2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
21 #include <linux/module.h>
23 #include <linux/highmem.h>
27 #include "segment_descriptor.h"
29 #define MSR_IA32_FEATURE_CONTROL 0x03a
31 MODULE_AUTHOR("Qumranet");
32 MODULE_LICENSE("GPL");
34 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
35 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
43 static struct vmcs_descriptor {
49 #define VMX_SEGMENT_FIELD(seg) \
50 [VCPU_SREG_##seg] = { \
51 .selector = GUEST_##seg##_SELECTOR, \
52 .base = GUEST_##seg##_BASE, \
53 .limit = GUEST_##seg##_LIMIT, \
54 .ar_bytes = GUEST_##seg##_AR_BYTES, \
57 static struct kvm_vmx_segment_field {
62 } kvm_vmx_segment_fields[] = {
63 VMX_SEGMENT_FIELD(CS),
64 VMX_SEGMENT_FIELD(DS),
65 VMX_SEGMENT_FIELD(ES),
66 VMX_SEGMENT_FIELD(FS),
67 VMX_SEGMENT_FIELD(GS),
68 VMX_SEGMENT_FIELD(SS),
69 VMX_SEGMENT_FIELD(TR),
70 VMX_SEGMENT_FIELD(LDTR),
73 static const u32 vmx_msr_index[] = {
75 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
77 MSR_EFER, MSR_K6_STAR,
79 #define NR_VMX_MSR (sizeof(vmx_msr_index) / sizeof(*vmx_msr_index))
81 struct vmx_msr_entry *find_msr_entry(struct kvm_vcpu *vcpu, u32 msr);
83 static inline int is_page_fault(u32 intr_info)
85 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
86 INTR_INFO_VALID_MASK)) ==
87 (INTR_TYPE_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
90 static inline int is_external_interrupt(u32 intr_info)
92 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
93 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
96 static void vmcs_clear(struct vmcs *vmcs)
98 u64 phys_addr = __pa(vmcs);
101 asm volatile (ASM_VMX_VMCLEAR_RAX "; setna %0"
102 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
105 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
109 static void __vcpu_clear(void *arg)
111 struct kvm_vcpu *vcpu = arg;
112 int cpu = smp_processor_id();
114 if (vcpu->cpu == cpu)
115 vmcs_clear(vcpu->vmcs);
116 if (per_cpu(current_vmcs, cpu) == vcpu->vmcs)
117 per_cpu(current_vmcs, cpu) = NULL;
120 static unsigned long vmcs_readl(unsigned long field)
124 asm volatile (ASM_VMX_VMREAD_RDX_RAX
125 : "=a"(value) : "d"(field) : "cc");
129 static u16 vmcs_read16(unsigned long field)
131 return vmcs_readl(field);
134 static u32 vmcs_read32(unsigned long field)
136 return vmcs_readl(field);
139 static u64 vmcs_read64(unsigned long field)
142 return vmcs_readl(field);
144 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
148 static void vmcs_writel(unsigned long field, unsigned long value)
152 asm volatile (ASM_VMX_VMWRITE_RAX_RDX "; setna %0"
153 : "=q"(error) : "a"(value), "d"(field) : "cc" );
155 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
156 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
159 static void vmcs_write16(unsigned long field, u16 value)
161 vmcs_writel(field, value);
164 static void vmcs_write32(unsigned long field, u32 value)
166 vmcs_writel(field, value);
169 static void vmcs_write64(unsigned long field, u64 value)
172 vmcs_writel(field, value);
174 vmcs_writel(field, value);
176 vmcs_writel(field+1, value >> 32);
181 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
182 * vcpu mutex is already taken.
184 static struct kvm_vcpu *vmx_vcpu_load(struct kvm_vcpu *vcpu)
186 u64 phys_addr = __pa(vcpu->vmcs);
191 if (vcpu->cpu != cpu) {
192 smp_call_function(__vcpu_clear, vcpu, 0, 1);
196 if (per_cpu(current_vmcs, cpu) != vcpu->vmcs) {
199 per_cpu(current_vmcs, cpu) = vcpu->vmcs;
200 asm volatile (ASM_VMX_VMPTRLD_RAX "; setna %0"
201 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
204 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
205 vcpu->vmcs, phys_addr);
208 if (vcpu->cpu != cpu) {
209 struct descriptor_table dt;
210 unsigned long sysenter_esp;
214 * Linux uses per-cpu TSS and GDT, so set these when switching
217 vmcs_writel(HOST_TR_BASE, read_tr_base()); /* 22.2.4 */
219 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
221 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
222 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
227 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
232 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
234 return vmcs_readl(GUEST_RFLAGS);
237 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
239 vmcs_writel(GUEST_RFLAGS, rflags);
242 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
245 u32 interruptibility;
247 rip = vmcs_readl(GUEST_RIP);
248 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
249 vmcs_writel(GUEST_RIP, rip);
252 * We emulated an instruction, so temporary interrupt blocking
253 * should be removed, if set.
255 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
256 if (interruptibility & 3)
257 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
258 interruptibility & ~3);
261 static void vmx_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
263 printk(KERN_DEBUG "inject_general_protection: rip 0x%lx\n",
264 vmcs_readl(GUEST_RIP));
265 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
266 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
268 INTR_TYPE_EXCEPTION |
269 INTR_INFO_DELIEVER_CODE_MASK |
270 INTR_INFO_VALID_MASK);
274 * reads and returns guest's timestamp counter "register"
275 * guest_tsc = host_tsc + tsc_offset -- 21.3
277 static u64 guest_read_tsc(void)
279 u64 host_tsc, tsc_offset;
282 tsc_offset = vmcs_read64(TSC_OFFSET);
283 return host_tsc + tsc_offset;
287 * writes 'guest_tsc' into guest's timestamp counter "register"
288 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
290 static void guest_write_tsc(u64 guest_tsc)
295 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
298 static void reload_tss(void)
300 #ifndef CONFIG_X86_64
303 * VT restores TR but not its size. Useless.
305 struct descriptor_table gdt;
306 struct segment_descriptor *descs;
309 descs = (void *)gdt.base;
310 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
316 * Reads an msr value (of 'msr_index') into 'pdata'.
317 * Returns 0 on success, non-0 otherwise.
318 * Assumes vcpu_load() was already called.
320 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
323 struct vmx_msr_entry *msr;
326 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
333 data = vmcs_readl(GUEST_FS_BASE);
336 data = vmcs_readl(GUEST_GS_BASE);
339 data = vcpu->shadow_efer;
342 case MSR_IA32_TIME_STAMP_COUNTER:
343 data = guest_read_tsc();
345 case MSR_IA32_SYSENTER_CS:
346 data = vmcs_read32(GUEST_SYSENTER_CS);
348 case MSR_IA32_SYSENTER_EIP:
349 data = vmcs_read32(GUEST_SYSENTER_EIP);
351 case MSR_IA32_SYSENTER_ESP:
352 data = vmcs_read32(GUEST_SYSENTER_ESP);
354 case MSR_IA32_MC0_CTL:
355 case MSR_IA32_MCG_STATUS:
356 case MSR_IA32_MCG_CAP:
357 case MSR_IA32_MC0_MISC:
358 case MSR_IA32_MC0_MISC+4:
359 case MSR_IA32_MC0_MISC+8:
360 case MSR_IA32_MC0_MISC+12:
361 case MSR_IA32_MC0_MISC+16:
362 case MSR_IA32_UCODE_REV:
365 case 0x200 ... 0x2ff:
368 case MSR_IA32_APICBASE:
369 data = vcpu->apic_base;
372 msr = find_msr_entry(vcpu, msr_index);
374 printk(KERN_ERR "kvm: unhandled rdmsr: %x\n", msr_index);
386 * Writes msr value into into the appropriate "register".
387 * Returns 0 on success, non-0 otherwise.
388 * Assumes vcpu_load() was already called.
390 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
392 struct vmx_msr_entry *msr;
396 vmcs_writel(GUEST_FS_BASE, data);
399 vmcs_writel(GUEST_GS_BASE, data);
402 case MSR_IA32_SYSENTER_CS:
403 vmcs_write32(GUEST_SYSENTER_CS, data);
405 case MSR_IA32_SYSENTER_EIP:
406 vmcs_write32(GUEST_SYSENTER_EIP, data);
408 case MSR_IA32_SYSENTER_ESP:
409 vmcs_write32(GUEST_SYSENTER_ESP, data);
413 set_efer(vcpu, data);
415 case MSR_IA32_MC0_STATUS:
416 printk(KERN_WARNING "%s: MSR_IA32_MC0_STATUS 0x%llx, nop\n"
417 , __FUNCTION__, data);
420 case MSR_IA32_TIME_STAMP_COUNTER: {
421 guest_write_tsc(data);
424 case MSR_IA32_UCODE_REV:
425 case MSR_IA32_UCODE_WRITE:
426 case 0x200 ... 0x2ff: /* MTRRs */
428 case MSR_IA32_APICBASE:
429 vcpu->apic_base = data;
432 msr = find_msr_entry(vcpu, msr_index);
434 printk(KERN_ERR "kvm: unhandled wrmsr: 0x%x\n", msr_index);
445 * Sync the rsp and rip registers into the vcpu structure. This allows
446 * registers to be accessed by indexing vcpu->regs.
448 static void vcpu_load_rsp_rip(struct kvm_vcpu *vcpu)
450 vcpu->regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
451 vcpu->rip = vmcs_readl(GUEST_RIP);
455 * Syncs rsp and rip back into the vmcs. Should be called after possible
458 static void vcpu_put_rsp_rip(struct kvm_vcpu *vcpu)
460 vmcs_writel(GUEST_RSP, vcpu->regs[VCPU_REGS_RSP]);
461 vmcs_writel(GUEST_RIP, vcpu->rip);
464 static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
466 unsigned long dr7 = 0x400;
467 u32 exception_bitmap;
470 exception_bitmap = vmcs_read32(EXCEPTION_BITMAP);
471 old_singlestep = vcpu->guest_debug.singlestep;
473 vcpu->guest_debug.enabled = dbg->enabled;
474 if (vcpu->guest_debug.enabled) {
477 dr7 |= 0x200; /* exact */
478 for (i = 0; i < 4; ++i) {
479 if (!dbg->breakpoints[i].enabled)
481 vcpu->guest_debug.bp[i] = dbg->breakpoints[i].address;
482 dr7 |= 2 << (i*2); /* global enable */
483 dr7 |= 0 << (i*4+16); /* execution breakpoint */
486 exception_bitmap |= (1u << 1); /* Trap debug exceptions */
488 vcpu->guest_debug.singlestep = dbg->singlestep;
490 exception_bitmap &= ~(1u << 1); /* Ignore debug exceptions */
491 vcpu->guest_debug.singlestep = 0;
494 if (old_singlestep && !vcpu->guest_debug.singlestep) {
497 flags = vmcs_readl(GUEST_RFLAGS);
498 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
499 vmcs_writel(GUEST_RFLAGS, flags);
502 vmcs_write32(EXCEPTION_BITMAP, exception_bitmap);
503 vmcs_writel(GUEST_DR7, dr7);
508 static __init int cpu_has_kvm_support(void)
510 unsigned long ecx = cpuid_ecx(1);
511 return test_bit(5, &ecx); /* CPUID.1:ECX.VMX[bit 5] -> VT */
514 static __init int vmx_disabled_by_bios(void)
518 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
519 return (msr & 5) == 1; /* locked but not enabled */
522 static __init void hardware_enable(void *garbage)
524 int cpu = raw_smp_processor_id();
525 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
528 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
530 /* enable and lock */
531 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | 5);
532 write_cr4(read_cr4() | CR4_VMXE); /* FIXME: not cpu hotplug safe */
533 asm volatile (ASM_VMX_VMXON_RAX : : "a"(&phys_addr), "m"(phys_addr)
537 static void hardware_disable(void *garbage)
539 asm volatile (ASM_VMX_VMXOFF : : : "cc");
542 static __init void setup_vmcs_descriptor(void)
544 u32 vmx_msr_low, vmx_msr_high;
546 rdmsr(MSR_IA32_VMX_BASIC_MSR, vmx_msr_low, vmx_msr_high);
547 vmcs_descriptor.size = vmx_msr_high & 0x1fff;
548 vmcs_descriptor.order = get_order(vmcs_descriptor.size);
549 vmcs_descriptor.revision_id = vmx_msr_low;
552 static struct vmcs *alloc_vmcs_cpu(int cpu)
554 int node = cpu_to_node(cpu);
558 pages = alloc_pages_node(node, GFP_KERNEL, vmcs_descriptor.order);
561 vmcs = page_address(pages);
562 memset(vmcs, 0, vmcs_descriptor.size);
563 vmcs->revision_id = vmcs_descriptor.revision_id; /* vmcs revision id */
567 static struct vmcs *alloc_vmcs(void)
569 return alloc_vmcs_cpu(smp_processor_id());
572 static void free_vmcs(struct vmcs *vmcs)
574 free_pages((unsigned long)vmcs, vmcs_descriptor.order);
577 static __exit void free_kvm_area(void)
581 for_each_online_cpu(cpu)
582 free_vmcs(per_cpu(vmxarea, cpu));
585 extern struct vmcs *alloc_vmcs_cpu(int cpu);
587 static __init int alloc_kvm_area(void)
591 for_each_online_cpu(cpu) {
594 vmcs = alloc_vmcs_cpu(cpu);
600 per_cpu(vmxarea, cpu) = vmcs;
605 static __init int hardware_setup(void)
607 setup_vmcs_descriptor();
608 return alloc_kvm_area();
611 static __exit void hardware_unsetup(void)
616 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
618 if (vcpu->rmode.active)
619 vmcs_write32(EXCEPTION_BITMAP, ~0);
621 vmcs_write32(EXCEPTION_BITMAP, 1 << PF_VECTOR);
624 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
626 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
628 if (vmcs_readl(sf->base) == save->base) {
629 vmcs_write16(sf->selector, save->selector);
630 vmcs_writel(sf->base, save->base);
631 vmcs_write32(sf->limit, save->limit);
632 vmcs_write32(sf->ar_bytes, save->ar);
634 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
636 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
640 static void enter_pmode(struct kvm_vcpu *vcpu)
644 vcpu->rmode.active = 0;
646 vmcs_writel(GUEST_TR_BASE, vcpu->rmode.tr.base);
647 vmcs_write32(GUEST_TR_LIMIT, vcpu->rmode.tr.limit);
648 vmcs_write32(GUEST_TR_AR_BYTES, vcpu->rmode.tr.ar);
650 flags = vmcs_readl(GUEST_RFLAGS);
651 flags &= ~(IOPL_MASK | X86_EFLAGS_VM);
652 flags |= (vcpu->rmode.save_iopl << IOPL_SHIFT);
653 vmcs_writel(GUEST_RFLAGS, flags);
655 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~CR4_VME_MASK) |
656 (vmcs_readl(CR4_READ_SHADOW) & CR4_VME_MASK));
658 update_exception_bitmap(vcpu);
660 fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->rmode.es);
661 fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->rmode.ds);
662 fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->rmode.gs);
663 fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->rmode.fs);
665 vmcs_write16(GUEST_SS_SELECTOR, 0);
666 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
668 vmcs_write16(GUEST_CS_SELECTOR,
669 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
670 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
673 static int rmode_tss_base(struct kvm* kvm)
675 gfn_t base_gfn = kvm->memslots[0].base_gfn + kvm->memslots[0].npages - 3;
676 return base_gfn << PAGE_SHIFT;
679 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
681 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
683 save->selector = vmcs_read16(sf->selector);
684 save->base = vmcs_readl(sf->base);
685 save->limit = vmcs_read32(sf->limit);
686 save->ar = vmcs_read32(sf->ar_bytes);
687 vmcs_write16(sf->selector, vmcs_readl(sf->base) >> 4);
688 vmcs_write32(sf->limit, 0xffff);
689 vmcs_write32(sf->ar_bytes, 0xf3);
692 static void enter_rmode(struct kvm_vcpu *vcpu)
696 vcpu->rmode.active = 1;
698 vcpu->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
699 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
701 vcpu->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
702 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
704 vcpu->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
705 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
707 flags = vmcs_readl(GUEST_RFLAGS);
708 vcpu->rmode.save_iopl = (flags & IOPL_MASK) >> IOPL_SHIFT;
710 flags |= IOPL_MASK | X86_EFLAGS_VM;
712 vmcs_writel(GUEST_RFLAGS, flags);
713 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | CR4_VME_MASK);
714 update_exception_bitmap(vcpu);
716 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
717 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
718 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
720 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
721 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
723 fix_rmode_seg(VCPU_SREG_ES, &vcpu->rmode.es);
724 fix_rmode_seg(VCPU_SREG_DS, &vcpu->rmode.ds);
725 fix_rmode_seg(VCPU_SREG_GS, &vcpu->rmode.gs);
726 fix_rmode_seg(VCPU_SREG_FS, &vcpu->rmode.fs);
731 static void enter_lmode(struct kvm_vcpu *vcpu)
735 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
736 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
737 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
739 vmcs_write32(GUEST_TR_AR_BYTES,
740 (guest_tr_ar & ~AR_TYPE_MASK)
741 | AR_TYPE_BUSY_64_TSS);
744 vcpu->shadow_efer |= EFER_LMA;
746 find_msr_entry(vcpu, MSR_EFER)->data |= EFER_LMA | EFER_LME;
747 vmcs_write32(VM_ENTRY_CONTROLS,
748 vmcs_read32(VM_ENTRY_CONTROLS)
749 | VM_ENTRY_CONTROLS_IA32E_MASK);
752 static void exit_lmode(struct kvm_vcpu *vcpu)
754 vcpu->shadow_efer &= ~EFER_LMA;
756 vmcs_write32(VM_ENTRY_CONTROLS,
757 vmcs_read32(VM_ENTRY_CONTROLS)
758 & ~VM_ENTRY_CONTROLS_IA32E_MASK);
763 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
765 if (vcpu->rmode.active && (cr0 & CR0_PE_MASK))
768 if (!vcpu->rmode.active && !(cr0 & CR0_PE_MASK))
772 if (vcpu->shadow_efer & EFER_LME) {
773 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK))
775 if (is_paging(vcpu) && !(cr0 & CR0_PG_MASK))
780 vmcs_writel(CR0_READ_SHADOW, cr0);
781 vmcs_writel(GUEST_CR0,
782 (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON);
787 * Used when restoring the VM to avoid corrupting segment registers
789 static void vmx_set_cr0_no_modeswitch(struct kvm_vcpu *vcpu, unsigned long cr0)
791 vcpu->rmode.active = ((cr0 & CR0_PE_MASK) == 0);
792 update_exception_bitmap(vcpu);
793 vmcs_writel(CR0_READ_SHADOW, cr0);
794 vmcs_writel(GUEST_CR0,
795 (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON);
799 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
801 vmcs_writel(GUEST_CR3, cr3);
804 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
806 vmcs_writel(CR4_READ_SHADOW, cr4);
807 vmcs_writel(GUEST_CR4, cr4 | (vcpu->rmode.active ?
808 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON));
814 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
816 struct vmx_msr_entry *msr = find_msr_entry(vcpu, MSR_EFER);
818 vcpu->shadow_efer = efer;
819 if (efer & EFER_LMA) {
820 vmcs_write32(VM_ENTRY_CONTROLS,
821 vmcs_read32(VM_ENTRY_CONTROLS) |
822 VM_ENTRY_CONTROLS_IA32E_MASK);
826 vmcs_write32(VM_ENTRY_CONTROLS,
827 vmcs_read32(VM_ENTRY_CONTROLS) &
828 ~VM_ENTRY_CONTROLS_IA32E_MASK);
830 msr->data = efer & ~EFER_LME;
836 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
838 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
840 return vmcs_readl(sf->base);
843 static void vmx_get_segment(struct kvm_vcpu *vcpu,
844 struct kvm_segment *var, int seg)
846 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
849 var->base = vmcs_readl(sf->base);
850 var->limit = vmcs_read32(sf->limit);
851 var->selector = vmcs_read16(sf->selector);
852 ar = vmcs_read32(sf->ar_bytes);
853 if (ar & AR_UNUSABLE_MASK)
856 var->s = (ar >> 4) & 1;
857 var->dpl = (ar >> 5) & 3;
858 var->present = (ar >> 7) & 1;
859 var->avl = (ar >> 12) & 1;
860 var->l = (ar >> 13) & 1;
861 var->db = (ar >> 14) & 1;
862 var->g = (ar >> 15) & 1;
863 var->unusable = (ar >> 16) & 1;
866 static void vmx_set_segment(struct kvm_vcpu *vcpu,
867 struct kvm_segment *var, int seg)
869 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
872 vmcs_writel(sf->base, var->base);
873 vmcs_write32(sf->limit, var->limit);
874 vmcs_write16(sf->selector, var->selector);
879 ar |= (var->s & 1) << 4;
880 ar |= (var->dpl & 3) << 5;
881 ar |= (var->present & 1) << 7;
882 ar |= (var->avl & 1) << 12;
883 ar |= (var->l & 1) << 13;
884 ar |= (var->db & 1) << 14;
885 ar |= (var->g & 1) << 15;
887 if (ar == 0) /* a 0 value means unusable */
888 ar = AR_UNUSABLE_MASK;
889 vmcs_write32(sf->ar_bytes, ar);
892 static int vmx_is_long_mode(struct kvm_vcpu *vcpu)
894 return vmcs_read32(VM_ENTRY_CONTROLS) & VM_ENTRY_CONTROLS_IA32E_MASK;
897 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
899 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
901 *db = (ar >> 14) & 1;
905 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
907 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
908 dt->base = vmcs_readl(GUEST_IDTR_BASE);
911 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
913 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
914 vmcs_writel(GUEST_IDTR_BASE, dt->base);
917 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
919 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
920 dt->base = vmcs_readl(GUEST_GDTR_BASE);
923 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
925 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
926 vmcs_writel(GUEST_GDTR_BASE, dt->base);
929 static int init_rmode_tss(struct kvm* kvm)
931 struct page *p1, *p2, *p3;
932 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
935 p1 = _gfn_to_page(kvm, fn++);
936 p2 = _gfn_to_page(kvm, fn++);
937 p3 = _gfn_to_page(kvm, fn);
939 if (!p1 || !p2 || !p3) {
940 kvm_printf(kvm,"%s: gfn_to_page failed\n", __FUNCTION__);
944 page = kmap_atomic(p1, KM_USER0);
945 memset(page, 0, PAGE_SIZE);
946 *(u16*)(page + 0x66) = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
947 kunmap_atomic(page, KM_USER0);
949 page = kmap_atomic(p2, KM_USER0);
950 memset(page, 0, PAGE_SIZE);
951 kunmap_atomic(page, KM_USER0);
953 page = kmap_atomic(p3, KM_USER0);
954 memset(page, 0, PAGE_SIZE);
955 *(page + RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1) = ~0;
956 kunmap_atomic(page, KM_USER0);
961 static void vmcs_write32_fixedbits(u32 msr, u32 vmcs_field, u32 val)
963 u32 msr_high, msr_low;
965 rdmsr(msr, msr_low, msr_high);
969 vmcs_write32(vmcs_field, val);
972 static void seg_setup(int seg)
974 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
976 vmcs_write16(sf->selector, 0);
977 vmcs_writel(sf->base, 0);
978 vmcs_write32(sf->limit, 0xffff);
979 vmcs_write32(sf->ar_bytes, 0x93);
983 * Sets up the vmcs for emulated real mode.
985 static int vmx_vcpu_setup(struct kvm_vcpu *vcpu)
987 u32 host_sysenter_cs;
990 struct descriptor_table dt;
994 extern asmlinkage void kvm_vmx_return(void);
996 if (!init_rmode_tss(vcpu->kvm)) {
1001 memset(vcpu->regs, 0, sizeof(vcpu->regs));
1002 vcpu->regs[VCPU_REGS_RDX] = get_rdx_init_val();
1004 vcpu->apic_base = 0xfee00000 |
1005 /*for vcpu 0*/ MSR_IA32_APICBASE_BSP |
1006 MSR_IA32_APICBASE_ENABLE;
1011 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
1012 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
1014 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
1015 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
1016 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1017 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1019 seg_setup(VCPU_SREG_DS);
1020 seg_setup(VCPU_SREG_ES);
1021 seg_setup(VCPU_SREG_FS);
1022 seg_setup(VCPU_SREG_GS);
1023 seg_setup(VCPU_SREG_SS);
1025 vmcs_write16(GUEST_TR_SELECTOR, 0);
1026 vmcs_writel(GUEST_TR_BASE, 0);
1027 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
1028 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1030 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
1031 vmcs_writel(GUEST_LDTR_BASE, 0);
1032 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
1033 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
1035 vmcs_write32(GUEST_SYSENTER_CS, 0);
1036 vmcs_writel(GUEST_SYSENTER_ESP, 0);
1037 vmcs_writel(GUEST_SYSENTER_EIP, 0);
1039 vmcs_writel(GUEST_RFLAGS, 0x02);
1040 vmcs_writel(GUEST_RIP, 0xfff0);
1041 vmcs_writel(GUEST_RSP, 0);
1043 vmcs_writel(GUEST_CR3, 0);
1045 //todo: dr0 = dr1 = dr2 = dr3 = 0; dr6 = 0xffff0ff0
1046 vmcs_writel(GUEST_DR7, 0x400);
1048 vmcs_writel(GUEST_GDTR_BASE, 0);
1049 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
1051 vmcs_writel(GUEST_IDTR_BASE, 0);
1052 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
1054 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
1055 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
1056 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
1059 vmcs_write64(IO_BITMAP_A, 0);
1060 vmcs_write64(IO_BITMAP_B, 0);
1064 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
1066 /* Special registers */
1067 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
1070 vmcs_write32_fixedbits(MSR_IA32_VMX_PINBASED_CTLS_MSR,
1071 PIN_BASED_VM_EXEC_CONTROL,
1072 PIN_BASED_EXT_INTR_MASK /* 20.6.1 */
1073 | PIN_BASED_NMI_EXITING /* 20.6.1 */
1075 vmcs_write32_fixedbits(MSR_IA32_VMX_PROCBASED_CTLS_MSR,
1076 CPU_BASED_VM_EXEC_CONTROL,
1077 CPU_BASED_HLT_EXITING /* 20.6.2 */
1078 | CPU_BASED_CR8_LOAD_EXITING /* 20.6.2 */
1079 | CPU_BASED_CR8_STORE_EXITING /* 20.6.2 */
1080 | CPU_BASED_UNCOND_IO_EXITING /* 20.6.2 */
1081 | CPU_BASED_INVDPG_EXITING
1082 | CPU_BASED_MOV_DR_EXITING
1083 | CPU_BASED_USE_TSC_OFFSETING /* 21.3 */
1086 vmcs_write32(EXCEPTION_BITMAP, 1 << PF_VECTOR);
1087 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
1088 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
1089 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
1091 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
1092 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
1093 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
1095 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
1096 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1097 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1098 vmcs_write16(HOST_FS_SELECTOR, read_fs()); /* 22.2.4 */
1099 vmcs_write16(HOST_GS_SELECTOR, read_gs()); /* 22.2.4 */
1100 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1101 #ifdef CONFIG_X86_64
1102 rdmsrl(MSR_FS_BASE, a);
1103 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
1104 rdmsrl(MSR_GS_BASE, a);
1105 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
1107 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
1108 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
1111 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
1114 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
1117 vmcs_writel(HOST_RIP, (unsigned long)kvm_vmx_return); /* 22.2.5 */
1119 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
1120 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
1121 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
1122 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
1123 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
1124 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
1127 vcpu->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
1128 if (!vcpu->guest_msrs)
1130 vcpu->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
1131 if (!vcpu->host_msrs)
1132 goto out_free_guest_msrs;
1134 for (i = 0; i < NR_VMX_MSR; ++i) {
1135 u32 index = vmx_msr_index[i];
1136 u32 data_low, data_high;
1138 int j = vcpu->nmsrs;
1140 if (rdmsr_safe(index, &data_low, &data_high) < 0)
1142 data = data_low | ((u64)data_high << 32);
1143 vcpu->host_msrs[j].index = index;
1144 vcpu->host_msrs[j].reserved = 0;
1145 vcpu->host_msrs[j].data = data;
1146 vcpu->guest_msrs[j] = vcpu->host_msrs[j];
1149 printk(KERN_DEBUG "kvm: msrs: %d\n", vcpu->nmsrs);
1151 nr_good_msrs = vcpu->nmsrs - NR_BAD_MSRS;
1152 vmcs_writel(VM_ENTRY_MSR_LOAD_ADDR,
1153 virt_to_phys(vcpu->guest_msrs + NR_BAD_MSRS));
1154 vmcs_writel(VM_EXIT_MSR_STORE_ADDR,
1155 virt_to_phys(vcpu->guest_msrs + NR_BAD_MSRS));
1156 vmcs_writel(VM_EXIT_MSR_LOAD_ADDR,
1157 virt_to_phys(vcpu->host_msrs + NR_BAD_MSRS));
1158 vmcs_write32_fixedbits(MSR_IA32_VMX_EXIT_CTLS_MSR, VM_EXIT_CONTROLS,
1159 (HOST_IS_64 << 9)); /* 22.2,1, 20.7.1 */
1160 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, nr_good_msrs); /* 22.2.2 */
1161 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, nr_good_msrs); /* 22.2.2 */
1162 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, nr_good_msrs); /* 22.2.2 */
1165 /* 22.2.1, 20.8.1 */
1166 vmcs_write32_fixedbits(MSR_IA32_VMX_ENTRY_CTLS_MSR,
1167 VM_ENTRY_CONTROLS, 0);
1168 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
1170 vmcs_writel(VIRTUAL_APIC_PAGE_ADDR, 0);
1171 vmcs_writel(TPR_THRESHOLD, 0);
1173 vmcs_writel(CR0_GUEST_HOST_MASK, KVM_GUEST_CR0_MASK);
1174 vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
1176 vcpu->cr0 = 0x60000010;
1177 vmx_set_cr0(vcpu, vcpu->cr0); // enter rmode
1178 vmx_set_cr4(vcpu, 0);
1179 #ifdef CONFIG_X86_64
1180 vmx_set_efer(vcpu, 0);
1185 out_free_guest_msrs:
1186 kfree(vcpu->guest_msrs);
1191 static void inject_rmode_irq(struct kvm_vcpu *vcpu, int irq)
1196 unsigned long flags;
1197 unsigned long ss_base = vmcs_readl(GUEST_SS_BASE);
1198 u16 sp = vmcs_readl(GUEST_RSP);
1199 u32 ss_limit = vmcs_read32(GUEST_SS_LIMIT);
1201 if (sp > ss_limit || sp - 6 > sp) {
1202 vcpu_printf(vcpu, "%s: #SS, rsp 0x%lx ss 0x%lx limit 0x%x\n",
1204 vmcs_readl(GUEST_RSP),
1205 vmcs_readl(GUEST_SS_BASE),
1206 vmcs_read32(GUEST_SS_LIMIT));
1210 if (kvm_read_guest(vcpu, irq * sizeof(ent), sizeof(ent), &ent) !=
1212 vcpu_printf(vcpu, "%s: read guest err\n", __FUNCTION__);
1216 flags = vmcs_readl(GUEST_RFLAGS);
1217 cs = vmcs_readl(GUEST_CS_BASE) >> 4;
1218 ip = vmcs_readl(GUEST_RIP);
1221 if (kvm_write_guest(vcpu, ss_base + sp - 2, 2, &flags) != 2 ||
1222 kvm_write_guest(vcpu, ss_base + sp - 4, 2, &cs) != 2 ||
1223 kvm_write_guest(vcpu, ss_base + sp - 6, 2, &ip) != 2) {
1224 vcpu_printf(vcpu, "%s: write guest err\n", __FUNCTION__);
1228 vmcs_writel(GUEST_RFLAGS, flags &
1229 ~( X86_EFLAGS_IF | X86_EFLAGS_AC | X86_EFLAGS_TF));
1230 vmcs_write16(GUEST_CS_SELECTOR, ent[1]) ;
1231 vmcs_writel(GUEST_CS_BASE, ent[1] << 4);
1232 vmcs_writel(GUEST_RIP, ent[0]);
1233 vmcs_writel(GUEST_RSP, (vmcs_readl(GUEST_RSP) & ~0xffff) | (sp - 6));
1236 static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1238 int word_index = __ffs(vcpu->irq_summary);
1239 int bit_index = __ffs(vcpu->irq_pending[word_index]);
1240 int irq = word_index * BITS_PER_LONG + bit_index;
1242 clear_bit(bit_index, &vcpu->irq_pending[word_index]);
1243 if (!vcpu->irq_pending[word_index])
1244 clear_bit(word_index, &vcpu->irq_summary);
1246 if (vcpu->rmode.active) {
1247 inject_rmode_irq(vcpu, irq);
1250 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1251 irq | INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1254 static void kvm_try_inject_irq(struct kvm_vcpu *vcpu)
1256 if ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF)
1257 && (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0)
1259 * Interrupts enabled, and not blocked by sti or mov ss. Good.
1261 kvm_do_inject_irq(vcpu);
1264 * Interrupts blocked. Wait for unblock.
1266 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1267 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL)
1268 | CPU_BASED_VIRTUAL_INTR_PENDING);
1271 static void kvm_guest_debug_pre(struct kvm_vcpu *vcpu)
1273 struct kvm_guest_debug *dbg = &vcpu->guest_debug;
1275 set_debugreg(dbg->bp[0], 0);
1276 set_debugreg(dbg->bp[1], 1);
1277 set_debugreg(dbg->bp[2], 2);
1278 set_debugreg(dbg->bp[3], 3);
1280 if (dbg->singlestep) {
1281 unsigned long flags;
1283 flags = vmcs_readl(GUEST_RFLAGS);
1284 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1285 vmcs_writel(GUEST_RFLAGS, flags);
1289 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
1290 int vec, u32 err_code)
1292 if (!vcpu->rmode.active)
1295 if (vec == GP_VECTOR && err_code == 0)
1296 if (emulate_instruction(vcpu, NULL, 0, 0) == EMULATE_DONE)
1301 static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1303 u32 intr_info, error_code;
1304 unsigned long cr2, rip;
1306 enum emulation_result er;
1308 vect_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
1309 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
1311 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
1312 !is_page_fault(intr_info)) {
1313 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
1314 "intr info 0x%x\n", __FUNCTION__, vect_info, intr_info);
1317 if (is_external_interrupt(vect_info)) {
1318 int irq = vect_info & VECTORING_INFO_VECTOR_MASK;
1319 set_bit(irq, vcpu->irq_pending);
1320 set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
1323 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) { /* nmi */
1328 rip = vmcs_readl(GUEST_RIP);
1329 if (intr_info & INTR_INFO_DELIEVER_CODE_MASK)
1330 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
1331 if (is_page_fault(intr_info)) {
1332 cr2 = vmcs_readl(EXIT_QUALIFICATION);
1334 spin_lock(&vcpu->kvm->lock);
1335 if (!vcpu->mmu.page_fault(vcpu, cr2, error_code)) {
1336 spin_unlock(&vcpu->kvm->lock);
1340 er = emulate_instruction(vcpu, kvm_run, cr2, error_code);
1341 spin_unlock(&vcpu->kvm->lock);
1346 case EMULATE_DO_MMIO:
1347 ++kvm_stat.mmio_exits;
1348 kvm_run->exit_reason = KVM_EXIT_MMIO;
1351 vcpu_printf(vcpu, "%s: emulate fail\n", __FUNCTION__);
1358 if (vcpu->rmode.active &&
1359 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
1363 if ((intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK)) == (INTR_TYPE_EXCEPTION | 1)) {
1364 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1367 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
1368 kvm_run->ex.exception = intr_info & INTR_INFO_VECTOR_MASK;
1369 kvm_run->ex.error_code = error_code;
1373 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
1374 struct kvm_run *kvm_run)
1376 ++kvm_stat.irq_exits;
1381 static int get_io_count(struct kvm_vcpu *vcpu, u64 *count)
1388 if ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_VM)) {
1391 u32 cs_ar = vmcs_read32(GUEST_CS_AR_BYTES);
1393 countr_size = (cs_ar & AR_L_MASK) ? 8:
1394 (cs_ar & AR_DB_MASK) ? 4: 2;
1397 rip = vmcs_readl(GUEST_RIP);
1398 if (countr_size != 8)
1399 rip += vmcs_readl(GUEST_CS_BASE);
1401 n = kvm_read_guest(vcpu, rip, sizeof(inst), &inst);
1403 for (i = 0; i < n; i++) {
1404 switch (((u8*)&inst)[i]) {
1417 countr_size = (countr_size == 2) ? 4: (countr_size >> 1);
1425 *count = vcpu->regs[VCPU_REGS_RCX] & (~0ULL >> (64 - countr_size));
1429 static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1431 u64 exit_qualification;
1433 ++kvm_stat.io_exits;
1434 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
1435 kvm_run->exit_reason = KVM_EXIT_IO;
1436 if (exit_qualification & 8)
1437 kvm_run->io.direction = KVM_EXIT_IO_IN;
1439 kvm_run->io.direction = KVM_EXIT_IO_OUT;
1440 kvm_run->io.size = (exit_qualification & 7) + 1;
1441 kvm_run->io.string = (exit_qualification & 16) != 0;
1442 kvm_run->io.string_down
1443 = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_DF) != 0;
1444 kvm_run->io.rep = (exit_qualification & 32) != 0;
1445 kvm_run->io.port = exit_qualification >> 16;
1446 if (kvm_run->io.string) {
1447 if (!get_io_count(vcpu, &kvm_run->io.count))
1449 kvm_run->io.address = vmcs_readl(GUEST_LINEAR_ADDRESS);
1451 kvm_run->io.value = vcpu->regs[VCPU_REGS_RAX]; /* rax */
1455 static int handle_invlpg(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1457 u64 address = vmcs_read64(EXIT_QUALIFICATION);
1458 int instruction_length = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1459 spin_lock(&vcpu->kvm->lock);
1460 vcpu->mmu.inval_page(vcpu, address);
1461 spin_unlock(&vcpu->kvm->lock);
1462 vmcs_writel(GUEST_RIP, vmcs_readl(GUEST_RIP) + instruction_length);
1466 static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1468 u64 exit_qualification;
1472 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
1473 cr = exit_qualification & 15;
1474 reg = (exit_qualification >> 8) & 15;
1475 switch ((exit_qualification >> 4) & 3) {
1476 case 0: /* mov to cr */
1479 vcpu_load_rsp_rip(vcpu);
1480 set_cr0(vcpu, vcpu->regs[reg]);
1481 skip_emulated_instruction(vcpu);
1484 vcpu_load_rsp_rip(vcpu);
1485 set_cr3(vcpu, vcpu->regs[reg]);
1486 skip_emulated_instruction(vcpu);
1489 vcpu_load_rsp_rip(vcpu);
1490 set_cr4(vcpu, vcpu->regs[reg]);
1491 skip_emulated_instruction(vcpu);
1494 vcpu_load_rsp_rip(vcpu);
1495 set_cr8(vcpu, vcpu->regs[reg]);
1496 skip_emulated_instruction(vcpu);
1500 case 1: /*mov from cr*/
1503 vcpu_load_rsp_rip(vcpu);
1504 vcpu->regs[reg] = vcpu->cr3;
1505 vcpu_put_rsp_rip(vcpu);
1506 skip_emulated_instruction(vcpu);
1509 printk(KERN_DEBUG "handle_cr: read CR8 "
1510 "cpu erratum AA15\n");
1511 vcpu_load_rsp_rip(vcpu);
1512 vcpu->regs[reg] = vcpu->cr8;
1513 vcpu_put_rsp_rip(vcpu);
1514 skip_emulated_instruction(vcpu);
1519 lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
1521 skip_emulated_instruction(vcpu);
1526 kvm_run->exit_reason = 0;
1527 printk(KERN_ERR "kvm: unhandled control register: op %d cr %d\n",
1528 (int)(exit_qualification >> 4) & 3, cr);
1532 static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1534 u64 exit_qualification;
1539 * FIXME: this code assumes the host is debugging the guest.
1540 * need to deal with guest debugging itself too.
1542 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
1543 dr = exit_qualification & 7;
1544 reg = (exit_qualification >> 8) & 15;
1545 vcpu_load_rsp_rip(vcpu);
1546 if (exit_qualification & 16) {
1558 vcpu->regs[reg] = val;
1562 vcpu_put_rsp_rip(vcpu);
1563 skip_emulated_instruction(vcpu);
1567 static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1569 kvm_run->exit_reason = KVM_EXIT_CPUID;
1573 static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1575 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1578 if (vmx_get_msr(vcpu, ecx, &data)) {
1579 vmx_inject_gp(vcpu, 0);
1583 /* FIXME: handling of bits 32:63 of rax, rdx */
1584 vcpu->regs[VCPU_REGS_RAX] = data & -1u;
1585 vcpu->regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
1586 skip_emulated_instruction(vcpu);
1590 static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1592 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1593 u64 data = (vcpu->regs[VCPU_REGS_RAX] & -1u)
1594 | ((u64)(vcpu->regs[VCPU_REGS_RDX] & -1u) << 32);
1596 if (vmx_set_msr(vcpu, ecx, data) != 0) {
1597 vmx_inject_gp(vcpu, 0);
1601 skip_emulated_instruction(vcpu);
1605 static int handle_interrupt_window(struct kvm_vcpu *vcpu,
1606 struct kvm_run *kvm_run)
1608 /* Turn off interrupt window reporting. */
1609 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1610 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL)
1611 & ~CPU_BASED_VIRTUAL_INTR_PENDING);
1615 static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1617 skip_emulated_instruction(vcpu);
1618 if (vcpu->irq_summary && (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF))
1621 kvm_run->exit_reason = KVM_EXIT_HLT;
1626 * The exit handlers return 1 if the exit was handled fully and guest execution
1627 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
1628 * to be done to userspace and return 0.
1630 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
1631 struct kvm_run *kvm_run) = {
1632 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
1633 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
1634 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
1635 [EXIT_REASON_INVLPG] = handle_invlpg,
1636 [EXIT_REASON_CR_ACCESS] = handle_cr,
1637 [EXIT_REASON_DR_ACCESS] = handle_dr,
1638 [EXIT_REASON_CPUID] = handle_cpuid,
1639 [EXIT_REASON_MSR_READ] = handle_rdmsr,
1640 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
1641 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
1642 [EXIT_REASON_HLT] = handle_halt,
1645 static const int kvm_vmx_max_exit_handlers =
1646 sizeof(kvm_vmx_exit_handlers) / sizeof(*kvm_vmx_exit_handlers);
1649 * The guest has exited. See if we can fix it or if we need userspace
1652 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1654 u32 vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
1655 u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
1657 if ( (vectoring_info & VECTORING_INFO_VALID_MASK) &&
1658 exit_reason != EXIT_REASON_EXCEPTION_NMI )
1659 printk(KERN_WARNING "%s: unexpected, valid vectoring info and "
1660 "exit reason is 0x%x\n", __FUNCTION__, exit_reason);
1661 kvm_run->instruction_length = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1662 if (exit_reason < kvm_vmx_max_exit_handlers
1663 && kvm_vmx_exit_handlers[exit_reason])
1664 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
1666 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1667 kvm_run->hw.hardware_exit_reason = exit_reason;
1672 static int vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1675 u16 fs_sel, gs_sel, ldt_sel;
1676 int fs_gs_ldt_reload_needed;
1680 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1681 * allow segment selectors with cpl > 0 or ti == 1.
1685 ldt_sel = read_ldt();
1686 fs_gs_ldt_reload_needed = (fs_sel & 7) | (gs_sel & 7) | ldt_sel;
1687 if (!fs_gs_ldt_reload_needed) {
1688 vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1689 vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1691 vmcs_write16(HOST_FS_SELECTOR, 0);
1692 vmcs_write16(HOST_GS_SELECTOR, 0);
1695 #ifdef CONFIG_X86_64
1696 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1697 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1699 vmcs_writel(HOST_FS_BASE, segment_base(fs_sel));
1700 vmcs_writel(HOST_GS_BASE, segment_base(gs_sel));
1703 if (vcpu->irq_summary &&
1704 !(vmcs_read32(VM_ENTRY_INTR_INFO_FIELD) & INTR_INFO_VALID_MASK))
1705 kvm_try_inject_irq(vcpu);
1707 if (vcpu->guest_debug.enabled)
1708 kvm_guest_debug_pre(vcpu);
1710 fx_save(vcpu->host_fx_image);
1711 fx_restore(vcpu->guest_fx_image);
1713 save_msrs(vcpu->host_msrs, vcpu->nmsrs);
1714 load_msrs(vcpu->guest_msrs, NR_BAD_MSRS);
1717 /* Store host registers */
1719 #ifdef CONFIG_X86_64
1720 "push %%rax; push %%rbx; push %%rdx;"
1721 "push %%rsi; push %%rdi; push %%rbp;"
1722 "push %%r8; push %%r9; push %%r10; push %%r11;"
1723 "push %%r12; push %%r13; push %%r14; push %%r15;"
1725 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
1727 "pusha; push %%ecx \n\t"
1728 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
1730 /* Check if vmlaunch of vmresume is needed */
1732 /* Load guest registers. Don't clobber flags. */
1733 #ifdef CONFIG_X86_64
1734 "mov %c[cr2](%3), %%rax \n\t"
1735 "mov %%rax, %%cr2 \n\t"
1736 "mov %c[rax](%3), %%rax \n\t"
1737 "mov %c[rbx](%3), %%rbx \n\t"
1738 "mov %c[rdx](%3), %%rdx \n\t"
1739 "mov %c[rsi](%3), %%rsi \n\t"
1740 "mov %c[rdi](%3), %%rdi \n\t"
1741 "mov %c[rbp](%3), %%rbp \n\t"
1742 "mov %c[r8](%3), %%r8 \n\t"
1743 "mov %c[r9](%3), %%r9 \n\t"
1744 "mov %c[r10](%3), %%r10 \n\t"
1745 "mov %c[r11](%3), %%r11 \n\t"
1746 "mov %c[r12](%3), %%r12 \n\t"
1747 "mov %c[r13](%3), %%r13 \n\t"
1748 "mov %c[r14](%3), %%r14 \n\t"
1749 "mov %c[r15](%3), %%r15 \n\t"
1750 "mov %c[rcx](%3), %%rcx \n\t" /* kills %3 (rcx) */
1752 "mov %c[cr2](%3), %%eax \n\t"
1753 "mov %%eax, %%cr2 \n\t"
1754 "mov %c[rax](%3), %%eax \n\t"
1755 "mov %c[rbx](%3), %%ebx \n\t"
1756 "mov %c[rdx](%3), %%edx \n\t"
1757 "mov %c[rsi](%3), %%esi \n\t"
1758 "mov %c[rdi](%3), %%edi \n\t"
1759 "mov %c[rbp](%3), %%ebp \n\t"
1760 "mov %c[rcx](%3), %%ecx \n\t" /* kills %3 (ecx) */
1762 /* Enter guest mode */
1764 ASM_VMX_VMLAUNCH "\n\t"
1765 "jmp kvm_vmx_return \n\t"
1766 "launched: " ASM_VMX_VMRESUME "\n\t"
1767 ".globl kvm_vmx_return \n\t"
1769 /* Save guest registers, load host registers, keep flags */
1770 #ifdef CONFIG_X86_64
1771 "xchg %3, 0(%%rsp) \n\t"
1772 "mov %%rax, %c[rax](%3) \n\t"
1773 "mov %%rbx, %c[rbx](%3) \n\t"
1774 "pushq 0(%%rsp); popq %c[rcx](%3) \n\t"
1775 "mov %%rdx, %c[rdx](%3) \n\t"
1776 "mov %%rsi, %c[rsi](%3) \n\t"
1777 "mov %%rdi, %c[rdi](%3) \n\t"
1778 "mov %%rbp, %c[rbp](%3) \n\t"
1779 "mov %%r8, %c[r8](%3) \n\t"
1780 "mov %%r9, %c[r9](%3) \n\t"
1781 "mov %%r10, %c[r10](%3) \n\t"
1782 "mov %%r11, %c[r11](%3) \n\t"
1783 "mov %%r12, %c[r12](%3) \n\t"
1784 "mov %%r13, %c[r13](%3) \n\t"
1785 "mov %%r14, %c[r14](%3) \n\t"
1786 "mov %%r15, %c[r15](%3) \n\t"
1787 "mov %%cr2, %%rax \n\t"
1788 "mov %%rax, %c[cr2](%3) \n\t"
1789 "mov 0(%%rsp), %3 \n\t"
1791 "pop %%rcx; pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
1792 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
1793 "pop %%rbp; pop %%rdi; pop %%rsi;"
1794 "pop %%rdx; pop %%rbx; pop %%rax \n\t"
1796 "xchg %3, 0(%%esp) \n\t"
1797 "mov %%eax, %c[rax](%3) \n\t"
1798 "mov %%ebx, %c[rbx](%3) \n\t"
1799 "pushl 0(%%esp); popl %c[rcx](%3) \n\t"
1800 "mov %%edx, %c[rdx](%3) \n\t"
1801 "mov %%esi, %c[rsi](%3) \n\t"
1802 "mov %%edi, %c[rdi](%3) \n\t"
1803 "mov %%ebp, %c[rbp](%3) \n\t"
1804 "mov %%cr2, %%eax \n\t"
1805 "mov %%eax, %c[cr2](%3) \n\t"
1806 "mov 0(%%esp), %3 \n\t"
1808 "pop %%ecx; popa \n\t"
1813 : "r"(vcpu->launched), "d"((unsigned long)HOST_RSP),
1815 [rax]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RAX])),
1816 [rbx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBX])),
1817 [rcx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RCX])),
1818 [rdx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDX])),
1819 [rsi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RSI])),
1820 [rdi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDI])),
1821 [rbp]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBP])),
1822 #ifdef CONFIG_X86_64
1823 [r8 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R8 ])),
1824 [r9 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R9 ])),
1825 [r10]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R10])),
1826 [r11]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R11])),
1827 [r12]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R12])),
1828 [r13]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R13])),
1829 [r14]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R14])),
1830 [r15]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R15])),
1832 [cr2]"i"(offsetof(struct kvm_vcpu, cr2))
1837 save_msrs(vcpu->guest_msrs, NR_BAD_MSRS);
1838 load_msrs(vcpu->host_msrs, NR_BAD_MSRS);
1840 fx_save(vcpu->guest_fx_image);
1841 fx_restore(vcpu->host_fx_image);
1843 #ifndef CONFIG_X86_64
1844 asm ("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
1847 kvm_run->exit_type = 0;
1849 kvm_run->exit_type = KVM_EXIT_TYPE_FAIL_ENTRY;
1850 kvm_run->exit_reason = vmcs_read32(VM_INSTRUCTION_ERROR);
1852 if (fs_gs_ldt_reload_needed) {
1856 * If we have to reload gs, we must take care to
1857 * preserve our gs base.
1859 local_irq_disable();
1861 #ifdef CONFIG_X86_64
1862 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
1869 kvm_run->exit_type = KVM_EXIT_TYPE_VM_EXIT;
1870 if (kvm_handle_exit(kvm_run, vcpu)) {
1871 /* Give scheduler a change to reschedule. */
1872 if (signal_pending(current)) {
1873 ++kvm_stat.signal_exits;
1883 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1885 vmcs_writel(GUEST_CR3, vmcs_readl(GUEST_CR3));
1888 static void vmx_inject_page_fault(struct kvm_vcpu *vcpu,
1892 u32 vect_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
1894 ++kvm_stat.pf_guest;
1896 if (is_page_fault(vect_info)) {
1897 printk(KERN_DEBUG "inject_page_fault: "
1898 "double fault 0x%lx @ 0x%lx\n",
1899 addr, vmcs_readl(GUEST_RIP));
1900 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, 0);
1901 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1903 INTR_TYPE_EXCEPTION |
1904 INTR_INFO_DELIEVER_CODE_MASK |
1905 INTR_INFO_VALID_MASK);
1909 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, err_code);
1910 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1912 INTR_TYPE_EXCEPTION |
1913 INTR_INFO_DELIEVER_CODE_MASK |
1914 INTR_INFO_VALID_MASK);
1918 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
1921 on_each_cpu(__vcpu_clear, vcpu, 0, 1);
1922 free_vmcs(vcpu->vmcs);
1927 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
1929 vmx_free_vmcs(vcpu);
1932 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
1936 vmcs = alloc_vmcs();
1945 static struct kvm_arch_ops vmx_arch_ops = {
1946 .cpu_has_kvm_support = cpu_has_kvm_support,
1947 .disabled_by_bios = vmx_disabled_by_bios,
1948 .hardware_setup = hardware_setup,
1949 .hardware_unsetup = hardware_unsetup,
1950 .hardware_enable = hardware_enable,
1951 .hardware_disable = hardware_disable,
1953 .vcpu_create = vmx_create_vcpu,
1954 .vcpu_free = vmx_free_vcpu,
1956 .vcpu_load = vmx_vcpu_load,
1957 .vcpu_put = vmx_vcpu_put,
1959 .set_guest_debug = set_guest_debug,
1960 .get_msr = vmx_get_msr,
1961 .set_msr = vmx_set_msr,
1962 .get_segment_base = vmx_get_segment_base,
1963 .get_segment = vmx_get_segment,
1964 .set_segment = vmx_set_segment,
1965 .is_long_mode = vmx_is_long_mode,
1966 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
1967 .set_cr0 = vmx_set_cr0,
1968 .set_cr0_no_modeswitch = vmx_set_cr0_no_modeswitch,
1969 .set_cr3 = vmx_set_cr3,
1970 .set_cr4 = vmx_set_cr4,
1971 #ifdef CONFIG_X86_64
1972 .set_efer = vmx_set_efer,
1974 .get_idt = vmx_get_idt,
1975 .set_idt = vmx_set_idt,
1976 .get_gdt = vmx_get_gdt,
1977 .set_gdt = vmx_set_gdt,
1978 .cache_regs = vcpu_load_rsp_rip,
1979 .decache_regs = vcpu_put_rsp_rip,
1980 .get_rflags = vmx_get_rflags,
1981 .set_rflags = vmx_set_rflags,
1983 .tlb_flush = vmx_flush_tlb,
1984 .inject_page_fault = vmx_inject_page_fault,
1986 .inject_gp = vmx_inject_gp,
1988 .run = vmx_vcpu_run,
1989 .skip_emulated_instruction = skip_emulated_instruction,
1990 .vcpu_setup = vmx_vcpu_setup,
1993 static int __init vmx_init(void)
1995 kvm_init_arch(&vmx_arch_ops, THIS_MODULE);
1999 static void __exit vmx_exit(void)
2004 module_init(vmx_init)
2005 module_exit(vmx_exit)