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.
20 #include <linux/module.h>
21 #include <linux/kernel.h>
23 #include <linux/highmem.h>
24 #include <linux/profile.h>
25 #include <linux/sched.h>
29 #include "segment_descriptor.h"
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);
37 static struct page *vmx_io_bitmap_a;
38 static struct page *vmx_io_bitmap_b;
46 static struct vmcs_descriptor {
52 #define VMX_SEGMENT_FIELD(seg) \
53 [VCPU_SREG_##seg] = { \
54 .selector = GUEST_##seg##_SELECTOR, \
55 .base = GUEST_##seg##_BASE, \
56 .limit = GUEST_##seg##_LIMIT, \
57 .ar_bytes = GUEST_##seg##_AR_BYTES, \
60 static struct kvm_vmx_segment_field {
65 } kvm_vmx_segment_fields[] = {
66 VMX_SEGMENT_FIELD(CS),
67 VMX_SEGMENT_FIELD(DS),
68 VMX_SEGMENT_FIELD(ES),
69 VMX_SEGMENT_FIELD(FS),
70 VMX_SEGMENT_FIELD(GS),
71 VMX_SEGMENT_FIELD(SS),
72 VMX_SEGMENT_FIELD(TR),
73 VMX_SEGMENT_FIELD(LDTR),
77 * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
78 * away by decrementing the array size.
80 static const u32 vmx_msr_index[] = {
82 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
84 MSR_EFER, MSR_K6_STAR,
86 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
89 static unsigned msr_offset_kernel_gs_base;
90 #define NR_64BIT_MSRS 4
92 * avoid save/load MSR_SYSCALL_MASK and MSR_LSTAR by std vt
93 * mechanism (cpu bug AA24)
97 #define NR_64BIT_MSRS 0
101 static inline int is_page_fault(u32 intr_info)
103 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
104 INTR_INFO_VALID_MASK)) ==
105 (INTR_TYPE_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
108 static inline int is_no_device(u32 intr_info)
110 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
111 INTR_INFO_VALID_MASK)) ==
112 (INTR_TYPE_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
115 static inline int is_external_interrupt(u32 intr_info)
117 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
118 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
121 static struct vmx_msr_entry *find_msr_entry(struct kvm_vcpu *vcpu, u32 msr)
125 for (i = 0; i < vcpu->nmsrs; ++i)
126 if (vcpu->guest_msrs[i].index == msr)
127 return &vcpu->guest_msrs[i];
131 static void vmcs_clear(struct vmcs *vmcs)
133 u64 phys_addr = __pa(vmcs);
136 asm volatile (ASM_VMX_VMCLEAR_RAX "; setna %0"
137 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
140 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
144 static void __vcpu_clear(void *arg)
146 struct kvm_vcpu *vcpu = arg;
147 int cpu = raw_smp_processor_id();
149 if (vcpu->cpu == cpu)
150 vmcs_clear(vcpu->vmcs);
151 if (per_cpu(current_vmcs, cpu) == vcpu->vmcs)
152 per_cpu(current_vmcs, cpu) = NULL;
155 static void vcpu_clear(struct kvm_vcpu *vcpu)
157 if (vcpu->cpu != raw_smp_processor_id() && vcpu->cpu != -1)
158 smp_call_function_single(vcpu->cpu, __vcpu_clear, vcpu, 0, 1);
164 static unsigned long vmcs_readl(unsigned long field)
168 asm volatile (ASM_VMX_VMREAD_RDX_RAX
169 : "=a"(value) : "d"(field) : "cc");
173 static u16 vmcs_read16(unsigned long field)
175 return vmcs_readl(field);
178 static u32 vmcs_read32(unsigned long field)
180 return vmcs_readl(field);
183 static u64 vmcs_read64(unsigned long field)
186 return vmcs_readl(field);
188 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
192 static noinline void vmwrite_error(unsigned long field, unsigned long value)
194 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
195 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
199 static void vmcs_writel(unsigned long field, unsigned long value)
203 asm volatile (ASM_VMX_VMWRITE_RAX_RDX "; setna %0"
204 : "=q"(error) : "a"(value), "d"(field) : "cc" );
206 vmwrite_error(field, value);
209 static void vmcs_write16(unsigned long field, u16 value)
211 vmcs_writel(field, value);
214 static void vmcs_write32(unsigned long field, u32 value)
216 vmcs_writel(field, value);
219 static void vmcs_write64(unsigned long field, u64 value)
222 vmcs_writel(field, value);
224 vmcs_writel(field, value);
226 vmcs_writel(field+1, value >> 32);
230 static void vmcs_clear_bits(unsigned long field, u32 mask)
232 vmcs_writel(field, vmcs_readl(field) & ~mask);
235 static void vmcs_set_bits(unsigned long field, u32 mask)
237 vmcs_writel(field, vmcs_readl(field) | mask);
241 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
242 * vcpu mutex is already taken.
244 static void vmx_vcpu_load(struct kvm_vcpu *vcpu)
246 u64 phys_addr = __pa(vcpu->vmcs);
251 if (vcpu->cpu != cpu)
254 if (per_cpu(current_vmcs, cpu) != vcpu->vmcs) {
257 per_cpu(current_vmcs, cpu) = vcpu->vmcs;
258 asm volatile (ASM_VMX_VMPTRLD_RAX "; setna %0"
259 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
262 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
263 vcpu->vmcs, phys_addr);
266 if (vcpu->cpu != cpu) {
267 struct descriptor_table dt;
268 unsigned long sysenter_esp;
272 * Linux uses per-cpu TSS and GDT, so set these when switching
275 vmcs_writel(HOST_TR_BASE, read_tr_base()); /* 22.2.4 */
277 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
279 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
280 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
284 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
286 kvm_put_guest_fpu(vcpu);
290 static void vmx_vcpu_decache(struct kvm_vcpu *vcpu)
295 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
297 return vmcs_readl(GUEST_RFLAGS);
300 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
302 vmcs_writel(GUEST_RFLAGS, rflags);
305 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
308 u32 interruptibility;
310 rip = vmcs_readl(GUEST_RIP);
311 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
312 vmcs_writel(GUEST_RIP, rip);
315 * We emulated an instruction, so temporary interrupt blocking
316 * should be removed, if set.
318 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
319 if (interruptibility & 3)
320 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
321 interruptibility & ~3);
322 vcpu->interrupt_window_open = 1;
325 static void vmx_inject_gp(struct kvm_vcpu *vcpu, unsigned error_code)
327 printk(KERN_DEBUG "inject_general_protection: rip 0x%lx\n",
328 vmcs_readl(GUEST_RIP));
329 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
330 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
332 INTR_TYPE_EXCEPTION |
333 INTR_INFO_DELIEVER_CODE_MASK |
334 INTR_INFO_VALID_MASK);
338 * Set up the vmcs to automatically save and restore system
339 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
340 * mode, as fiddling with msrs is very expensive.
342 static void setup_msrs(struct kvm_vcpu *vcpu)
344 int nr_skip, nr_good_msrs;
346 if (is_long_mode(vcpu))
347 nr_skip = NR_BAD_MSRS;
349 nr_skip = NR_64BIT_MSRS;
350 nr_good_msrs = vcpu->nmsrs - nr_skip;
353 * MSR_K6_STAR is only needed on long mode guests, and only
354 * if efer.sce is enabled.
356 if (find_msr_entry(vcpu, MSR_K6_STAR)) {
359 if (is_long_mode(vcpu) && (vcpu->shadow_efer & EFER_SCE))
364 vmcs_writel(VM_ENTRY_MSR_LOAD_ADDR,
365 virt_to_phys(vcpu->guest_msrs + nr_skip));
366 vmcs_writel(VM_EXIT_MSR_STORE_ADDR,
367 virt_to_phys(vcpu->guest_msrs + nr_skip));
368 vmcs_writel(VM_EXIT_MSR_LOAD_ADDR,
369 virt_to_phys(vcpu->host_msrs + nr_skip));
370 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, nr_good_msrs); /* 22.2.2 */
371 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, nr_good_msrs); /* 22.2.2 */
372 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, nr_good_msrs); /* 22.2.2 */
376 * reads and returns guest's timestamp counter "register"
377 * guest_tsc = host_tsc + tsc_offset -- 21.3
379 static u64 guest_read_tsc(void)
381 u64 host_tsc, tsc_offset;
384 tsc_offset = vmcs_read64(TSC_OFFSET);
385 return host_tsc + tsc_offset;
389 * writes 'guest_tsc' into guest's timestamp counter "register"
390 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
392 static void guest_write_tsc(u64 guest_tsc)
397 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
400 static void reload_tss(void)
402 #ifndef CONFIG_X86_64
405 * VT restores TR but not its size. Useless.
407 struct descriptor_table gdt;
408 struct segment_descriptor *descs;
411 descs = (void *)gdt.base;
412 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
418 * Reads an msr value (of 'msr_index') into 'pdata'.
419 * Returns 0 on success, non-0 otherwise.
420 * Assumes vcpu_load() was already called.
422 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
425 struct vmx_msr_entry *msr;
428 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
435 data = vmcs_readl(GUEST_FS_BASE);
438 data = vmcs_readl(GUEST_GS_BASE);
441 return kvm_get_msr_common(vcpu, msr_index, pdata);
443 case MSR_IA32_TIME_STAMP_COUNTER:
444 data = guest_read_tsc();
446 case MSR_IA32_SYSENTER_CS:
447 data = vmcs_read32(GUEST_SYSENTER_CS);
449 case MSR_IA32_SYSENTER_EIP:
450 data = vmcs_readl(GUEST_SYSENTER_EIP);
452 case MSR_IA32_SYSENTER_ESP:
453 data = vmcs_readl(GUEST_SYSENTER_ESP);
456 msr = find_msr_entry(vcpu, msr_index);
461 return kvm_get_msr_common(vcpu, msr_index, pdata);
469 * Writes msr value into into the appropriate "register".
470 * Returns 0 on success, non-0 otherwise.
471 * Assumes vcpu_load() was already called.
473 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
475 struct vmx_msr_entry *msr;
479 return kvm_set_msr_common(vcpu, msr_index, data);
481 vmcs_writel(GUEST_FS_BASE, data);
484 vmcs_writel(GUEST_GS_BASE, data);
487 case MSR_IA32_SYSENTER_CS:
488 vmcs_write32(GUEST_SYSENTER_CS, data);
490 case MSR_IA32_SYSENTER_EIP:
491 vmcs_writel(GUEST_SYSENTER_EIP, data);
493 case MSR_IA32_SYSENTER_ESP:
494 vmcs_writel(GUEST_SYSENTER_ESP, data);
496 case MSR_IA32_TIME_STAMP_COUNTER:
497 guest_write_tsc(data);
500 msr = find_msr_entry(vcpu, msr_index);
505 return kvm_set_msr_common(vcpu, msr_index, data);
514 * Sync the rsp and rip registers into the vcpu structure. This allows
515 * registers to be accessed by indexing vcpu->regs.
517 static void vcpu_load_rsp_rip(struct kvm_vcpu *vcpu)
519 vcpu->regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
520 vcpu->rip = vmcs_readl(GUEST_RIP);
524 * Syncs rsp and rip back into the vmcs. Should be called after possible
527 static void vcpu_put_rsp_rip(struct kvm_vcpu *vcpu)
529 vmcs_writel(GUEST_RSP, vcpu->regs[VCPU_REGS_RSP]);
530 vmcs_writel(GUEST_RIP, vcpu->rip);
533 static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
535 unsigned long dr7 = 0x400;
536 u32 exception_bitmap;
539 exception_bitmap = vmcs_read32(EXCEPTION_BITMAP);
540 old_singlestep = vcpu->guest_debug.singlestep;
542 vcpu->guest_debug.enabled = dbg->enabled;
543 if (vcpu->guest_debug.enabled) {
546 dr7 |= 0x200; /* exact */
547 for (i = 0; i < 4; ++i) {
548 if (!dbg->breakpoints[i].enabled)
550 vcpu->guest_debug.bp[i] = dbg->breakpoints[i].address;
551 dr7 |= 2 << (i*2); /* global enable */
552 dr7 |= 0 << (i*4+16); /* execution breakpoint */
555 exception_bitmap |= (1u << 1); /* Trap debug exceptions */
557 vcpu->guest_debug.singlestep = dbg->singlestep;
559 exception_bitmap &= ~(1u << 1); /* Ignore debug exceptions */
560 vcpu->guest_debug.singlestep = 0;
563 if (old_singlestep && !vcpu->guest_debug.singlestep) {
566 flags = vmcs_readl(GUEST_RFLAGS);
567 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
568 vmcs_writel(GUEST_RFLAGS, flags);
571 vmcs_write32(EXCEPTION_BITMAP, exception_bitmap);
572 vmcs_writel(GUEST_DR7, dr7);
577 static __init int cpu_has_kvm_support(void)
579 unsigned long ecx = cpuid_ecx(1);
580 return test_bit(5, &ecx); /* CPUID.1:ECX.VMX[bit 5] -> VT */
583 static __init int vmx_disabled_by_bios(void)
587 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
588 return (msr & 5) == 1; /* locked but not enabled */
591 static void hardware_enable(void *garbage)
593 int cpu = raw_smp_processor_id();
594 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
597 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
599 /* enable and lock */
600 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | 5);
601 write_cr4(read_cr4() | CR4_VMXE); /* FIXME: not cpu hotplug safe */
602 asm volatile (ASM_VMX_VMXON_RAX : : "a"(&phys_addr), "m"(phys_addr)
606 static void hardware_disable(void *garbage)
608 asm volatile (ASM_VMX_VMXOFF : : : "cc");
611 static __init void setup_vmcs_descriptor(void)
613 u32 vmx_msr_low, vmx_msr_high;
615 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
616 vmcs_descriptor.size = vmx_msr_high & 0x1fff;
617 vmcs_descriptor.order = get_order(vmcs_descriptor.size);
618 vmcs_descriptor.revision_id = vmx_msr_low;
621 static struct vmcs *alloc_vmcs_cpu(int cpu)
623 int node = cpu_to_node(cpu);
627 pages = alloc_pages_node(node, GFP_KERNEL, vmcs_descriptor.order);
630 vmcs = page_address(pages);
631 memset(vmcs, 0, vmcs_descriptor.size);
632 vmcs->revision_id = vmcs_descriptor.revision_id; /* vmcs revision id */
636 static struct vmcs *alloc_vmcs(void)
638 return alloc_vmcs_cpu(raw_smp_processor_id());
641 static void free_vmcs(struct vmcs *vmcs)
643 free_pages((unsigned long)vmcs, vmcs_descriptor.order);
646 static void free_kvm_area(void)
650 for_each_online_cpu(cpu)
651 free_vmcs(per_cpu(vmxarea, cpu));
654 extern struct vmcs *alloc_vmcs_cpu(int cpu);
656 static __init int alloc_kvm_area(void)
660 for_each_online_cpu(cpu) {
663 vmcs = alloc_vmcs_cpu(cpu);
669 per_cpu(vmxarea, cpu) = vmcs;
674 static __init int hardware_setup(void)
676 setup_vmcs_descriptor();
677 return alloc_kvm_area();
680 static __exit void hardware_unsetup(void)
685 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
687 if (vcpu->rmode.active)
688 vmcs_write32(EXCEPTION_BITMAP, ~0);
690 vmcs_write32(EXCEPTION_BITMAP, 1 << PF_VECTOR);
693 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
695 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
697 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
698 vmcs_write16(sf->selector, save->selector);
699 vmcs_writel(sf->base, save->base);
700 vmcs_write32(sf->limit, save->limit);
701 vmcs_write32(sf->ar_bytes, save->ar);
703 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
705 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
709 static void enter_pmode(struct kvm_vcpu *vcpu)
713 vcpu->rmode.active = 0;
715 vmcs_writel(GUEST_TR_BASE, vcpu->rmode.tr.base);
716 vmcs_write32(GUEST_TR_LIMIT, vcpu->rmode.tr.limit);
717 vmcs_write32(GUEST_TR_AR_BYTES, vcpu->rmode.tr.ar);
719 flags = vmcs_readl(GUEST_RFLAGS);
720 flags &= ~(IOPL_MASK | X86_EFLAGS_VM);
721 flags |= (vcpu->rmode.save_iopl << IOPL_SHIFT);
722 vmcs_writel(GUEST_RFLAGS, flags);
724 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~CR4_VME_MASK) |
725 (vmcs_readl(CR4_READ_SHADOW) & CR4_VME_MASK));
727 update_exception_bitmap(vcpu);
729 fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->rmode.es);
730 fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->rmode.ds);
731 fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->rmode.gs);
732 fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->rmode.fs);
734 vmcs_write16(GUEST_SS_SELECTOR, 0);
735 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
737 vmcs_write16(GUEST_CS_SELECTOR,
738 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
739 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
742 static int rmode_tss_base(struct kvm* kvm)
744 gfn_t base_gfn = kvm->memslots[0].base_gfn + kvm->memslots[0].npages - 3;
745 return base_gfn << PAGE_SHIFT;
748 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
750 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
752 save->selector = vmcs_read16(sf->selector);
753 save->base = vmcs_readl(sf->base);
754 save->limit = vmcs_read32(sf->limit);
755 save->ar = vmcs_read32(sf->ar_bytes);
756 vmcs_write16(sf->selector, vmcs_readl(sf->base) >> 4);
757 vmcs_write32(sf->limit, 0xffff);
758 vmcs_write32(sf->ar_bytes, 0xf3);
761 static void enter_rmode(struct kvm_vcpu *vcpu)
765 vcpu->rmode.active = 1;
767 vcpu->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
768 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
770 vcpu->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
771 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
773 vcpu->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
774 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
776 flags = vmcs_readl(GUEST_RFLAGS);
777 vcpu->rmode.save_iopl = (flags & IOPL_MASK) >> IOPL_SHIFT;
779 flags |= IOPL_MASK | X86_EFLAGS_VM;
781 vmcs_writel(GUEST_RFLAGS, flags);
782 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | CR4_VME_MASK);
783 update_exception_bitmap(vcpu);
785 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
786 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
787 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
789 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
790 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
791 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
792 vmcs_writel(GUEST_CS_BASE, 0xf0000);
793 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
795 fix_rmode_seg(VCPU_SREG_ES, &vcpu->rmode.es);
796 fix_rmode_seg(VCPU_SREG_DS, &vcpu->rmode.ds);
797 fix_rmode_seg(VCPU_SREG_GS, &vcpu->rmode.gs);
798 fix_rmode_seg(VCPU_SREG_FS, &vcpu->rmode.fs);
803 static void enter_lmode(struct kvm_vcpu *vcpu)
807 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
808 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
809 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
811 vmcs_write32(GUEST_TR_AR_BYTES,
812 (guest_tr_ar & ~AR_TYPE_MASK)
813 | AR_TYPE_BUSY_64_TSS);
816 vcpu->shadow_efer |= EFER_LMA;
818 find_msr_entry(vcpu, MSR_EFER)->data |= EFER_LMA | EFER_LME;
819 vmcs_write32(VM_ENTRY_CONTROLS,
820 vmcs_read32(VM_ENTRY_CONTROLS)
821 | VM_ENTRY_CONTROLS_IA32E_MASK);
824 static void exit_lmode(struct kvm_vcpu *vcpu)
826 vcpu->shadow_efer &= ~EFER_LMA;
828 vmcs_write32(VM_ENTRY_CONTROLS,
829 vmcs_read32(VM_ENTRY_CONTROLS)
830 & ~VM_ENTRY_CONTROLS_IA32E_MASK);
835 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
837 vcpu->cr4 &= KVM_GUEST_CR4_MASK;
838 vcpu->cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
841 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
843 if (vcpu->rmode.active && (cr0 & CR0_PE_MASK))
846 if (!vcpu->rmode.active && !(cr0 & CR0_PE_MASK))
850 if (vcpu->shadow_efer & EFER_LME) {
851 if (!is_paging(vcpu) && (cr0 & CR0_PG_MASK))
853 if (is_paging(vcpu) && !(cr0 & CR0_PG_MASK))
858 if (!(cr0 & CR0_TS_MASK)) {
859 vcpu->fpu_active = 1;
860 vmcs_clear_bits(EXCEPTION_BITMAP, CR0_TS_MASK);
863 vmcs_writel(CR0_READ_SHADOW, cr0);
864 vmcs_writel(GUEST_CR0,
865 (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON);
869 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
871 vmcs_writel(GUEST_CR3, cr3);
873 if (!(vcpu->cr0 & CR0_TS_MASK)) {
874 vcpu->fpu_active = 0;
875 vmcs_set_bits(GUEST_CR0, CR0_TS_MASK);
876 vmcs_set_bits(EXCEPTION_BITMAP, 1 << NM_VECTOR);
880 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
882 vmcs_writel(CR4_READ_SHADOW, cr4);
883 vmcs_writel(GUEST_CR4, cr4 | (vcpu->rmode.active ?
884 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON));
890 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
892 struct vmx_msr_entry *msr = find_msr_entry(vcpu, MSR_EFER);
894 vcpu->shadow_efer = efer;
895 if (efer & EFER_LMA) {
896 vmcs_write32(VM_ENTRY_CONTROLS,
897 vmcs_read32(VM_ENTRY_CONTROLS) |
898 VM_ENTRY_CONTROLS_IA32E_MASK);
902 vmcs_write32(VM_ENTRY_CONTROLS,
903 vmcs_read32(VM_ENTRY_CONTROLS) &
904 ~VM_ENTRY_CONTROLS_IA32E_MASK);
906 msr->data = efer & ~EFER_LME;
913 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
915 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
917 return vmcs_readl(sf->base);
920 static void vmx_get_segment(struct kvm_vcpu *vcpu,
921 struct kvm_segment *var, int seg)
923 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
926 var->base = vmcs_readl(sf->base);
927 var->limit = vmcs_read32(sf->limit);
928 var->selector = vmcs_read16(sf->selector);
929 ar = vmcs_read32(sf->ar_bytes);
930 if (ar & AR_UNUSABLE_MASK)
933 var->s = (ar >> 4) & 1;
934 var->dpl = (ar >> 5) & 3;
935 var->present = (ar >> 7) & 1;
936 var->avl = (ar >> 12) & 1;
937 var->l = (ar >> 13) & 1;
938 var->db = (ar >> 14) & 1;
939 var->g = (ar >> 15) & 1;
940 var->unusable = (ar >> 16) & 1;
943 static void vmx_set_segment(struct kvm_vcpu *vcpu,
944 struct kvm_segment *var, int seg)
946 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
949 vmcs_writel(sf->base, var->base);
950 vmcs_write32(sf->limit, var->limit);
951 vmcs_write16(sf->selector, var->selector);
952 if (vcpu->rmode.active && var->s) {
954 * Hack real-mode segments into vm86 compatibility.
956 if (var->base == 0xffff0000 && var->selector == 0xf000)
957 vmcs_writel(sf->base, 0xf0000);
959 } else if (var->unusable)
963 ar |= (var->s & 1) << 4;
964 ar |= (var->dpl & 3) << 5;
965 ar |= (var->present & 1) << 7;
966 ar |= (var->avl & 1) << 12;
967 ar |= (var->l & 1) << 13;
968 ar |= (var->db & 1) << 14;
969 ar |= (var->g & 1) << 15;
971 if (ar == 0) /* a 0 value means unusable */
972 ar = AR_UNUSABLE_MASK;
973 vmcs_write32(sf->ar_bytes, ar);
976 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
978 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
980 *db = (ar >> 14) & 1;
984 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
986 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
987 dt->base = vmcs_readl(GUEST_IDTR_BASE);
990 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
992 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
993 vmcs_writel(GUEST_IDTR_BASE, dt->base);
996 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
998 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
999 dt->base = vmcs_readl(GUEST_GDTR_BASE);
1002 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1004 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1005 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1008 static int init_rmode_tss(struct kvm* kvm)
1010 struct page *p1, *p2, *p3;
1011 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
1014 p1 = gfn_to_page(kvm, fn++);
1015 p2 = gfn_to_page(kvm, fn++);
1016 p3 = gfn_to_page(kvm, fn);
1018 if (!p1 || !p2 || !p3) {
1019 kvm_printf(kvm,"%s: gfn_to_page failed\n", __FUNCTION__);
1023 page = kmap_atomic(p1, KM_USER0);
1024 memset(page, 0, PAGE_SIZE);
1025 *(u16*)(page + 0x66) = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
1026 kunmap_atomic(page, KM_USER0);
1028 page = kmap_atomic(p2, KM_USER0);
1029 memset(page, 0, PAGE_SIZE);
1030 kunmap_atomic(page, KM_USER0);
1032 page = kmap_atomic(p3, KM_USER0);
1033 memset(page, 0, PAGE_SIZE);
1034 *(page + RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1) = ~0;
1035 kunmap_atomic(page, KM_USER0);
1040 static void vmcs_write32_fixedbits(u32 msr, u32 vmcs_field, u32 val)
1042 u32 msr_high, msr_low;
1044 rdmsr(msr, msr_low, msr_high);
1048 vmcs_write32(vmcs_field, val);
1051 static void seg_setup(int seg)
1053 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1055 vmcs_write16(sf->selector, 0);
1056 vmcs_writel(sf->base, 0);
1057 vmcs_write32(sf->limit, 0xffff);
1058 vmcs_write32(sf->ar_bytes, 0x93);
1062 * Sets up the vmcs for emulated real mode.
1064 static int vmx_vcpu_setup(struct kvm_vcpu *vcpu)
1066 u32 host_sysenter_cs;
1069 struct descriptor_table dt;
1072 extern asmlinkage void kvm_vmx_return(void);
1074 if (!init_rmode_tss(vcpu->kvm)) {
1079 memset(vcpu->regs, 0, sizeof(vcpu->regs));
1080 vcpu->regs[VCPU_REGS_RDX] = get_rdx_init_val();
1082 vcpu->apic_base = 0xfee00000 |
1083 /*for vcpu 0*/ MSR_IA32_APICBASE_BSP |
1084 MSR_IA32_APICBASE_ENABLE;
1089 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
1090 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
1092 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
1093 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
1094 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1095 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1097 seg_setup(VCPU_SREG_DS);
1098 seg_setup(VCPU_SREG_ES);
1099 seg_setup(VCPU_SREG_FS);
1100 seg_setup(VCPU_SREG_GS);
1101 seg_setup(VCPU_SREG_SS);
1103 vmcs_write16(GUEST_TR_SELECTOR, 0);
1104 vmcs_writel(GUEST_TR_BASE, 0);
1105 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
1106 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1108 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
1109 vmcs_writel(GUEST_LDTR_BASE, 0);
1110 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
1111 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
1113 vmcs_write32(GUEST_SYSENTER_CS, 0);
1114 vmcs_writel(GUEST_SYSENTER_ESP, 0);
1115 vmcs_writel(GUEST_SYSENTER_EIP, 0);
1117 vmcs_writel(GUEST_RFLAGS, 0x02);
1118 vmcs_writel(GUEST_RIP, 0xfff0);
1119 vmcs_writel(GUEST_RSP, 0);
1121 //todo: dr0 = dr1 = dr2 = dr3 = 0; dr6 = 0xffff0ff0
1122 vmcs_writel(GUEST_DR7, 0x400);
1124 vmcs_writel(GUEST_GDTR_BASE, 0);
1125 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
1127 vmcs_writel(GUEST_IDTR_BASE, 0);
1128 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
1130 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
1131 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
1132 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
1135 vmcs_write64(IO_BITMAP_A, page_to_phys(vmx_io_bitmap_a));
1136 vmcs_write64(IO_BITMAP_B, page_to_phys(vmx_io_bitmap_b));
1140 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
1142 /* Special registers */
1143 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
1146 vmcs_write32_fixedbits(MSR_IA32_VMX_PINBASED_CTLS,
1147 PIN_BASED_VM_EXEC_CONTROL,
1148 PIN_BASED_EXT_INTR_MASK /* 20.6.1 */
1149 | PIN_BASED_NMI_EXITING /* 20.6.1 */
1151 vmcs_write32_fixedbits(MSR_IA32_VMX_PROCBASED_CTLS,
1152 CPU_BASED_VM_EXEC_CONTROL,
1153 CPU_BASED_HLT_EXITING /* 20.6.2 */
1154 | CPU_BASED_CR8_LOAD_EXITING /* 20.6.2 */
1155 | CPU_BASED_CR8_STORE_EXITING /* 20.6.2 */
1156 | CPU_BASED_ACTIVATE_IO_BITMAP /* 20.6.2 */
1157 | CPU_BASED_MOV_DR_EXITING
1158 | CPU_BASED_USE_TSC_OFFSETING /* 21.3 */
1161 vmcs_write32(EXCEPTION_BITMAP, 1 << PF_VECTOR);
1162 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
1163 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
1164 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
1166 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
1167 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
1168 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
1170 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
1171 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1172 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1173 vmcs_write16(HOST_FS_SELECTOR, read_fs()); /* 22.2.4 */
1174 vmcs_write16(HOST_GS_SELECTOR, read_gs()); /* 22.2.4 */
1175 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1176 #ifdef CONFIG_X86_64
1177 rdmsrl(MSR_FS_BASE, a);
1178 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
1179 rdmsrl(MSR_GS_BASE, a);
1180 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
1182 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
1183 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
1186 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
1189 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
1192 vmcs_writel(HOST_RIP, (unsigned long)kvm_vmx_return); /* 22.2.5 */
1194 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
1195 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
1196 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
1197 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
1198 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
1199 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
1201 for (i = 0; i < NR_VMX_MSR; ++i) {
1202 u32 index = vmx_msr_index[i];
1203 u32 data_low, data_high;
1205 int j = vcpu->nmsrs;
1207 if (rdmsr_safe(index, &data_low, &data_high) < 0)
1209 if (wrmsr_safe(index, data_low, data_high) < 0)
1211 data = data_low | ((u64)data_high << 32);
1212 vcpu->host_msrs[j].index = index;
1213 vcpu->host_msrs[j].reserved = 0;
1214 vcpu->host_msrs[j].data = data;
1215 vcpu->guest_msrs[j] = vcpu->host_msrs[j];
1216 #ifdef CONFIG_X86_64
1217 if (index == MSR_KERNEL_GS_BASE)
1218 msr_offset_kernel_gs_base = j;
1225 vmcs_write32_fixedbits(MSR_IA32_VMX_EXIT_CTLS, VM_EXIT_CONTROLS,
1226 (HOST_IS_64 << 9)); /* 22.2,1, 20.7.1 */
1228 /* 22.2.1, 20.8.1 */
1229 vmcs_write32_fixedbits(MSR_IA32_VMX_ENTRY_CTLS,
1230 VM_ENTRY_CONTROLS, 0);
1231 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
1233 #ifdef CONFIG_X86_64
1234 vmcs_writel(VIRTUAL_APIC_PAGE_ADDR, 0);
1235 vmcs_writel(TPR_THRESHOLD, 0);
1238 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
1239 vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
1241 vcpu->cr0 = 0x60000010;
1242 vmx_set_cr0(vcpu, vcpu->cr0); // enter rmode
1243 vmx_set_cr4(vcpu, 0);
1244 #ifdef CONFIG_X86_64
1245 vmx_set_efer(vcpu, 0);
1254 static void inject_rmode_irq(struct kvm_vcpu *vcpu, int irq)
1259 unsigned long flags;
1260 unsigned long ss_base = vmcs_readl(GUEST_SS_BASE);
1261 u16 sp = vmcs_readl(GUEST_RSP);
1262 u32 ss_limit = vmcs_read32(GUEST_SS_LIMIT);
1264 if (sp > ss_limit || sp < 6 ) {
1265 vcpu_printf(vcpu, "%s: #SS, rsp 0x%lx ss 0x%lx limit 0x%x\n",
1267 vmcs_readl(GUEST_RSP),
1268 vmcs_readl(GUEST_SS_BASE),
1269 vmcs_read32(GUEST_SS_LIMIT));
1273 if (kvm_read_guest(vcpu, irq * sizeof(ent), sizeof(ent), &ent) !=
1275 vcpu_printf(vcpu, "%s: read guest err\n", __FUNCTION__);
1279 flags = vmcs_readl(GUEST_RFLAGS);
1280 cs = vmcs_readl(GUEST_CS_BASE) >> 4;
1281 ip = vmcs_readl(GUEST_RIP);
1284 if (kvm_write_guest(vcpu, ss_base + sp - 2, 2, &flags) != 2 ||
1285 kvm_write_guest(vcpu, ss_base + sp - 4, 2, &cs) != 2 ||
1286 kvm_write_guest(vcpu, ss_base + sp - 6, 2, &ip) != 2) {
1287 vcpu_printf(vcpu, "%s: write guest err\n", __FUNCTION__);
1291 vmcs_writel(GUEST_RFLAGS, flags &
1292 ~( X86_EFLAGS_IF | X86_EFLAGS_AC | X86_EFLAGS_TF));
1293 vmcs_write16(GUEST_CS_SELECTOR, ent[1]) ;
1294 vmcs_writel(GUEST_CS_BASE, ent[1] << 4);
1295 vmcs_writel(GUEST_RIP, ent[0]);
1296 vmcs_writel(GUEST_RSP, (vmcs_readl(GUEST_RSP) & ~0xffff) | (sp - 6));
1299 static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1301 int word_index = __ffs(vcpu->irq_summary);
1302 int bit_index = __ffs(vcpu->irq_pending[word_index]);
1303 int irq = word_index * BITS_PER_LONG + bit_index;
1305 clear_bit(bit_index, &vcpu->irq_pending[word_index]);
1306 if (!vcpu->irq_pending[word_index])
1307 clear_bit(word_index, &vcpu->irq_summary);
1309 if (vcpu->rmode.active) {
1310 inject_rmode_irq(vcpu, irq);
1313 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1314 irq | INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1318 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1319 struct kvm_run *kvm_run)
1321 u32 cpu_based_vm_exec_control;
1323 vcpu->interrupt_window_open =
1324 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
1325 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
1327 if (vcpu->interrupt_window_open &&
1328 vcpu->irq_summary &&
1329 !(vmcs_read32(VM_ENTRY_INTR_INFO_FIELD) & INTR_INFO_VALID_MASK))
1331 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1333 kvm_do_inject_irq(vcpu);
1335 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
1336 if (!vcpu->interrupt_window_open &&
1337 (vcpu->irq_summary || kvm_run->request_interrupt_window))
1339 * Interrupts blocked. Wait for unblock.
1341 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
1343 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
1344 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
1347 static void kvm_guest_debug_pre(struct kvm_vcpu *vcpu)
1349 struct kvm_guest_debug *dbg = &vcpu->guest_debug;
1351 set_debugreg(dbg->bp[0], 0);
1352 set_debugreg(dbg->bp[1], 1);
1353 set_debugreg(dbg->bp[2], 2);
1354 set_debugreg(dbg->bp[3], 3);
1356 if (dbg->singlestep) {
1357 unsigned long flags;
1359 flags = vmcs_readl(GUEST_RFLAGS);
1360 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1361 vmcs_writel(GUEST_RFLAGS, flags);
1365 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
1366 int vec, u32 err_code)
1368 if (!vcpu->rmode.active)
1371 if (vec == GP_VECTOR && err_code == 0)
1372 if (emulate_instruction(vcpu, NULL, 0, 0) == EMULATE_DONE)
1377 static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1379 u32 intr_info, error_code;
1380 unsigned long cr2, rip;
1382 enum emulation_result er;
1385 vect_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
1386 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
1388 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
1389 !is_page_fault(intr_info)) {
1390 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
1391 "intr info 0x%x\n", __FUNCTION__, vect_info, intr_info);
1394 if (is_external_interrupt(vect_info)) {
1395 int irq = vect_info & VECTORING_INFO_VECTOR_MASK;
1396 set_bit(irq, vcpu->irq_pending);
1397 set_bit(irq / BITS_PER_LONG, &vcpu->irq_summary);
1400 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) { /* nmi */
1405 if (is_no_device(intr_info)) {
1406 vcpu->fpu_active = 1;
1407 vmcs_clear_bits(EXCEPTION_BITMAP, 1 << NM_VECTOR);
1408 if (!(vcpu->cr0 & CR0_TS_MASK))
1409 vmcs_clear_bits(GUEST_CR0, CR0_TS_MASK);
1414 rip = vmcs_readl(GUEST_RIP);
1415 if (intr_info & INTR_INFO_DELIEVER_CODE_MASK)
1416 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
1417 if (is_page_fault(intr_info)) {
1418 cr2 = vmcs_readl(EXIT_QUALIFICATION);
1420 spin_lock(&vcpu->kvm->lock);
1421 r = kvm_mmu_page_fault(vcpu, cr2, error_code);
1423 spin_unlock(&vcpu->kvm->lock);
1427 spin_unlock(&vcpu->kvm->lock);
1431 er = emulate_instruction(vcpu, kvm_run, cr2, error_code);
1432 spin_unlock(&vcpu->kvm->lock);
1437 case EMULATE_DO_MMIO:
1438 ++vcpu->stat.mmio_exits;
1439 kvm_run->exit_reason = KVM_EXIT_MMIO;
1442 vcpu_printf(vcpu, "%s: emulate fail\n", __FUNCTION__);
1449 if (vcpu->rmode.active &&
1450 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
1454 if ((intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK)) == (INTR_TYPE_EXCEPTION | 1)) {
1455 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1458 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
1459 kvm_run->ex.exception = intr_info & INTR_INFO_VECTOR_MASK;
1460 kvm_run->ex.error_code = error_code;
1464 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
1465 struct kvm_run *kvm_run)
1467 ++vcpu->stat.irq_exits;
1471 static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1473 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1477 static int get_io_count(struct kvm_vcpu *vcpu, unsigned long *count)
1484 if ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_VM)) {
1487 u32 cs_ar = vmcs_read32(GUEST_CS_AR_BYTES);
1489 countr_size = (cs_ar & AR_L_MASK) ? 8:
1490 (cs_ar & AR_DB_MASK) ? 4: 2;
1493 rip = vmcs_readl(GUEST_RIP);
1494 if (countr_size != 8)
1495 rip += vmcs_readl(GUEST_CS_BASE);
1497 n = kvm_read_guest(vcpu, rip, sizeof(inst), &inst);
1499 for (i = 0; i < n; i++) {
1500 switch (((u8*)&inst)[i]) {
1513 countr_size = (countr_size == 2) ? 4: (countr_size >> 1);
1521 *count = vcpu->regs[VCPU_REGS_RCX] & (~0ULL >> (64 - countr_size));
1522 //printk("cx: %lx\n", vcpu->regs[VCPU_REGS_RCX]);
1526 static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1528 u64 exit_qualification;
1529 int size, down, in, string, rep;
1531 unsigned long count;
1534 ++vcpu->stat.io_exits;
1535 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
1536 in = (exit_qualification & 8) != 0;
1537 size = (exit_qualification & 7) + 1;
1538 string = (exit_qualification & 16) != 0;
1539 down = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_DF) != 0;
1541 rep = (exit_qualification & 32) != 0;
1542 port = exit_qualification >> 16;
1545 if (rep && !get_io_count(vcpu, &count))
1547 address = vmcs_readl(GUEST_LINEAR_ADDRESS);
1549 return kvm_setup_pio(vcpu, kvm_run, in, size, count, string, down,
1550 address, rep, port);
1554 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1557 * Patch in the VMCALL instruction:
1559 hypercall[0] = 0x0f;
1560 hypercall[1] = 0x01;
1561 hypercall[2] = 0xc1;
1562 hypercall[3] = 0xc3;
1565 static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1567 u64 exit_qualification;
1571 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
1572 cr = exit_qualification & 15;
1573 reg = (exit_qualification >> 8) & 15;
1574 switch ((exit_qualification >> 4) & 3) {
1575 case 0: /* mov to cr */
1578 vcpu_load_rsp_rip(vcpu);
1579 set_cr0(vcpu, vcpu->regs[reg]);
1580 skip_emulated_instruction(vcpu);
1583 vcpu_load_rsp_rip(vcpu);
1584 set_cr3(vcpu, vcpu->regs[reg]);
1585 skip_emulated_instruction(vcpu);
1588 vcpu_load_rsp_rip(vcpu);
1589 set_cr4(vcpu, vcpu->regs[reg]);
1590 skip_emulated_instruction(vcpu);
1593 vcpu_load_rsp_rip(vcpu);
1594 set_cr8(vcpu, vcpu->regs[reg]);
1595 skip_emulated_instruction(vcpu);
1600 vcpu_load_rsp_rip(vcpu);
1601 vcpu->fpu_active = 1;
1602 vmcs_clear_bits(EXCEPTION_BITMAP, 1 << NM_VECTOR);
1603 vmcs_clear_bits(GUEST_CR0, CR0_TS_MASK);
1604 vcpu->cr0 &= ~CR0_TS_MASK;
1605 vmcs_writel(CR0_READ_SHADOW, vcpu->cr0);
1606 skip_emulated_instruction(vcpu);
1608 case 1: /*mov from cr*/
1611 vcpu_load_rsp_rip(vcpu);
1612 vcpu->regs[reg] = vcpu->cr3;
1613 vcpu_put_rsp_rip(vcpu);
1614 skip_emulated_instruction(vcpu);
1617 vcpu_load_rsp_rip(vcpu);
1618 vcpu->regs[reg] = vcpu->cr8;
1619 vcpu_put_rsp_rip(vcpu);
1620 skip_emulated_instruction(vcpu);
1625 lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
1627 skip_emulated_instruction(vcpu);
1632 kvm_run->exit_reason = 0;
1633 printk(KERN_ERR "kvm: unhandled control register: op %d cr %d\n",
1634 (int)(exit_qualification >> 4) & 3, cr);
1638 static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1640 u64 exit_qualification;
1645 * FIXME: this code assumes the host is debugging the guest.
1646 * need to deal with guest debugging itself too.
1648 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
1649 dr = exit_qualification & 7;
1650 reg = (exit_qualification >> 8) & 15;
1651 vcpu_load_rsp_rip(vcpu);
1652 if (exit_qualification & 16) {
1664 vcpu->regs[reg] = val;
1668 vcpu_put_rsp_rip(vcpu);
1669 skip_emulated_instruction(vcpu);
1673 static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1675 kvm_emulate_cpuid(vcpu);
1679 static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1681 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1684 if (vmx_get_msr(vcpu, ecx, &data)) {
1685 vmx_inject_gp(vcpu, 0);
1689 /* FIXME: handling of bits 32:63 of rax, rdx */
1690 vcpu->regs[VCPU_REGS_RAX] = data & -1u;
1691 vcpu->regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
1692 skip_emulated_instruction(vcpu);
1696 static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1698 u32 ecx = vcpu->regs[VCPU_REGS_RCX];
1699 u64 data = (vcpu->regs[VCPU_REGS_RAX] & -1u)
1700 | ((u64)(vcpu->regs[VCPU_REGS_RDX] & -1u) << 32);
1702 if (vmx_set_msr(vcpu, ecx, data) != 0) {
1703 vmx_inject_gp(vcpu, 0);
1707 skip_emulated_instruction(vcpu);
1711 static void post_kvm_run_save(struct kvm_vcpu *vcpu,
1712 struct kvm_run *kvm_run)
1714 kvm_run->if_flag = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) != 0;
1715 kvm_run->cr8 = vcpu->cr8;
1716 kvm_run->apic_base = vcpu->apic_base;
1717 kvm_run->ready_for_interrupt_injection = (vcpu->interrupt_window_open &&
1718 vcpu->irq_summary == 0);
1721 static int handle_interrupt_window(struct kvm_vcpu *vcpu,
1722 struct kvm_run *kvm_run)
1725 * If the user space waits to inject interrupts, exit as soon as
1728 if (kvm_run->request_interrupt_window &&
1729 !vcpu->irq_summary) {
1730 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
1731 ++vcpu->stat.irq_window_exits;
1737 static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1739 skip_emulated_instruction(vcpu);
1740 if (vcpu->irq_summary)
1743 kvm_run->exit_reason = KVM_EXIT_HLT;
1744 ++vcpu->stat.halt_exits;
1748 static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1750 skip_emulated_instruction(vcpu);
1751 return kvm_hypercall(vcpu, kvm_run);
1755 * The exit handlers return 1 if the exit was handled fully and guest execution
1756 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
1757 * to be done to userspace and return 0.
1759 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
1760 struct kvm_run *kvm_run) = {
1761 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
1762 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
1763 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
1764 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
1765 [EXIT_REASON_CR_ACCESS] = handle_cr,
1766 [EXIT_REASON_DR_ACCESS] = handle_dr,
1767 [EXIT_REASON_CPUID] = handle_cpuid,
1768 [EXIT_REASON_MSR_READ] = handle_rdmsr,
1769 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
1770 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
1771 [EXIT_REASON_HLT] = handle_halt,
1772 [EXIT_REASON_VMCALL] = handle_vmcall,
1775 static const int kvm_vmx_max_exit_handlers =
1776 sizeof(kvm_vmx_exit_handlers) / sizeof(*kvm_vmx_exit_handlers);
1779 * The guest has exited. See if we can fix it or if we need userspace
1782 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
1784 u32 vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
1785 u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
1787 if ( (vectoring_info & VECTORING_INFO_VALID_MASK) &&
1788 exit_reason != EXIT_REASON_EXCEPTION_NMI )
1789 printk(KERN_WARNING "%s: unexpected, valid vectoring info and "
1790 "exit reason is 0x%x\n", __FUNCTION__, exit_reason);
1791 if (exit_reason < kvm_vmx_max_exit_handlers
1792 && kvm_vmx_exit_handlers[exit_reason])
1793 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
1795 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
1796 kvm_run->hw.hardware_exit_reason = exit_reason;
1802 * Check if userspace requested an interrupt window, and that the
1803 * interrupt window is open.
1805 * No need to exit to userspace if we already have an interrupt queued.
1807 static int dm_request_for_irq_injection(struct kvm_vcpu *vcpu,
1808 struct kvm_run *kvm_run)
1810 return (!vcpu->irq_summary &&
1811 kvm_run->request_interrupt_window &&
1812 vcpu->interrupt_window_open &&
1813 (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF));
1816 static int vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1819 u16 fs_sel, gs_sel, ldt_sel;
1820 int fs_gs_ldt_reload_needed;
1825 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1826 * allow segment selectors with cpl > 0 or ti == 1.
1830 ldt_sel = read_ldt();
1831 fs_gs_ldt_reload_needed = (fs_sel & 7) | (gs_sel & 7) | ldt_sel;
1832 if (!fs_gs_ldt_reload_needed) {
1833 vmcs_write16(HOST_FS_SELECTOR, fs_sel);
1834 vmcs_write16(HOST_GS_SELECTOR, gs_sel);
1836 vmcs_write16(HOST_FS_SELECTOR, 0);
1837 vmcs_write16(HOST_GS_SELECTOR, 0);
1840 #ifdef CONFIG_X86_64
1841 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1842 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1844 vmcs_writel(HOST_FS_BASE, segment_base(fs_sel));
1845 vmcs_writel(HOST_GS_BASE, segment_base(gs_sel));
1848 if (!vcpu->mmio_read_completed)
1849 do_interrupt_requests(vcpu, kvm_run);
1851 if (vcpu->guest_debug.enabled)
1852 kvm_guest_debug_pre(vcpu);
1854 kvm_load_guest_fpu(vcpu);
1857 * Loading guest fpu may have cleared host cr0.ts
1859 vmcs_writel(HOST_CR0, read_cr0());
1861 #ifdef CONFIG_X86_64
1862 if (is_long_mode(vcpu)) {
1863 save_msrs(vcpu->host_msrs + msr_offset_kernel_gs_base, 1);
1864 load_msrs(vcpu->guest_msrs, NR_BAD_MSRS);
1869 /* Store host registers */
1871 #ifdef CONFIG_X86_64
1872 "push %%rax; push %%rbx; push %%rdx;"
1873 "push %%rsi; push %%rdi; push %%rbp;"
1874 "push %%r8; push %%r9; push %%r10; push %%r11;"
1875 "push %%r12; push %%r13; push %%r14; push %%r15;"
1877 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
1879 "pusha; push %%ecx \n\t"
1880 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
1882 /* Check if vmlaunch of vmresume is needed */
1884 /* Load guest registers. Don't clobber flags. */
1885 #ifdef CONFIG_X86_64
1886 "mov %c[cr2](%3), %%rax \n\t"
1887 "mov %%rax, %%cr2 \n\t"
1888 "mov %c[rax](%3), %%rax \n\t"
1889 "mov %c[rbx](%3), %%rbx \n\t"
1890 "mov %c[rdx](%3), %%rdx \n\t"
1891 "mov %c[rsi](%3), %%rsi \n\t"
1892 "mov %c[rdi](%3), %%rdi \n\t"
1893 "mov %c[rbp](%3), %%rbp \n\t"
1894 "mov %c[r8](%3), %%r8 \n\t"
1895 "mov %c[r9](%3), %%r9 \n\t"
1896 "mov %c[r10](%3), %%r10 \n\t"
1897 "mov %c[r11](%3), %%r11 \n\t"
1898 "mov %c[r12](%3), %%r12 \n\t"
1899 "mov %c[r13](%3), %%r13 \n\t"
1900 "mov %c[r14](%3), %%r14 \n\t"
1901 "mov %c[r15](%3), %%r15 \n\t"
1902 "mov %c[rcx](%3), %%rcx \n\t" /* kills %3 (rcx) */
1904 "mov %c[cr2](%3), %%eax \n\t"
1905 "mov %%eax, %%cr2 \n\t"
1906 "mov %c[rax](%3), %%eax \n\t"
1907 "mov %c[rbx](%3), %%ebx \n\t"
1908 "mov %c[rdx](%3), %%edx \n\t"
1909 "mov %c[rsi](%3), %%esi \n\t"
1910 "mov %c[rdi](%3), %%edi \n\t"
1911 "mov %c[rbp](%3), %%ebp \n\t"
1912 "mov %c[rcx](%3), %%ecx \n\t" /* kills %3 (ecx) */
1914 /* Enter guest mode */
1916 ASM_VMX_VMLAUNCH "\n\t"
1917 "jmp kvm_vmx_return \n\t"
1918 "launched: " ASM_VMX_VMRESUME "\n\t"
1919 ".globl kvm_vmx_return \n\t"
1921 /* Save guest registers, load host registers, keep flags */
1922 #ifdef CONFIG_X86_64
1923 "xchg %3, (%%rsp) \n\t"
1924 "mov %%rax, %c[rax](%3) \n\t"
1925 "mov %%rbx, %c[rbx](%3) \n\t"
1926 "pushq (%%rsp); popq %c[rcx](%3) \n\t"
1927 "mov %%rdx, %c[rdx](%3) \n\t"
1928 "mov %%rsi, %c[rsi](%3) \n\t"
1929 "mov %%rdi, %c[rdi](%3) \n\t"
1930 "mov %%rbp, %c[rbp](%3) \n\t"
1931 "mov %%r8, %c[r8](%3) \n\t"
1932 "mov %%r9, %c[r9](%3) \n\t"
1933 "mov %%r10, %c[r10](%3) \n\t"
1934 "mov %%r11, %c[r11](%3) \n\t"
1935 "mov %%r12, %c[r12](%3) \n\t"
1936 "mov %%r13, %c[r13](%3) \n\t"
1937 "mov %%r14, %c[r14](%3) \n\t"
1938 "mov %%r15, %c[r15](%3) \n\t"
1939 "mov %%cr2, %%rax \n\t"
1940 "mov %%rax, %c[cr2](%3) \n\t"
1941 "mov (%%rsp), %3 \n\t"
1943 "pop %%rcx; pop %%r15; pop %%r14; pop %%r13; pop %%r12;"
1944 "pop %%r11; pop %%r10; pop %%r9; pop %%r8;"
1945 "pop %%rbp; pop %%rdi; pop %%rsi;"
1946 "pop %%rdx; pop %%rbx; pop %%rax \n\t"
1948 "xchg %3, (%%esp) \n\t"
1949 "mov %%eax, %c[rax](%3) \n\t"
1950 "mov %%ebx, %c[rbx](%3) \n\t"
1951 "pushl (%%esp); popl %c[rcx](%3) \n\t"
1952 "mov %%edx, %c[rdx](%3) \n\t"
1953 "mov %%esi, %c[rsi](%3) \n\t"
1954 "mov %%edi, %c[rdi](%3) \n\t"
1955 "mov %%ebp, %c[rbp](%3) \n\t"
1956 "mov %%cr2, %%eax \n\t"
1957 "mov %%eax, %c[cr2](%3) \n\t"
1958 "mov (%%esp), %3 \n\t"
1960 "pop %%ecx; popa \n\t"
1965 : "r"(vcpu->launched), "d"((unsigned long)HOST_RSP),
1967 [rax]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RAX])),
1968 [rbx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBX])),
1969 [rcx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RCX])),
1970 [rdx]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDX])),
1971 [rsi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RSI])),
1972 [rdi]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RDI])),
1973 [rbp]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_RBP])),
1974 #ifdef CONFIG_X86_64
1975 [r8 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R8 ])),
1976 [r9 ]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R9 ])),
1977 [r10]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R10])),
1978 [r11]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R11])),
1979 [r12]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R12])),
1980 [r13]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R13])),
1981 [r14]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R14])),
1982 [r15]"i"(offsetof(struct kvm_vcpu, regs[VCPU_REGS_R15])),
1984 [cr2]"i"(offsetof(struct kvm_vcpu, cr2))
1988 * Reload segment selectors ASAP. (it's needed for a functional
1989 * kernel: x86 relies on having __KERNEL_PDA in %fs and x86_64
1990 * relies on having 0 in %gs for the CPU PDA to work.)
1992 if (fs_gs_ldt_reload_needed) {
1996 * If we have to reload gs, we must take care to
1997 * preserve our gs base.
1999 local_irq_disable();
2001 #ifdef CONFIG_X86_64
2002 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
2010 #ifdef CONFIG_X86_64
2011 if (is_long_mode(vcpu)) {
2012 save_msrs(vcpu->guest_msrs, NR_BAD_MSRS);
2013 load_msrs(vcpu->host_msrs, NR_BAD_MSRS);
2017 vcpu->interrupt_window_open = (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0;
2019 asm ("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
2022 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2023 kvm_run->fail_entry.hardware_entry_failure_reason
2024 = vmcs_read32(VM_INSTRUCTION_ERROR);
2028 * Profile KVM exit RIPs:
2030 if (unlikely(prof_on == KVM_PROFILING))
2031 profile_hit(KVM_PROFILING, (void *)vmcs_readl(GUEST_RIP));
2034 r = kvm_handle_exit(kvm_run, vcpu);
2036 /* Give scheduler a change to reschedule. */
2037 if (signal_pending(current)) {
2038 ++vcpu->stat.signal_exits;
2039 post_kvm_run_save(vcpu, kvm_run);
2040 kvm_run->exit_reason = KVM_EXIT_INTR;
2044 if (dm_request_for_irq_injection(vcpu, kvm_run)) {
2045 ++vcpu->stat.request_irq_exits;
2046 post_kvm_run_save(vcpu, kvm_run);
2047 kvm_run->exit_reason = KVM_EXIT_INTR;
2056 post_kvm_run_save(vcpu, kvm_run);
2060 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
2062 vmcs_writel(GUEST_CR3, vmcs_readl(GUEST_CR3));
2065 static void vmx_inject_page_fault(struct kvm_vcpu *vcpu,
2069 u32 vect_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
2071 ++vcpu->stat.pf_guest;
2073 if (is_page_fault(vect_info)) {
2074 printk(KERN_DEBUG "inject_page_fault: "
2075 "double fault 0x%lx @ 0x%lx\n",
2076 addr, vmcs_readl(GUEST_RIP));
2077 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, 0);
2078 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2080 INTR_TYPE_EXCEPTION |
2081 INTR_INFO_DELIEVER_CODE_MASK |
2082 INTR_INFO_VALID_MASK);
2086 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, err_code);
2087 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2089 INTR_TYPE_EXCEPTION |
2090 INTR_INFO_DELIEVER_CODE_MASK |
2091 INTR_INFO_VALID_MASK);
2095 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
2098 on_each_cpu(__vcpu_clear, vcpu, 0, 1);
2099 free_vmcs(vcpu->vmcs);
2104 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
2106 vmx_free_vmcs(vcpu);
2109 static int vmx_create_vcpu(struct kvm_vcpu *vcpu)
2113 vcpu->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2114 if (!vcpu->guest_msrs)
2117 vcpu->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2118 if (!vcpu->host_msrs)
2119 goto out_free_guest_msrs;
2121 vmcs = alloc_vmcs();
2128 vcpu->fpu_active = 1;
2133 kfree(vcpu->host_msrs);
2134 vcpu->host_msrs = NULL;
2136 out_free_guest_msrs:
2137 kfree(vcpu->guest_msrs);
2138 vcpu->guest_msrs = NULL;
2143 static struct kvm_arch_ops vmx_arch_ops = {
2144 .cpu_has_kvm_support = cpu_has_kvm_support,
2145 .disabled_by_bios = vmx_disabled_by_bios,
2146 .hardware_setup = hardware_setup,
2147 .hardware_unsetup = hardware_unsetup,
2148 .hardware_enable = hardware_enable,
2149 .hardware_disable = hardware_disable,
2151 .vcpu_create = vmx_create_vcpu,
2152 .vcpu_free = vmx_free_vcpu,
2154 .vcpu_load = vmx_vcpu_load,
2155 .vcpu_put = vmx_vcpu_put,
2156 .vcpu_decache = vmx_vcpu_decache,
2158 .set_guest_debug = set_guest_debug,
2159 .get_msr = vmx_get_msr,
2160 .set_msr = vmx_set_msr,
2161 .get_segment_base = vmx_get_segment_base,
2162 .get_segment = vmx_get_segment,
2163 .set_segment = vmx_set_segment,
2164 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
2165 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
2166 .set_cr0 = vmx_set_cr0,
2167 .set_cr3 = vmx_set_cr3,
2168 .set_cr4 = vmx_set_cr4,
2169 #ifdef CONFIG_X86_64
2170 .set_efer = vmx_set_efer,
2172 .get_idt = vmx_get_idt,
2173 .set_idt = vmx_set_idt,
2174 .get_gdt = vmx_get_gdt,
2175 .set_gdt = vmx_set_gdt,
2176 .cache_regs = vcpu_load_rsp_rip,
2177 .decache_regs = vcpu_put_rsp_rip,
2178 .get_rflags = vmx_get_rflags,
2179 .set_rflags = vmx_set_rflags,
2181 .tlb_flush = vmx_flush_tlb,
2182 .inject_page_fault = vmx_inject_page_fault,
2184 .inject_gp = vmx_inject_gp,
2186 .run = vmx_vcpu_run,
2187 .skip_emulated_instruction = skip_emulated_instruction,
2188 .vcpu_setup = vmx_vcpu_setup,
2189 .patch_hypercall = vmx_patch_hypercall,
2192 static int __init vmx_init(void)
2197 vmx_io_bitmap_a = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2198 if (!vmx_io_bitmap_a)
2201 vmx_io_bitmap_b = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2202 if (!vmx_io_bitmap_b) {
2208 * Allow direct access to the PC debug port (it is often used for I/O
2209 * delays, but the vmexits simply slow things down).
2211 iova = kmap(vmx_io_bitmap_a);
2212 memset(iova, 0xff, PAGE_SIZE);
2213 clear_bit(0x80, iova);
2216 iova = kmap(vmx_io_bitmap_b);
2217 memset(iova, 0xff, PAGE_SIZE);
2220 r = kvm_init_arch(&vmx_arch_ops, THIS_MODULE);
2227 __free_page(vmx_io_bitmap_b);
2229 __free_page(vmx_io_bitmap_a);
2233 static void __exit vmx_exit(void)
2235 __free_page(vmx_io_bitmap_b);
2236 __free_page(vmx_io_bitmap_a);
2241 module_init(vmx_init)
2242 module_exit(vmx_exit)