blocked. Wait for unblock.
1770 */
1771 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
1772 else
1773 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
1774 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
1775}
1776
1777static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
1778{
1779 int ret;
1780 struct kvm_userspace_memory_region tss_mem = {
1781 .slot = 8,
1782 .guest_phys_addr = addr,
1783 .memory_size = PAGE_SIZE * 3,
1784 .flags = 0,
1785 };
1786
1787 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
1788 if (ret)
1789 return ret;
1790 kvm->arch.tss_addr = addr;
1791 return 0;
1792}
1793
1794static void kvm_guest_debug_pre(struct kvm_vcpu *vcpu)
1795{
1796 struct kvm_guest_debug *dbg = &vcpu->guest_debug;
1797
1798 set_debugreg(dbg->bp[0], 0);
1799 set_debugreg(dbg->bp[1], 1);
1800 set_debugreg(dbg->bp[2], 2);
1801 set_debugreg(dbg->bp[3], 3);
1802
1803 if (dbg->singlestep) {
1804 unsigned long flags;
1805
1806 flags = vmcs_readl(GUEST_RFLAGS);
1807 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1808 vmcs_writel(GUEST_RFLAGS, flags);
1809 }
1810}
1811
1812static int handle_rmode_exception(struct kvm_vcpu *vcpu,
1813 int vec, u32 err_code)
1814{
1815 if (!vcpu->arch.rmode.active)
1816 return 0;
1817
1818 /*
1819 * Instruction with address size override prefix opcode 0x67
1820 * Cause the #SS fault with 0 error code in VM86 mode.
1821 */
1822 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
1823 if (emulate_instruction(vcpu, NULL, 0, 0, 0) == EMULATE_DONE)
1824 return 1;
1825 return 0;
1826}
1827
1828static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1829{
1830 struct vcpu_vmx *vmx = to_vmx(vcpu);
1831 u32 intr_info, error_code;
1832 unsigned long cr2, rip;
1833 u32 vect_info;
1834 enum emulation_result er;
1835
1836 vect_info = vmx->idt_vectoring_info;
1837 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
1838
1839 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
1840 !is_page_fault(intr_info))
1841 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
1842 "intr info 0x%x\n", __FUNCTION__, vect_info, intr_info);
1843
1844 if (!irqchip_in_kernel(vcpu->kvm) && is_external_interrupt(vect_info)) {
1845 int irq = vect_info & VECTORING_INFO_VECTOR_MASK;
1846 set_bit(irq, vcpu->arch.irq_pending);
1847 set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1848 }
1849
1850 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) /* nmi */
1851 return 1; /* already handled by vmx_vcpu_run() */
1852
1853 if (is_no_device(intr_info)) {
1854 vmx_fpu_activate(vcpu);
1855 return 1;
1856 }
1857
1858 if (is_invalid_opcode(intr_info)) {
1859 er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
1860 if (er != EMULATE_DONE)
1861 kvm_queue_exception(vcpu, UD_VECTOR);
1862 return 1;
1863 }
1864
1865 error_code = 0;
1866 rip = vmcs_readl(GUEST_RIP);
1867 if (intr_info & INTR_INFO_DELIEVER_CODE_MASK)
1868 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
1869 if (is_page_fault(intr_info)) {
1870 cr2 = vmcs_readl(EXIT_QUALIFICATION);
1871 return kvm_mmu_page_fault(vcpu, cr2, error_code);
1872 }
1873
1874 if (vcpu->arch.rmode.active &&
1875 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
1876 error_code)) {
1877 if (vcpu->arch.halt_request) {
1878 vcpu->arch.halt_request = 0;
1879 return kvm_emulate_halt(vcpu);
1880 }
1881 return 1;
1882 }
1883
1884 if ((intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK)) ==
1885 (INTR_TYPE_EXCEPTION | 1)) {
1886 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1887 return 0;
1888 }
1889 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
1890 kvm_run->ex.exception = intr_info & INTR_INFO_VECTOR_MASK;
1891 kvm_run->ex.error_code = error_code;
1892 return 0;
1893}
1894
1895static int handle_external_interrupt(struct kvm_vcpu *vcpu,
1896 struct kvm_run *kvm_run)
1897{
1898 ++vcpu->stat.irq_exits;
1899 return 1;
1900}
1901
1902static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1903{
1904 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1905 return 0;
1906}
1907
1908static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1909{
1910 unsigned long exit_qualification;
1911 int size, down, in, string, rep;
1912 unsigned port;
1913
1914 ++vcpu->stat.io_exits;
1915 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
1916 string = (exit_qualification & 16) != 0;
1917
1918 if (string) {
1919 if (emulate_instruction(vcpu,
1920 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
1921 return 0;
1922 return 1;
1923 }
1924
1925 size = (exit_qualification & 7) + 1;
1926 in = (exit_qualification & 8) != 0;
1927 down = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_DF) != 0;
1928 rep = (exit_qualification & 32) != 0;
1929 port = exit_qualification >> 16;
1930
1931 return kvm_emulate_pio(vcpu, kvm_run, in, size, port);
1932}
1933
1934static void
1935vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1936{
1937 /*
1938 * Patch in the VMCALL instruction:
1939 */
1940 hypercall[0] = 0x0f;
1941 hypercall[1] = 0x01;
1942 hypercall[2] = 0xc1;
1943}
1944
1945static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1946{
1947 unsigned long exit_qualification;
1948 int cr;
1949 int reg;
1950
1951 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
1952 cr = exit_qualification & 15;
1953 reg = (exit_qualification >> 8) & 15;
1954 switch ((exit_qualification >> 4) & 3) {
1955 case 0: /* mov to cr */
1956 switch (cr) {
1957 case 0:
1958 vcpu_load_rsp_rip(vcpu);
1959 set_cr0(vcpu, vcpu->arch.regs[reg]);
1960 skip_emulated_instruction(vcpu);
1961 return 1;
1962 case 3:
1963 vcpu_load_rsp_rip(vcpu);
1964 set_cr3(vcpu, vcpu->arch.regs[reg]);
1965 skip_emulated_instruction(vcpu);
1966 return 1;
1967 case 4:
1968 vcpu_load_rsp_rip(vcpu);
1969 set_cr4(vcpu, vcpu->arch.regs[reg]);
1970 skip_emulated_instruction(vcpu);
1971 return 1;
1972 case 8:
1973 vcpu_load_rsp_rip(vcpu);
1974 set_cr8(vcpu, vcpu->arch.regs[reg]);
1975 skip_emulated_instruction(vcpu);
1976 if (irqchip_in_kernel(vcpu->kvm))
1977 return 1;
1978 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
1979 return 0;
1980 };
1981 break;
1982 case 2: /* clts */
1983 vcpu_load_rsp_rip(vcpu);
1984 vmx_fpu_deactivate(vcpu);
1985 vcpu->arch.cr0 &= ~X86_CR0_TS;
1986 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1987 vmx_fpu_activate(vcpu);
1988 skip_emulated_instruction(vcpu);
1989 return 1;
1990 case 1: /*mov from cr*/
1991 switch (cr) {
1992 case 3:
1993 vcpu_load_rsp_rip(vcpu);
1994 vcpu->arch.regs[reg] = vcpu->arch.cr3;
1995 vcpu_put_rsp_rip(vcpu);
1996 skip_emulated_instruction(vcpu);
1997 return 1;
1998 case 8:
1999 vcpu_load_rsp_rip(vcpu);
2000 vcpu->arch.regs[reg] = get_cr8(vcpu);
2001 vcpu_put_rsp_rip(vcpu);
2002 skip_emulated_instruction(vcpu);
2003 return 1;
2004 }
2005 break;
2006 case 3: /* lmsw */
2007 lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
2008
2009 skip_emulated_instruction(vcpu);
2010 return 1;
2011 default:
2012 break;
2013 }
2014 kvm_run->exit_reason = 0;
2015 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
2016 (int)(exit_qualification >> 4) & 3, cr);
2017 return 0;
2018}
2019
2020static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2021{
2022 unsigned long exit_qualification;
2023 unsigned long val;
2024 int dr, reg;
2025
2026 /*
2027 * FIXME: this code assumes the host is debugging the guest.
2028 * need to deal with guest debugging itself too.
2029 */
2030 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2031 dr = exit_qualification & 7;
2032 reg = (exit_qualification >> 8) & 15;
2033 vcpu_load_rsp_rip(vcpu);
2034 if (exit_qualification & 16) {
2035 /* mov from dr */
2036 switch (dr) {
2037 case 6:
2038 val = 0xffff0ff0;
2039 break;
2040 case 7:
2041 val = 0x400;
2042 break;
2043 default:
2044 val = 0;
2045 }
2046 vcpu->arch.regs[reg] = val;
2047 } else {
2048 /* mov to dr */
2049 }
2050 vcpu_put_rsp_rip(vcpu);
2051 skip_emulated_instruction(vcpu);
2052 return 1;
2053}
2054
2055static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2056{
2057 kvm_emulate_cpuid(vcpu);
2058 return 1;
2059}
2060
2061static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2062{
2063 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2064 u64 data;
2065
2066 if (vmx_get_msr(vcpu, ecx, &data)) {
2067 kvm_inject_gp(vcpu, 0);
2068 return 1;
2069 }
2070
2071 /* FIXME: handling of bits 32:63 of rax, rdx */
2072 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
2073 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
2074 skip_emulated_instruction(vcpu);
2075 return 1;
2076}
2077
2078static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2079{
2080 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2081 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
2082 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2083
2084 if (vmx_set_msr(vcpu, ecx, data) != 0) {
2085 kvm_inject_gp(vcpu, 0);
2086 return 1;
2087 }
2088
2089 skip_emulated_instruction(vcpu);
2090 return 1;
2091}
2092
2093static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu,
2094 struct kvm_run *kvm_run)
2095{
2096 return 1;
2097}
2098
2099static int handle_interrupt_window(struct kvm_vcpu *vcpu,
2100 struct kvm_run *kvm_run)
2101{
2102 u32 cpu_based_vm_exec_control;
2103
2104 /* clear pending irq */
2105 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2106 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2107 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2108 /*
2109 * If the user space waits to inject interrupts, exit as soon as
2110 * possible
2111 */
2112 if (kvm_run->request_interrupt_window &&
2113 !vcpu->arch.irq_summary) {
2114 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2115 ++vcpu->stat.irq_window_exits;
2116 return 0;
2117 }
2118 return 1;
2119}
2120
2121static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2122{
2123 skip_emulated_instruction(vcpu);
2124 return kvm_emulate_halt(vcpu);
2125}
2126
2127static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2128{
2129 skip_emulated_instruction(vcpu);
2130 kvm_emulate_hypercall(vcpu);
2131 return 1;
2132}
2133
2134static int handle_wbinvd(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2135{
2136 skip_emulated_instruction(vcpu);
2137 /* TODO: Add support for VT-d/pass-through device */
2138 return 1;
2139}
2140
2141static int handle_apic_access(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2142{
2143 u64 exit_qualification;
2144 enum emulation_result er;
2145 unsigned long offset;
2146
2147 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
2148 offset = exit_qualification & 0xffful;
2149
2150 er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
2151
2152 if (er != EMULATE_DONE) {
2153 printk(KERN_ERR
2154 "Fail to handle apic access vmexit! Offset is 0x%lx\n",
2155 offset);
2156 return -ENOTSUPP;
2157 }
2158 return 1;
2159}
2160
2161/*
2162 * The exit handlers return 1 if the exit was handled fully and guest execution
2163 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
2164 * to be done to userspace and return 0.
2165 */
2166static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
2167 struct kvm_run *kvm_run) = {
2168 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
2169 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
2170 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
2171 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
2172 [EXIT_REASON_CR_ACCESS] = handle_cr,
2173 [EXIT_REASON_DR_ACCESS] = handle_dr,
2174 [EXIT_REASON_CPUID] = handle_cpuid,
2175 [EXIT_REASON_MSR_READ] = handle_rdmsr,
2176 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
2177 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
2178 [EXIT_REASON_HLT] = handle_halt,
2179 [EXIT_REASON_VMCALL] = handle_vmcall,
2180 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
2181 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
2182 [EXIT_REASON_WBINVD] = handle_wbinvd,
2183};
2184
2185static const int kvm_vmx_max_exit_handlers =
2186 ARRAY_SIZE(kvm_vmx_exit_handlers);
2187
2188/*
2189 * The guest has exited. See if we can fix it or if we need userspace
2190 * assistance.
2191 */
2192static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2193{
2194 u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
2195 struct vcpu_vmx *vmx = to_vmx(vcpu);
2196 u32 vectoring_info = vmx->idt_vectoring_info;
2197
2198 if (unlikely(vmx->fail)) {
2199 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2200 kvm_run->fail_entry.hardware_entry_failure_reason
2201 = vmcs_read32(VM_INSTRUCTION_ERROR);
2202 return 0;
2203 }
2204
2205 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
2206 exit_reason != EXIT_REASON_EXCEPTION_NMI)
2207 printk(KERN_WARNING "%s: unexpected, valid vectoring info and "
2208 "exit reason is 0x%x\n", __FUNCTION__, exit_reason);
2209 if (exit_reason < kvm_vmx_max_exit_handlers
2210 && kvm_vmx_exit_handlers[exit_reason])
2211 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
2212 else {
2213 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2214 kvm_run->hw.hardware_exit_reason = exit_reason;
2215 }
2216 return 0;
2217}
2218
2219static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
2220{
2221}
2222
2223static void update_tpr_threshold(struct kvm_vcpu *vcpu)
2224{
2225 int max_irr, tpr;
2226
2227 if (!vm_need_tpr_shadow(vcpu->kvm))
2228 return;
2229
2230 if (!kvm_lapic_enabled(vcpu) ||
2231 ((max_irr = kvm_lapic_find_highest_irr(vcpu)) == -1)) {
2232 vmcs_write32(TPR_THRESHOLD, 0);
2233 return;
2234 }
2235
2236 tpr = (kvm_lapic_get_cr8(vcpu) & 0x0f) << 4;
2237 vmcs_write32(TPR_THRESHOLD, (max_irr > tpr) ? tpr >> 4 : max_irr >> 4);
2238}
2239
2240static void enable_irq_window(struct kvm_vcpu *vcpu)
2241{
2242 u32 cpu_based_vm_exec_control;
2243
2244 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2245 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2246 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2247}
2248
2249static void vmx_intr_assist(struct kvm_vcpu *vcpu)
2250{
2251 struct vcpu_vmx *vmx = to_vmx(vcpu);
2252 u32 idtv_info_field, intr_info_field;
2253 int has_ext_irq, interrupt_window_open;
2254 int vector;
2255
2256 update_tpr_threshold(vcpu);
2257
2258 has_ext_irq = kvm_cpu_has_interrupt(vcpu);
2259 intr_info_field = vmcs_read32(VM_ENTRY_INTR_INFO_FIELD);
2260 idtv_info_field = vmx->idt_vectoring_info;
2261 if (intr_info_field & INTR_INFO_VALID_MASK) {
2262 if (idtv_info_field & INTR_INFO_VALID_MASK) {
2263 /* TODO: fault when IDT_Vectoring */
2264 if (printk_ratelimit())
2265 printk(KERN_ERR "Fault when IDT_Vectoring\n");
2266 }
2267 if (has_ext_irq)
2268 enable_irq_window(vcpu);
2269 return;
2270 }
2271 if (unlikely(idtv_info_field & INTR_INFO_VALID_MASK)) {
2272 if ((idtv_info_field & VECTORING_INFO_TYPE_MASK)
2273 == INTR_TYPE_EXT_INTR
2274 && vcpu->arch.rmode.active) {
2275 u8 vect = idtv_info_field & VECTORING_INFO_VECTOR_MASK;
2276
2277 vmx_inject_irq(vcpu, vect);
2278 if (unlikely(has_ext_irq))
2279 enable_irq_window(vcpu);
2280 return;
2281 }
2282
2283 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, idtv_info_field);
2284 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2285 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
2286
2287 if (unlikely(idtv_info_field & INTR_INFO_DELIEVER_CODE_MASK))
2288 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2289 vmcs_read32(IDT_VECTORING_ERROR_CODE));
2290 if (unlikely(has_ext_irq))
2291 enable_irq_window(vcpu);
2292 return;
2293 }
2294 if (!has_ext_irq)
2295 return;
2296 interrupt_window_open =
2297 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2298 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
2299 if (interrupt_window_open) {
2300 vector = kvm_cpu_get_interrupt(vcpu);
2301 vmx_inject_irq(vcpu, vector);
2302 kvm_timer_intr_post(vcpu, vector);
2303 } else
2304 enable_irq_window(vcpu);
2305}
2306
2307/*
2308 * Failure to inject an interrupt should give us the information
2309 * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs
2310 * when fetching the interrupt redirection bitmap in the real-mode
2311 * tss, this doesn't happen. So we do it ourselves.
2312 */
2313static void fixup_rmode_irq(struct vcpu_vmx *vmx)
2314{
2315 vmx->rmode.irq.pending = 0;
2316 if (vmcs_readl(GUEST_RIP) + 1 != vmx->rmode.irq.rip)
2317 return;
2318 vmcs_writel(GUEST_RIP, vmx->rmode.irq.rip);
2319 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
2320 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
2321 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
2322 return;
2323 }
2324 vmx->idt_vectoring_info =
2325 VECTORING_INFO_VALID_MASK
2326 | INTR_TYPE_EXT_INTR
2327 | vmx->rmode.irq.vector;
2328}
2329
2330static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2331{
2332 struct vcpu_vmx *vmx = to_vmx(vcpu);
2333 u32 intr_info;
2334
2335 /*
2336 * Loading guest fpu may have cleared host cr0.ts
2337 */
2338 vmcs_writel(HOST_CR0, read_cr0());
2339
2340 asm(
2341 /* Store host registers */
2342#ifdef CONFIG_X86_64
2343 "push %%rdx; push %%rbp;"
2344 "push %%rcx \n\t"
2345#else
2346 "push %%edx; push %%ebp;"
2347 "push %%ecx \n\t"
2348#endif
2349 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
2350 /* Check if vmlaunch of vmresume is needed */
2351 "cmpl $0, %c[launched](%0) \n\t"
2352 /* Load guest registers. Don't clobber flags. */
2353#ifdef CONFIG_X86_64
2354 "mov %c[cr2](%0), %%rax \n\t"
2355 "mov %%rax, %%cr2 \n\t"
2356 "mov %c[rax](%0), %%rax \n\t"
2357 "mov %c[rbx](%0), %%rbx \n\t"
2358 "mov %c[rdx](%0), %%rdx \n\t"
2359 "mov %c[rsi](%0), %%rsi \n\t"
2360 "mov %c[rdi](%0), %%rdi \n\t"
2361 "mov %c[rbp](%0), %%rbp \n\t"
2362 "mov %c[r8](%0), %%r8 \n\t"
2363 "mov %c[r9](%0), %%r9 \n\t"
2364 "mov %c[r10](%0), %%r10 \n\t"
2365 "mov %c[r11](%0), %%r11 \n\t"
2366 "mov %c[r12](%0), %%r12 \n\t"
2367 "mov %c[r13](%0), %%r13 \n\t"
2368 "mov %c[r14](%0), %%r14 \n\t"
2369 "mov %c[r15](%0), %%r15 \n\t"
2370 "mov %c[rcx](%0), %%rcx \n\t" /* kills %0 (rcx) */
2371#else
2372 "mov %c[cr2](%0), %%eax \n\t"
2373 "mov %%eax, %%cr2 \n\t"
2374 "mov %c[rax](%0), %%eax \n\t"
2375 "mov %c[rbx](%0), %%ebx \n\t"
2376 "mov %c[rdx](%0), %%edx \n\t"
2377 "mov %c[rsi](%0), %%esi \n\t"
2378 "mov %c[rdi](%0), %%edi \n\t"
2379 "mov %c[rbp](%0), %%ebp \n\t"
2380 "mov %c[rcx](%0), %%ecx \n\t" /* kills %0 (ecx) */
2381#endif
2382 /* Enter guest mode */
2383 "jne .Llaunched \n\t"
2384 ASM_VMX_VMLAUNCH "\n\t"
2385 "jmp .Lkvm_vmx_return \n\t"
2386 ".Llaunched: " ASM_VMX_VMRESUME "\n\t"
2387 ".Lkvm_vmx_return: "
2388 /* Save guest registers, load host registers, keep flags */
2389#ifdef CONFIG_X86_64
2390 "xchg %0, (%%rsp) \n\t"
2391 "mov %%rax, %c[rax](%0) \n\t"
2392 "mov %%rbx, %c[rbx](%0) \n\t"
2393 "pushq (%%rsp); popq %c[rcx](%0) \n\t"
2394 "mov %%rdx, %c[rdx](%0) \n\t"
2395 "mov %%rsi, %c[rsi](%0) \n\t"
2396 "mov %%rdi, %c[rdi](%0) \n\t"
2397 "mov %%rbp, %c[rbp](%0) \n\t"
2398 "mov %%r8, %c[r8](%0) \n\t"
2399 "mov %%r9, %c[r9](%0) \n\t"
2400 "mov %%r10, %c[r10](%0) \n\t"
2401 "mov %%r11, %c[r11](%0) \n\t"
2402 "mov %%r12, %c[r12](%0) \n\t"
2403 "mov %%r13, %c[r13](%0) \n\t"
2404 "mov %%r14, %c[r14](%0) \n\t"
2405 "mov %%r15, %c[r15](%0) \n\t"
2406 "mov %%cr2, %%rax \n\t"
2407 "mov %%rax, %c[cr2](%0) \n\t"
2408
2409 "pop %%rbp; pop %%rbp; pop %%rdx \n\t"
2410#else
2411 "xchg %0, (%%esp) \n\t"
2412 "mov %%eax, %c[rax](%0) \n\t"
2413 "mov %%ebx, %c[rbx](%0) \n\t"
2414 "pushl (%%esp); popl %c[rcx](%0) \n\t"
2415 "mov %%edx, %c[rdx](%0) \n\t"
2416 "mov %%esi, %c[rsi](%0) \n\t"
2417 "mov %%edi, %c[rdi](%0) \n\t"
2418 "mov %%ebp, %c[rbp](%0) \n\t"
2419 "mov %%cr2, %%eax \n\t"
2420 "mov %%eax, %c[cr2](%0) \n\t"
2421
2422 "pop %%ebp; pop %%ebp; pop %%edx \n\t"
2423#endif
2424 "setbe %c[fail](%0) \n\t"
2425 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
2426 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
2427 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
2428 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
2429 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
2430 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
2431 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
2432 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
2433 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
2434 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
2435#ifdef CONFIG_X86_64
2436 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
2437 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
2438 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
2439 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
2440 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
2441 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
2442 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
2443 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
2444#endif
2445 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
2446 : "cc", "memory"
2447#ifdef CONFIG_X86_64
2448 , "rbx", "rdi", "rsi"
2449 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
2450#else
2451 , "ebx", "edi", "rsi"
2452#endif
2453 );
2454
2455 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
2456 if (vmx->rmode.irq.pending)
2457 fixup_rmode_irq(vmx);
2458
2459 vcpu->arch.interrupt_window_open =
2460 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0;
2461
2462 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
2463 vmx->launched = 1;
2464
2465 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2466
2467 /* We need to handle NMIs before interrupts are enabled */
2468 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) /* nmi */
2469 asm("int $2");
2470}
2471
2472static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
2473{
2474 struct vcpu_vmx *vmx = to_vmx(vcpu);
2475
2476 if (vmx->vmcs) {
2477 on_each_cpu(__vcpu_clear, vmx, 0, 1);
2478 free_vmcs(vmx->vmcs);
2479 vmx->vmcs = NULL;
2480 }
2481}
2482
2483static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
2484{
2485 struct vcpu_vmx *vmx = to_vmx(vcpu);
2486
2487 vmx_free_vmcs(vcpu);
2488 kfree(vmx->host_msrs);
2489 kfree(vmx->guest_msrs);
2490 kvm_vcpu_uninit(vcpu);
2491 kmem_cache_free(kvm_vcpu_cache, vmx);
2492}
2493
2494static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
2495{
2496 int err;
2497 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2498 int cpu;
2499
2500 if (!vmx)
2501 return ERR_PTR(-ENOMEM);
2502
2503 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
2504 if (err)
2505 goto free_vcpu;
2506
2507 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2508 if (!vmx->guest_msrs) {
2509 err = -ENOMEM;
2510 goto uninit_vcpu;
2511 }
2512
2513 vmx->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2514 if (!vmx->host_msrs)
2515 goto free_guest_msrs;
2516
2517 vmx->vmcs = alloc_vmcs();
2518 if (!vmx->vmcs)
2519 goto free_msrs;
2520
2521 vmcs_clear(vmx->vmcs);
2522
2523 cpu = get_cpu();
2524 vmx_vcpu_load(&vmx->vcpu, cpu);
2525 err = vmx_vcpu_setup(vmx);
2526 vmx_vcpu_put(&vmx->vcpu);
2527 put_cpu();
2528 if (err)
2529 goto free_vmcs;
2530
2531 return &vmx->vcpu;
2532
2533free_vmcs:
2534 free_vmcs(vmx->vmcs);
2535free_msrs:
2536 kfree(vmx->host_msrs);
2537free_guest_msrs:
2538 kfree(vmx->guest_msrs);
2539uninit_vcpu:
2540 kvm_vcpu_uninit(&vmx->vcpu);
2541free_vcpu:
2542 kmem_cache_free(kvm_vcpu_cache, vmx);
2543 return ERR_PTR(err);
2544}
2545
2546static void __init vmx_check_processor_compat(void *rtn)
2547{
2548 struct vmcs_config vmcs_conf;
2549
2550 *(int *)rtn = 0;
2551 if (setup_vmcs_config(&vmcs_conf) < 0)
2552 *(int *)rtn = -EIO;
2553 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
2554 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
2555 smp_processor_id());
2556 *(int *)rtn = -EIO;
2557 }
2558}
2559
2560static struct kvm_x86_ops vmx_x86_ops = {
2561 .cpu_has_kvm_support = cpu_has_kvm_support,
2562 .disabled_by_bios = vmx_disabled_by_bios,
2563 .hardware_setup = hardware_setup,
2564 .hardware_unsetup = hardware_unsetup,
2565 .check_processor_compatibility = vmx_check_processor_compat,
2566 .hardware_enable = hardware_enable,
2567 .hardware_disable = hardware_disable,
2568
2569 .vcpu_create = vmx_create_vcpu,
2570 .vcpu_free = vmx_free_vcpu,
2571 .vcpu_reset = vmx_vcpu_reset,
2572
2573 .prepare_guest_switch = vmx_save_host_state,
2574 .vcpu_load = vmx_vcpu_load,
2575 .vcpu_put = vmx_vcpu_put,
2576 .vcpu_decache = vmx_vcpu_decache,
2577
2578 .set_guest_debug = set_guest_debug,
2579 .guest_debug_pre = kvm_guest_debug_pre,
2580 .get_msr = vmx_get_msr,
2581 .set_msr = vmx_set_msr,
2582 .get_segment_base = vmx_get_segment_base,
2583 .get_segment = vmx_get_segment,
2584 .set_segment = vmx_set_segment,
2585 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
2586 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
2587 .set_cr0 = vmx_set_cr0,
2588 .set_cr3 = vmx_set_cr3,
2589 .set_cr4 = vmx_set_cr4,
2590#ifdef CONFIG_X86_64
2591 .set_efer = vmx_set_efer,
2592#endif
2593 .get_idt = vmx_get_idt,
2594 .set_idt = vmx_set_idt,
2595 .get_gdt = vmx_get_gdt,
2596 .set_gdt = vmx_set_gdt,
2597 .cache_regs = vcpu_load_rsp_rip,
2598 .decache_regs = vcpu_put_rsp_rip,
2599 .get_rflags = vmx_get_rflags,
2600 .set_rflags = vmx_set_rflags,
2601
2602 .tlb_flush = vmx_flush_tlb,
2603
2604 .run = vmx_vcpu_run,
2605 .handle_exit = kvm_handle_exit,
2606 .skip_emulated_instruction = skip_emulated_instruction,
2607 .patch_hypercall = vmx_patch_hypercall,
2608 .get_irq = vmx_get_irq,
2609 .set_irq = vmx_inject_irq,
2610 .queue_exception = vmx_queue_exception,
2611 .exception_injected = vmx_exception_injected,
2612 .inject_pending_irq = vmx_intr_assist,
2613 .inject_pending_vectors = do_interrupt_requests,
2614
2615 .set_tss_addr = vmx_set_tss_addr,
2616};
2617
2618static int __init vmx_init(void)
2619{
2620 void *iova;
2621 int r;
2622
2623 vmx_io_bitmap_a = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2624 if (!vmx_io_bitmap_a)
2625 return -ENOMEM;
2626
2627 vmx_io_bitmap_b = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2628 if (!vmx_io_bitmap_b) {
2629 r = -ENOMEM;
2630 goto out;
2631 }
2632
2633 /*
2634 * Allow direct access to the PC debug port (it is often used for I/O
2635 * delays, but the vmexits simply slow things down).
2636 */
2637 iova = kmap(vmx_io_bitmap_a);
2638 memset(iova, 0xff, PAGE_SIZE);
2639 clear_bit(0x80, iova);
2640 kunmap(vmx_io_bitmap_a);
2641
2642 iova = kmap(vmx_io_bitmap_b);
2643 memset(iova, 0xff, PAGE_SIZE);
2644 kunmap(vmx_io_bitmap_b);
2645
2646 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
2647 if (r)
2648 goto out1;
2649
2650 if (bypass_guest_pf)
2651 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
2652
2653 return 0;
2654
2655out1:
2656 __free_page(vmx_io_bitmap_b);
2657out:
2658 __free_page(vmx_io_bitmap_a);
2659 return r;
2660}
2661
2662static void __exit vmx_exit(void)
2663{
2664 __free_page(vmx_io_bitmap_b);
2665 __free_page(vmx_io_bitmap_a);
2666
2667 kvm_exit();
2668}
2669
2670module_init(vmx_init)
2671module_exit(vmx_exit)
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-28 03:57:20 -0500