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/kvm_host.h>
22 #include <linux/module.h>
23 #include <linux/kernel.h>
25 #include <linux/highmem.h>
26 #include <linux/sched.h>
27 #include <linux/moduleparam.h>
28 #include <linux/ftrace_event.h>
29 #include "kvm_cache_regs.h"
35 #include <asm/virtext.h>
40 #define __ex(x) __kvm_handle_fault_on_reboot(x)
42 MODULE_AUTHOR("Qumranet");
43 MODULE_LICENSE("GPL");
45 static int __read_mostly bypass_guest_pf = 1;
46 module_param(bypass_guest_pf, bool, S_IRUGO);
48 static int __read_mostly enable_vpid = 1;
49 module_param_named(vpid, enable_vpid, bool, 0444);
51 static int __read_mostly flexpriority_enabled = 1;
52 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
54 static int __read_mostly enable_ept = 1;
55 module_param_named(ept, enable_ept, bool, S_IRUGO);
57 static int __read_mostly enable_unrestricted_guest = 1;
58 module_param_named(unrestricted_guest,
59 enable_unrestricted_guest, bool, S_IRUGO);
61 static int __read_mostly emulate_invalid_guest_state = 0;
62 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
65 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
66 * ple_gap: upper bound on the amount of time between two successive
67 * executions of PAUSE in a loop. Also indicate if ple enabled.
68 * According to test, this time is usually small than 41 cycles.
69 * ple_window: upper bound on the amount of time a guest is allowed to execute
70 * in a PAUSE loop. Tests indicate that most spinlocks are held for
71 * less than 2^12 cycles
72 * Time is measured based on a counter that runs at the same rate as the TSC,
73 * refer SDM volume 3b section 21.6.13 & 22.1.3.
75 #define KVM_VMX_DEFAULT_PLE_GAP 41
76 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
77 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
78 module_param(ple_gap, int, S_IRUGO);
80 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
81 module_param(ple_window, int, S_IRUGO);
89 struct shared_msr_entry {
96 struct list_head local_vcpus_link;
97 unsigned long host_rsp;
100 u32 idt_vectoring_info;
101 struct shared_msr_entry *guest_msrs;
105 u64 msr_host_kernel_gs_base;
106 u64 msr_guest_kernel_gs_base;
111 u16 fs_sel, gs_sel, ldt_sel;
112 int gs_ldt_reload_needed;
113 int fs_reload_needed;
118 struct kvm_save_segment {
123 } tr, es, ds, fs, gs;
131 bool emulation_required;
133 /* Support for vnmi-less CPUs */
134 int soft_vnmi_blocked;
136 s64 vnmi_blocked_time;
140 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
142 return container_of(vcpu, struct vcpu_vmx, vcpu);
145 static int init_rmode(struct kvm *kvm);
146 static u64 construct_eptp(unsigned long root_hpa);
148 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
149 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
150 static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
152 static unsigned long *vmx_io_bitmap_a;
153 static unsigned long *vmx_io_bitmap_b;
154 static unsigned long *vmx_msr_bitmap_legacy;
155 static unsigned long *vmx_msr_bitmap_longmode;
157 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
158 static DEFINE_SPINLOCK(vmx_vpid_lock);
160 static struct vmcs_config {
164 u32 pin_based_exec_ctrl;
165 u32 cpu_based_exec_ctrl;
166 u32 cpu_based_2nd_exec_ctrl;
171 static struct vmx_capability {
176 #define VMX_SEGMENT_FIELD(seg) \
177 [VCPU_SREG_##seg] = { \
178 .selector = GUEST_##seg##_SELECTOR, \
179 .base = GUEST_##seg##_BASE, \
180 .limit = GUEST_##seg##_LIMIT, \
181 .ar_bytes = GUEST_##seg##_AR_BYTES, \
184 static struct kvm_vmx_segment_field {
189 } kvm_vmx_segment_fields[] = {
190 VMX_SEGMENT_FIELD(CS),
191 VMX_SEGMENT_FIELD(DS),
192 VMX_SEGMENT_FIELD(ES),
193 VMX_SEGMENT_FIELD(FS),
194 VMX_SEGMENT_FIELD(GS),
195 VMX_SEGMENT_FIELD(SS),
196 VMX_SEGMENT_FIELD(TR),
197 VMX_SEGMENT_FIELD(LDTR),
200 static u64 host_efer;
202 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
205 * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
206 * away by decrementing the array size.
208 static const u32 vmx_msr_index[] = {
210 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
212 MSR_EFER, MSR_K6_STAR,
214 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
216 static inline int is_page_fault(u32 intr_info)
218 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
219 INTR_INFO_VALID_MASK)) ==
220 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
223 static inline int is_no_device(u32 intr_info)
225 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
226 INTR_INFO_VALID_MASK)) ==
227 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
230 static inline int is_invalid_opcode(u32 intr_info)
232 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
233 INTR_INFO_VALID_MASK)) ==
234 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
237 static inline int is_external_interrupt(u32 intr_info)
239 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
240 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
243 static inline int is_machine_check(u32 intr_info)
245 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
246 INTR_INFO_VALID_MASK)) ==
247 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
250 static inline int cpu_has_vmx_msr_bitmap(void)
252 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
255 static inline int cpu_has_vmx_tpr_shadow(void)
257 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
260 static inline int vm_need_tpr_shadow(struct kvm *kvm)
262 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
265 static inline int cpu_has_secondary_exec_ctrls(void)
267 return vmcs_config.cpu_based_exec_ctrl &
268 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
271 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
273 return vmcs_config.cpu_based_2nd_exec_ctrl &
274 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
277 static inline bool cpu_has_vmx_flexpriority(void)
279 return cpu_has_vmx_tpr_shadow() &&
280 cpu_has_vmx_virtualize_apic_accesses();
283 static inline bool cpu_has_vmx_ept_execute_only(void)
285 return !!(vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT);
288 static inline bool cpu_has_vmx_eptp_uncacheable(void)
290 return !!(vmx_capability.ept & VMX_EPTP_UC_BIT);
293 static inline bool cpu_has_vmx_eptp_writeback(void)
295 return !!(vmx_capability.ept & VMX_EPTP_WB_BIT);
298 static inline bool cpu_has_vmx_ept_2m_page(void)
300 return !!(vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT);
303 static inline int cpu_has_vmx_invept_individual_addr(void)
305 return !!(vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT);
308 static inline int cpu_has_vmx_invept_context(void)
310 return !!(vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT);
313 static inline int cpu_has_vmx_invept_global(void)
315 return !!(vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT);
318 static inline int cpu_has_vmx_ept(void)
320 return vmcs_config.cpu_based_2nd_exec_ctrl &
321 SECONDARY_EXEC_ENABLE_EPT;
324 static inline int cpu_has_vmx_unrestricted_guest(void)
326 return vmcs_config.cpu_based_2nd_exec_ctrl &
327 SECONDARY_EXEC_UNRESTRICTED_GUEST;
330 static inline int cpu_has_vmx_ple(void)
332 return vmcs_config.cpu_based_2nd_exec_ctrl &
333 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
336 static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
338 return flexpriority_enabled &&
339 (cpu_has_vmx_virtualize_apic_accesses()) &&
340 (irqchip_in_kernel(kvm));
343 static inline int cpu_has_vmx_vpid(void)
345 return vmcs_config.cpu_based_2nd_exec_ctrl &
346 SECONDARY_EXEC_ENABLE_VPID;
349 static inline int cpu_has_virtual_nmis(void)
351 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
354 static inline bool report_flexpriority(void)
356 return flexpriority_enabled;
359 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
363 for (i = 0; i < vmx->nmsrs; ++i)
364 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
369 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
375 } operand = { vpid, 0, gva };
377 asm volatile (__ex(ASM_VMX_INVVPID)
378 /* CF==1 or ZF==1 --> rc = -1 */
380 : : "a"(&operand), "c"(ext) : "cc", "memory");
383 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
387 } operand = {eptp, gpa};
389 asm volatile (__ex(ASM_VMX_INVEPT)
390 /* CF==1 or ZF==1 --> rc = -1 */
391 "; ja 1f ; ud2 ; 1:\n"
392 : : "a" (&operand), "c" (ext) : "cc", "memory");
395 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
399 i = __find_msr_index(vmx, msr);
401 return &vmx->guest_msrs[i];
405 static void vmcs_clear(struct vmcs *vmcs)
407 u64 phys_addr = __pa(vmcs);
410 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
411 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
414 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
418 static void __vcpu_clear(void *arg)
420 struct vcpu_vmx *vmx = arg;
421 int cpu = raw_smp_processor_id();
423 if (vmx->vcpu.cpu == cpu)
424 vmcs_clear(vmx->vmcs);
425 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
426 per_cpu(current_vmcs, cpu) = NULL;
427 rdtscll(vmx->vcpu.arch.host_tsc);
428 list_del(&vmx->local_vcpus_link);
433 static void vcpu_clear(struct vcpu_vmx *vmx)
435 if (vmx->vcpu.cpu == -1)
437 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
440 static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
445 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
448 static inline void ept_sync_global(void)
450 if (cpu_has_vmx_invept_global())
451 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
454 static inline void ept_sync_context(u64 eptp)
457 if (cpu_has_vmx_invept_context())
458 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
464 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
467 if (cpu_has_vmx_invept_individual_addr())
468 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
471 ept_sync_context(eptp);
475 static unsigned long vmcs_readl(unsigned long field)
479 asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
480 : "=a"(value) : "d"(field) : "cc");
484 static u16 vmcs_read16(unsigned long field)
486 return vmcs_readl(field);
489 static u32 vmcs_read32(unsigned long field)
491 return vmcs_readl(field);
494 static u64 vmcs_read64(unsigned long field)
497 return vmcs_readl(field);
499 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
503 static noinline void vmwrite_error(unsigned long field, unsigned long value)
505 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
506 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
510 static void vmcs_writel(unsigned long field, unsigned long value)
514 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
515 : "=q"(error) : "a"(value), "d"(field) : "cc");
517 vmwrite_error(field, value);
520 static void vmcs_write16(unsigned long field, u16 value)
522 vmcs_writel(field, value);
525 static void vmcs_write32(unsigned long field, u32 value)
527 vmcs_writel(field, value);
530 static void vmcs_write64(unsigned long field, u64 value)
532 vmcs_writel(field, value);
533 #ifndef CONFIG_X86_64
535 vmcs_writel(field+1, value >> 32);
539 static void vmcs_clear_bits(unsigned long field, u32 mask)
541 vmcs_writel(field, vmcs_readl(field) & ~mask);
544 static void vmcs_set_bits(unsigned long field, u32 mask)
546 vmcs_writel(field, vmcs_readl(field) | mask);
549 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
553 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR);
554 if (!vcpu->fpu_active)
555 eb |= 1u << NM_VECTOR;
557 * Unconditionally intercept #DB so we can maintain dr6 without
558 * reading it every exit.
560 eb |= 1u << DB_VECTOR;
561 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
562 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
563 eb |= 1u << BP_VECTOR;
565 if (to_vmx(vcpu)->rmode.vm86_active)
568 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
569 vmcs_write32(EXCEPTION_BITMAP, eb);
572 static void reload_tss(void)
575 * VT restores TR but not its size. Useless.
577 struct descriptor_table gdt;
578 struct desc_struct *descs;
581 descs = (void *)gdt.base;
582 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
586 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
591 guest_efer = vmx->vcpu.arch.shadow_efer;
594 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
597 ignore_bits = EFER_NX | EFER_SCE;
599 ignore_bits |= EFER_LMA | EFER_LME;
600 /* SCE is meaningful only in long mode on Intel */
601 if (guest_efer & EFER_LMA)
602 ignore_bits &= ~(u64)EFER_SCE;
604 if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
607 guest_efer &= ~ignore_bits;
608 guest_efer |= host_efer & ignore_bits;
609 vmx->guest_msrs[efer_offset].data = guest_efer;
613 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
615 struct vcpu_vmx *vmx = to_vmx(vcpu);
618 if (vmx->host_state.loaded)
621 vmx->host_state.loaded = 1;
623 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
624 * allow segment selectors with cpl > 0 or ti == 1.
626 vmx->host_state.ldt_sel = kvm_read_ldt();
627 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
628 vmx->host_state.fs_sel = kvm_read_fs();
629 if (!(vmx->host_state.fs_sel & 7)) {
630 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
631 vmx->host_state.fs_reload_needed = 0;
633 vmcs_write16(HOST_FS_SELECTOR, 0);
634 vmx->host_state.fs_reload_needed = 1;
636 vmx->host_state.gs_sel = kvm_read_gs();
637 if (!(vmx->host_state.gs_sel & 7))
638 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
640 vmcs_write16(HOST_GS_SELECTOR, 0);
641 vmx->host_state.gs_ldt_reload_needed = 1;
645 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
646 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
648 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
649 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
653 if (is_long_mode(&vmx->vcpu)) {
654 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
655 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
658 for (i = 0; i < vmx->save_nmsrs; ++i)
659 kvm_set_shared_msr(vmx->guest_msrs[i].index,
660 vmx->guest_msrs[i].data);
663 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
667 if (!vmx->host_state.loaded)
670 ++vmx->vcpu.stat.host_state_reload;
671 vmx->host_state.loaded = 0;
672 if (vmx->host_state.fs_reload_needed)
673 kvm_load_fs(vmx->host_state.fs_sel);
674 if (vmx->host_state.gs_ldt_reload_needed) {
675 kvm_load_ldt(vmx->host_state.ldt_sel);
677 * If we have to reload gs, we must take care to
678 * preserve our gs base.
680 local_irq_save(flags);
681 kvm_load_gs(vmx->host_state.gs_sel);
683 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
685 local_irq_restore(flags);
689 if (is_long_mode(&vmx->vcpu)) {
690 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
691 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
696 static void vmx_load_host_state(struct vcpu_vmx *vmx)
699 __vmx_load_host_state(vmx);
704 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
705 * vcpu mutex is already taken.
707 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
709 struct vcpu_vmx *vmx = to_vmx(vcpu);
710 u64 phys_addr = __pa(vmx->vmcs);
711 u64 tsc_this, delta, new_offset;
713 if (vcpu->cpu != cpu) {
715 kvm_migrate_timers(vcpu);
716 set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests);
718 list_add(&vmx->local_vcpus_link,
719 &per_cpu(vcpus_on_cpu, cpu));
723 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
726 per_cpu(current_vmcs, cpu) = vmx->vmcs;
727 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
728 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
731 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
732 vmx->vmcs, phys_addr);
735 if (vcpu->cpu != cpu) {
736 struct descriptor_table dt;
737 unsigned long sysenter_esp;
741 * Linux uses per-cpu TSS and GDT, so set these when switching
744 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
746 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
748 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
749 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
752 * Make sure the time stamp counter is monotonous.
755 if (tsc_this < vcpu->arch.host_tsc) {
756 delta = vcpu->arch.host_tsc - tsc_this;
757 new_offset = vmcs_read64(TSC_OFFSET) + delta;
758 vmcs_write64(TSC_OFFSET, new_offset);
763 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
765 __vmx_load_host_state(to_vmx(vcpu));
768 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
770 if (vcpu->fpu_active)
772 vcpu->fpu_active = 1;
773 vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
774 if (vcpu->arch.cr0 & X86_CR0_TS)
775 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
776 update_exception_bitmap(vcpu);
779 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
781 if (!vcpu->fpu_active)
783 vcpu->fpu_active = 0;
784 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
785 update_exception_bitmap(vcpu);
788 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
790 unsigned long rflags;
792 rflags = vmcs_readl(GUEST_RFLAGS);
793 if (to_vmx(vcpu)->rmode.vm86_active)
794 rflags &= ~(unsigned long)(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
798 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
800 if (to_vmx(vcpu)->rmode.vm86_active)
801 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
802 vmcs_writel(GUEST_RFLAGS, rflags);
805 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
807 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
810 if (interruptibility & GUEST_INTR_STATE_STI)
811 ret |= X86_SHADOW_INT_STI;
812 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
813 ret |= X86_SHADOW_INT_MOV_SS;
818 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
820 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
821 u32 interruptibility = interruptibility_old;
823 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
825 if (mask & X86_SHADOW_INT_MOV_SS)
826 interruptibility |= GUEST_INTR_STATE_MOV_SS;
827 if (mask & X86_SHADOW_INT_STI)
828 interruptibility |= GUEST_INTR_STATE_STI;
830 if ((interruptibility != interruptibility_old))
831 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
834 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
838 rip = kvm_rip_read(vcpu);
839 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
840 kvm_rip_write(vcpu, rip);
842 /* skipping an emulated instruction also counts */
843 vmx_set_interrupt_shadow(vcpu, 0);
846 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
847 bool has_error_code, u32 error_code)
849 struct vcpu_vmx *vmx = to_vmx(vcpu);
850 u32 intr_info = nr | INTR_INFO_VALID_MASK;
852 if (has_error_code) {
853 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
854 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
857 if (vmx->rmode.vm86_active) {
858 vmx->rmode.irq.pending = true;
859 vmx->rmode.irq.vector = nr;
860 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
861 if (kvm_exception_is_soft(nr))
862 vmx->rmode.irq.rip +=
863 vmx->vcpu.arch.event_exit_inst_len;
864 intr_info |= INTR_TYPE_SOFT_INTR;
865 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
866 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
867 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
871 if (kvm_exception_is_soft(nr)) {
872 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
873 vmx->vcpu.arch.event_exit_inst_len);
874 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
876 intr_info |= INTR_TYPE_HARD_EXCEPTION;
878 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
882 * Swap MSR entry in host/guest MSR entry array.
884 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
886 struct shared_msr_entry tmp;
888 tmp = vmx->guest_msrs[to];
889 vmx->guest_msrs[to] = vmx->guest_msrs[from];
890 vmx->guest_msrs[from] = tmp;
894 * Set up the vmcs to automatically save and restore system
895 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
896 * mode, as fiddling with msrs is very expensive.
898 static void setup_msrs(struct vcpu_vmx *vmx)
900 int save_nmsrs, index;
901 unsigned long *msr_bitmap;
903 vmx_load_host_state(vmx);
906 if (is_long_mode(&vmx->vcpu)) {
907 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
909 move_msr_up(vmx, index, save_nmsrs++);
910 index = __find_msr_index(vmx, MSR_LSTAR);
912 move_msr_up(vmx, index, save_nmsrs++);
913 index = __find_msr_index(vmx, MSR_CSTAR);
915 move_msr_up(vmx, index, save_nmsrs++);
917 * MSR_K6_STAR is only needed on long mode guests, and only
918 * if efer.sce is enabled.
920 index = __find_msr_index(vmx, MSR_K6_STAR);
921 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
922 move_msr_up(vmx, index, save_nmsrs++);
925 index = __find_msr_index(vmx, MSR_EFER);
926 if (index >= 0 && update_transition_efer(vmx, index))
927 move_msr_up(vmx, index, save_nmsrs++);
929 vmx->save_nmsrs = save_nmsrs;
931 if (cpu_has_vmx_msr_bitmap()) {
932 if (is_long_mode(&vmx->vcpu))
933 msr_bitmap = vmx_msr_bitmap_longmode;
935 msr_bitmap = vmx_msr_bitmap_legacy;
937 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
942 * reads and returns guest's timestamp counter "register"
943 * guest_tsc = host_tsc + tsc_offset -- 21.3
945 static u64 guest_read_tsc(void)
947 u64 host_tsc, tsc_offset;
950 tsc_offset = vmcs_read64(TSC_OFFSET);
951 return host_tsc + tsc_offset;
955 * writes 'guest_tsc' into guest's timestamp counter "register"
956 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
958 static void guest_write_tsc(u64 guest_tsc, u64 host_tsc)
960 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
964 * Reads an msr value (of 'msr_index') into 'pdata'.
965 * Returns 0 on success, non-0 otherwise.
966 * Assumes vcpu_load() was already called.
968 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
971 struct shared_msr_entry *msr;
974 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
981 data = vmcs_readl(GUEST_FS_BASE);
984 data = vmcs_readl(GUEST_GS_BASE);
986 case MSR_KERNEL_GS_BASE:
987 vmx_load_host_state(to_vmx(vcpu));
988 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
992 return kvm_get_msr_common(vcpu, msr_index, pdata);
994 data = guest_read_tsc();
996 case MSR_IA32_SYSENTER_CS:
997 data = vmcs_read32(GUEST_SYSENTER_CS);
999 case MSR_IA32_SYSENTER_EIP:
1000 data = vmcs_readl(GUEST_SYSENTER_EIP);
1002 case MSR_IA32_SYSENTER_ESP:
1003 data = vmcs_readl(GUEST_SYSENTER_ESP);
1006 vmx_load_host_state(to_vmx(vcpu));
1007 msr = find_msr_entry(to_vmx(vcpu), msr_index);
1009 vmx_load_host_state(to_vmx(vcpu));
1013 return kvm_get_msr_common(vcpu, msr_index, pdata);
1021 * Writes msr value into into the appropriate "register".
1022 * Returns 0 on success, non-0 otherwise.
1023 * Assumes vcpu_load() was already called.
1025 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1027 struct vcpu_vmx *vmx = to_vmx(vcpu);
1028 struct shared_msr_entry *msr;
1032 switch (msr_index) {
1034 vmx_load_host_state(vmx);
1035 ret = kvm_set_msr_common(vcpu, msr_index, data);
1037 #ifdef CONFIG_X86_64
1039 vmcs_writel(GUEST_FS_BASE, data);
1042 vmcs_writel(GUEST_GS_BASE, data);
1044 case MSR_KERNEL_GS_BASE:
1045 vmx_load_host_state(vmx);
1046 vmx->msr_guest_kernel_gs_base = data;
1049 case MSR_IA32_SYSENTER_CS:
1050 vmcs_write32(GUEST_SYSENTER_CS, data);
1052 case MSR_IA32_SYSENTER_EIP:
1053 vmcs_writel(GUEST_SYSENTER_EIP, data);
1055 case MSR_IA32_SYSENTER_ESP:
1056 vmcs_writel(GUEST_SYSENTER_ESP, data);
1060 guest_write_tsc(data, host_tsc);
1062 case MSR_IA32_CR_PAT:
1063 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1064 vmcs_write64(GUEST_IA32_PAT, data);
1065 vcpu->arch.pat = data;
1068 /* Otherwise falls through to kvm_set_msr_common */
1070 msr = find_msr_entry(vmx, msr_index);
1072 vmx_load_host_state(vmx);
1076 ret = kvm_set_msr_common(vcpu, msr_index, data);
1082 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1084 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1087 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1090 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1092 case VCPU_EXREG_PDPTR:
1094 ept_save_pdptrs(vcpu);
1101 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1103 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1104 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1106 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1108 update_exception_bitmap(vcpu);
1111 static __init int cpu_has_kvm_support(void)
1113 return cpu_has_vmx();
1116 static __init int vmx_disabled_by_bios(void)
1120 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1121 return (msr & (FEATURE_CONTROL_LOCKED |
1122 FEATURE_CONTROL_VMXON_ENABLED))
1123 == FEATURE_CONTROL_LOCKED;
1124 /* locked but not enabled */
1127 static int hardware_enable(void *garbage)
1129 int cpu = raw_smp_processor_id();
1130 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1133 if (read_cr4() & X86_CR4_VMXE)
1136 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1137 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1138 if ((old & (FEATURE_CONTROL_LOCKED |
1139 FEATURE_CONTROL_VMXON_ENABLED))
1140 != (FEATURE_CONTROL_LOCKED |
1141 FEATURE_CONTROL_VMXON_ENABLED))
1142 /* enable and lock */
1143 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
1144 FEATURE_CONTROL_LOCKED |
1145 FEATURE_CONTROL_VMXON_ENABLED);
1146 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1147 asm volatile (ASM_VMX_VMXON_RAX
1148 : : "a"(&phys_addr), "m"(phys_addr)
1156 static void vmclear_local_vcpus(void)
1158 int cpu = raw_smp_processor_id();
1159 struct vcpu_vmx *vmx, *n;
1161 list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1167 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1170 static void kvm_cpu_vmxoff(void)
1172 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1173 write_cr4(read_cr4() & ~X86_CR4_VMXE);
1176 static void hardware_disable(void *garbage)
1178 vmclear_local_vcpus();
1182 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1183 u32 msr, u32 *result)
1185 u32 vmx_msr_low, vmx_msr_high;
1186 u32 ctl = ctl_min | ctl_opt;
1188 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1190 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1191 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
1193 /* Ensure minimum (required) set of control bits are supported. */
1201 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1203 u32 vmx_msr_low, vmx_msr_high;
1204 u32 min, opt, min2, opt2;
1205 u32 _pin_based_exec_control = 0;
1206 u32 _cpu_based_exec_control = 0;
1207 u32 _cpu_based_2nd_exec_control = 0;
1208 u32 _vmexit_control = 0;
1209 u32 _vmentry_control = 0;
1211 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1212 opt = PIN_BASED_VIRTUAL_NMIS;
1213 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1214 &_pin_based_exec_control) < 0)
1217 min = CPU_BASED_HLT_EXITING |
1218 #ifdef CONFIG_X86_64
1219 CPU_BASED_CR8_LOAD_EXITING |
1220 CPU_BASED_CR8_STORE_EXITING |
1222 CPU_BASED_CR3_LOAD_EXITING |
1223 CPU_BASED_CR3_STORE_EXITING |
1224 CPU_BASED_USE_IO_BITMAPS |
1225 CPU_BASED_MOV_DR_EXITING |
1226 CPU_BASED_USE_TSC_OFFSETING |
1227 CPU_BASED_INVLPG_EXITING;
1228 opt = CPU_BASED_TPR_SHADOW |
1229 CPU_BASED_USE_MSR_BITMAPS |
1230 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1231 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1232 &_cpu_based_exec_control) < 0)
1234 #ifdef CONFIG_X86_64
1235 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1236 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1237 ~CPU_BASED_CR8_STORE_EXITING;
1239 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1241 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1242 SECONDARY_EXEC_WBINVD_EXITING |
1243 SECONDARY_EXEC_ENABLE_VPID |
1244 SECONDARY_EXEC_ENABLE_EPT |
1245 SECONDARY_EXEC_UNRESTRICTED_GUEST |
1246 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1247 if (adjust_vmx_controls(min2, opt2,
1248 MSR_IA32_VMX_PROCBASED_CTLS2,
1249 &_cpu_based_2nd_exec_control) < 0)
1252 #ifndef CONFIG_X86_64
1253 if (!(_cpu_based_2nd_exec_control &
1254 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1255 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1257 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1258 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1260 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
1261 CPU_BASED_CR3_STORE_EXITING |
1262 CPU_BASED_INVLPG_EXITING);
1263 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1264 vmx_capability.ept, vmx_capability.vpid);
1268 #ifdef CONFIG_X86_64
1269 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1271 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1272 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1273 &_vmexit_control) < 0)
1277 opt = VM_ENTRY_LOAD_IA32_PAT;
1278 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1279 &_vmentry_control) < 0)
1282 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1284 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1285 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1288 #ifdef CONFIG_X86_64
1289 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1290 if (vmx_msr_high & (1u<<16))
1294 /* Require Write-Back (WB) memory type for VMCS accesses. */
1295 if (((vmx_msr_high >> 18) & 15) != 6)
1298 vmcs_conf->size = vmx_msr_high & 0x1fff;
1299 vmcs_conf->order = get_order(vmcs_config.size);
1300 vmcs_conf->revision_id = vmx_msr_low;
1302 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1303 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1304 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1305 vmcs_conf->vmexit_ctrl = _vmexit_control;
1306 vmcs_conf->vmentry_ctrl = _vmentry_control;
1311 static struct vmcs *alloc_vmcs_cpu(int cpu)
1313 int node = cpu_to_node(cpu);
1317 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
1320 vmcs = page_address(pages);
1321 memset(vmcs, 0, vmcs_config.size);
1322 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1326 static struct vmcs *alloc_vmcs(void)
1328 return alloc_vmcs_cpu(raw_smp_processor_id());
1331 static void free_vmcs(struct vmcs *vmcs)
1333 free_pages((unsigned long)vmcs, vmcs_config.order);
1336 static void free_kvm_area(void)
1340 for_each_possible_cpu(cpu) {
1341 free_vmcs(per_cpu(vmxarea, cpu));
1342 per_cpu(vmxarea, cpu) = NULL;
1346 static __init int alloc_kvm_area(void)
1350 for_each_possible_cpu(cpu) {
1353 vmcs = alloc_vmcs_cpu(cpu);
1359 per_cpu(vmxarea, cpu) = vmcs;
1364 static __init int hardware_setup(void)
1366 if (setup_vmcs_config(&vmcs_config) < 0)
1369 if (boot_cpu_has(X86_FEATURE_NX))
1370 kvm_enable_efer_bits(EFER_NX);
1372 if (!cpu_has_vmx_vpid())
1375 if (!cpu_has_vmx_ept()) {
1377 enable_unrestricted_guest = 0;
1380 if (!cpu_has_vmx_unrestricted_guest())
1381 enable_unrestricted_guest = 0;
1383 if (!cpu_has_vmx_flexpriority())
1384 flexpriority_enabled = 0;
1386 if (!cpu_has_vmx_tpr_shadow())
1387 kvm_x86_ops->update_cr8_intercept = NULL;
1389 if (enable_ept && !cpu_has_vmx_ept_2m_page())
1390 kvm_disable_largepages();
1392 if (!cpu_has_vmx_ple())
1395 return alloc_kvm_area();
1398 static __exit void hardware_unsetup(void)
1403 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1405 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1407 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1408 vmcs_write16(sf->selector, save->selector);
1409 vmcs_writel(sf->base, save->base);
1410 vmcs_write32(sf->limit, save->limit);
1411 vmcs_write32(sf->ar_bytes, save->ar);
1413 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1415 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1419 static void enter_pmode(struct kvm_vcpu *vcpu)
1421 unsigned long flags;
1422 struct vcpu_vmx *vmx = to_vmx(vcpu);
1424 vmx->emulation_required = 1;
1425 vmx->rmode.vm86_active = 0;
1427 vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
1428 vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
1429 vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
1431 flags = vmcs_readl(GUEST_RFLAGS);
1432 flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
1433 flags |= (vmx->rmode.save_iopl << IOPL_SHIFT);
1434 vmcs_writel(GUEST_RFLAGS, flags);
1436 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1437 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1439 update_exception_bitmap(vcpu);
1441 if (emulate_invalid_guest_state)
1444 fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
1445 fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
1446 fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
1447 fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
1449 vmcs_write16(GUEST_SS_SELECTOR, 0);
1450 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1452 vmcs_write16(GUEST_CS_SELECTOR,
1453 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1454 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1457 static gva_t rmode_tss_base(struct kvm *kvm)
1459 if (!kvm->arch.tss_addr) {
1460 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1461 kvm->memslots[0].npages - 3;
1462 return base_gfn << PAGE_SHIFT;
1464 return kvm->arch.tss_addr;
1467 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1469 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1471 save->selector = vmcs_read16(sf->selector);
1472 save->base = vmcs_readl(sf->base);
1473 save->limit = vmcs_read32(sf->limit);
1474 save->ar = vmcs_read32(sf->ar_bytes);
1475 vmcs_write16(sf->selector, save->base >> 4);
1476 vmcs_write32(sf->base, save->base & 0xfffff);
1477 vmcs_write32(sf->limit, 0xffff);
1478 vmcs_write32(sf->ar_bytes, 0xf3);
1481 static void enter_rmode(struct kvm_vcpu *vcpu)
1483 unsigned long flags;
1484 struct vcpu_vmx *vmx = to_vmx(vcpu);
1486 if (enable_unrestricted_guest)
1489 vmx->emulation_required = 1;
1490 vmx->rmode.vm86_active = 1;
1492 vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1493 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1495 vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1496 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1498 vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1499 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1501 flags = vmcs_readl(GUEST_RFLAGS);
1502 vmx->rmode.save_iopl
1503 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
1505 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1507 vmcs_writel(GUEST_RFLAGS, flags);
1508 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1509 update_exception_bitmap(vcpu);
1511 if (emulate_invalid_guest_state)
1512 goto continue_rmode;
1514 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1515 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1516 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1518 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1519 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1520 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1521 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1522 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1524 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
1525 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
1526 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
1527 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
1530 kvm_mmu_reset_context(vcpu);
1531 init_rmode(vcpu->kvm);
1534 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1536 struct vcpu_vmx *vmx = to_vmx(vcpu);
1537 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1543 * Force kernel_gs_base reloading before EFER changes, as control
1544 * of this msr depends on is_long_mode().
1546 vmx_load_host_state(to_vmx(vcpu));
1547 vcpu->arch.shadow_efer = efer;
1550 if (efer & EFER_LMA) {
1551 vmcs_write32(VM_ENTRY_CONTROLS,
1552 vmcs_read32(VM_ENTRY_CONTROLS) |
1553 VM_ENTRY_IA32E_MODE);
1556 vmcs_write32(VM_ENTRY_CONTROLS,
1557 vmcs_read32(VM_ENTRY_CONTROLS) &
1558 ~VM_ENTRY_IA32E_MODE);
1560 msr->data = efer & ~EFER_LME;
1565 #ifdef CONFIG_X86_64
1567 static void enter_lmode(struct kvm_vcpu *vcpu)
1571 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1572 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1573 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1575 vmcs_write32(GUEST_TR_AR_BYTES,
1576 (guest_tr_ar & ~AR_TYPE_MASK)
1577 | AR_TYPE_BUSY_64_TSS);
1579 vcpu->arch.shadow_efer |= EFER_LMA;
1580 vmx_set_efer(vcpu, vcpu->arch.shadow_efer);
1583 static void exit_lmode(struct kvm_vcpu *vcpu)
1585 vcpu->arch.shadow_efer &= ~EFER_LMA;
1587 vmcs_write32(VM_ENTRY_CONTROLS,
1588 vmcs_read32(VM_ENTRY_CONTROLS)
1589 & ~VM_ENTRY_IA32E_MODE);
1594 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1596 vpid_sync_vcpu_all(to_vmx(vcpu));
1598 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1601 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1603 vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK;
1604 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
1607 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1609 if (!test_bit(VCPU_EXREG_PDPTR,
1610 (unsigned long *)&vcpu->arch.regs_dirty))
1613 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1614 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1615 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1616 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1617 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1621 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
1623 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1624 vcpu->arch.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
1625 vcpu->arch.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
1626 vcpu->arch.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
1627 vcpu->arch.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
1630 __set_bit(VCPU_EXREG_PDPTR,
1631 (unsigned long *)&vcpu->arch.regs_avail);
1632 __set_bit(VCPU_EXREG_PDPTR,
1633 (unsigned long *)&vcpu->arch.regs_dirty);
1636 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1638 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1640 struct kvm_vcpu *vcpu)
1642 if (!(cr0 & X86_CR0_PG)) {
1643 /* From paging/starting to nonpaging */
1644 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1645 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1646 (CPU_BASED_CR3_LOAD_EXITING |
1647 CPU_BASED_CR3_STORE_EXITING));
1648 vcpu->arch.cr0 = cr0;
1649 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1650 } else if (!is_paging(vcpu)) {
1651 /* From nonpaging to paging */
1652 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1653 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1654 ~(CPU_BASED_CR3_LOAD_EXITING |
1655 CPU_BASED_CR3_STORE_EXITING));
1656 vcpu->arch.cr0 = cr0;
1657 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1660 if (!(cr0 & X86_CR0_WP))
1661 *hw_cr0 &= ~X86_CR0_WP;
1664 static void ept_update_paging_mode_cr4(unsigned long *hw_cr4,
1665 struct kvm_vcpu *vcpu)
1667 if (!is_paging(vcpu)) {
1668 *hw_cr4 &= ~X86_CR4_PAE;
1669 *hw_cr4 |= X86_CR4_PSE;
1670 } else if (!(vcpu->arch.cr4 & X86_CR4_PAE))
1671 *hw_cr4 &= ~X86_CR4_PAE;
1674 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1676 struct vcpu_vmx *vmx = to_vmx(vcpu);
1677 unsigned long hw_cr0;
1679 if (enable_unrestricted_guest)
1680 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
1681 | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
1683 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
1685 vmx_fpu_deactivate(vcpu);
1687 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
1690 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
1693 #ifdef CONFIG_X86_64
1694 if (vcpu->arch.shadow_efer & EFER_LME) {
1695 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1697 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1703 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1705 vmcs_writel(CR0_READ_SHADOW, cr0);
1706 vmcs_writel(GUEST_CR0, hw_cr0);
1707 vcpu->arch.cr0 = cr0;
1709 if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
1710 vmx_fpu_activate(vcpu);
1713 static u64 construct_eptp(unsigned long root_hpa)
1717 /* TODO write the value reading from MSR */
1718 eptp = VMX_EPT_DEFAULT_MT |
1719 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1720 eptp |= (root_hpa & PAGE_MASK);
1725 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1727 unsigned long guest_cr3;
1732 eptp = construct_eptp(cr3);
1733 vmcs_write64(EPT_POINTER, eptp);
1734 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1735 vcpu->kvm->arch.ept_identity_map_addr;
1736 ept_load_pdptrs(vcpu);
1739 vmx_flush_tlb(vcpu);
1740 vmcs_writel(GUEST_CR3, guest_cr3);
1741 if (vcpu->arch.cr0 & X86_CR0_PE)
1742 vmx_fpu_deactivate(vcpu);
1745 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1747 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
1748 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1750 vcpu->arch.cr4 = cr4;
1752 ept_update_paging_mode_cr4(&hw_cr4, vcpu);
1754 vmcs_writel(CR4_READ_SHADOW, cr4);
1755 vmcs_writel(GUEST_CR4, hw_cr4);
1758 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1760 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1762 return vmcs_readl(sf->base);
1765 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1766 struct kvm_segment *var, int seg)
1768 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1771 var->base = vmcs_readl(sf->base);
1772 var->limit = vmcs_read32(sf->limit);
1773 var->selector = vmcs_read16(sf->selector);
1774 ar = vmcs_read32(sf->ar_bytes);
1775 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
1777 var->type = ar & 15;
1778 var->s = (ar >> 4) & 1;
1779 var->dpl = (ar >> 5) & 3;
1780 var->present = (ar >> 7) & 1;
1781 var->avl = (ar >> 12) & 1;
1782 var->l = (ar >> 13) & 1;
1783 var->db = (ar >> 14) & 1;
1784 var->g = (ar >> 15) & 1;
1785 var->unusable = (ar >> 16) & 1;
1788 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
1790 if (!(vcpu->arch.cr0 & X86_CR0_PE)) /* if real mode */
1793 if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
1796 return vmcs_read16(GUEST_CS_SELECTOR) & 3;
1799 static u32 vmx_segment_access_rights(struct kvm_segment *var)
1806 ar = var->type & 15;
1807 ar |= (var->s & 1) << 4;
1808 ar |= (var->dpl & 3) << 5;
1809 ar |= (var->present & 1) << 7;
1810 ar |= (var->avl & 1) << 12;
1811 ar |= (var->l & 1) << 13;
1812 ar |= (var->db & 1) << 14;
1813 ar |= (var->g & 1) << 15;
1815 if (ar == 0) /* a 0 value means unusable */
1816 ar = AR_UNUSABLE_MASK;
1821 static void vmx_set_segment(struct kvm_vcpu *vcpu,
1822 struct kvm_segment *var, int seg)
1824 struct vcpu_vmx *vmx = to_vmx(vcpu);
1825 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1828 if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
1829 vmx->rmode.tr.selector = var->selector;
1830 vmx->rmode.tr.base = var->base;
1831 vmx->rmode.tr.limit = var->limit;
1832 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
1835 vmcs_writel(sf->base, var->base);
1836 vmcs_write32(sf->limit, var->limit);
1837 vmcs_write16(sf->selector, var->selector);
1838 if (vmx->rmode.vm86_active && var->s) {
1840 * Hack real-mode segments into vm86 compatibility.
1842 if (var->base == 0xffff0000 && var->selector == 0xf000)
1843 vmcs_writel(sf->base, 0xf0000);
1846 ar = vmx_segment_access_rights(var);
1849 * Fix the "Accessed" bit in AR field of segment registers for older
1851 * IA32 arch specifies that at the time of processor reset the
1852 * "Accessed" bit in the AR field of segment registers is 1. And qemu
1853 * is setting it to 0 in the usedland code. This causes invalid guest
1854 * state vmexit when "unrestricted guest" mode is turned on.
1855 * Fix for this setup issue in cpu_reset is being pushed in the qemu
1856 * tree. Newer qemu binaries with that qemu fix would not need this
1859 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
1860 ar |= 0x1; /* Accessed */
1862 vmcs_write32(sf->ar_bytes, ar);
1865 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1867 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1869 *db = (ar >> 14) & 1;
1870 *l = (ar >> 13) & 1;
1873 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1875 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1876 dt->base = vmcs_readl(GUEST_IDTR_BASE);
1879 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1881 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1882 vmcs_writel(GUEST_IDTR_BASE, dt->base);
1885 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1887 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1888 dt->base = vmcs_readl(GUEST_GDTR_BASE);
1891 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1893 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1894 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1897 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
1899 struct kvm_segment var;
1902 vmx_get_segment(vcpu, &var, seg);
1903 ar = vmx_segment_access_rights(&var);
1905 if (var.base != (var.selector << 4))
1907 if (var.limit != 0xffff)
1915 static bool code_segment_valid(struct kvm_vcpu *vcpu)
1917 struct kvm_segment cs;
1918 unsigned int cs_rpl;
1920 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1921 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
1925 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
1929 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
1930 if (cs.dpl > cs_rpl)
1933 if (cs.dpl != cs_rpl)
1939 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
1943 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
1945 struct kvm_segment ss;
1946 unsigned int ss_rpl;
1948 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1949 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
1953 if (ss.type != 3 && ss.type != 7)
1957 if (ss.dpl != ss_rpl) /* DPL != RPL */
1965 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
1967 struct kvm_segment var;
1970 vmx_get_segment(vcpu, &var, seg);
1971 rpl = var.selector & SELECTOR_RPL_MASK;
1979 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
1980 if (var.dpl < rpl) /* DPL < RPL */
1984 /* TODO: Add other members to kvm_segment_field to allow checking for other access
1990 static bool tr_valid(struct kvm_vcpu *vcpu)
1992 struct kvm_segment tr;
1994 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
1998 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2000 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
2008 static bool ldtr_valid(struct kvm_vcpu *vcpu)
2010 struct kvm_segment ldtr;
2012 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
2016 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2026 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
2028 struct kvm_segment cs, ss;
2030 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2031 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2033 return ((cs.selector & SELECTOR_RPL_MASK) ==
2034 (ss.selector & SELECTOR_RPL_MASK));
2038 * Check if guest state is valid. Returns true if valid, false if
2040 * We assume that registers are always usable
2042 static bool guest_state_valid(struct kvm_vcpu *vcpu)
2044 /* real mode guest state checks */
2045 if (!(vcpu->arch.cr0 & X86_CR0_PE)) {
2046 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
2048 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
2050 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
2052 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
2054 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
2056 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
2059 /* protected mode guest state checks */
2060 if (!cs_ss_rpl_check(vcpu))
2062 if (!code_segment_valid(vcpu))
2064 if (!stack_segment_valid(vcpu))
2066 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
2068 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
2070 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
2072 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2074 if (!tr_valid(vcpu))
2076 if (!ldtr_valid(vcpu))
2080 * - Add checks on RIP
2081 * - Add checks on RFLAGS
2087 static int init_rmode_tss(struct kvm *kvm)
2089 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
2094 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2097 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
2098 r = kvm_write_guest_page(kvm, fn++, &data,
2099 TSS_IOPB_BASE_OFFSET, sizeof(u16));
2102 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2105 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2109 r = kvm_write_guest_page(kvm, fn, &data,
2110 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2120 static int init_rmode_identity_map(struct kvm *kvm)
2123 pfn_t identity_map_pfn;
2128 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2129 printk(KERN_ERR "EPT: identity-mapping pagetable "
2130 "haven't been allocated!\n");
2133 if (likely(kvm->arch.ept_identity_pagetable_done))
2136 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
2137 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2140 /* Set up identity-mapping pagetable for EPT in real mode */
2141 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2142 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2143 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2144 r = kvm_write_guest_page(kvm, identity_map_pfn,
2145 &tmp, i * sizeof(tmp), sizeof(tmp));
2149 kvm->arch.ept_identity_pagetable_done = true;
2155 static void seg_setup(int seg)
2157 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2160 vmcs_write16(sf->selector, 0);
2161 vmcs_writel(sf->base, 0);
2162 vmcs_write32(sf->limit, 0xffff);
2163 if (enable_unrestricted_guest) {
2165 if (seg == VCPU_SREG_CS)
2166 ar |= 0x08; /* code segment */
2170 vmcs_write32(sf->ar_bytes, ar);
2173 static int alloc_apic_access_page(struct kvm *kvm)
2175 struct kvm_userspace_memory_region kvm_userspace_mem;
2178 down_write(&kvm->slots_lock);
2179 if (kvm->arch.apic_access_page)
2181 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2182 kvm_userspace_mem.flags = 0;
2183 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2184 kvm_userspace_mem.memory_size = PAGE_SIZE;
2185 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2189 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2191 up_write(&kvm->slots_lock);
2195 static int alloc_identity_pagetable(struct kvm *kvm)
2197 struct kvm_userspace_memory_region kvm_userspace_mem;
2200 down_write(&kvm->slots_lock);
2201 if (kvm->arch.ept_identity_pagetable)
2203 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2204 kvm_userspace_mem.flags = 0;
2205 kvm_userspace_mem.guest_phys_addr =
2206 kvm->arch.ept_identity_map_addr;
2207 kvm_userspace_mem.memory_size = PAGE_SIZE;
2208 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2212 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2213 kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
2215 up_write(&kvm->slots_lock);
2219 static void allocate_vpid(struct vcpu_vmx *vmx)
2226 spin_lock(&vmx_vpid_lock);
2227 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2228 if (vpid < VMX_NR_VPIDS) {
2230 __set_bit(vpid, vmx_vpid_bitmap);
2232 spin_unlock(&vmx_vpid_lock);
2235 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
2237 int f = sizeof(unsigned long);
2239 if (!cpu_has_vmx_msr_bitmap())
2243 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2244 * have the write-low and read-high bitmap offsets the wrong way round.
2245 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2247 if (msr <= 0x1fff) {
2248 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2249 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
2250 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2252 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2253 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
2257 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2260 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2261 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2265 * Sets up the vmcs for emulated real mode.
2267 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2269 u32 host_sysenter_cs, msr_low, msr_high;
2271 u64 host_pat, tsc_this, tsc_base;
2273 struct descriptor_table dt;
2275 unsigned long kvm_vmx_return;
2279 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2280 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
2282 if (cpu_has_vmx_msr_bitmap())
2283 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
2285 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2288 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2289 vmcs_config.pin_based_exec_ctrl);
2291 exec_control = vmcs_config.cpu_based_exec_ctrl;
2292 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2293 exec_control &= ~CPU_BASED_TPR_SHADOW;
2294 #ifdef CONFIG_X86_64
2295 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2296 CPU_BASED_CR8_LOAD_EXITING;
2300 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2301 CPU_BASED_CR3_LOAD_EXITING |
2302 CPU_BASED_INVLPG_EXITING;
2303 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2305 if (cpu_has_secondary_exec_ctrls()) {
2306 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2307 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2309 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2311 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2313 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2314 if (!enable_unrestricted_guest)
2315 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2317 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
2318 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2322 vmcs_write32(PLE_GAP, ple_gap);
2323 vmcs_write32(PLE_WINDOW, ple_window);
2326 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2327 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2328 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
2330 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
2331 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
2332 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
2334 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
2335 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2336 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2337 vmcs_write16(HOST_FS_SELECTOR, kvm_read_fs()); /* 22.2.4 */
2338 vmcs_write16(HOST_GS_SELECTOR, kvm_read_gs()); /* 22.2.4 */
2339 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2340 #ifdef CONFIG_X86_64
2341 rdmsrl(MSR_FS_BASE, a);
2342 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2343 rdmsrl(MSR_GS_BASE, a);
2344 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2346 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2347 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2350 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
2353 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
2355 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2356 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2357 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2358 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2359 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2361 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2362 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2363 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2364 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
2365 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2366 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
2368 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2369 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2370 host_pat = msr_low | ((u64) msr_high << 32);
2371 vmcs_write64(HOST_IA32_PAT, host_pat);
2373 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2374 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2375 host_pat = msr_low | ((u64) msr_high << 32);
2376 /* Write the default value follow host pat */
2377 vmcs_write64(GUEST_IA32_PAT, host_pat);
2378 /* Keep arch.pat sync with GUEST_IA32_PAT */
2379 vmx->vcpu.arch.pat = host_pat;
2382 for (i = 0; i < NR_VMX_MSR; ++i) {
2383 u32 index = vmx_msr_index[i];
2384 u32 data_low, data_high;
2388 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2390 if (wrmsr_safe(index, data_low, data_high) < 0)
2392 data = data_low | ((u64)data_high << 32);
2393 vmx->guest_msrs[j].index = i;
2394 vmx->guest_msrs[j].data = 0;
2398 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2400 /* 22.2.1, 20.8.1 */
2401 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2403 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2404 vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
2406 tsc_base = vmx->vcpu.kvm->arch.vm_init_tsc;
2408 if (tsc_this < vmx->vcpu.kvm->arch.vm_init_tsc)
2409 tsc_base = tsc_this;
2411 guest_write_tsc(0, tsc_base);
2416 static int init_rmode(struct kvm *kvm)
2418 if (!init_rmode_tss(kvm))
2420 if (!init_rmode_identity_map(kvm))
2425 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2427 struct vcpu_vmx *vmx = to_vmx(vcpu);
2431 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2432 down_read(&vcpu->kvm->slots_lock);
2433 if (!init_rmode(vmx->vcpu.kvm)) {
2438 vmx->rmode.vm86_active = 0;
2440 vmx->soft_vnmi_blocked = 0;
2442 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2443 kvm_set_cr8(&vmx->vcpu, 0);
2444 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2445 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2446 msr |= MSR_IA32_APICBASE_BSP;
2447 kvm_set_apic_base(&vmx->vcpu, msr);
2449 fx_init(&vmx->vcpu);
2451 seg_setup(VCPU_SREG_CS);
2453 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2454 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
2456 if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
2457 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2458 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2460 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2461 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2464 seg_setup(VCPU_SREG_DS);
2465 seg_setup(VCPU_SREG_ES);
2466 seg_setup(VCPU_SREG_FS);
2467 seg_setup(VCPU_SREG_GS);
2468 seg_setup(VCPU_SREG_SS);
2470 vmcs_write16(GUEST_TR_SELECTOR, 0);
2471 vmcs_writel(GUEST_TR_BASE, 0);
2472 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2473 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2475 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2476 vmcs_writel(GUEST_LDTR_BASE, 0);
2477 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2478 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2480 vmcs_write32(GUEST_SYSENTER_CS, 0);
2481 vmcs_writel(GUEST_SYSENTER_ESP, 0);
2482 vmcs_writel(GUEST_SYSENTER_EIP, 0);
2484 vmcs_writel(GUEST_RFLAGS, 0x02);
2485 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2486 kvm_rip_write(vcpu, 0xfff0);
2488 kvm_rip_write(vcpu, 0);
2489 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2491 vmcs_writel(GUEST_DR7, 0x400);
2493 vmcs_writel(GUEST_GDTR_BASE, 0);
2494 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2496 vmcs_writel(GUEST_IDTR_BASE, 0);
2497 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2499 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2500 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2501 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2503 /* Special registers */
2504 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2508 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
2510 if (cpu_has_vmx_tpr_shadow()) {
2511 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2512 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2513 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2514 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2515 vmcs_write32(TPR_THRESHOLD, 0);
2518 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2519 vmcs_write64(APIC_ACCESS_ADDR,
2520 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2523 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2525 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
2526 vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
2527 vmx_set_cr4(&vmx->vcpu, 0);
2528 vmx_set_efer(&vmx->vcpu, 0);
2529 vmx_fpu_activate(&vmx->vcpu);
2530 update_exception_bitmap(&vmx->vcpu);
2532 vpid_sync_vcpu_all(vmx);
2536 /* HACK: Don't enable emulation on guest boot/reset */
2537 vmx->emulation_required = 0;
2540 up_read(&vcpu->kvm->slots_lock);
2544 static void enable_irq_window(struct kvm_vcpu *vcpu)
2546 u32 cpu_based_vm_exec_control;
2548 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2549 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2550 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2553 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2555 u32 cpu_based_vm_exec_control;
2557 if (!cpu_has_virtual_nmis()) {
2558 enable_irq_window(vcpu);
2562 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2563 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2564 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2567 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
2569 struct vcpu_vmx *vmx = to_vmx(vcpu);
2571 int irq = vcpu->arch.interrupt.nr;
2573 trace_kvm_inj_virq(irq);
2575 ++vcpu->stat.irq_injections;
2576 if (vmx->rmode.vm86_active) {
2577 vmx->rmode.irq.pending = true;
2578 vmx->rmode.irq.vector = irq;
2579 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2580 if (vcpu->arch.interrupt.soft)
2581 vmx->rmode.irq.rip +=
2582 vmx->vcpu.arch.event_exit_inst_len;
2583 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2584 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2585 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2586 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2589 intr = irq | INTR_INFO_VALID_MASK;
2590 if (vcpu->arch.interrupt.soft) {
2591 intr |= INTR_TYPE_SOFT_INTR;
2592 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2593 vmx->vcpu.arch.event_exit_inst_len);
2595 intr |= INTR_TYPE_EXT_INTR;
2596 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
2599 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2601 struct vcpu_vmx *vmx = to_vmx(vcpu);
2603 if (!cpu_has_virtual_nmis()) {
2605 * Tracking the NMI-blocked state in software is built upon
2606 * finding the next open IRQ window. This, in turn, depends on
2607 * well-behaving guests: They have to keep IRQs disabled at
2608 * least as long as the NMI handler runs. Otherwise we may
2609 * cause NMI nesting, maybe breaking the guest. But as this is
2610 * highly unlikely, we can live with the residual risk.
2612 vmx->soft_vnmi_blocked = 1;
2613 vmx->vnmi_blocked_time = 0;
2616 ++vcpu->stat.nmi_injections;
2617 if (vmx->rmode.vm86_active) {
2618 vmx->rmode.irq.pending = true;
2619 vmx->rmode.irq.vector = NMI_VECTOR;
2620 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2621 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2622 NMI_VECTOR | INTR_TYPE_SOFT_INTR |
2623 INTR_INFO_VALID_MASK);
2624 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2625 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2628 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2629 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2632 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
2634 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2637 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2638 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS |
2639 GUEST_INTR_STATE_NMI));
2642 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2644 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2645 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2646 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
2649 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2652 struct kvm_userspace_memory_region tss_mem = {
2653 .slot = TSS_PRIVATE_MEMSLOT,
2654 .guest_phys_addr = addr,
2655 .memory_size = PAGE_SIZE * 3,
2659 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2662 kvm->arch.tss_addr = addr;
2666 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2667 int vec, u32 err_code)
2670 * Instruction with address size override prefix opcode 0x67
2671 * Cause the #SS fault with 0 error code in VM86 mode.
2673 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
2674 if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE)
2677 * Forward all other exceptions that are valid in real mode.
2678 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2679 * the required debugging infrastructure rework.
2683 if (vcpu->guest_debug &
2684 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
2686 kvm_queue_exception(vcpu, vec);
2689 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2700 kvm_queue_exception(vcpu, vec);
2707 * Trigger machine check on the host. We assume all the MSRs are already set up
2708 * by the CPU and that we still run on the same CPU as the MCE occurred on.
2709 * We pass a fake environment to the machine check handler because we want
2710 * the guest to be always treated like user space, no matter what context
2711 * it used internally.
2713 static void kvm_machine_check(void)
2715 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
2716 struct pt_regs regs = {
2717 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
2718 .flags = X86_EFLAGS_IF,
2721 do_machine_check(®s, 0);
2725 static int handle_machine_check(struct kvm_vcpu *vcpu)
2727 /* already handled by vcpu_run */
2731 static int handle_exception(struct kvm_vcpu *vcpu)
2733 struct vcpu_vmx *vmx = to_vmx(vcpu);
2734 struct kvm_run *kvm_run = vcpu->run;
2735 u32 intr_info, ex_no, error_code;
2736 unsigned long cr2, rip, dr6;
2738 enum emulation_result er;
2740 vect_info = vmx->idt_vectoring_info;
2741 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2743 if (is_machine_check(intr_info))
2744 return handle_machine_check(vcpu);
2746 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
2747 !is_page_fault(intr_info))
2748 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
2749 "intr info 0x%x\n", __func__, vect_info, intr_info);
2751 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
2752 return 1; /* already handled by vmx_vcpu_run() */
2754 if (is_no_device(intr_info)) {
2755 vmx_fpu_activate(vcpu);
2759 if (is_invalid_opcode(intr_info)) {
2760 er = emulate_instruction(vcpu, 0, 0, EMULTYPE_TRAP_UD);
2761 if (er != EMULATE_DONE)
2762 kvm_queue_exception(vcpu, UD_VECTOR);
2767 rip = kvm_rip_read(vcpu);
2768 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
2769 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
2770 if (is_page_fault(intr_info)) {
2771 /* EPT won't cause page fault directly */
2774 cr2 = vmcs_readl(EXIT_QUALIFICATION);
2775 trace_kvm_page_fault(cr2, error_code);
2777 if (kvm_event_needs_reinjection(vcpu))
2778 kvm_mmu_unprotect_page_virt(vcpu, cr2);
2779 return kvm_mmu_page_fault(vcpu, cr2, error_code);
2782 if (vmx->rmode.vm86_active &&
2783 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
2785 if (vcpu->arch.halt_request) {
2786 vcpu->arch.halt_request = 0;
2787 return kvm_emulate_halt(vcpu);
2792 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
2795 dr6 = vmcs_readl(EXIT_QUALIFICATION);
2796 if (!(vcpu->guest_debug &
2797 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
2798 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
2799 kvm_queue_exception(vcpu, DB_VECTOR);
2802 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
2803 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
2806 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2807 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
2808 kvm_run->debug.arch.exception = ex_no;
2811 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
2812 kvm_run->ex.exception = ex_no;
2813 kvm_run->ex.error_code = error_code;
2819 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
2821 ++vcpu->stat.irq_exits;
2825 static int handle_triple_fault(struct kvm_vcpu *vcpu)
2827 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
2831 static int handle_io(struct kvm_vcpu *vcpu)
2833 unsigned long exit_qualification;
2834 int size, in, string;
2837 ++vcpu->stat.io_exits;
2838 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2839 string = (exit_qualification & 16) != 0;
2842 if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DO_MMIO)
2847 size = (exit_qualification & 7) + 1;
2848 in = (exit_qualification & 8) != 0;
2849 port = exit_qualification >> 16;
2851 skip_emulated_instruction(vcpu);
2852 return kvm_emulate_pio(vcpu, in, size, port);
2856 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2859 * Patch in the VMCALL instruction:
2861 hypercall[0] = 0x0f;
2862 hypercall[1] = 0x01;
2863 hypercall[2] = 0xc1;
2866 static int handle_cr(struct kvm_vcpu *vcpu)
2868 unsigned long exit_qualification, val;
2872 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2873 cr = exit_qualification & 15;
2874 reg = (exit_qualification >> 8) & 15;
2875 switch ((exit_qualification >> 4) & 3) {
2876 case 0: /* mov to cr */
2877 val = kvm_register_read(vcpu, reg);
2878 trace_kvm_cr_write(cr, val);
2881 kvm_set_cr0(vcpu, val);
2882 skip_emulated_instruction(vcpu);
2885 kvm_set_cr3(vcpu, val);
2886 skip_emulated_instruction(vcpu);
2889 kvm_set_cr4(vcpu, val);
2890 skip_emulated_instruction(vcpu);
2893 u8 cr8_prev = kvm_get_cr8(vcpu);
2894 u8 cr8 = kvm_register_read(vcpu, reg);
2895 kvm_set_cr8(vcpu, cr8);
2896 skip_emulated_instruction(vcpu);
2897 if (irqchip_in_kernel(vcpu->kvm))
2899 if (cr8_prev <= cr8)
2901 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
2907 vmx_fpu_deactivate(vcpu);
2908 vcpu->arch.cr0 &= ~X86_CR0_TS;
2909 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2910 vmx_fpu_activate(vcpu);
2911 skip_emulated_instruction(vcpu);
2913 case 1: /*mov from cr*/
2916 kvm_register_write(vcpu, reg, vcpu->arch.cr3);
2917 trace_kvm_cr_read(cr, vcpu->arch.cr3);
2918 skip_emulated_instruction(vcpu);
2921 val = kvm_get_cr8(vcpu);
2922 kvm_register_write(vcpu, reg, val);
2923 trace_kvm_cr_read(cr, val);
2924 skip_emulated_instruction(vcpu);
2929 kvm_lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
2931 skip_emulated_instruction(vcpu);
2936 vcpu->run->exit_reason = 0;
2937 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
2938 (int)(exit_qualification >> 4) & 3, cr);
2942 static int handle_dr(struct kvm_vcpu *vcpu)
2944 unsigned long exit_qualification;
2948 if (!kvm_require_cpl(vcpu, 0))
2950 dr = vmcs_readl(GUEST_DR7);
2953 * As the vm-exit takes precedence over the debug trap, we
2954 * need to emulate the latter, either for the host or the
2955 * guest debugging itself.
2957 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
2958 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
2959 vcpu->run->debug.arch.dr7 = dr;
2960 vcpu->run->debug.arch.pc =
2961 vmcs_readl(GUEST_CS_BASE) +
2962 vmcs_readl(GUEST_RIP);
2963 vcpu->run->debug.arch.exception = DB_VECTOR;
2964 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
2967 vcpu->arch.dr7 &= ~DR7_GD;
2968 vcpu->arch.dr6 |= DR6_BD;
2969 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2970 kvm_queue_exception(vcpu, DB_VECTOR);
2975 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2976 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
2977 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
2978 if (exit_qualification & TYPE_MOV_FROM_DR) {
2981 val = vcpu->arch.db[dr];
2984 val = vcpu->arch.dr6;
2987 val = vcpu->arch.dr7;
2992 kvm_register_write(vcpu, reg, val);
2994 val = vcpu->arch.regs[reg];
2997 vcpu->arch.db[dr] = val;
2998 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
2999 vcpu->arch.eff_db[dr] = val;
3002 if (vcpu->arch.cr4 & X86_CR4_DE)
3003 kvm_queue_exception(vcpu, UD_VECTOR);
3006 if (val & 0xffffffff00000000ULL) {
3007 kvm_queue_exception(vcpu, GP_VECTOR);
3010 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
3013 if (val & 0xffffffff00000000ULL) {
3014 kvm_queue_exception(vcpu, GP_VECTOR);
3017 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
3018 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
3019 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3020 vcpu->arch.switch_db_regs =
3021 (val & DR7_BP_EN_MASK);
3026 skip_emulated_instruction(vcpu);
3030 static int handle_cpuid(struct kvm_vcpu *vcpu)
3032 kvm_emulate_cpuid(vcpu);
3036 static int handle_rdmsr(struct kvm_vcpu *vcpu)
3038 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3041 if (vmx_get_msr(vcpu, ecx, &data)) {
3042 kvm_inject_gp(vcpu, 0);
3046 trace_kvm_msr_read(ecx, data);
3048 /* FIXME: handling of bits 32:63 of rax, rdx */
3049 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
3050 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
3051 skip_emulated_instruction(vcpu);
3055 static int handle_wrmsr(struct kvm_vcpu *vcpu)
3057 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3058 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
3059 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3061 trace_kvm_msr_write(ecx, data);
3063 if (vmx_set_msr(vcpu, ecx, data) != 0) {
3064 kvm_inject_gp(vcpu, 0);
3068 skip_emulated_instruction(vcpu);
3072 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
3077 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
3079 u32 cpu_based_vm_exec_control;
3081 /* clear pending irq */
3082 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3083 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3084 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3086 ++vcpu->stat.irq_window_exits;
3089 * If the user space waits to inject interrupts, exit as soon as
3092 if (!irqchip_in_kernel(vcpu->kvm) &&
3093 vcpu->run->request_interrupt_window &&
3094 !kvm_cpu_has_interrupt(vcpu)) {
3095 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3101 static int handle_halt(struct kvm_vcpu *vcpu)
3103 skip_emulated_instruction(vcpu);
3104 return kvm_emulate_halt(vcpu);
3107 static int handle_vmcall(struct kvm_vcpu *vcpu)
3109 skip_emulated_instruction(vcpu);
3110 kvm_emulate_hypercall(vcpu);
3114 static int handle_vmx_insn(struct kvm_vcpu *vcpu)
3116 kvm_queue_exception(vcpu, UD_VECTOR);
3120 static int handle_invlpg(struct kvm_vcpu *vcpu)
3122 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3124 kvm_mmu_invlpg(vcpu, exit_qualification);
3125 skip_emulated_instruction(vcpu);
3129 static int handle_wbinvd(struct kvm_vcpu *vcpu)
3131 skip_emulated_instruction(vcpu);
3132 /* TODO: Add support for VT-d/pass-through device */
3136 static int handle_apic_access(struct kvm_vcpu *vcpu)
3138 unsigned long exit_qualification;
3139 enum emulation_result er;
3140 unsigned long offset;
3142 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3143 offset = exit_qualification & 0xffful;
3145 er = emulate_instruction(vcpu, 0, 0, 0);
3147 if (er != EMULATE_DONE) {
3149 "Fail to handle apic access vmexit! Offset is 0x%lx\n",
3156 static int handle_task_switch(struct kvm_vcpu *vcpu)
3158 struct vcpu_vmx *vmx = to_vmx(vcpu);
3159 unsigned long exit_qualification;
3161 int reason, type, idt_v;
3163 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3164 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
3166 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3168 reason = (u32)exit_qualification >> 30;
3169 if (reason == TASK_SWITCH_GATE && idt_v) {
3171 case INTR_TYPE_NMI_INTR:
3172 vcpu->arch.nmi_injected = false;
3173 if (cpu_has_virtual_nmis())
3174 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3175 GUEST_INTR_STATE_NMI);
3177 case INTR_TYPE_EXT_INTR:
3178 case INTR_TYPE_SOFT_INTR:
3179 kvm_clear_interrupt_queue(vcpu);
3181 case INTR_TYPE_HARD_EXCEPTION:
3182 case INTR_TYPE_SOFT_EXCEPTION:
3183 kvm_clear_exception_queue(vcpu);
3189 tss_selector = exit_qualification;
3191 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3192 type != INTR_TYPE_EXT_INTR &&
3193 type != INTR_TYPE_NMI_INTR))
3194 skip_emulated_instruction(vcpu);
3196 if (!kvm_task_switch(vcpu, tss_selector, reason))
3199 /* clear all local breakpoint enable flags */
3200 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3203 * TODO: What about debug traps on tss switch?
3204 * Are we supposed to inject them and update dr6?
3210 static int handle_ept_violation(struct kvm_vcpu *vcpu)
3212 unsigned long exit_qualification;
3216 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3218 if (exit_qualification & (1 << 6)) {
3219 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3223 gla_validity = (exit_qualification >> 7) & 0x3;
3224 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3225 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3226 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3227 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3228 vmcs_readl(GUEST_LINEAR_ADDRESS));
3229 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3230 (long unsigned int)exit_qualification);
3231 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3232 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
3236 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3237 trace_kvm_page_fault(gpa, exit_qualification);
3238 return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
3241 static u64 ept_rsvd_mask(u64 spte, int level)
3246 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
3247 mask |= (1ULL << i);
3250 /* bits 7:3 reserved */
3252 else if (level == 2) {
3253 if (spte & (1ULL << 7))
3254 /* 2MB ref, bits 20:12 reserved */
3257 /* bits 6:3 reserved */
3264 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
3267 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
3269 /* 010b (write-only) */
3270 WARN_ON((spte & 0x7) == 0x2);
3272 /* 110b (write/execute) */
3273 WARN_ON((spte & 0x7) == 0x6);
3275 /* 100b (execute-only) and value not supported by logical processor */
3276 if (!cpu_has_vmx_ept_execute_only())
3277 WARN_ON((spte & 0x7) == 0x4);
3281 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
3283 if (rsvd_bits != 0) {
3284 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
3285 __func__, rsvd_bits);
3289 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
3290 u64 ept_mem_type = (spte & 0x38) >> 3;
3292 if (ept_mem_type == 2 || ept_mem_type == 3 ||
3293 ept_mem_type == 7) {
3294 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
3295 __func__, ept_mem_type);
3302 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
3308 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3310 printk(KERN_ERR "EPT: Misconfiguration.\n");
3311 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
3313 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
3315 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
3316 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
3318 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3319 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
3324 static int handle_nmi_window(struct kvm_vcpu *vcpu)
3326 u32 cpu_based_vm_exec_control;
3328 /* clear pending NMI */
3329 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3330 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3331 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3332 ++vcpu->stat.nmi_window_exits;
3337 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
3339 struct vcpu_vmx *vmx = to_vmx(vcpu);
3340 enum emulation_result err = EMULATE_DONE;
3343 while (!guest_state_valid(vcpu)) {
3344 err = emulate_instruction(vcpu, 0, 0, 0);
3346 if (err == EMULATE_DO_MMIO) {
3351 if (err != EMULATE_DONE) {
3352 kvm_report_emulation_failure(vcpu, "emulation failure");
3353 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3354 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3359 if (signal_pending(current))
3365 vmx->emulation_required = 0;
3371 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
3372 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
3374 static int handle_pause(struct kvm_vcpu *vcpu)
3376 skip_emulated_instruction(vcpu);
3377 kvm_vcpu_on_spin(vcpu);
3383 * The exit handlers return 1 if the exit was handled fully and guest execution
3384 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
3385 * to be done to userspace and return 0.
3387 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
3388 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
3389 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
3390 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
3391 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
3392 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
3393 [EXIT_REASON_CR_ACCESS] = handle_cr,
3394 [EXIT_REASON_DR_ACCESS] = handle_dr,
3395 [EXIT_REASON_CPUID] = handle_cpuid,
3396 [EXIT_REASON_MSR_READ] = handle_rdmsr,
3397 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
3398 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
3399 [EXIT_REASON_HLT] = handle_halt,
3400 [EXIT_REASON_INVLPG] = handle_invlpg,
3401 [EXIT_REASON_VMCALL] = handle_vmcall,
3402 [EXIT_REASON_VMCLEAR] = handle_vmx_insn,
3403 [EXIT_REASON_VMLAUNCH] = handle_vmx_insn,
3404 [EXIT_REASON_VMPTRLD] = handle_vmx_insn,
3405 [EXIT_REASON_VMPTRST] = handle_vmx_insn,
3406 [EXIT_REASON_VMREAD] = handle_vmx_insn,
3407 [EXIT_REASON_VMRESUME] = handle_vmx_insn,
3408 [EXIT_REASON_VMWRITE] = handle_vmx_insn,
3409 [EXIT_REASON_VMOFF] = handle_vmx_insn,
3410 [EXIT_REASON_VMON] = handle_vmx_insn,
3411 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
3412 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
3413 [EXIT_REASON_WBINVD] = handle_wbinvd,
3414 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
3415 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
3416 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
3417 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
3418 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
3421 static const int kvm_vmx_max_exit_handlers =
3422 ARRAY_SIZE(kvm_vmx_exit_handlers);
3425 * The guest has exited. See if we can fix it or if we need userspace
3428 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
3430 struct vcpu_vmx *vmx = to_vmx(vcpu);
3431 u32 exit_reason = vmx->exit_reason;
3432 u32 vectoring_info = vmx->idt_vectoring_info;
3434 trace_kvm_exit(exit_reason, kvm_rip_read(vcpu));
3436 /* If guest state is invalid, start emulating */
3437 if (vmx->emulation_required && emulate_invalid_guest_state)
3438 return handle_invalid_guest_state(vcpu);
3440 /* Access CR3 don't cause VMExit in paging mode, so we need
3441 * to sync with guest real CR3. */
3442 if (enable_ept && is_paging(vcpu))
3443 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3445 if (unlikely(vmx->fail)) {
3446 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3447 vcpu->run->fail_entry.hardware_entry_failure_reason
3448 = vmcs_read32(VM_INSTRUCTION_ERROR);
3452 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3453 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3454 exit_reason != EXIT_REASON_EPT_VIOLATION &&
3455 exit_reason != EXIT_REASON_TASK_SWITCH))
3456 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3457 "(0x%x) and exit reason is 0x%x\n",
3458 __func__, vectoring_info, exit_reason);
3460 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3461 if (vmx_interrupt_allowed(vcpu)) {
3462 vmx->soft_vnmi_blocked = 0;
3463 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3464 vcpu->arch.nmi_pending) {
3466 * This CPU don't support us in finding the end of an
3467 * NMI-blocked window if the guest runs with IRQs
3468 * disabled. So we pull the trigger after 1 s of
3469 * futile waiting, but inform the user about this.
3471 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3472 "state on VCPU %d after 1 s timeout\n",
3473 __func__, vcpu->vcpu_id);
3474 vmx->soft_vnmi_blocked = 0;
3478 if (exit_reason < kvm_vmx_max_exit_handlers
3479 && kvm_vmx_exit_handlers[exit_reason])
3480 return kvm_vmx_exit_handlers[exit_reason](vcpu);
3482 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3483 vcpu->run->hw.hardware_exit_reason = exit_reason;
3488 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3490 if (irr == -1 || tpr < irr) {
3491 vmcs_write32(TPR_THRESHOLD, 0);
3495 vmcs_write32(TPR_THRESHOLD, irr);
3498 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3501 u32 idt_vectoring_info = vmx->idt_vectoring_info;
3505 bool idtv_info_valid;
3507 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3509 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
3511 /* Handle machine checks before interrupts are enabled */
3512 if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3513 || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3514 && is_machine_check(exit_intr_info)))
3515 kvm_machine_check();
3517 /* We need to handle NMIs before interrupts are enabled */
3518 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3519 (exit_intr_info & INTR_INFO_VALID_MASK))
3522 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3524 if (cpu_has_virtual_nmis()) {
3525 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3526 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3528 * SDM 3: 27.7.1.2 (September 2008)
3529 * Re-set bit "block by NMI" before VM entry if vmexit caused by
3530 * a guest IRET fault.
3531 * SDM 3: 23.2.2 (September 2008)
3532 * Bit 12 is undefined in any of the following cases:
3533 * If the VM exit sets the valid bit in the IDT-vectoring
3534 * information field.
3535 * If the VM exit is due to a double fault.
3537 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3538 vector != DF_VECTOR && !idtv_info_valid)
3539 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3540 GUEST_INTR_STATE_NMI);
3541 } else if (unlikely(vmx->soft_vnmi_blocked))
3542 vmx->vnmi_blocked_time +=
3543 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3545 vmx->vcpu.arch.nmi_injected = false;
3546 kvm_clear_exception_queue(&vmx->vcpu);
3547 kvm_clear_interrupt_queue(&vmx->vcpu);
3549 if (!idtv_info_valid)
3552 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3553 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3556 case INTR_TYPE_NMI_INTR:
3557 vmx->vcpu.arch.nmi_injected = true;
3559 * SDM 3: 27.7.1.2 (September 2008)
3560 * Clear bit "block by NMI" before VM entry if a NMI
3563 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3564 GUEST_INTR_STATE_NMI);
3566 case INTR_TYPE_SOFT_EXCEPTION:
3567 vmx->vcpu.arch.event_exit_inst_len =
3568 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3570 case INTR_TYPE_HARD_EXCEPTION:
3571 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3572 u32 err = vmcs_read32(IDT_VECTORING_ERROR_CODE);
3573 kvm_queue_exception_e(&vmx->vcpu, vector, err);
3575 kvm_queue_exception(&vmx->vcpu, vector);
3577 case INTR_TYPE_SOFT_INTR:
3578 vmx->vcpu.arch.event_exit_inst_len =
3579 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3581 case INTR_TYPE_EXT_INTR:
3582 kvm_queue_interrupt(&vmx->vcpu, vector,
3583 type == INTR_TYPE_SOFT_INTR);
3591 * Failure to inject an interrupt should give us the information
3592 * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs
3593 * when fetching the interrupt redirection bitmap in the real-mode
3594 * tss, this doesn't happen. So we do it ourselves.
3596 static void fixup_rmode_irq(struct vcpu_vmx *vmx)
3598 vmx->rmode.irq.pending = 0;
3599 if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip)
3601 kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip);
3602 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
3603 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
3604 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
3607 vmx->idt_vectoring_info =
3608 VECTORING_INFO_VALID_MASK
3609 | INTR_TYPE_EXT_INTR
3610 | vmx->rmode.irq.vector;
3613 #ifdef CONFIG_X86_64
3621 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
3623 struct vcpu_vmx *vmx = to_vmx(vcpu);
3625 /* Record the guest's net vcpu time for enforced NMI injections. */
3626 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3627 vmx->entry_time = ktime_get();
3629 /* Don't enter VMX if guest state is invalid, let the exit handler
3630 start emulation until we arrive back to a valid state */
3631 if (vmx->emulation_required && emulate_invalid_guest_state)
3634 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3635 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3636 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3637 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3639 /* When single-stepping over STI and MOV SS, we must clear the
3640 * corresponding interruptibility bits in the guest state. Otherwise
3641 * vmentry fails as it then expects bit 14 (BS) in pending debug
3642 * exceptions being set, but that's not correct for the guest debugging
3644 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3645 vmx_set_interrupt_shadow(vcpu, 0);
3648 * Loading guest fpu may have cleared host cr0.ts
3650 vmcs_writel(HOST_CR0, read_cr0());
3652 if (vcpu->arch.switch_db_regs)
3653 set_debugreg(vcpu->arch.dr6, 6);
3656 /* Store host registers */
3657 "push %%"R"dx; push %%"R"bp;"
3659 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3661 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
3662 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
3664 /* Reload cr2 if changed */
3665 "mov %c[cr2](%0), %%"R"ax \n\t"
3666 "mov %%cr2, %%"R"dx \n\t"
3667 "cmp %%"R"ax, %%"R"dx \n\t"
3669 "mov %%"R"ax, %%cr2 \n\t"
3671 /* Check if vmlaunch of vmresume is needed */
3672 "cmpl $0, %c[launched](%0) \n\t"
3673 /* Load guest registers. Don't clobber flags. */
3674 "mov %c[rax](%0), %%"R"ax \n\t"
3675 "mov %c[rbx](%0), %%"R"bx \n\t"
3676 "mov %c[rdx](%0), %%"R"dx \n\t"
3677 "mov %c[rsi](%0), %%"R"si \n\t"
3678 "mov %c[rdi](%0), %%"R"di \n\t"
3679 "mov %c[rbp](%0), %%"R"bp \n\t"
3680 #ifdef CONFIG_X86_64
3681 "mov %c[r8](%0), %%r8 \n\t"
3682 "mov %c[r9](%0), %%r9 \n\t"
3683 "mov %c[r10](%0), %%r10 \n\t"
3684 "mov %c[r11](%0), %%r11 \n\t"
3685 "mov %c[r12](%0), %%r12 \n\t"
3686 "mov %c[r13](%0), %%r13 \n\t"
3687 "mov %c[r14](%0), %%r14 \n\t"
3688 "mov %c[r15](%0), %%r15 \n\t"
3690 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3692 /* Enter guest mode */
3693 "jne .Llaunched \n\t"
3694 __ex(ASM_VMX_VMLAUNCH) "\n\t"
3695 "jmp .Lkvm_vmx_return \n\t"
3696 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
3697 ".Lkvm_vmx_return: "
3698 /* Save guest registers, load host registers, keep flags */
3699 "xchg %0, (%%"R"sp) \n\t"
3700 "mov %%"R"ax, %c[rax](%0) \n\t"
3701 "mov %%"R"bx, %c[rbx](%0) \n\t"
3702 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3703 "mov %%"R"dx, %c[rdx](%0) \n\t"
3704 "mov %%"R"si, %c[rsi](%0) \n\t"
3705 "mov %%"R"di, %c[rdi](%0) \n\t"
3706 "mov %%"R"bp, %c[rbp](%0) \n\t"
3707 #ifdef CONFIG_X86_64
3708 "mov %%r8, %c[r8](%0) \n\t"
3709 "mov %%r9, %c[r9](%0) \n\t"
3710 "mov %%r10, %c[r10](%0) \n\t"
3711 "mov %%r11, %c[r11](%0) \n\t"
3712 "mov %%r12, %c[r12](%0) \n\t"
3713 "mov %%r13, %c[r13](%0) \n\t"
3714 "mov %%r14, %c[r14](%0) \n\t"
3715 "mov %%r15, %c[r15](%0) \n\t"
3717 "mov %%cr2, %%"R"ax \n\t"
3718 "mov %%"R"ax, %c[cr2](%0) \n\t"
3720 "pop %%"R"bp; pop %%"R"bp; pop %%"R"dx \n\t"
3721 "setbe %c[fail](%0) \n\t"
3722 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
3723 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
3724 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
3725 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
3726 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
3727 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
3728 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
3729 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
3730 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
3731 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
3732 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
3733 #ifdef CONFIG_X86_64
3734 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
3735 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
3736 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
3737 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
3738 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
3739 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
3740 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
3741 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
3743 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
3745 , R"bx", R"di", R"si"
3746 #ifdef CONFIG_X86_64
3747 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
3751 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
3752 | (1 << VCPU_EXREG_PDPTR));
3753 vcpu->arch.regs_dirty = 0;
3755 if (vcpu->arch.switch_db_regs)
3756 get_debugreg(vcpu->arch.dr6, 6);
3758 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
3759 if (vmx->rmode.irq.pending)
3760 fixup_rmode_irq(vmx);
3762 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
3765 vmx_complete_interrupts(vmx);
3771 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
3773 struct vcpu_vmx *vmx = to_vmx(vcpu);
3777 free_vmcs(vmx->vmcs);
3782 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
3784 struct vcpu_vmx *vmx = to_vmx(vcpu);
3786 spin_lock(&vmx_vpid_lock);
3788 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3789 spin_unlock(&vmx_vpid_lock);
3790 vmx_free_vmcs(vcpu);
3791 kfree(vmx->guest_msrs);
3792 kvm_vcpu_uninit(vcpu);
3793 kmem_cache_free(kvm_vcpu_cache, vmx);
3796 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
3799 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3803 return ERR_PTR(-ENOMEM);
3807 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
3811 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3812 if (!vmx->guest_msrs) {
3817 vmx->vmcs = alloc_vmcs();
3821 vmcs_clear(vmx->vmcs);
3824 vmx_vcpu_load(&vmx->vcpu, cpu);
3825 err = vmx_vcpu_setup(vmx);
3826 vmx_vcpu_put(&vmx->vcpu);
3830 if (vm_need_virtualize_apic_accesses(kvm))
3831 if (alloc_apic_access_page(kvm) != 0)
3835 if (!kvm->arch.ept_identity_map_addr)
3836 kvm->arch.ept_identity_map_addr =
3837 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3838 if (alloc_identity_pagetable(kvm) != 0)
3845 free_vmcs(vmx->vmcs);
3847 kfree(vmx->guest_msrs);
3849 kvm_vcpu_uninit(&vmx->vcpu);
3851 kmem_cache_free(kvm_vcpu_cache, vmx);
3852 return ERR_PTR(err);
3855 static void __init vmx_check_processor_compat(void *rtn)
3857 struct vmcs_config vmcs_conf;
3860 if (setup_vmcs_config(&vmcs_conf) < 0)
3862 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
3863 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
3864 smp_processor_id());
3869 static int get_ept_level(void)
3871 return VMX_EPT_DEFAULT_GAW + 1;
3874 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3878 /* For VT-d and EPT combination
3879 * 1. MMIO: always map as UC
3881 * a. VT-d without snooping control feature: can't guarantee the
3882 * result, try to trust guest.
3883 * b. VT-d with snooping control feature: snooping control feature of
3884 * VT-d engine can guarantee the cache correctness. Just set it
3885 * to WB to keep consistent with host. So the same as item 3.
3886 * 3. EPT without VT-d: always map as WB and set IGMT=1 to keep
3887 * consistent with host MTRR
3890 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
3891 else if (vcpu->kvm->arch.iommu_domain &&
3892 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
3893 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
3894 VMX_EPT_MT_EPTE_SHIFT;
3896 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
3902 static const struct trace_print_flags vmx_exit_reasons_str[] = {
3903 { EXIT_REASON_EXCEPTION_NMI, "exception" },
3904 { EXIT_REASON_EXTERNAL_INTERRUPT, "ext_irq" },
3905 { EXIT_REASON_TRIPLE_FAULT, "triple_fault" },
3906 { EXIT_REASON_NMI_WINDOW, "nmi_window" },
3907 { EXIT_REASON_IO_INSTRUCTION, "io_instruction" },
3908 { EXIT_REASON_CR_ACCESS, "cr_access" },
3909 { EXIT_REASON_DR_ACCESS, "dr_access" },
3910 { EXIT_REASON_CPUID, "cpuid" },
3911 { EXIT_REASON_MSR_READ, "rdmsr" },
3912 { EXIT_REASON_MSR_WRITE, "wrmsr" },
3913 { EXIT_REASON_PENDING_INTERRUPT, "interrupt_window" },
3914 { EXIT_REASON_HLT, "halt" },
3915 { EXIT_REASON_INVLPG, "invlpg" },
3916 { EXIT_REASON_VMCALL, "hypercall" },
3917 { EXIT_REASON_TPR_BELOW_THRESHOLD, "tpr_below_thres" },
3918 { EXIT_REASON_APIC_ACCESS, "apic_access" },
3919 { EXIT_REASON_WBINVD, "wbinvd" },
3920 { EXIT_REASON_TASK_SWITCH, "task_switch" },
3921 { EXIT_REASON_EPT_VIOLATION, "ept_violation" },
3925 static bool vmx_gb_page_enable(void)
3930 static struct kvm_x86_ops vmx_x86_ops = {
3931 .cpu_has_kvm_support = cpu_has_kvm_support,
3932 .disabled_by_bios = vmx_disabled_by_bios,
3933 .hardware_setup = hardware_setup,
3934 .hardware_unsetup = hardware_unsetup,
3935 .check_processor_compatibility = vmx_check_processor_compat,
3936 .hardware_enable = hardware_enable,
3937 .hardware_disable = hardware_disable,
3938 .cpu_has_accelerated_tpr = report_flexpriority,
3940 .vcpu_create = vmx_create_vcpu,
3941 .vcpu_free = vmx_free_vcpu,
3942 .vcpu_reset = vmx_vcpu_reset,
3944 .prepare_guest_switch = vmx_save_host_state,
3945 .vcpu_load = vmx_vcpu_load,
3946 .vcpu_put = vmx_vcpu_put,
3948 .set_guest_debug = set_guest_debug,
3949 .get_msr = vmx_get_msr,
3950 .set_msr = vmx_set_msr,
3951 .get_segment_base = vmx_get_segment_base,
3952 .get_segment = vmx_get_segment,
3953 .set_segment = vmx_set_segment,
3954 .get_cpl = vmx_get_cpl,
3955 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
3956 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
3957 .set_cr0 = vmx_set_cr0,
3958 .set_cr3 = vmx_set_cr3,
3959 .set_cr4 = vmx_set_cr4,
3960 .set_efer = vmx_set_efer,
3961 .get_idt = vmx_get_idt,
3962 .set_idt = vmx_set_idt,
3963 .get_gdt = vmx_get_gdt,
3964 .set_gdt = vmx_set_gdt,
3965 .cache_reg = vmx_cache_reg,
3966 .get_rflags = vmx_get_rflags,
3967 .set_rflags = vmx_set_rflags,
3969 .tlb_flush = vmx_flush_tlb,
3971 .run = vmx_vcpu_run,
3972 .handle_exit = vmx_handle_exit,
3973 .skip_emulated_instruction = skip_emulated_instruction,
3974 .set_interrupt_shadow = vmx_set_interrupt_shadow,
3975 .get_interrupt_shadow = vmx_get_interrupt_shadow,
3976 .patch_hypercall = vmx_patch_hypercall,
3977 .set_irq = vmx_inject_irq,
3978 .set_nmi = vmx_inject_nmi,
3979 .queue_exception = vmx_queue_exception,
3980 .interrupt_allowed = vmx_interrupt_allowed,
3981 .nmi_allowed = vmx_nmi_allowed,
3982 .enable_nmi_window = enable_nmi_window,
3983 .enable_irq_window = enable_irq_window,
3984 .update_cr8_intercept = update_cr8_intercept,
3986 .set_tss_addr = vmx_set_tss_addr,
3987 .get_tdp_level = get_ept_level,
3988 .get_mt_mask = vmx_get_mt_mask,
3990 .exit_reasons_str = vmx_exit_reasons_str,
3991 .gb_page_enable = vmx_gb_page_enable,
3994 static int __init vmx_init(void)
3998 rdmsrl_safe(MSR_EFER, &host_efer);
4000 for (i = 0; i < NR_VMX_MSR; ++i)
4001 kvm_define_shared_msr(i, vmx_msr_index[i]);
4003 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
4004 if (!vmx_io_bitmap_a)
4007 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
4008 if (!vmx_io_bitmap_b) {
4013 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
4014 if (!vmx_msr_bitmap_legacy) {
4019 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
4020 if (!vmx_msr_bitmap_longmode) {
4026 * Allow direct access to the PC debug port (it is often used for I/O
4027 * delays, but the vmexits simply slow things down).
4029 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
4030 clear_bit(0x80, vmx_io_bitmap_a);
4032 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
4034 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
4035 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
4037 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
4039 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
4043 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
4044 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
4045 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
4046 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
4047 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
4048 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
4051 bypass_guest_pf = 0;
4052 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
4053 VMX_EPT_WRITABLE_MASK);
4054 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4055 VMX_EPT_EXECUTABLE_MASK);
4060 if (bypass_guest_pf)
4061 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
4066 free_page((unsigned long)vmx_msr_bitmap_longmode);
4068 free_page((unsigned long)vmx_msr_bitmap_legacy);
4070 free_page((unsigned long)vmx_io_bitmap_b);
4072 free_page((unsigned long)vmx_io_bitmap_a);
4076 static void __exit vmx_exit(void)
4078 free_page((unsigned long)vmx_msr_bitmap_legacy);
4079 free_page((unsigned long)vmx_msr_bitmap_longmode);
4080 free_page((unsigned long)vmx_io_bitmap_b);
4081 free_page((unsigned long)vmx_io_bitmap_a);
4086 module_init(vmx_init)
4087 module_exit(vmx_exit)