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;
106 u64 msr_host_kernel_gs_base;
107 u64 msr_guest_kernel_gs_base;
112 u16 fs_sel, gs_sel, ldt_sel;
113 int gs_ldt_reload_needed;
114 int fs_reload_needed;
119 struct kvm_save_segment {
124 } tr, es, ds, fs, gs;
132 bool emulation_required;
134 /* Support for vnmi-less CPUs */
135 int soft_vnmi_blocked;
137 s64 vnmi_blocked_time;
141 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
143 return container_of(vcpu, struct vcpu_vmx, vcpu);
146 static int init_rmode(struct kvm *kvm);
147 static u64 construct_eptp(unsigned long root_hpa);
149 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
150 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
151 static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
153 static unsigned long *vmx_io_bitmap_a;
154 static unsigned long *vmx_io_bitmap_b;
155 static unsigned long *vmx_msr_bitmap_legacy;
156 static unsigned long *vmx_msr_bitmap_longmode;
158 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
159 static DEFINE_SPINLOCK(vmx_vpid_lock);
161 static struct vmcs_config {
165 u32 pin_based_exec_ctrl;
166 u32 cpu_based_exec_ctrl;
167 u32 cpu_based_2nd_exec_ctrl;
172 static struct vmx_capability {
177 #define VMX_SEGMENT_FIELD(seg) \
178 [VCPU_SREG_##seg] = { \
179 .selector = GUEST_##seg##_SELECTOR, \
180 .base = GUEST_##seg##_BASE, \
181 .limit = GUEST_##seg##_LIMIT, \
182 .ar_bytes = GUEST_##seg##_AR_BYTES, \
185 static struct kvm_vmx_segment_field {
190 } kvm_vmx_segment_fields[] = {
191 VMX_SEGMENT_FIELD(CS),
192 VMX_SEGMENT_FIELD(DS),
193 VMX_SEGMENT_FIELD(ES),
194 VMX_SEGMENT_FIELD(FS),
195 VMX_SEGMENT_FIELD(GS),
196 VMX_SEGMENT_FIELD(SS),
197 VMX_SEGMENT_FIELD(TR),
198 VMX_SEGMENT_FIELD(LDTR),
201 static u64 host_efer;
203 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
206 * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
207 * away by decrementing the array size.
209 static const u32 vmx_msr_index[] = {
211 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
213 MSR_EFER, MSR_K6_STAR,
215 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
217 static inline int is_page_fault(u32 intr_info)
219 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
220 INTR_INFO_VALID_MASK)) ==
221 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
224 static inline int is_no_device(u32 intr_info)
226 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
227 INTR_INFO_VALID_MASK)) ==
228 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
231 static inline int is_invalid_opcode(u32 intr_info)
233 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
234 INTR_INFO_VALID_MASK)) ==
235 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
238 static inline int is_external_interrupt(u32 intr_info)
240 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
241 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
244 static inline int is_machine_check(u32 intr_info)
246 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
247 INTR_INFO_VALID_MASK)) ==
248 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
251 static inline int cpu_has_vmx_msr_bitmap(void)
253 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
256 static inline int cpu_has_vmx_tpr_shadow(void)
258 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
261 static inline int vm_need_tpr_shadow(struct kvm *kvm)
263 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
266 static inline int cpu_has_secondary_exec_ctrls(void)
268 return vmcs_config.cpu_based_exec_ctrl &
269 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
272 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
274 return vmcs_config.cpu_based_2nd_exec_ctrl &
275 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
278 static inline bool cpu_has_vmx_flexpriority(void)
280 return cpu_has_vmx_tpr_shadow() &&
281 cpu_has_vmx_virtualize_apic_accesses();
284 static inline bool cpu_has_vmx_ept_execute_only(void)
286 return !!(vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT);
289 static inline bool cpu_has_vmx_eptp_uncacheable(void)
291 return !!(vmx_capability.ept & VMX_EPTP_UC_BIT);
294 static inline bool cpu_has_vmx_eptp_writeback(void)
296 return !!(vmx_capability.ept & VMX_EPTP_WB_BIT);
299 static inline bool cpu_has_vmx_ept_2m_page(void)
301 return !!(vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT);
304 static inline int cpu_has_vmx_invept_individual_addr(void)
306 return !!(vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT);
309 static inline int cpu_has_vmx_invept_context(void)
311 return !!(vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT);
314 static inline int cpu_has_vmx_invept_global(void)
316 return !!(vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT);
319 static inline int cpu_has_vmx_ept(void)
321 return vmcs_config.cpu_based_2nd_exec_ctrl &
322 SECONDARY_EXEC_ENABLE_EPT;
325 static inline int cpu_has_vmx_unrestricted_guest(void)
327 return vmcs_config.cpu_based_2nd_exec_ctrl &
328 SECONDARY_EXEC_UNRESTRICTED_GUEST;
331 static inline int cpu_has_vmx_ple(void)
333 return vmcs_config.cpu_based_2nd_exec_ctrl &
334 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
337 static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
339 return flexpriority_enabled &&
340 (cpu_has_vmx_virtualize_apic_accesses()) &&
341 (irqchip_in_kernel(kvm));
344 static inline int cpu_has_vmx_vpid(void)
346 return vmcs_config.cpu_based_2nd_exec_ctrl &
347 SECONDARY_EXEC_ENABLE_VPID;
350 static inline int cpu_has_virtual_nmis(void)
352 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
355 static inline bool report_flexpriority(void)
357 return flexpriority_enabled;
360 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
364 for (i = 0; i < vmx->nmsrs; ++i)
365 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
370 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
376 } operand = { vpid, 0, gva };
378 asm volatile (__ex(ASM_VMX_INVVPID)
379 /* CF==1 or ZF==1 --> rc = -1 */
381 : : "a"(&operand), "c"(ext) : "cc", "memory");
384 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
388 } operand = {eptp, gpa};
390 asm volatile (__ex(ASM_VMX_INVEPT)
391 /* CF==1 or ZF==1 --> rc = -1 */
392 "; ja 1f ; ud2 ; 1:\n"
393 : : "a" (&operand), "c" (ext) : "cc", "memory");
396 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
400 i = __find_msr_index(vmx, msr);
402 return &vmx->guest_msrs[i];
406 static void vmcs_clear(struct vmcs *vmcs)
408 u64 phys_addr = __pa(vmcs);
411 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
412 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
415 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
419 static void __vcpu_clear(void *arg)
421 struct vcpu_vmx *vmx = arg;
422 int cpu = raw_smp_processor_id();
424 if (vmx->vcpu.cpu == cpu)
425 vmcs_clear(vmx->vmcs);
426 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
427 per_cpu(current_vmcs, cpu) = NULL;
428 rdtscll(vmx->vcpu.arch.host_tsc);
429 list_del(&vmx->local_vcpus_link);
434 static void vcpu_clear(struct vcpu_vmx *vmx)
436 if (vmx->vcpu.cpu == -1)
438 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
441 static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
446 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
449 static inline void ept_sync_global(void)
451 if (cpu_has_vmx_invept_global())
452 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
455 static inline void ept_sync_context(u64 eptp)
458 if (cpu_has_vmx_invept_context())
459 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
465 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
468 if (cpu_has_vmx_invept_individual_addr())
469 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
472 ept_sync_context(eptp);
476 static unsigned long vmcs_readl(unsigned long field)
480 asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
481 : "=a"(value) : "d"(field) : "cc");
485 static u16 vmcs_read16(unsigned long field)
487 return vmcs_readl(field);
490 static u32 vmcs_read32(unsigned long field)
492 return vmcs_readl(field);
495 static u64 vmcs_read64(unsigned long field)
498 return vmcs_readl(field);
500 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
504 static noinline void vmwrite_error(unsigned long field, unsigned long value)
506 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
507 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
511 static void vmcs_writel(unsigned long field, unsigned long value)
515 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
516 : "=q"(error) : "a"(value), "d"(field) : "cc");
518 vmwrite_error(field, value);
521 static void vmcs_write16(unsigned long field, u16 value)
523 vmcs_writel(field, value);
526 static void vmcs_write32(unsigned long field, u32 value)
528 vmcs_writel(field, value);
531 static void vmcs_write64(unsigned long field, u64 value)
533 vmcs_writel(field, value);
534 #ifndef CONFIG_X86_64
536 vmcs_writel(field+1, value >> 32);
540 static void vmcs_clear_bits(unsigned long field, u32 mask)
542 vmcs_writel(field, vmcs_readl(field) & ~mask);
545 static void vmcs_set_bits(unsigned long field, u32 mask)
547 vmcs_writel(field, vmcs_readl(field) | mask);
550 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
554 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR);
555 if (!vcpu->fpu_active)
556 eb |= 1u << NM_VECTOR;
558 * Unconditionally intercept #DB so we can maintain dr6 without
559 * reading it every exit.
561 eb |= 1u << DB_VECTOR;
562 if (vcpu->guest_debug & KVM_GUESTDBG_ENABLE) {
563 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
564 eb |= 1u << BP_VECTOR;
566 if (to_vmx(vcpu)->rmode.vm86_active)
569 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
570 vmcs_write32(EXCEPTION_BITMAP, eb);
573 static void reload_tss(void)
576 * VT restores TR but not its size. Useless.
578 struct descriptor_table gdt;
579 struct desc_struct *descs;
582 descs = (void *)gdt.base;
583 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
587 static bool update_transition_efer(struct vcpu_vmx *vmx)
589 int efer_offset = vmx->msr_offset_efer;
595 guest_efer = vmx->vcpu.arch.shadow_efer;
598 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
601 ignore_bits = EFER_NX | EFER_SCE;
603 ignore_bits |= EFER_LMA | EFER_LME;
604 /* SCE is meaningful only in long mode on Intel */
605 if (guest_efer & EFER_LMA)
606 ignore_bits &= ~(u64)EFER_SCE;
608 if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
611 guest_efer &= ~ignore_bits;
612 guest_efer |= host_efer & ignore_bits;
613 vmx->guest_msrs[efer_offset].data = guest_efer;
617 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
619 struct vcpu_vmx *vmx = to_vmx(vcpu);
622 if (vmx->host_state.loaded)
625 vmx->host_state.loaded = 1;
627 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
628 * allow segment selectors with cpl > 0 or ti == 1.
630 vmx->host_state.ldt_sel = kvm_read_ldt();
631 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
632 vmx->host_state.fs_sel = kvm_read_fs();
633 if (!(vmx->host_state.fs_sel & 7)) {
634 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
635 vmx->host_state.fs_reload_needed = 0;
637 vmcs_write16(HOST_FS_SELECTOR, 0);
638 vmx->host_state.fs_reload_needed = 1;
640 vmx->host_state.gs_sel = kvm_read_gs();
641 if (!(vmx->host_state.gs_sel & 7))
642 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
644 vmcs_write16(HOST_GS_SELECTOR, 0);
645 vmx->host_state.gs_ldt_reload_needed = 1;
649 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
650 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
652 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
653 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
657 if (is_long_mode(&vmx->vcpu)) {
658 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
659 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
662 for (i = 0; i < vmx->save_nmsrs; ++i)
663 kvm_set_shared_msr(vmx->guest_msrs[i].index,
664 vmx->guest_msrs[i].data);
667 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
671 if (!vmx->host_state.loaded)
674 ++vmx->vcpu.stat.host_state_reload;
675 vmx->host_state.loaded = 0;
676 if (vmx->host_state.fs_reload_needed)
677 kvm_load_fs(vmx->host_state.fs_sel);
678 if (vmx->host_state.gs_ldt_reload_needed) {
679 kvm_load_ldt(vmx->host_state.ldt_sel);
681 * If we have to reload gs, we must take care to
682 * preserve our gs base.
684 local_irq_save(flags);
685 kvm_load_gs(vmx->host_state.gs_sel);
687 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
689 local_irq_restore(flags);
693 if (is_long_mode(&vmx->vcpu)) {
694 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
695 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
700 static void vmx_load_host_state(struct vcpu_vmx *vmx)
703 __vmx_load_host_state(vmx);
708 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
709 * vcpu mutex is already taken.
711 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
713 struct vcpu_vmx *vmx = to_vmx(vcpu);
714 u64 phys_addr = __pa(vmx->vmcs);
715 u64 tsc_this, delta, new_offset;
717 if (vcpu->cpu != cpu) {
719 kvm_migrate_timers(vcpu);
720 set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests);
722 list_add(&vmx->local_vcpus_link,
723 &per_cpu(vcpus_on_cpu, cpu));
727 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
730 per_cpu(current_vmcs, cpu) = vmx->vmcs;
731 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
732 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
735 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
736 vmx->vmcs, phys_addr);
739 if (vcpu->cpu != cpu) {
740 struct descriptor_table dt;
741 unsigned long sysenter_esp;
745 * Linux uses per-cpu TSS and GDT, so set these when switching
748 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
750 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
752 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
753 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
756 * Make sure the time stamp counter is monotonous.
759 if (tsc_this < vcpu->arch.host_tsc) {
760 delta = vcpu->arch.host_tsc - tsc_this;
761 new_offset = vmcs_read64(TSC_OFFSET) + delta;
762 vmcs_write64(TSC_OFFSET, new_offset);
767 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
769 __vmx_load_host_state(to_vmx(vcpu));
772 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
774 if (vcpu->fpu_active)
776 vcpu->fpu_active = 1;
777 vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
778 if (vcpu->arch.cr0 & X86_CR0_TS)
779 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
780 update_exception_bitmap(vcpu);
783 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
785 if (!vcpu->fpu_active)
787 vcpu->fpu_active = 0;
788 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
789 update_exception_bitmap(vcpu);
792 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
794 unsigned long rflags;
796 rflags = vmcs_readl(GUEST_RFLAGS);
797 if (to_vmx(vcpu)->rmode.vm86_active)
798 rflags &= ~(unsigned long)(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
802 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
804 if (to_vmx(vcpu)->rmode.vm86_active)
805 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
806 vmcs_writel(GUEST_RFLAGS, rflags);
809 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
811 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
814 if (interruptibility & GUEST_INTR_STATE_STI)
815 ret |= X86_SHADOW_INT_STI;
816 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
817 ret |= X86_SHADOW_INT_MOV_SS;
822 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
824 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
825 u32 interruptibility = interruptibility_old;
827 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
829 if (mask & X86_SHADOW_INT_MOV_SS)
830 interruptibility |= GUEST_INTR_STATE_MOV_SS;
831 if (mask & X86_SHADOW_INT_STI)
832 interruptibility |= GUEST_INTR_STATE_STI;
834 if ((interruptibility != interruptibility_old))
835 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
838 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
842 rip = kvm_rip_read(vcpu);
843 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
844 kvm_rip_write(vcpu, rip);
846 /* skipping an emulated instruction also counts */
847 vmx_set_interrupt_shadow(vcpu, 0);
850 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
851 bool has_error_code, u32 error_code)
853 struct vcpu_vmx *vmx = to_vmx(vcpu);
854 u32 intr_info = nr | INTR_INFO_VALID_MASK;
856 if (has_error_code) {
857 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
858 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
861 if (vmx->rmode.vm86_active) {
862 vmx->rmode.irq.pending = true;
863 vmx->rmode.irq.vector = nr;
864 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
865 if (kvm_exception_is_soft(nr))
866 vmx->rmode.irq.rip +=
867 vmx->vcpu.arch.event_exit_inst_len;
868 intr_info |= INTR_TYPE_SOFT_INTR;
869 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
870 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
871 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
875 if (kvm_exception_is_soft(nr)) {
876 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
877 vmx->vcpu.arch.event_exit_inst_len);
878 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
880 intr_info |= INTR_TYPE_HARD_EXCEPTION;
882 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
886 * Swap MSR entry in host/guest MSR entry array.
888 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
890 struct shared_msr_entry tmp;
892 tmp = vmx->guest_msrs[to];
893 vmx->guest_msrs[to] = vmx->guest_msrs[from];
894 vmx->guest_msrs[from] = tmp;
898 * Set up the vmcs to automatically save and restore system
899 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
900 * mode, as fiddling with msrs is very expensive.
902 static void setup_msrs(struct vcpu_vmx *vmx)
904 int save_nmsrs, index;
905 unsigned long *msr_bitmap;
907 vmx_load_host_state(vmx);
910 if (is_long_mode(&vmx->vcpu)) {
911 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
913 move_msr_up(vmx, index, save_nmsrs++);
914 index = __find_msr_index(vmx, MSR_LSTAR);
916 move_msr_up(vmx, index, save_nmsrs++);
917 index = __find_msr_index(vmx, MSR_CSTAR);
919 move_msr_up(vmx, index, save_nmsrs++);
921 * MSR_K6_STAR is only needed on long mode guests, and only
922 * if efer.sce is enabled.
924 index = __find_msr_index(vmx, MSR_K6_STAR);
925 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
926 move_msr_up(vmx, index, save_nmsrs++);
929 vmx->msr_offset_efer = index = __find_msr_index(vmx, MSR_EFER);
930 if (index >= 0 && update_transition_efer(vmx))
931 move_msr_up(vmx, index, save_nmsrs++);
933 vmx->save_nmsrs = save_nmsrs;
935 if (cpu_has_vmx_msr_bitmap()) {
936 if (is_long_mode(&vmx->vcpu))
937 msr_bitmap = vmx_msr_bitmap_longmode;
939 msr_bitmap = vmx_msr_bitmap_legacy;
941 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
946 * reads and returns guest's timestamp counter "register"
947 * guest_tsc = host_tsc + tsc_offset -- 21.3
949 static u64 guest_read_tsc(void)
951 u64 host_tsc, tsc_offset;
954 tsc_offset = vmcs_read64(TSC_OFFSET);
955 return host_tsc + tsc_offset;
959 * writes 'guest_tsc' into guest's timestamp counter "register"
960 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
962 static void guest_write_tsc(u64 guest_tsc, u64 host_tsc)
964 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
968 * Reads an msr value (of 'msr_index') into 'pdata'.
969 * Returns 0 on success, non-0 otherwise.
970 * Assumes vcpu_load() was already called.
972 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
975 struct shared_msr_entry *msr;
978 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
985 data = vmcs_readl(GUEST_FS_BASE);
988 data = vmcs_readl(GUEST_GS_BASE);
990 case MSR_KERNEL_GS_BASE:
991 vmx_load_host_state(to_vmx(vcpu));
992 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
996 return kvm_get_msr_common(vcpu, msr_index, pdata);
998 data = guest_read_tsc();
1000 case MSR_IA32_SYSENTER_CS:
1001 data = vmcs_read32(GUEST_SYSENTER_CS);
1003 case MSR_IA32_SYSENTER_EIP:
1004 data = vmcs_readl(GUEST_SYSENTER_EIP);
1006 case MSR_IA32_SYSENTER_ESP:
1007 data = vmcs_readl(GUEST_SYSENTER_ESP);
1010 vmx_load_host_state(to_vmx(vcpu));
1011 msr = find_msr_entry(to_vmx(vcpu), msr_index);
1013 vmx_load_host_state(to_vmx(vcpu));
1017 return kvm_get_msr_common(vcpu, msr_index, pdata);
1025 * Writes msr value into into the appropriate "register".
1026 * Returns 0 on success, non-0 otherwise.
1027 * Assumes vcpu_load() was already called.
1029 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1031 struct vcpu_vmx *vmx = to_vmx(vcpu);
1032 struct shared_msr_entry *msr;
1036 switch (msr_index) {
1038 vmx_load_host_state(vmx);
1039 ret = kvm_set_msr_common(vcpu, msr_index, data);
1041 #ifdef CONFIG_X86_64
1043 vmcs_writel(GUEST_FS_BASE, data);
1046 vmcs_writel(GUEST_GS_BASE, data);
1048 case MSR_KERNEL_GS_BASE:
1049 vmx_load_host_state(vmx);
1050 vmx->msr_guest_kernel_gs_base = data;
1053 case MSR_IA32_SYSENTER_CS:
1054 vmcs_write32(GUEST_SYSENTER_CS, data);
1056 case MSR_IA32_SYSENTER_EIP:
1057 vmcs_writel(GUEST_SYSENTER_EIP, data);
1059 case MSR_IA32_SYSENTER_ESP:
1060 vmcs_writel(GUEST_SYSENTER_ESP, data);
1064 guest_write_tsc(data, host_tsc);
1066 case MSR_IA32_CR_PAT:
1067 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1068 vmcs_write64(GUEST_IA32_PAT, data);
1069 vcpu->arch.pat = data;
1072 /* Otherwise falls through to kvm_set_msr_common */
1074 msr = find_msr_entry(vmx, msr_index);
1076 vmx_load_host_state(vmx);
1080 ret = kvm_set_msr_common(vcpu, msr_index, data);
1086 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1088 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1091 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1094 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1096 case VCPU_EXREG_PDPTR:
1098 ept_save_pdptrs(vcpu);
1105 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1107 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1108 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1110 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1112 update_exception_bitmap(vcpu);
1115 static __init int cpu_has_kvm_support(void)
1117 return cpu_has_vmx();
1120 static __init int vmx_disabled_by_bios(void)
1124 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1125 return (msr & (FEATURE_CONTROL_LOCKED |
1126 FEATURE_CONTROL_VMXON_ENABLED))
1127 == FEATURE_CONTROL_LOCKED;
1128 /* locked but not enabled */
1131 static int hardware_enable(void *garbage)
1133 int cpu = raw_smp_processor_id();
1134 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1137 if (read_cr4() & X86_CR4_VMXE)
1140 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1141 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1142 if ((old & (FEATURE_CONTROL_LOCKED |
1143 FEATURE_CONTROL_VMXON_ENABLED))
1144 != (FEATURE_CONTROL_LOCKED |
1145 FEATURE_CONTROL_VMXON_ENABLED))
1146 /* enable and lock */
1147 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
1148 FEATURE_CONTROL_LOCKED |
1149 FEATURE_CONTROL_VMXON_ENABLED);
1150 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1151 asm volatile (ASM_VMX_VMXON_RAX
1152 : : "a"(&phys_addr), "m"(phys_addr)
1160 static void vmclear_local_vcpus(void)
1162 int cpu = raw_smp_processor_id();
1163 struct vcpu_vmx *vmx, *n;
1165 list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1171 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1174 static void kvm_cpu_vmxoff(void)
1176 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1177 write_cr4(read_cr4() & ~X86_CR4_VMXE);
1180 static void hardware_disable(void *garbage)
1182 vmclear_local_vcpus();
1186 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1187 u32 msr, u32 *result)
1189 u32 vmx_msr_low, vmx_msr_high;
1190 u32 ctl = ctl_min | ctl_opt;
1192 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1194 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1195 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
1197 /* Ensure minimum (required) set of control bits are supported. */
1205 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1207 u32 vmx_msr_low, vmx_msr_high;
1208 u32 min, opt, min2, opt2;
1209 u32 _pin_based_exec_control = 0;
1210 u32 _cpu_based_exec_control = 0;
1211 u32 _cpu_based_2nd_exec_control = 0;
1212 u32 _vmexit_control = 0;
1213 u32 _vmentry_control = 0;
1215 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1216 opt = PIN_BASED_VIRTUAL_NMIS;
1217 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1218 &_pin_based_exec_control) < 0)
1221 min = CPU_BASED_HLT_EXITING |
1222 #ifdef CONFIG_X86_64
1223 CPU_BASED_CR8_LOAD_EXITING |
1224 CPU_BASED_CR8_STORE_EXITING |
1226 CPU_BASED_CR3_LOAD_EXITING |
1227 CPU_BASED_CR3_STORE_EXITING |
1228 CPU_BASED_USE_IO_BITMAPS |
1229 CPU_BASED_MOV_DR_EXITING |
1230 CPU_BASED_USE_TSC_OFFSETING |
1231 CPU_BASED_INVLPG_EXITING;
1232 opt = CPU_BASED_TPR_SHADOW |
1233 CPU_BASED_USE_MSR_BITMAPS |
1234 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1235 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1236 &_cpu_based_exec_control) < 0)
1238 #ifdef CONFIG_X86_64
1239 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1240 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1241 ~CPU_BASED_CR8_STORE_EXITING;
1243 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1245 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1246 SECONDARY_EXEC_WBINVD_EXITING |
1247 SECONDARY_EXEC_ENABLE_VPID |
1248 SECONDARY_EXEC_ENABLE_EPT |
1249 SECONDARY_EXEC_UNRESTRICTED_GUEST |
1250 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1251 if (adjust_vmx_controls(min2, opt2,
1252 MSR_IA32_VMX_PROCBASED_CTLS2,
1253 &_cpu_based_2nd_exec_control) < 0)
1256 #ifndef CONFIG_X86_64
1257 if (!(_cpu_based_2nd_exec_control &
1258 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1259 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1261 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1262 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1264 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
1265 CPU_BASED_CR3_STORE_EXITING |
1266 CPU_BASED_INVLPG_EXITING);
1267 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1268 vmx_capability.ept, vmx_capability.vpid);
1272 #ifdef CONFIG_X86_64
1273 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1275 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1276 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1277 &_vmexit_control) < 0)
1281 opt = VM_ENTRY_LOAD_IA32_PAT;
1282 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1283 &_vmentry_control) < 0)
1286 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1288 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1289 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1292 #ifdef CONFIG_X86_64
1293 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1294 if (vmx_msr_high & (1u<<16))
1298 /* Require Write-Back (WB) memory type for VMCS accesses. */
1299 if (((vmx_msr_high >> 18) & 15) != 6)
1302 vmcs_conf->size = vmx_msr_high & 0x1fff;
1303 vmcs_conf->order = get_order(vmcs_config.size);
1304 vmcs_conf->revision_id = vmx_msr_low;
1306 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1307 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1308 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1309 vmcs_conf->vmexit_ctrl = _vmexit_control;
1310 vmcs_conf->vmentry_ctrl = _vmentry_control;
1315 static struct vmcs *alloc_vmcs_cpu(int cpu)
1317 int node = cpu_to_node(cpu);
1321 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
1324 vmcs = page_address(pages);
1325 memset(vmcs, 0, vmcs_config.size);
1326 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1330 static struct vmcs *alloc_vmcs(void)
1332 return alloc_vmcs_cpu(raw_smp_processor_id());
1335 static void free_vmcs(struct vmcs *vmcs)
1337 free_pages((unsigned long)vmcs, vmcs_config.order);
1340 static void free_kvm_area(void)
1344 for_each_possible_cpu(cpu) {
1345 free_vmcs(per_cpu(vmxarea, cpu));
1346 per_cpu(vmxarea, cpu) = NULL;
1350 static __init int alloc_kvm_area(void)
1354 for_each_possible_cpu(cpu) {
1357 vmcs = alloc_vmcs_cpu(cpu);
1363 per_cpu(vmxarea, cpu) = vmcs;
1368 static __init int hardware_setup(void)
1370 if (setup_vmcs_config(&vmcs_config) < 0)
1373 if (boot_cpu_has(X86_FEATURE_NX))
1374 kvm_enable_efer_bits(EFER_NX);
1376 if (!cpu_has_vmx_vpid())
1379 if (!cpu_has_vmx_ept()) {
1381 enable_unrestricted_guest = 0;
1384 if (!cpu_has_vmx_unrestricted_guest())
1385 enable_unrestricted_guest = 0;
1387 if (!cpu_has_vmx_flexpriority())
1388 flexpriority_enabled = 0;
1390 if (!cpu_has_vmx_tpr_shadow())
1391 kvm_x86_ops->update_cr8_intercept = NULL;
1393 if (enable_ept && !cpu_has_vmx_ept_2m_page())
1394 kvm_disable_largepages();
1396 if (!cpu_has_vmx_ple())
1399 return alloc_kvm_area();
1402 static __exit void hardware_unsetup(void)
1407 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1409 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1411 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1412 vmcs_write16(sf->selector, save->selector);
1413 vmcs_writel(sf->base, save->base);
1414 vmcs_write32(sf->limit, save->limit);
1415 vmcs_write32(sf->ar_bytes, save->ar);
1417 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1419 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1423 static void enter_pmode(struct kvm_vcpu *vcpu)
1425 unsigned long flags;
1426 struct vcpu_vmx *vmx = to_vmx(vcpu);
1428 vmx->emulation_required = 1;
1429 vmx->rmode.vm86_active = 0;
1431 vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
1432 vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
1433 vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
1435 flags = vmcs_readl(GUEST_RFLAGS);
1436 flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
1437 flags |= (vmx->rmode.save_iopl << IOPL_SHIFT);
1438 vmcs_writel(GUEST_RFLAGS, flags);
1440 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1441 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1443 update_exception_bitmap(vcpu);
1445 if (emulate_invalid_guest_state)
1448 fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
1449 fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
1450 fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
1451 fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
1453 vmcs_write16(GUEST_SS_SELECTOR, 0);
1454 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1456 vmcs_write16(GUEST_CS_SELECTOR,
1457 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1458 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1461 static gva_t rmode_tss_base(struct kvm *kvm)
1463 if (!kvm->arch.tss_addr) {
1464 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1465 kvm->memslots[0].npages - 3;
1466 return base_gfn << PAGE_SHIFT;
1468 return kvm->arch.tss_addr;
1471 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1473 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1475 save->selector = vmcs_read16(sf->selector);
1476 save->base = vmcs_readl(sf->base);
1477 save->limit = vmcs_read32(sf->limit);
1478 save->ar = vmcs_read32(sf->ar_bytes);
1479 vmcs_write16(sf->selector, save->base >> 4);
1480 vmcs_write32(sf->base, save->base & 0xfffff);
1481 vmcs_write32(sf->limit, 0xffff);
1482 vmcs_write32(sf->ar_bytes, 0xf3);
1485 static void enter_rmode(struct kvm_vcpu *vcpu)
1487 unsigned long flags;
1488 struct vcpu_vmx *vmx = to_vmx(vcpu);
1490 if (enable_unrestricted_guest)
1493 vmx->emulation_required = 1;
1494 vmx->rmode.vm86_active = 1;
1496 vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1497 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1499 vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1500 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1502 vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1503 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1505 flags = vmcs_readl(GUEST_RFLAGS);
1506 vmx->rmode.save_iopl
1507 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
1509 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1511 vmcs_writel(GUEST_RFLAGS, flags);
1512 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1513 update_exception_bitmap(vcpu);
1515 if (emulate_invalid_guest_state)
1516 goto continue_rmode;
1518 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1519 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1520 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1522 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1523 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1524 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1525 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1526 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1528 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
1529 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
1530 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
1531 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
1534 kvm_mmu_reset_context(vcpu);
1535 init_rmode(vcpu->kvm);
1538 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1540 struct vcpu_vmx *vmx = to_vmx(vcpu);
1541 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1547 * Force kernel_gs_base reloading before EFER changes, as control
1548 * of this msr depends on is_long_mode().
1550 vmx_load_host_state(to_vmx(vcpu));
1551 vcpu->arch.shadow_efer = efer;
1554 if (efer & EFER_LMA) {
1555 vmcs_write32(VM_ENTRY_CONTROLS,
1556 vmcs_read32(VM_ENTRY_CONTROLS) |
1557 VM_ENTRY_IA32E_MODE);
1560 vmcs_write32(VM_ENTRY_CONTROLS,
1561 vmcs_read32(VM_ENTRY_CONTROLS) &
1562 ~VM_ENTRY_IA32E_MODE);
1564 msr->data = efer & ~EFER_LME;
1569 #ifdef CONFIG_X86_64
1571 static void enter_lmode(struct kvm_vcpu *vcpu)
1575 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1576 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1577 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1579 vmcs_write32(GUEST_TR_AR_BYTES,
1580 (guest_tr_ar & ~AR_TYPE_MASK)
1581 | AR_TYPE_BUSY_64_TSS);
1583 vcpu->arch.shadow_efer |= EFER_LMA;
1584 vmx_set_efer(vcpu, vcpu->arch.shadow_efer);
1587 static void exit_lmode(struct kvm_vcpu *vcpu)
1589 vcpu->arch.shadow_efer &= ~EFER_LMA;
1591 vmcs_write32(VM_ENTRY_CONTROLS,
1592 vmcs_read32(VM_ENTRY_CONTROLS)
1593 & ~VM_ENTRY_IA32E_MODE);
1598 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1600 vpid_sync_vcpu_all(to_vmx(vcpu));
1602 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1605 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1607 vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK;
1608 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
1611 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1613 if (!test_bit(VCPU_EXREG_PDPTR,
1614 (unsigned long *)&vcpu->arch.regs_dirty))
1617 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1618 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1619 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1620 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1621 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1625 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
1627 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1628 vcpu->arch.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
1629 vcpu->arch.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
1630 vcpu->arch.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
1631 vcpu->arch.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
1634 __set_bit(VCPU_EXREG_PDPTR,
1635 (unsigned long *)&vcpu->arch.regs_avail);
1636 __set_bit(VCPU_EXREG_PDPTR,
1637 (unsigned long *)&vcpu->arch.regs_dirty);
1640 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1642 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1644 struct kvm_vcpu *vcpu)
1646 if (!(cr0 & X86_CR0_PG)) {
1647 /* From paging/starting to nonpaging */
1648 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1649 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1650 (CPU_BASED_CR3_LOAD_EXITING |
1651 CPU_BASED_CR3_STORE_EXITING));
1652 vcpu->arch.cr0 = cr0;
1653 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1654 } else if (!is_paging(vcpu)) {
1655 /* From nonpaging to paging */
1656 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1657 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1658 ~(CPU_BASED_CR3_LOAD_EXITING |
1659 CPU_BASED_CR3_STORE_EXITING));
1660 vcpu->arch.cr0 = cr0;
1661 vmx_set_cr4(vcpu, vcpu->arch.cr4);
1664 if (!(cr0 & X86_CR0_WP))
1665 *hw_cr0 &= ~X86_CR0_WP;
1668 static void ept_update_paging_mode_cr4(unsigned long *hw_cr4,
1669 struct kvm_vcpu *vcpu)
1671 if (!is_paging(vcpu)) {
1672 *hw_cr4 &= ~X86_CR4_PAE;
1673 *hw_cr4 |= X86_CR4_PSE;
1674 } else if (!(vcpu->arch.cr4 & X86_CR4_PAE))
1675 *hw_cr4 &= ~X86_CR4_PAE;
1678 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1680 struct vcpu_vmx *vmx = to_vmx(vcpu);
1681 unsigned long hw_cr0;
1683 if (enable_unrestricted_guest)
1684 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
1685 | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
1687 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
1689 vmx_fpu_deactivate(vcpu);
1691 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
1694 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
1697 #ifdef CONFIG_X86_64
1698 if (vcpu->arch.shadow_efer & EFER_LME) {
1699 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1701 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1707 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1709 vmcs_writel(CR0_READ_SHADOW, cr0);
1710 vmcs_writel(GUEST_CR0, hw_cr0);
1711 vcpu->arch.cr0 = cr0;
1713 if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
1714 vmx_fpu_activate(vcpu);
1717 static u64 construct_eptp(unsigned long root_hpa)
1721 /* TODO write the value reading from MSR */
1722 eptp = VMX_EPT_DEFAULT_MT |
1723 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1724 eptp |= (root_hpa & PAGE_MASK);
1729 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1731 unsigned long guest_cr3;
1736 eptp = construct_eptp(cr3);
1737 vmcs_write64(EPT_POINTER, eptp);
1738 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1739 vcpu->kvm->arch.ept_identity_map_addr;
1740 ept_load_pdptrs(vcpu);
1743 vmx_flush_tlb(vcpu);
1744 vmcs_writel(GUEST_CR3, guest_cr3);
1745 if (vcpu->arch.cr0 & X86_CR0_PE)
1746 vmx_fpu_deactivate(vcpu);
1749 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1751 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
1752 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1754 vcpu->arch.cr4 = cr4;
1756 ept_update_paging_mode_cr4(&hw_cr4, vcpu);
1758 vmcs_writel(CR4_READ_SHADOW, cr4);
1759 vmcs_writel(GUEST_CR4, hw_cr4);
1762 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1764 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1766 return vmcs_readl(sf->base);
1769 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1770 struct kvm_segment *var, int seg)
1772 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1775 var->base = vmcs_readl(sf->base);
1776 var->limit = vmcs_read32(sf->limit);
1777 var->selector = vmcs_read16(sf->selector);
1778 ar = vmcs_read32(sf->ar_bytes);
1779 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
1781 var->type = ar & 15;
1782 var->s = (ar >> 4) & 1;
1783 var->dpl = (ar >> 5) & 3;
1784 var->present = (ar >> 7) & 1;
1785 var->avl = (ar >> 12) & 1;
1786 var->l = (ar >> 13) & 1;
1787 var->db = (ar >> 14) & 1;
1788 var->g = (ar >> 15) & 1;
1789 var->unusable = (ar >> 16) & 1;
1792 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
1794 if (!(vcpu->arch.cr0 & X86_CR0_PE)) /* if real mode */
1797 if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
1800 return vmcs_read16(GUEST_CS_SELECTOR) & 3;
1803 static u32 vmx_segment_access_rights(struct kvm_segment *var)
1810 ar = var->type & 15;
1811 ar |= (var->s & 1) << 4;
1812 ar |= (var->dpl & 3) << 5;
1813 ar |= (var->present & 1) << 7;
1814 ar |= (var->avl & 1) << 12;
1815 ar |= (var->l & 1) << 13;
1816 ar |= (var->db & 1) << 14;
1817 ar |= (var->g & 1) << 15;
1819 if (ar == 0) /* a 0 value means unusable */
1820 ar = AR_UNUSABLE_MASK;
1825 static void vmx_set_segment(struct kvm_vcpu *vcpu,
1826 struct kvm_segment *var, int seg)
1828 struct vcpu_vmx *vmx = to_vmx(vcpu);
1829 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1832 if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
1833 vmx->rmode.tr.selector = var->selector;
1834 vmx->rmode.tr.base = var->base;
1835 vmx->rmode.tr.limit = var->limit;
1836 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
1839 vmcs_writel(sf->base, var->base);
1840 vmcs_write32(sf->limit, var->limit);
1841 vmcs_write16(sf->selector, var->selector);
1842 if (vmx->rmode.vm86_active && var->s) {
1844 * Hack real-mode segments into vm86 compatibility.
1846 if (var->base == 0xffff0000 && var->selector == 0xf000)
1847 vmcs_writel(sf->base, 0xf0000);
1850 ar = vmx_segment_access_rights(var);
1853 * Fix the "Accessed" bit in AR field of segment registers for older
1855 * IA32 arch specifies that at the time of processor reset the
1856 * "Accessed" bit in the AR field of segment registers is 1. And qemu
1857 * is setting it to 0 in the usedland code. This causes invalid guest
1858 * state vmexit when "unrestricted guest" mode is turned on.
1859 * Fix for this setup issue in cpu_reset is being pushed in the qemu
1860 * tree. Newer qemu binaries with that qemu fix would not need this
1863 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
1864 ar |= 0x1; /* Accessed */
1866 vmcs_write32(sf->ar_bytes, ar);
1869 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1871 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1873 *db = (ar >> 14) & 1;
1874 *l = (ar >> 13) & 1;
1877 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1879 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1880 dt->base = vmcs_readl(GUEST_IDTR_BASE);
1883 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1885 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1886 vmcs_writel(GUEST_IDTR_BASE, dt->base);
1889 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1891 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1892 dt->base = vmcs_readl(GUEST_GDTR_BASE);
1895 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1897 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1898 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1901 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
1903 struct kvm_segment var;
1906 vmx_get_segment(vcpu, &var, seg);
1907 ar = vmx_segment_access_rights(&var);
1909 if (var.base != (var.selector << 4))
1911 if (var.limit != 0xffff)
1919 static bool code_segment_valid(struct kvm_vcpu *vcpu)
1921 struct kvm_segment cs;
1922 unsigned int cs_rpl;
1924 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1925 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
1929 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
1933 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
1934 if (cs.dpl > cs_rpl)
1937 if (cs.dpl != cs_rpl)
1943 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
1947 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
1949 struct kvm_segment ss;
1950 unsigned int ss_rpl;
1952 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
1953 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
1957 if (ss.type != 3 && ss.type != 7)
1961 if (ss.dpl != ss_rpl) /* DPL != RPL */
1969 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
1971 struct kvm_segment var;
1974 vmx_get_segment(vcpu, &var, seg);
1975 rpl = var.selector & SELECTOR_RPL_MASK;
1983 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
1984 if (var.dpl < rpl) /* DPL < RPL */
1988 /* TODO: Add other members to kvm_segment_field to allow checking for other access
1994 static bool tr_valid(struct kvm_vcpu *vcpu)
1996 struct kvm_segment tr;
1998 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
2002 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2004 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
2012 static bool ldtr_valid(struct kvm_vcpu *vcpu)
2014 struct kvm_segment ldtr;
2016 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
2020 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2030 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
2032 struct kvm_segment cs, ss;
2034 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2035 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2037 return ((cs.selector & SELECTOR_RPL_MASK) ==
2038 (ss.selector & SELECTOR_RPL_MASK));
2042 * Check if guest state is valid. Returns true if valid, false if
2044 * We assume that registers are always usable
2046 static bool guest_state_valid(struct kvm_vcpu *vcpu)
2048 /* real mode guest state checks */
2049 if (!(vcpu->arch.cr0 & X86_CR0_PE)) {
2050 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
2052 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
2054 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
2056 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
2058 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
2060 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
2063 /* protected mode guest state checks */
2064 if (!cs_ss_rpl_check(vcpu))
2066 if (!code_segment_valid(vcpu))
2068 if (!stack_segment_valid(vcpu))
2070 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
2072 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
2074 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
2076 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2078 if (!tr_valid(vcpu))
2080 if (!ldtr_valid(vcpu))
2084 * - Add checks on RIP
2085 * - Add checks on RFLAGS
2091 static int init_rmode_tss(struct kvm *kvm)
2093 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
2098 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2101 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
2102 r = kvm_write_guest_page(kvm, fn++, &data,
2103 TSS_IOPB_BASE_OFFSET, sizeof(u16));
2106 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2109 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2113 r = kvm_write_guest_page(kvm, fn, &data,
2114 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2124 static int init_rmode_identity_map(struct kvm *kvm)
2127 pfn_t identity_map_pfn;
2132 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2133 printk(KERN_ERR "EPT: identity-mapping pagetable "
2134 "haven't been allocated!\n");
2137 if (likely(kvm->arch.ept_identity_pagetable_done))
2140 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
2141 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2144 /* Set up identity-mapping pagetable for EPT in real mode */
2145 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2146 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2147 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2148 r = kvm_write_guest_page(kvm, identity_map_pfn,
2149 &tmp, i * sizeof(tmp), sizeof(tmp));
2153 kvm->arch.ept_identity_pagetable_done = true;
2159 static void seg_setup(int seg)
2161 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2164 vmcs_write16(sf->selector, 0);
2165 vmcs_writel(sf->base, 0);
2166 vmcs_write32(sf->limit, 0xffff);
2167 if (enable_unrestricted_guest) {
2169 if (seg == VCPU_SREG_CS)
2170 ar |= 0x08; /* code segment */
2174 vmcs_write32(sf->ar_bytes, ar);
2177 static int alloc_apic_access_page(struct kvm *kvm)
2179 struct kvm_userspace_memory_region kvm_userspace_mem;
2182 down_write(&kvm->slots_lock);
2183 if (kvm->arch.apic_access_page)
2185 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2186 kvm_userspace_mem.flags = 0;
2187 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2188 kvm_userspace_mem.memory_size = PAGE_SIZE;
2189 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2193 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2195 up_write(&kvm->slots_lock);
2199 static int alloc_identity_pagetable(struct kvm *kvm)
2201 struct kvm_userspace_memory_region kvm_userspace_mem;
2204 down_write(&kvm->slots_lock);
2205 if (kvm->arch.ept_identity_pagetable)
2207 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2208 kvm_userspace_mem.flags = 0;
2209 kvm_userspace_mem.guest_phys_addr =
2210 kvm->arch.ept_identity_map_addr;
2211 kvm_userspace_mem.memory_size = PAGE_SIZE;
2212 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2216 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2217 kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
2219 up_write(&kvm->slots_lock);
2223 static void allocate_vpid(struct vcpu_vmx *vmx)
2230 spin_lock(&vmx_vpid_lock);
2231 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2232 if (vpid < VMX_NR_VPIDS) {
2234 __set_bit(vpid, vmx_vpid_bitmap);
2236 spin_unlock(&vmx_vpid_lock);
2239 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
2241 int f = sizeof(unsigned long);
2243 if (!cpu_has_vmx_msr_bitmap())
2247 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2248 * have the write-low and read-high bitmap offsets the wrong way round.
2249 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2251 if (msr <= 0x1fff) {
2252 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2253 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
2254 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2256 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2257 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
2261 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2264 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2265 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2269 * Sets up the vmcs for emulated real mode.
2271 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2273 u32 host_sysenter_cs, msr_low, msr_high;
2275 u64 host_pat, tsc_this, tsc_base;
2277 struct descriptor_table dt;
2279 unsigned long kvm_vmx_return;
2283 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2284 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
2286 if (cpu_has_vmx_msr_bitmap())
2287 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
2289 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2292 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2293 vmcs_config.pin_based_exec_ctrl);
2295 exec_control = vmcs_config.cpu_based_exec_ctrl;
2296 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2297 exec_control &= ~CPU_BASED_TPR_SHADOW;
2298 #ifdef CONFIG_X86_64
2299 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2300 CPU_BASED_CR8_LOAD_EXITING;
2304 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2305 CPU_BASED_CR3_LOAD_EXITING |
2306 CPU_BASED_INVLPG_EXITING;
2307 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2309 if (cpu_has_secondary_exec_ctrls()) {
2310 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2311 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2313 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2315 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2317 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2318 if (!enable_unrestricted_guest)
2319 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2321 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
2322 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2326 vmcs_write32(PLE_GAP, ple_gap);
2327 vmcs_write32(PLE_WINDOW, ple_window);
2330 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2331 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2332 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
2334 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
2335 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
2336 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
2338 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
2339 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2340 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2341 vmcs_write16(HOST_FS_SELECTOR, kvm_read_fs()); /* 22.2.4 */
2342 vmcs_write16(HOST_GS_SELECTOR, kvm_read_gs()); /* 22.2.4 */
2343 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2344 #ifdef CONFIG_X86_64
2345 rdmsrl(MSR_FS_BASE, a);
2346 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2347 rdmsrl(MSR_GS_BASE, a);
2348 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2350 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2351 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2354 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
2357 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
2359 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2360 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2361 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2362 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2363 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2365 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2366 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2367 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2368 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
2369 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2370 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
2372 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2373 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2374 host_pat = msr_low | ((u64) msr_high << 32);
2375 vmcs_write64(HOST_IA32_PAT, host_pat);
2377 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2378 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2379 host_pat = msr_low | ((u64) msr_high << 32);
2380 /* Write the default value follow host pat */
2381 vmcs_write64(GUEST_IA32_PAT, host_pat);
2382 /* Keep arch.pat sync with GUEST_IA32_PAT */
2383 vmx->vcpu.arch.pat = host_pat;
2386 for (i = 0; i < NR_VMX_MSR; ++i) {
2387 u32 index = vmx_msr_index[i];
2388 u32 data_low, data_high;
2392 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2394 if (wrmsr_safe(index, data_low, data_high) < 0)
2396 data = data_low | ((u64)data_high << 32);
2397 vmx->guest_msrs[j].index = i;
2398 vmx->guest_msrs[j].data = 0;
2402 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2404 /* 22.2.1, 20.8.1 */
2405 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2407 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2408 vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
2410 tsc_base = vmx->vcpu.kvm->arch.vm_init_tsc;
2412 if (tsc_this < vmx->vcpu.kvm->arch.vm_init_tsc)
2413 tsc_base = tsc_this;
2415 guest_write_tsc(0, tsc_base);
2420 static int init_rmode(struct kvm *kvm)
2422 if (!init_rmode_tss(kvm))
2424 if (!init_rmode_identity_map(kvm))
2429 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2431 struct vcpu_vmx *vmx = to_vmx(vcpu);
2435 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2436 down_read(&vcpu->kvm->slots_lock);
2437 if (!init_rmode(vmx->vcpu.kvm)) {
2442 vmx->rmode.vm86_active = 0;
2444 vmx->soft_vnmi_blocked = 0;
2446 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2447 kvm_set_cr8(&vmx->vcpu, 0);
2448 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2449 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2450 msr |= MSR_IA32_APICBASE_BSP;
2451 kvm_set_apic_base(&vmx->vcpu, msr);
2453 fx_init(&vmx->vcpu);
2455 seg_setup(VCPU_SREG_CS);
2457 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2458 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
2460 if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
2461 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2462 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2464 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2465 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2468 seg_setup(VCPU_SREG_DS);
2469 seg_setup(VCPU_SREG_ES);
2470 seg_setup(VCPU_SREG_FS);
2471 seg_setup(VCPU_SREG_GS);
2472 seg_setup(VCPU_SREG_SS);
2474 vmcs_write16(GUEST_TR_SELECTOR, 0);
2475 vmcs_writel(GUEST_TR_BASE, 0);
2476 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2477 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2479 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2480 vmcs_writel(GUEST_LDTR_BASE, 0);
2481 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2482 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2484 vmcs_write32(GUEST_SYSENTER_CS, 0);
2485 vmcs_writel(GUEST_SYSENTER_ESP, 0);
2486 vmcs_writel(GUEST_SYSENTER_EIP, 0);
2488 vmcs_writel(GUEST_RFLAGS, 0x02);
2489 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2490 kvm_rip_write(vcpu, 0xfff0);
2492 kvm_rip_write(vcpu, 0);
2493 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2495 vmcs_writel(GUEST_DR7, 0x400);
2497 vmcs_writel(GUEST_GDTR_BASE, 0);
2498 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2500 vmcs_writel(GUEST_IDTR_BASE, 0);
2501 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2503 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2504 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2505 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2507 /* Special registers */
2508 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2512 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
2514 if (cpu_has_vmx_tpr_shadow()) {
2515 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2516 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2517 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2518 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2519 vmcs_write32(TPR_THRESHOLD, 0);
2522 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2523 vmcs_write64(APIC_ACCESS_ADDR,
2524 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2527 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2529 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
2530 vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
2531 vmx_set_cr4(&vmx->vcpu, 0);
2532 vmx_set_efer(&vmx->vcpu, 0);
2533 vmx_fpu_activate(&vmx->vcpu);
2534 update_exception_bitmap(&vmx->vcpu);
2536 vpid_sync_vcpu_all(vmx);
2540 /* HACK: Don't enable emulation on guest boot/reset */
2541 vmx->emulation_required = 0;
2544 up_read(&vcpu->kvm->slots_lock);
2548 static void enable_irq_window(struct kvm_vcpu *vcpu)
2550 u32 cpu_based_vm_exec_control;
2552 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2553 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2554 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2557 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2559 u32 cpu_based_vm_exec_control;
2561 if (!cpu_has_virtual_nmis()) {
2562 enable_irq_window(vcpu);
2566 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2567 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2568 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2571 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
2573 struct vcpu_vmx *vmx = to_vmx(vcpu);
2575 int irq = vcpu->arch.interrupt.nr;
2577 trace_kvm_inj_virq(irq);
2579 ++vcpu->stat.irq_injections;
2580 if (vmx->rmode.vm86_active) {
2581 vmx->rmode.irq.pending = true;
2582 vmx->rmode.irq.vector = irq;
2583 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2584 if (vcpu->arch.interrupt.soft)
2585 vmx->rmode.irq.rip +=
2586 vmx->vcpu.arch.event_exit_inst_len;
2587 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2588 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2589 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2590 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2593 intr = irq | INTR_INFO_VALID_MASK;
2594 if (vcpu->arch.interrupt.soft) {
2595 intr |= INTR_TYPE_SOFT_INTR;
2596 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2597 vmx->vcpu.arch.event_exit_inst_len);
2599 intr |= INTR_TYPE_EXT_INTR;
2600 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
2603 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2605 struct vcpu_vmx *vmx = to_vmx(vcpu);
2607 if (!cpu_has_virtual_nmis()) {
2609 * Tracking the NMI-blocked state in software is built upon
2610 * finding the next open IRQ window. This, in turn, depends on
2611 * well-behaving guests: They have to keep IRQs disabled at
2612 * least as long as the NMI handler runs. Otherwise we may
2613 * cause NMI nesting, maybe breaking the guest. But as this is
2614 * highly unlikely, we can live with the residual risk.
2616 vmx->soft_vnmi_blocked = 1;
2617 vmx->vnmi_blocked_time = 0;
2620 ++vcpu->stat.nmi_injections;
2621 if (vmx->rmode.vm86_active) {
2622 vmx->rmode.irq.pending = true;
2623 vmx->rmode.irq.vector = NMI_VECTOR;
2624 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2625 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2626 NMI_VECTOR | INTR_TYPE_SOFT_INTR |
2627 INTR_INFO_VALID_MASK);
2628 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2629 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2632 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2633 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2636 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
2638 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2641 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2642 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS |
2643 GUEST_INTR_STATE_NMI));
2646 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2648 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2649 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2650 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
2653 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2656 struct kvm_userspace_memory_region tss_mem = {
2657 .slot = TSS_PRIVATE_MEMSLOT,
2658 .guest_phys_addr = addr,
2659 .memory_size = PAGE_SIZE * 3,
2663 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2666 kvm->arch.tss_addr = addr;
2670 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2671 int vec, u32 err_code)
2674 * Instruction with address size override prefix opcode 0x67
2675 * Cause the #SS fault with 0 error code in VM86 mode.
2677 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
2678 if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE)
2681 * Forward all other exceptions that are valid in real mode.
2682 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2683 * the required debugging infrastructure rework.
2687 if (vcpu->guest_debug &
2688 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
2690 kvm_queue_exception(vcpu, vec);
2693 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2704 kvm_queue_exception(vcpu, vec);
2711 * Trigger machine check on the host. We assume all the MSRs are already set up
2712 * by the CPU and that we still run on the same CPU as the MCE occurred on.
2713 * We pass a fake environment to the machine check handler because we want
2714 * the guest to be always treated like user space, no matter what context
2715 * it used internally.
2717 static void kvm_machine_check(void)
2719 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
2720 struct pt_regs regs = {
2721 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
2722 .flags = X86_EFLAGS_IF,
2725 do_machine_check(®s, 0);
2729 static int handle_machine_check(struct kvm_vcpu *vcpu)
2731 /* already handled by vcpu_run */
2735 static int handle_exception(struct kvm_vcpu *vcpu)
2737 struct vcpu_vmx *vmx = to_vmx(vcpu);
2738 struct kvm_run *kvm_run = vcpu->run;
2739 u32 intr_info, ex_no, error_code;
2740 unsigned long cr2, rip, dr6;
2742 enum emulation_result er;
2744 vect_info = vmx->idt_vectoring_info;
2745 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2747 if (is_machine_check(intr_info))
2748 return handle_machine_check(vcpu);
2750 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
2751 !is_page_fault(intr_info))
2752 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
2753 "intr info 0x%x\n", __func__, vect_info, intr_info);
2755 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
2756 return 1; /* already handled by vmx_vcpu_run() */
2758 if (is_no_device(intr_info)) {
2759 vmx_fpu_activate(vcpu);
2763 if (is_invalid_opcode(intr_info)) {
2764 er = emulate_instruction(vcpu, 0, 0, EMULTYPE_TRAP_UD);
2765 if (er != EMULATE_DONE)
2766 kvm_queue_exception(vcpu, UD_VECTOR);
2771 rip = kvm_rip_read(vcpu);
2772 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
2773 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
2774 if (is_page_fault(intr_info)) {
2775 /* EPT won't cause page fault directly */
2778 cr2 = vmcs_readl(EXIT_QUALIFICATION);
2779 trace_kvm_page_fault(cr2, error_code);
2781 if (kvm_event_needs_reinjection(vcpu))
2782 kvm_mmu_unprotect_page_virt(vcpu, cr2);
2783 return kvm_mmu_page_fault(vcpu, cr2, error_code);
2786 if (vmx->rmode.vm86_active &&
2787 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
2789 if (vcpu->arch.halt_request) {
2790 vcpu->arch.halt_request = 0;
2791 return kvm_emulate_halt(vcpu);
2796 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
2799 dr6 = vmcs_readl(EXIT_QUALIFICATION);
2800 if (!(vcpu->guest_debug &
2801 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
2802 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
2803 kvm_queue_exception(vcpu, DB_VECTOR);
2806 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
2807 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
2810 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2811 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
2812 kvm_run->debug.arch.exception = ex_no;
2815 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
2816 kvm_run->ex.exception = ex_no;
2817 kvm_run->ex.error_code = error_code;
2823 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
2825 ++vcpu->stat.irq_exits;
2829 static int handle_triple_fault(struct kvm_vcpu *vcpu)
2831 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
2835 static int handle_io(struct kvm_vcpu *vcpu)
2837 unsigned long exit_qualification;
2838 int size, in, string;
2841 ++vcpu->stat.io_exits;
2842 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2843 string = (exit_qualification & 16) != 0;
2846 if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DO_MMIO)
2851 size = (exit_qualification & 7) + 1;
2852 in = (exit_qualification & 8) != 0;
2853 port = exit_qualification >> 16;
2855 skip_emulated_instruction(vcpu);
2856 return kvm_emulate_pio(vcpu, in, size, port);
2860 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2863 * Patch in the VMCALL instruction:
2865 hypercall[0] = 0x0f;
2866 hypercall[1] = 0x01;
2867 hypercall[2] = 0xc1;
2870 static int handle_cr(struct kvm_vcpu *vcpu)
2872 unsigned long exit_qualification, val;
2876 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2877 cr = exit_qualification & 15;
2878 reg = (exit_qualification >> 8) & 15;
2879 switch ((exit_qualification >> 4) & 3) {
2880 case 0: /* mov to cr */
2881 val = kvm_register_read(vcpu, reg);
2882 trace_kvm_cr_write(cr, val);
2885 kvm_set_cr0(vcpu, val);
2886 skip_emulated_instruction(vcpu);
2889 kvm_set_cr3(vcpu, val);
2890 skip_emulated_instruction(vcpu);
2893 kvm_set_cr4(vcpu, val);
2894 skip_emulated_instruction(vcpu);
2897 u8 cr8_prev = kvm_get_cr8(vcpu);
2898 u8 cr8 = kvm_register_read(vcpu, reg);
2899 kvm_set_cr8(vcpu, cr8);
2900 skip_emulated_instruction(vcpu);
2901 if (irqchip_in_kernel(vcpu->kvm))
2903 if (cr8_prev <= cr8)
2905 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
2911 vmx_fpu_deactivate(vcpu);
2912 vcpu->arch.cr0 &= ~X86_CR0_TS;
2913 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2914 vmx_fpu_activate(vcpu);
2915 skip_emulated_instruction(vcpu);
2917 case 1: /*mov from cr*/
2920 kvm_register_write(vcpu, reg, vcpu->arch.cr3);
2921 trace_kvm_cr_read(cr, vcpu->arch.cr3);
2922 skip_emulated_instruction(vcpu);
2925 val = kvm_get_cr8(vcpu);
2926 kvm_register_write(vcpu, reg, val);
2927 trace_kvm_cr_read(cr, val);
2928 skip_emulated_instruction(vcpu);
2933 kvm_lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
2935 skip_emulated_instruction(vcpu);
2940 vcpu->run->exit_reason = 0;
2941 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
2942 (int)(exit_qualification >> 4) & 3, cr);
2946 static int handle_dr(struct kvm_vcpu *vcpu)
2948 unsigned long exit_qualification;
2952 if (!kvm_require_cpl(vcpu, 0))
2954 dr = vmcs_readl(GUEST_DR7);
2957 * As the vm-exit takes precedence over the debug trap, we
2958 * need to emulate the latter, either for the host or the
2959 * guest debugging itself.
2961 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
2962 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
2963 vcpu->run->debug.arch.dr7 = dr;
2964 vcpu->run->debug.arch.pc =
2965 vmcs_readl(GUEST_CS_BASE) +
2966 vmcs_readl(GUEST_RIP);
2967 vcpu->run->debug.arch.exception = DB_VECTOR;
2968 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
2971 vcpu->arch.dr7 &= ~DR7_GD;
2972 vcpu->arch.dr6 |= DR6_BD;
2973 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2974 kvm_queue_exception(vcpu, DB_VECTOR);
2979 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2980 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
2981 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
2982 if (exit_qualification & TYPE_MOV_FROM_DR) {
2985 val = vcpu->arch.db[dr];
2988 val = vcpu->arch.dr6;
2991 val = vcpu->arch.dr7;
2996 kvm_register_write(vcpu, reg, val);
2998 val = vcpu->arch.regs[reg];
3001 vcpu->arch.db[dr] = val;
3002 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
3003 vcpu->arch.eff_db[dr] = val;
3006 if (vcpu->arch.cr4 & X86_CR4_DE)
3007 kvm_queue_exception(vcpu, UD_VECTOR);
3010 if (val & 0xffffffff00000000ULL) {
3011 kvm_queue_exception(vcpu, GP_VECTOR);
3014 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
3017 if (val & 0xffffffff00000000ULL) {
3018 kvm_queue_exception(vcpu, GP_VECTOR);
3021 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
3022 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
3023 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3024 vcpu->arch.switch_db_regs =
3025 (val & DR7_BP_EN_MASK);
3030 skip_emulated_instruction(vcpu);
3034 static int handle_cpuid(struct kvm_vcpu *vcpu)
3036 kvm_emulate_cpuid(vcpu);
3040 static int handle_rdmsr(struct kvm_vcpu *vcpu)
3042 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3045 if (vmx_get_msr(vcpu, ecx, &data)) {
3046 kvm_inject_gp(vcpu, 0);
3050 trace_kvm_msr_read(ecx, data);
3052 /* FIXME: handling of bits 32:63 of rax, rdx */
3053 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
3054 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
3055 skip_emulated_instruction(vcpu);
3059 static int handle_wrmsr(struct kvm_vcpu *vcpu)
3061 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3062 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
3063 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3065 trace_kvm_msr_write(ecx, data);
3067 if (vmx_set_msr(vcpu, ecx, data) != 0) {
3068 kvm_inject_gp(vcpu, 0);
3072 skip_emulated_instruction(vcpu);
3076 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
3081 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
3083 u32 cpu_based_vm_exec_control;
3085 /* clear pending irq */
3086 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3087 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3088 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3090 ++vcpu->stat.irq_window_exits;
3093 * If the user space waits to inject interrupts, exit as soon as
3096 if (!irqchip_in_kernel(vcpu->kvm) &&
3097 vcpu->run->request_interrupt_window &&
3098 !kvm_cpu_has_interrupt(vcpu)) {
3099 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3105 static int handle_halt(struct kvm_vcpu *vcpu)
3107 skip_emulated_instruction(vcpu);
3108 return kvm_emulate_halt(vcpu);
3111 static int handle_vmcall(struct kvm_vcpu *vcpu)
3113 skip_emulated_instruction(vcpu);
3114 kvm_emulate_hypercall(vcpu);
3118 static int handle_vmx_insn(struct kvm_vcpu *vcpu)
3120 kvm_queue_exception(vcpu, UD_VECTOR);
3124 static int handle_invlpg(struct kvm_vcpu *vcpu)
3126 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3128 kvm_mmu_invlpg(vcpu, exit_qualification);
3129 skip_emulated_instruction(vcpu);
3133 static int handle_wbinvd(struct kvm_vcpu *vcpu)
3135 skip_emulated_instruction(vcpu);
3136 /* TODO: Add support for VT-d/pass-through device */
3140 static int handle_apic_access(struct kvm_vcpu *vcpu)
3142 unsigned long exit_qualification;
3143 enum emulation_result er;
3144 unsigned long offset;
3146 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3147 offset = exit_qualification & 0xffful;
3149 er = emulate_instruction(vcpu, 0, 0, 0);
3151 if (er != EMULATE_DONE) {
3153 "Fail to handle apic access vmexit! Offset is 0x%lx\n",
3160 static int handle_task_switch(struct kvm_vcpu *vcpu)
3162 struct vcpu_vmx *vmx = to_vmx(vcpu);
3163 unsigned long exit_qualification;
3165 int reason, type, idt_v;
3167 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3168 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
3170 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3172 reason = (u32)exit_qualification >> 30;
3173 if (reason == TASK_SWITCH_GATE && idt_v) {
3175 case INTR_TYPE_NMI_INTR:
3176 vcpu->arch.nmi_injected = false;
3177 if (cpu_has_virtual_nmis())
3178 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3179 GUEST_INTR_STATE_NMI);
3181 case INTR_TYPE_EXT_INTR:
3182 case INTR_TYPE_SOFT_INTR:
3183 kvm_clear_interrupt_queue(vcpu);
3185 case INTR_TYPE_HARD_EXCEPTION:
3186 case INTR_TYPE_SOFT_EXCEPTION:
3187 kvm_clear_exception_queue(vcpu);
3193 tss_selector = exit_qualification;
3195 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3196 type != INTR_TYPE_EXT_INTR &&
3197 type != INTR_TYPE_NMI_INTR))
3198 skip_emulated_instruction(vcpu);
3200 if (!kvm_task_switch(vcpu, tss_selector, reason))
3203 /* clear all local breakpoint enable flags */
3204 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3207 * TODO: What about debug traps on tss switch?
3208 * Are we supposed to inject them and update dr6?
3214 static int handle_ept_violation(struct kvm_vcpu *vcpu)
3216 unsigned long exit_qualification;
3220 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3222 if (exit_qualification & (1 << 6)) {
3223 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3227 gla_validity = (exit_qualification >> 7) & 0x3;
3228 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3229 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3230 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3231 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3232 vmcs_readl(GUEST_LINEAR_ADDRESS));
3233 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3234 (long unsigned int)exit_qualification);
3235 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3236 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
3240 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3241 trace_kvm_page_fault(gpa, exit_qualification);
3242 return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
3245 static u64 ept_rsvd_mask(u64 spte, int level)
3250 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
3251 mask |= (1ULL << i);
3254 /* bits 7:3 reserved */
3256 else if (level == 2) {
3257 if (spte & (1ULL << 7))
3258 /* 2MB ref, bits 20:12 reserved */
3261 /* bits 6:3 reserved */
3268 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
3271 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
3273 /* 010b (write-only) */
3274 WARN_ON((spte & 0x7) == 0x2);
3276 /* 110b (write/execute) */
3277 WARN_ON((spte & 0x7) == 0x6);
3279 /* 100b (execute-only) and value not supported by logical processor */
3280 if (!cpu_has_vmx_ept_execute_only())
3281 WARN_ON((spte & 0x7) == 0x4);
3285 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
3287 if (rsvd_bits != 0) {
3288 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
3289 __func__, rsvd_bits);
3293 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
3294 u64 ept_mem_type = (spte & 0x38) >> 3;
3296 if (ept_mem_type == 2 || ept_mem_type == 3 ||
3297 ept_mem_type == 7) {
3298 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
3299 __func__, ept_mem_type);
3306 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
3312 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3314 printk(KERN_ERR "EPT: Misconfiguration.\n");
3315 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
3317 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
3319 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
3320 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
3322 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3323 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
3328 static int handle_nmi_window(struct kvm_vcpu *vcpu)
3330 u32 cpu_based_vm_exec_control;
3332 /* clear pending NMI */
3333 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3334 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3335 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3336 ++vcpu->stat.nmi_window_exits;
3341 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
3343 struct vcpu_vmx *vmx = to_vmx(vcpu);
3344 enum emulation_result err = EMULATE_DONE;
3347 while (!guest_state_valid(vcpu)) {
3348 err = emulate_instruction(vcpu, 0, 0, 0);
3350 if (err == EMULATE_DO_MMIO) {
3355 if (err != EMULATE_DONE) {
3356 kvm_report_emulation_failure(vcpu, "emulation failure");
3357 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3358 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3363 if (signal_pending(current))
3369 vmx->emulation_required = 0;
3375 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
3376 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
3378 static int handle_pause(struct kvm_vcpu *vcpu)
3380 skip_emulated_instruction(vcpu);
3381 kvm_vcpu_on_spin(vcpu);
3387 * The exit handlers return 1 if the exit was handled fully and guest execution
3388 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
3389 * to be done to userspace and return 0.
3391 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
3392 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
3393 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
3394 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
3395 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
3396 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
3397 [EXIT_REASON_CR_ACCESS] = handle_cr,
3398 [EXIT_REASON_DR_ACCESS] = handle_dr,
3399 [EXIT_REASON_CPUID] = handle_cpuid,
3400 [EXIT_REASON_MSR_READ] = handle_rdmsr,
3401 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
3402 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
3403 [EXIT_REASON_HLT] = handle_halt,
3404 [EXIT_REASON_INVLPG] = handle_invlpg,
3405 [EXIT_REASON_VMCALL] = handle_vmcall,
3406 [EXIT_REASON_VMCLEAR] = handle_vmx_insn,
3407 [EXIT_REASON_VMLAUNCH] = handle_vmx_insn,
3408 [EXIT_REASON_VMPTRLD] = handle_vmx_insn,
3409 [EXIT_REASON_VMPTRST] = handle_vmx_insn,
3410 [EXIT_REASON_VMREAD] = handle_vmx_insn,
3411 [EXIT_REASON_VMRESUME] = handle_vmx_insn,
3412 [EXIT_REASON_VMWRITE] = handle_vmx_insn,
3413 [EXIT_REASON_VMOFF] = handle_vmx_insn,
3414 [EXIT_REASON_VMON] = handle_vmx_insn,
3415 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
3416 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
3417 [EXIT_REASON_WBINVD] = handle_wbinvd,
3418 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
3419 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
3420 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
3421 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
3422 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
3425 static const int kvm_vmx_max_exit_handlers =
3426 ARRAY_SIZE(kvm_vmx_exit_handlers);
3429 * The guest has exited. See if we can fix it or if we need userspace
3432 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
3434 struct vcpu_vmx *vmx = to_vmx(vcpu);
3435 u32 exit_reason = vmx->exit_reason;
3436 u32 vectoring_info = vmx->idt_vectoring_info;
3438 trace_kvm_exit(exit_reason, kvm_rip_read(vcpu));
3440 /* If guest state is invalid, start emulating */
3441 if (vmx->emulation_required && emulate_invalid_guest_state)
3442 return handle_invalid_guest_state(vcpu);
3444 /* Access CR3 don't cause VMExit in paging mode, so we need
3445 * to sync with guest real CR3. */
3446 if (enable_ept && is_paging(vcpu))
3447 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3449 if (unlikely(vmx->fail)) {
3450 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3451 vcpu->run->fail_entry.hardware_entry_failure_reason
3452 = vmcs_read32(VM_INSTRUCTION_ERROR);
3456 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3457 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3458 exit_reason != EXIT_REASON_EPT_VIOLATION &&
3459 exit_reason != EXIT_REASON_TASK_SWITCH))
3460 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3461 "(0x%x) and exit reason is 0x%x\n",
3462 __func__, vectoring_info, exit_reason);
3464 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3465 if (vmx_interrupt_allowed(vcpu)) {
3466 vmx->soft_vnmi_blocked = 0;
3467 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3468 vcpu->arch.nmi_pending) {
3470 * This CPU don't support us in finding the end of an
3471 * NMI-blocked window if the guest runs with IRQs
3472 * disabled. So we pull the trigger after 1 s of
3473 * futile waiting, but inform the user about this.
3475 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3476 "state on VCPU %d after 1 s timeout\n",
3477 __func__, vcpu->vcpu_id);
3478 vmx->soft_vnmi_blocked = 0;
3482 if (exit_reason < kvm_vmx_max_exit_handlers
3483 && kvm_vmx_exit_handlers[exit_reason])
3484 return kvm_vmx_exit_handlers[exit_reason](vcpu);
3486 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3487 vcpu->run->hw.hardware_exit_reason = exit_reason;
3492 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3494 if (irr == -1 || tpr < irr) {
3495 vmcs_write32(TPR_THRESHOLD, 0);
3499 vmcs_write32(TPR_THRESHOLD, irr);
3502 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3505 u32 idt_vectoring_info = vmx->idt_vectoring_info;
3509 bool idtv_info_valid;
3511 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3513 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
3515 /* Handle machine checks before interrupts are enabled */
3516 if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3517 || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3518 && is_machine_check(exit_intr_info)))
3519 kvm_machine_check();
3521 /* We need to handle NMIs before interrupts are enabled */
3522 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3523 (exit_intr_info & INTR_INFO_VALID_MASK))
3526 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3528 if (cpu_has_virtual_nmis()) {
3529 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3530 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3532 * SDM 3: 27.7.1.2 (September 2008)
3533 * Re-set bit "block by NMI" before VM entry if vmexit caused by
3534 * a guest IRET fault.
3535 * SDM 3: 23.2.2 (September 2008)
3536 * Bit 12 is undefined in any of the following cases:
3537 * If the VM exit sets the valid bit in the IDT-vectoring
3538 * information field.
3539 * If the VM exit is due to a double fault.
3541 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3542 vector != DF_VECTOR && !idtv_info_valid)
3543 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3544 GUEST_INTR_STATE_NMI);
3545 } else if (unlikely(vmx->soft_vnmi_blocked))
3546 vmx->vnmi_blocked_time +=
3547 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3549 vmx->vcpu.arch.nmi_injected = false;
3550 kvm_clear_exception_queue(&vmx->vcpu);
3551 kvm_clear_interrupt_queue(&vmx->vcpu);
3553 if (!idtv_info_valid)
3556 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3557 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3560 case INTR_TYPE_NMI_INTR:
3561 vmx->vcpu.arch.nmi_injected = true;
3563 * SDM 3: 27.7.1.2 (September 2008)
3564 * Clear bit "block by NMI" before VM entry if a NMI
3567 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3568 GUEST_INTR_STATE_NMI);
3570 case INTR_TYPE_SOFT_EXCEPTION:
3571 vmx->vcpu.arch.event_exit_inst_len =
3572 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3574 case INTR_TYPE_HARD_EXCEPTION:
3575 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3576 u32 err = vmcs_read32(IDT_VECTORING_ERROR_CODE);
3577 kvm_queue_exception_e(&vmx->vcpu, vector, err);
3579 kvm_queue_exception(&vmx->vcpu, vector);
3581 case INTR_TYPE_SOFT_INTR:
3582 vmx->vcpu.arch.event_exit_inst_len =
3583 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3585 case INTR_TYPE_EXT_INTR:
3586 kvm_queue_interrupt(&vmx->vcpu, vector,
3587 type == INTR_TYPE_SOFT_INTR);
3595 * Failure to inject an interrupt should give us the information
3596 * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs
3597 * when fetching the interrupt redirection bitmap in the real-mode
3598 * tss, this doesn't happen. So we do it ourselves.
3600 static void fixup_rmode_irq(struct vcpu_vmx *vmx)
3602 vmx->rmode.irq.pending = 0;
3603 if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip)
3605 kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip);
3606 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
3607 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
3608 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
3611 vmx->idt_vectoring_info =
3612 VECTORING_INFO_VALID_MASK
3613 | INTR_TYPE_EXT_INTR
3614 | vmx->rmode.irq.vector;
3617 #ifdef CONFIG_X86_64
3625 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
3627 struct vcpu_vmx *vmx = to_vmx(vcpu);
3629 /* Record the guest's net vcpu time for enforced NMI injections. */
3630 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3631 vmx->entry_time = ktime_get();
3633 /* Don't enter VMX if guest state is invalid, let the exit handler
3634 start emulation until we arrive back to a valid state */
3635 if (vmx->emulation_required && emulate_invalid_guest_state)
3638 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3639 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3640 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3641 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3643 /* When single-stepping over STI and MOV SS, we must clear the
3644 * corresponding interruptibility bits in the guest state. Otherwise
3645 * vmentry fails as it then expects bit 14 (BS) in pending debug
3646 * exceptions being set, but that's not correct for the guest debugging
3648 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3649 vmx_set_interrupt_shadow(vcpu, 0);
3652 * Loading guest fpu may have cleared host cr0.ts
3654 vmcs_writel(HOST_CR0, read_cr0());
3656 if (vcpu->arch.switch_db_regs)
3657 set_debugreg(vcpu->arch.dr6, 6);
3660 /* Store host registers */
3661 "push %%"R"dx; push %%"R"bp;"
3663 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3665 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
3666 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
3668 /* Reload cr2 if changed */
3669 "mov %c[cr2](%0), %%"R"ax \n\t"
3670 "mov %%cr2, %%"R"dx \n\t"
3671 "cmp %%"R"ax, %%"R"dx \n\t"
3673 "mov %%"R"ax, %%cr2 \n\t"
3675 /* Check if vmlaunch of vmresume is needed */
3676 "cmpl $0, %c[launched](%0) \n\t"
3677 /* Load guest registers. Don't clobber flags. */
3678 "mov %c[rax](%0), %%"R"ax \n\t"
3679 "mov %c[rbx](%0), %%"R"bx \n\t"
3680 "mov %c[rdx](%0), %%"R"dx \n\t"
3681 "mov %c[rsi](%0), %%"R"si \n\t"
3682 "mov %c[rdi](%0), %%"R"di \n\t"
3683 "mov %c[rbp](%0), %%"R"bp \n\t"
3684 #ifdef CONFIG_X86_64
3685 "mov %c[r8](%0), %%r8 \n\t"
3686 "mov %c[r9](%0), %%r9 \n\t"
3687 "mov %c[r10](%0), %%r10 \n\t"
3688 "mov %c[r11](%0), %%r11 \n\t"
3689 "mov %c[r12](%0), %%r12 \n\t"
3690 "mov %c[r13](%0), %%r13 \n\t"
3691 "mov %c[r14](%0), %%r14 \n\t"
3692 "mov %c[r15](%0), %%r15 \n\t"
3694 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3696 /* Enter guest mode */
3697 "jne .Llaunched \n\t"
3698 __ex(ASM_VMX_VMLAUNCH) "\n\t"
3699 "jmp .Lkvm_vmx_return \n\t"
3700 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
3701 ".Lkvm_vmx_return: "
3702 /* Save guest registers, load host registers, keep flags */
3703 "xchg %0, (%%"R"sp) \n\t"
3704 "mov %%"R"ax, %c[rax](%0) \n\t"
3705 "mov %%"R"bx, %c[rbx](%0) \n\t"
3706 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3707 "mov %%"R"dx, %c[rdx](%0) \n\t"
3708 "mov %%"R"si, %c[rsi](%0) \n\t"
3709 "mov %%"R"di, %c[rdi](%0) \n\t"
3710 "mov %%"R"bp, %c[rbp](%0) \n\t"
3711 #ifdef CONFIG_X86_64
3712 "mov %%r8, %c[r8](%0) \n\t"
3713 "mov %%r9, %c[r9](%0) \n\t"
3714 "mov %%r10, %c[r10](%0) \n\t"
3715 "mov %%r11, %c[r11](%0) \n\t"
3716 "mov %%r12, %c[r12](%0) \n\t"
3717 "mov %%r13, %c[r13](%0) \n\t"
3718 "mov %%r14, %c[r14](%0) \n\t"
3719 "mov %%r15, %c[r15](%0) \n\t"
3721 "mov %%cr2, %%"R"ax \n\t"
3722 "mov %%"R"ax, %c[cr2](%0) \n\t"
3724 "pop %%"R"bp; pop %%"R"bp; pop %%"R"dx \n\t"
3725 "setbe %c[fail](%0) \n\t"
3726 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
3727 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
3728 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
3729 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
3730 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
3731 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
3732 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
3733 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
3734 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
3735 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
3736 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
3737 #ifdef CONFIG_X86_64
3738 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
3739 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
3740 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
3741 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
3742 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
3743 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
3744 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
3745 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
3747 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
3749 , R"bx", R"di", R"si"
3750 #ifdef CONFIG_X86_64
3751 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
3755 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
3756 | (1 << VCPU_EXREG_PDPTR));
3757 vcpu->arch.regs_dirty = 0;
3759 if (vcpu->arch.switch_db_regs)
3760 get_debugreg(vcpu->arch.dr6, 6);
3762 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
3763 if (vmx->rmode.irq.pending)
3764 fixup_rmode_irq(vmx);
3766 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
3769 vmx_complete_interrupts(vmx);
3775 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
3777 struct vcpu_vmx *vmx = to_vmx(vcpu);
3781 free_vmcs(vmx->vmcs);
3786 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
3788 struct vcpu_vmx *vmx = to_vmx(vcpu);
3790 spin_lock(&vmx_vpid_lock);
3792 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3793 spin_unlock(&vmx_vpid_lock);
3794 vmx_free_vmcs(vcpu);
3795 kfree(vmx->guest_msrs);
3796 kvm_vcpu_uninit(vcpu);
3797 kmem_cache_free(kvm_vcpu_cache, vmx);
3800 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
3803 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3807 return ERR_PTR(-ENOMEM);
3811 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
3815 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3816 if (!vmx->guest_msrs) {
3821 vmx->vmcs = alloc_vmcs();
3825 vmcs_clear(vmx->vmcs);
3828 vmx_vcpu_load(&vmx->vcpu, cpu);
3829 err = vmx_vcpu_setup(vmx);
3830 vmx_vcpu_put(&vmx->vcpu);
3834 if (vm_need_virtualize_apic_accesses(kvm))
3835 if (alloc_apic_access_page(kvm) != 0)
3839 if (!kvm->arch.ept_identity_map_addr)
3840 kvm->arch.ept_identity_map_addr =
3841 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3842 if (alloc_identity_pagetable(kvm) != 0)
3849 free_vmcs(vmx->vmcs);
3851 kfree(vmx->guest_msrs);
3853 kvm_vcpu_uninit(&vmx->vcpu);
3855 kmem_cache_free(kvm_vcpu_cache, vmx);
3856 return ERR_PTR(err);
3859 static void __init vmx_check_processor_compat(void *rtn)
3861 struct vmcs_config vmcs_conf;
3864 if (setup_vmcs_config(&vmcs_conf) < 0)
3866 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
3867 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
3868 smp_processor_id());
3873 static int get_ept_level(void)
3875 return VMX_EPT_DEFAULT_GAW + 1;
3878 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
3882 /* For VT-d and EPT combination
3883 * 1. MMIO: always map as UC
3885 * a. VT-d without snooping control feature: can't guarantee the
3886 * result, try to trust guest.
3887 * b. VT-d with snooping control feature: snooping control feature of
3888 * VT-d engine can guarantee the cache correctness. Just set it
3889 * to WB to keep consistent with host. So the same as item 3.
3890 * 3. EPT without VT-d: always map as WB and set IGMT=1 to keep
3891 * consistent with host MTRR
3894 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
3895 else if (vcpu->kvm->arch.iommu_domain &&
3896 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
3897 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
3898 VMX_EPT_MT_EPTE_SHIFT;
3900 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
3906 static const struct trace_print_flags vmx_exit_reasons_str[] = {
3907 { EXIT_REASON_EXCEPTION_NMI, "exception" },
3908 { EXIT_REASON_EXTERNAL_INTERRUPT, "ext_irq" },
3909 { EXIT_REASON_TRIPLE_FAULT, "triple_fault" },
3910 { EXIT_REASON_NMI_WINDOW, "nmi_window" },
3911 { EXIT_REASON_IO_INSTRUCTION, "io_instruction" },
3912 { EXIT_REASON_CR_ACCESS, "cr_access" },
3913 { EXIT_REASON_DR_ACCESS, "dr_access" },
3914 { EXIT_REASON_CPUID, "cpuid" },
3915 { EXIT_REASON_MSR_READ, "rdmsr" },
3916 { EXIT_REASON_MSR_WRITE, "wrmsr" },
3917 { EXIT_REASON_PENDING_INTERRUPT, "interrupt_window" },
3918 { EXIT_REASON_HLT, "halt" },
3919 { EXIT_REASON_INVLPG, "invlpg" },
3920 { EXIT_REASON_VMCALL, "hypercall" },
3921 { EXIT_REASON_TPR_BELOW_THRESHOLD, "tpr_below_thres" },
3922 { EXIT_REASON_APIC_ACCESS, "apic_access" },
3923 { EXIT_REASON_WBINVD, "wbinvd" },
3924 { EXIT_REASON_TASK_SWITCH, "task_switch" },
3925 { EXIT_REASON_EPT_VIOLATION, "ept_violation" },
3929 static bool vmx_gb_page_enable(void)
3934 static struct kvm_x86_ops vmx_x86_ops = {
3935 .cpu_has_kvm_support = cpu_has_kvm_support,
3936 .disabled_by_bios = vmx_disabled_by_bios,
3937 .hardware_setup = hardware_setup,
3938 .hardware_unsetup = hardware_unsetup,
3939 .check_processor_compatibility = vmx_check_processor_compat,
3940 .hardware_enable = hardware_enable,
3941 .hardware_disable = hardware_disable,
3942 .cpu_has_accelerated_tpr = report_flexpriority,
3944 .vcpu_create = vmx_create_vcpu,
3945 .vcpu_free = vmx_free_vcpu,
3946 .vcpu_reset = vmx_vcpu_reset,
3948 .prepare_guest_switch = vmx_save_host_state,
3949 .vcpu_load = vmx_vcpu_load,
3950 .vcpu_put = vmx_vcpu_put,
3952 .set_guest_debug = set_guest_debug,
3953 .get_msr = vmx_get_msr,
3954 .set_msr = vmx_set_msr,
3955 .get_segment_base = vmx_get_segment_base,
3956 .get_segment = vmx_get_segment,
3957 .set_segment = vmx_set_segment,
3958 .get_cpl = vmx_get_cpl,
3959 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
3960 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
3961 .set_cr0 = vmx_set_cr0,
3962 .set_cr3 = vmx_set_cr3,
3963 .set_cr4 = vmx_set_cr4,
3964 .set_efer = vmx_set_efer,
3965 .get_idt = vmx_get_idt,
3966 .set_idt = vmx_set_idt,
3967 .get_gdt = vmx_get_gdt,
3968 .set_gdt = vmx_set_gdt,
3969 .cache_reg = vmx_cache_reg,
3970 .get_rflags = vmx_get_rflags,
3971 .set_rflags = vmx_set_rflags,
3973 .tlb_flush = vmx_flush_tlb,
3975 .run = vmx_vcpu_run,
3976 .handle_exit = vmx_handle_exit,
3977 .skip_emulated_instruction = skip_emulated_instruction,
3978 .set_interrupt_shadow = vmx_set_interrupt_shadow,
3979 .get_interrupt_shadow = vmx_get_interrupt_shadow,
3980 .patch_hypercall = vmx_patch_hypercall,
3981 .set_irq = vmx_inject_irq,
3982 .set_nmi = vmx_inject_nmi,
3983 .queue_exception = vmx_queue_exception,
3984 .interrupt_allowed = vmx_interrupt_allowed,
3985 .nmi_allowed = vmx_nmi_allowed,
3986 .enable_nmi_window = enable_nmi_window,
3987 .enable_irq_window = enable_irq_window,
3988 .update_cr8_intercept = update_cr8_intercept,
3990 .set_tss_addr = vmx_set_tss_addr,
3991 .get_tdp_level = get_ept_level,
3992 .get_mt_mask = vmx_get_mt_mask,
3994 .exit_reasons_str = vmx_exit_reasons_str,
3995 .gb_page_enable = vmx_gb_page_enable,
3998 static int __init vmx_init(void)
4002 rdmsrl_safe(MSR_EFER, &host_efer);
4004 for (i = 0; i < NR_VMX_MSR; ++i)
4005 kvm_define_shared_msr(i, vmx_msr_index[i]);
4007 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
4008 if (!vmx_io_bitmap_a)
4011 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
4012 if (!vmx_io_bitmap_b) {
4017 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
4018 if (!vmx_msr_bitmap_legacy) {
4023 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
4024 if (!vmx_msr_bitmap_longmode) {
4030 * Allow direct access to the PC debug port (it is often used for I/O
4031 * delays, but the vmexits simply slow things down).
4033 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
4034 clear_bit(0x80, vmx_io_bitmap_a);
4036 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
4038 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
4039 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
4041 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
4043 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
4047 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
4048 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
4049 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
4050 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
4051 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
4052 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
4055 bypass_guest_pf = 0;
4056 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
4057 VMX_EPT_WRITABLE_MASK);
4058 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4059 VMX_EPT_EXECUTABLE_MASK);
4064 if (bypass_guest_pf)
4065 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
4070 free_page((unsigned long)vmx_msr_bitmap_longmode);
4072 free_page((unsigned long)vmx_msr_bitmap_legacy);
4074 free_page((unsigned long)vmx_io_bitmap_b);
4076 free_page((unsigned long)vmx_io_bitmap_a);
4080 static void __exit vmx_exit(void)
4082 free_page((unsigned long)vmx_msr_bitmap_legacy);
4083 free_page((unsigned long)vmx_msr_bitmap_longmode);
4084 free_page((unsigned long)vmx_io_bitmap_b);
4085 free_page((unsigned long)vmx_io_bitmap_a);
4090 module_init(vmx_init)
4091 module_exit(vmx_exit)