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 <linux/slab.h>
30 #include "kvm_cache_regs.h"
36 #include <asm/virtext.h>
41 #define __ex(x) __kvm_handle_fault_on_reboot(x)
43 MODULE_AUTHOR("Qumranet");
44 MODULE_LICENSE("GPL");
46 static int __read_mostly bypass_guest_pf = 1;
47 module_param(bypass_guest_pf, bool, S_IRUGO);
49 static int __read_mostly enable_vpid = 1;
50 module_param_named(vpid, enable_vpid, bool, 0444);
52 static int __read_mostly flexpriority_enabled = 1;
53 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
55 static int __read_mostly enable_ept = 1;
56 module_param_named(ept, enable_ept, bool, S_IRUGO);
58 static int __read_mostly enable_unrestricted_guest = 1;
59 module_param_named(unrestricted_guest,
60 enable_unrestricted_guest, bool, S_IRUGO);
62 static int __read_mostly emulate_invalid_guest_state = 0;
63 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
65 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST \
66 (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
67 #define KVM_GUEST_CR0_MASK \
68 (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
69 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST \
70 (X86_CR0_WP | X86_CR0_NE)
71 #define KVM_VM_CR0_ALWAYS_ON \
72 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
73 #define KVM_CR4_GUEST_OWNED_BITS \
74 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
77 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
78 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
80 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
83 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
84 * ple_gap: upper bound on the amount of time between two successive
85 * executions of PAUSE in a loop. Also indicate if ple enabled.
86 * According to test, this time is usually small than 41 cycles.
87 * ple_window: upper bound on the amount of time a guest is allowed to execute
88 * in a PAUSE loop. Tests indicate that most spinlocks are held for
89 * less than 2^12 cycles
90 * Time is measured based on a counter that runs at the same rate as the TSC,
91 * refer SDM volume 3b section 21.6.13 & 22.1.3.
93 #define KVM_VMX_DEFAULT_PLE_GAP 41
94 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
95 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
96 module_param(ple_gap, int, S_IRUGO);
98 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
99 module_param(ple_window, int, S_IRUGO);
107 struct shared_msr_entry {
114 struct kvm_vcpu vcpu;
115 struct list_head local_vcpus_link;
116 unsigned long host_rsp;
119 u32 idt_vectoring_info;
120 struct shared_msr_entry *guest_msrs;
124 u64 msr_host_kernel_gs_base;
125 u64 msr_guest_kernel_gs_base;
130 u16 fs_sel, gs_sel, ldt_sel;
131 int gs_ldt_reload_needed;
132 int fs_reload_needed;
137 struct kvm_save_segment {
142 } tr, es, ds, fs, gs;
150 bool emulation_required;
152 /* Support for vnmi-less CPUs */
153 int soft_vnmi_blocked;
155 s64 vnmi_blocked_time;
161 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
163 return container_of(vcpu, struct vcpu_vmx, vcpu);
166 static int init_rmode(struct kvm *kvm);
167 static u64 construct_eptp(unsigned long root_hpa);
169 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
170 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
171 static DEFINE_PER_CPU(struct list_head, vcpus_on_cpu);
173 static unsigned long *vmx_io_bitmap_a;
174 static unsigned long *vmx_io_bitmap_b;
175 static unsigned long *vmx_msr_bitmap_legacy;
176 static unsigned long *vmx_msr_bitmap_longmode;
178 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
179 static DEFINE_SPINLOCK(vmx_vpid_lock);
181 static struct vmcs_config {
185 u32 pin_based_exec_ctrl;
186 u32 cpu_based_exec_ctrl;
187 u32 cpu_based_2nd_exec_ctrl;
192 static struct vmx_capability {
197 #define VMX_SEGMENT_FIELD(seg) \
198 [VCPU_SREG_##seg] = { \
199 .selector = GUEST_##seg##_SELECTOR, \
200 .base = GUEST_##seg##_BASE, \
201 .limit = GUEST_##seg##_LIMIT, \
202 .ar_bytes = GUEST_##seg##_AR_BYTES, \
205 static struct kvm_vmx_segment_field {
210 } kvm_vmx_segment_fields[] = {
211 VMX_SEGMENT_FIELD(CS),
212 VMX_SEGMENT_FIELD(DS),
213 VMX_SEGMENT_FIELD(ES),
214 VMX_SEGMENT_FIELD(FS),
215 VMX_SEGMENT_FIELD(GS),
216 VMX_SEGMENT_FIELD(SS),
217 VMX_SEGMENT_FIELD(TR),
218 VMX_SEGMENT_FIELD(LDTR),
221 static u64 host_efer;
223 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
226 * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
227 * away by decrementing the array size.
229 static const u32 vmx_msr_index[] = {
231 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
233 MSR_EFER, MSR_TSC_AUX, MSR_K6_STAR,
235 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
237 static inline int is_page_fault(u32 intr_info)
239 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
240 INTR_INFO_VALID_MASK)) ==
241 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
244 static inline int is_no_device(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 | NM_VECTOR | INTR_INFO_VALID_MASK);
251 static inline int is_invalid_opcode(u32 intr_info)
253 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
254 INTR_INFO_VALID_MASK)) ==
255 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
258 static inline int is_external_interrupt(u32 intr_info)
260 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
261 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
264 static inline int is_machine_check(u32 intr_info)
266 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
267 INTR_INFO_VALID_MASK)) ==
268 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
271 static inline int cpu_has_vmx_msr_bitmap(void)
273 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
276 static inline int cpu_has_vmx_tpr_shadow(void)
278 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
281 static inline int vm_need_tpr_shadow(struct kvm *kvm)
283 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
286 static inline int cpu_has_secondary_exec_ctrls(void)
288 return vmcs_config.cpu_based_exec_ctrl &
289 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
292 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
294 return vmcs_config.cpu_based_2nd_exec_ctrl &
295 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
298 static inline bool cpu_has_vmx_flexpriority(void)
300 return cpu_has_vmx_tpr_shadow() &&
301 cpu_has_vmx_virtualize_apic_accesses();
304 static inline bool cpu_has_vmx_ept_execute_only(void)
306 return !!(vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT);
309 static inline bool cpu_has_vmx_eptp_uncacheable(void)
311 return !!(vmx_capability.ept & VMX_EPTP_UC_BIT);
314 static inline bool cpu_has_vmx_eptp_writeback(void)
316 return !!(vmx_capability.ept & VMX_EPTP_WB_BIT);
319 static inline bool cpu_has_vmx_ept_2m_page(void)
321 return !!(vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT);
324 static inline bool cpu_has_vmx_ept_1g_page(void)
326 return !!(vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT);
329 static inline int cpu_has_vmx_invept_individual_addr(void)
331 return !!(vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT);
334 static inline int cpu_has_vmx_invept_context(void)
336 return !!(vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT);
339 static inline int cpu_has_vmx_invept_global(void)
341 return !!(vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT);
344 static inline int cpu_has_vmx_ept(void)
346 return vmcs_config.cpu_based_2nd_exec_ctrl &
347 SECONDARY_EXEC_ENABLE_EPT;
350 static inline int cpu_has_vmx_unrestricted_guest(void)
352 return vmcs_config.cpu_based_2nd_exec_ctrl &
353 SECONDARY_EXEC_UNRESTRICTED_GUEST;
356 static inline int cpu_has_vmx_ple(void)
358 return vmcs_config.cpu_based_2nd_exec_ctrl &
359 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
362 static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
364 return flexpriority_enabled && irqchip_in_kernel(kvm);
367 static inline int cpu_has_vmx_vpid(void)
369 return vmcs_config.cpu_based_2nd_exec_ctrl &
370 SECONDARY_EXEC_ENABLE_VPID;
373 static inline int cpu_has_vmx_rdtscp(void)
375 return vmcs_config.cpu_based_2nd_exec_ctrl &
376 SECONDARY_EXEC_RDTSCP;
379 static inline int cpu_has_virtual_nmis(void)
381 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
384 static inline bool report_flexpriority(void)
386 return flexpriority_enabled;
389 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
393 for (i = 0; i < vmx->nmsrs; ++i)
394 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
399 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
405 } operand = { vpid, 0, gva };
407 asm volatile (__ex(ASM_VMX_INVVPID)
408 /* CF==1 or ZF==1 --> rc = -1 */
410 : : "a"(&operand), "c"(ext) : "cc", "memory");
413 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
417 } operand = {eptp, gpa};
419 asm volatile (__ex(ASM_VMX_INVEPT)
420 /* CF==1 or ZF==1 --> rc = -1 */
421 "; ja 1f ; ud2 ; 1:\n"
422 : : "a" (&operand), "c" (ext) : "cc", "memory");
425 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
429 i = __find_msr_index(vmx, msr);
431 return &vmx->guest_msrs[i];
435 static void vmcs_clear(struct vmcs *vmcs)
437 u64 phys_addr = __pa(vmcs);
440 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
441 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
444 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
448 static void __vcpu_clear(void *arg)
450 struct vcpu_vmx *vmx = arg;
451 int cpu = raw_smp_processor_id();
453 if (vmx->vcpu.cpu == cpu)
454 vmcs_clear(vmx->vmcs);
455 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
456 per_cpu(current_vmcs, cpu) = NULL;
457 rdtscll(vmx->vcpu.arch.host_tsc);
458 list_del(&vmx->local_vcpus_link);
463 static void vcpu_clear(struct vcpu_vmx *vmx)
465 if (vmx->vcpu.cpu == -1)
467 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
470 static inline void vpid_sync_vcpu_all(struct vcpu_vmx *vmx)
475 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
478 static inline void ept_sync_global(void)
480 if (cpu_has_vmx_invept_global())
481 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
484 static inline void ept_sync_context(u64 eptp)
487 if (cpu_has_vmx_invept_context())
488 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
494 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
497 if (cpu_has_vmx_invept_individual_addr())
498 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
501 ept_sync_context(eptp);
505 static unsigned long vmcs_readl(unsigned long field)
509 asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
510 : "=a"(value) : "d"(field) : "cc");
514 static u16 vmcs_read16(unsigned long field)
516 return vmcs_readl(field);
519 static u32 vmcs_read32(unsigned long field)
521 return vmcs_readl(field);
524 static u64 vmcs_read64(unsigned long field)
527 return vmcs_readl(field);
529 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
533 static noinline void vmwrite_error(unsigned long field, unsigned long value)
535 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
536 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
540 static void vmcs_writel(unsigned long field, unsigned long value)
544 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
545 : "=q"(error) : "a"(value), "d"(field) : "cc");
547 vmwrite_error(field, value);
550 static void vmcs_write16(unsigned long field, u16 value)
552 vmcs_writel(field, value);
555 static void vmcs_write32(unsigned long field, u32 value)
557 vmcs_writel(field, value);
560 static void vmcs_write64(unsigned long field, u64 value)
562 vmcs_writel(field, value);
563 #ifndef CONFIG_X86_64
565 vmcs_writel(field+1, value >> 32);
569 static void vmcs_clear_bits(unsigned long field, u32 mask)
571 vmcs_writel(field, vmcs_readl(field) & ~mask);
574 static void vmcs_set_bits(unsigned long field, u32 mask)
576 vmcs_writel(field, vmcs_readl(field) | mask);
579 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
583 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
584 (1u << NM_VECTOR) | (1u << DB_VECTOR);
585 if ((vcpu->guest_debug &
586 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
587 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
588 eb |= 1u << BP_VECTOR;
589 if (to_vmx(vcpu)->rmode.vm86_active)
592 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
593 if (vcpu->fpu_active)
594 eb &= ~(1u << NM_VECTOR);
595 vmcs_write32(EXCEPTION_BITMAP, eb);
598 static void reload_tss(void)
601 * VT restores TR but not its size. Useless.
603 struct descriptor_table gdt;
604 struct desc_struct *descs;
607 descs = (void *)gdt.base;
608 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
612 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
617 guest_efer = vmx->vcpu.arch.efer;
620 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
623 ignore_bits = EFER_NX | EFER_SCE;
625 ignore_bits |= EFER_LMA | EFER_LME;
626 /* SCE is meaningful only in long mode on Intel */
627 if (guest_efer & EFER_LMA)
628 ignore_bits &= ~(u64)EFER_SCE;
630 guest_efer &= ~ignore_bits;
631 guest_efer |= host_efer & ignore_bits;
632 vmx->guest_msrs[efer_offset].data = guest_efer;
633 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
637 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
639 struct vcpu_vmx *vmx = to_vmx(vcpu);
642 if (vmx->host_state.loaded)
645 vmx->host_state.loaded = 1;
647 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
648 * allow segment selectors with cpl > 0 or ti == 1.
650 vmx->host_state.ldt_sel = kvm_read_ldt();
651 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
652 vmx->host_state.fs_sel = kvm_read_fs();
653 if (!(vmx->host_state.fs_sel & 7)) {
654 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
655 vmx->host_state.fs_reload_needed = 0;
657 vmcs_write16(HOST_FS_SELECTOR, 0);
658 vmx->host_state.fs_reload_needed = 1;
660 vmx->host_state.gs_sel = kvm_read_gs();
661 if (!(vmx->host_state.gs_sel & 7))
662 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
664 vmcs_write16(HOST_GS_SELECTOR, 0);
665 vmx->host_state.gs_ldt_reload_needed = 1;
669 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
670 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
672 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
673 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
677 if (is_long_mode(&vmx->vcpu)) {
678 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
679 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
682 for (i = 0; i < vmx->save_nmsrs; ++i)
683 kvm_set_shared_msr(vmx->guest_msrs[i].index,
684 vmx->guest_msrs[i].data,
685 vmx->guest_msrs[i].mask);
688 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
692 if (!vmx->host_state.loaded)
695 ++vmx->vcpu.stat.host_state_reload;
696 vmx->host_state.loaded = 0;
697 if (vmx->host_state.fs_reload_needed)
698 kvm_load_fs(vmx->host_state.fs_sel);
699 if (vmx->host_state.gs_ldt_reload_needed) {
700 kvm_load_ldt(vmx->host_state.ldt_sel);
702 * If we have to reload gs, we must take care to
703 * preserve our gs base.
705 local_irq_save(flags);
706 kvm_load_gs(vmx->host_state.gs_sel);
708 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
710 local_irq_restore(flags);
714 if (is_long_mode(&vmx->vcpu)) {
715 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
716 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
721 static void vmx_load_host_state(struct vcpu_vmx *vmx)
724 __vmx_load_host_state(vmx);
729 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
730 * vcpu mutex is already taken.
732 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
734 struct vcpu_vmx *vmx = to_vmx(vcpu);
735 u64 phys_addr = __pa(vmx->vmcs);
736 u64 tsc_this, delta, new_offset;
738 if (vcpu->cpu != cpu) {
740 kvm_migrate_timers(vcpu);
741 set_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests);
743 list_add(&vmx->local_vcpus_link,
744 &per_cpu(vcpus_on_cpu, cpu));
748 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
751 per_cpu(current_vmcs, cpu) = vmx->vmcs;
752 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
753 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
756 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
757 vmx->vmcs, phys_addr);
760 if (vcpu->cpu != cpu) {
761 struct descriptor_table dt;
762 unsigned long sysenter_esp;
766 * Linux uses per-cpu TSS and GDT, so set these when switching
769 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
771 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
773 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
774 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
777 * Make sure the time stamp counter is monotonous.
780 if (tsc_this < vcpu->arch.host_tsc) {
781 delta = vcpu->arch.host_tsc - tsc_this;
782 new_offset = vmcs_read64(TSC_OFFSET) + delta;
783 vmcs_write64(TSC_OFFSET, new_offset);
788 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
790 __vmx_load_host_state(to_vmx(vcpu));
793 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
797 if (vcpu->fpu_active)
799 vcpu->fpu_active = 1;
800 cr0 = vmcs_readl(GUEST_CR0);
801 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
802 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
803 vmcs_writel(GUEST_CR0, cr0);
804 update_exception_bitmap(vcpu);
805 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
806 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
809 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
811 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
813 vmx_decache_cr0_guest_bits(vcpu);
814 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
815 update_exception_bitmap(vcpu);
816 vcpu->arch.cr0_guest_owned_bits = 0;
817 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
818 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
821 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
823 unsigned long rflags, save_rflags;
825 rflags = vmcs_readl(GUEST_RFLAGS);
826 if (to_vmx(vcpu)->rmode.vm86_active) {
827 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
828 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
829 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
834 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
836 if (to_vmx(vcpu)->rmode.vm86_active) {
837 to_vmx(vcpu)->rmode.save_rflags = rflags;
838 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
840 vmcs_writel(GUEST_RFLAGS, rflags);
843 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
845 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
848 if (interruptibility & GUEST_INTR_STATE_STI)
849 ret |= X86_SHADOW_INT_STI;
850 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
851 ret |= X86_SHADOW_INT_MOV_SS;
856 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
858 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
859 u32 interruptibility = interruptibility_old;
861 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
863 if (mask & X86_SHADOW_INT_MOV_SS)
864 interruptibility |= GUEST_INTR_STATE_MOV_SS;
865 if (mask & X86_SHADOW_INT_STI)
866 interruptibility |= GUEST_INTR_STATE_STI;
868 if ((interruptibility != interruptibility_old))
869 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
872 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
876 rip = kvm_rip_read(vcpu);
877 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
878 kvm_rip_write(vcpu, rip);
880 /* skipping an emulated instruction also counts */
881 vmx_set_interrupt_shadow(vcpu, 0);
884 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
885 bool has_error_code, u32 error_code)
887 struct vcpu_vmx *vmx = to_vmx(vcpu);
888 u32 intr_info = nr | INTR_INFO_VALID_MASK;
890 if (has_error_code) {
891 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
892 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
895 if (vmx->rmode.vm86_active) {
896 vmx->rmode.irq.pending = true;
897 vmx->rmode.irq.vector = nr;
898 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
899 if (kvm_exception_is_soft(nr))
900 vmx->rmode.irq.rip +=
901 vmx->vcpu.arch.event_exit_inst_len;
902 intr_info |= INTR_TYPE_SOFT_INTR;
903 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
904 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
905 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
909 if (kvm_exception_is_soft(nr)) {
910 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
911 vmx->vcpu.arch.event_exit_inst_len);
912 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
914 intr_info |= INTR_TYPE_HARD_EXCEPTION;
916 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
919 static bool vmx_rdtscp_supported(void)
921 return cpu_has_vmx_rdtscp();
925 * Swap MSR entry in host/guest MSR entry array.
927 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
929 struct shared_msr_entry tmp;
931 tmp = vmx->guest_msrs[to];
932 vmx->guest_msrs[to] = vmx->guest_msrs[from];
933 vmx->guest_msrs[from] = tmp;
937 * Set up the vmcs to automatically save and restore system
938 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
939 * mode, as fiddling with msrs is very expensive.
941 static void setup_msrs(struct vcpu_vmx *vmx)
943 int save_nmsrs, index;
944 unsigned long *msr_bitmap;
946 vmx_load_host_state(vmx);
949 if (is_long_mode(&vmx->vcpu)) {
950 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
952 move_msr_up(vmx, index, save_nmsrs++);
953 index = __find_msr_index(vmx, MSR_LSTAR);
955 move_msr_up(vmx, index, save_nmsrs++);
956 index = __find_msr_index(vmx, MSR_CSTAR);
958 move_msr_up(vmx, index, save_nmsrs++);
959 index = __find_msr_index(vmx, MSR_TSC_AUX);
960 if (index >= 0 && vmx->rdtscp_enabled)
961 move_msr_up(vmx, index, save_nmsrs++);
963 * MSR_K6_STAR is only needed on long mode guests, and only
964 * if efer.sce is enabled.
966 index = __find_msr_index(vmx, MSR_K6_STAR);
967 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
968 move_msr_up(vmx, index, save_nmsrs++);
971 index = __find_msr_index(vmx, MSR_EFER);
972 if (index >= 0 && update_transition_efer(vmx, index))
973 move_msr_up(vmx, index, save_nmsrs++);
975 vmx->save_nmsrs = save_nmsrs;
977 if (cpu_has_vmx_msr_bitmap()) {
978 if (is_long_mode(&vmx->vcpu))
979 msr_bitmap = vmx_msr_bitmap_longmode;
981 msr_bitmap = vmx_msr_bitmap_legacy;
983 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
988 * reads and returns guest's timestamp counter "register"
989 * guest_tsc = host_tsc + tsc_offset -- 21.3
991 static u64 guest_read_tsc(void)
993 u64 host_tsc, tsc_offset;
996 tsc_offset = vmcs_read64(TSC_OFFSET);
997 return host_tsc + tsc_offset;
1001 * writes 'guest_tsc' into guest's timestamp counter "register"
1002 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
1004 static void guest_write_tsc(u64 guest_tsc, u64 host_tsc)
1006 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
1010 * Reads an msr value (of 'msr_index') into 'pdata'.
1011 * Returns 0 on success, non-0 otherwise.
1012 * Assumes vcpu_load() was already called.
1014 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1017 struct shared_msr_entry *msr;
1020 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
1024 switch (msr_index) {
1025 #ifdef CONFIG_X86_64
1027 data = vmcs_readl(GUEST_FS_BASE);
1030 data = vmcs_readl(GUEST_GS_BASE);
1032 case MSR_KERNEL_GS_BASE:
1033 vmx_load_host_state(to_vmx(vcpu));
1034 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
1038 return kvm_get_msr_common(vcpu, msr_index, pdata);
1040 data = guest_read_tsc();
1042 case MSR_IA32_SYSENTER_CS:
1043 data = vmcs_read32(GUEST_SYSENTER_CS);
1045 case MSR_IA32_SYSENTER_EIP:
1046 data = vmcs_readl(GUEST_SYSENTER_EIP);
1048 case MSR_IA32_SYSENTER_ESP:
1049 data = vmcs_readl(GUEST_SYSENTER_ESP);
1052 if (!to_vmx(vcpu)->rdtscp_enabled)
1054 /* Otherwise falls through */
1056 vmx_load_host_state(to_vmx(vcpu));
1057 msr = find_msr_entry(to_vmx(vcpu), msr_index);
1059 vmx_load_host_state(to_vmx(vcpu));
1063 return kvm_get_msr_common(vcpu, msr_index, pdata);
1071 * Writes msr value into into the appropriate "register".
1072 * Returns 0 on success, non-0 otherwise.
1073 * Assumes vcpu_load() was already called.
1075 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1077 struct vcpu_vmx *vmx = to_vmx(vcpu);
1078 struct shared_msr_entry *msr;
1082 switch (msr_index) {
1084 vmx_load_host_state(vmx);
1085 ret = kvm_set_msr_common(vcpu, msr_index, data);
1087 #ifdef CONFIG_X86_64
1089 vmcs_writel(GUEST_FS_BASE, data);
1092 vmcs_writel(GUEST_GS_BASE, data);
1094 case MSR_KERNEL_GS_BASE:
1095 vmx_load_host_state(vmx);
1096 vmx->msr_guest_kernel_gs_base = data;
1099 case MSR_IA32_SYSENTER_CS:
1100 vmcs_write32(GUEST_SYSENTER_CS, data);
1102 case MSR_IA32_SYSENTER_EIP:
1103 vmcs_writel(GUEST_SYSENTER_EIP, data);
1105 case MSR_IA32_SYSENTER_ESP:
1106 vmcs_writel(GUEST_SYSENTER_ESP, data);
1110 guest_write_tsc(data, host_tsc);
1112 case MSR_IA32_CR_PAT:
1113 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1114 vmcs_write64(GUEST_IA32_PAT, data);
1115 vcpu->arch.pat = data;
1118 ret = kvm_set_msr_common(vcpu, msr_index, data);
1121 if (!vmx->rdtscp_enabled)
1123 /* Check reserved bit, higher 32 bits should be zero */
1124 if ((data >> 32) != 0)
1126 /* Otherwise falls through */
1128 msr = find_msr_entry(vmx, msr_index);
1130 vmx_load_host_state(vmx);
1134 ret = kvm_set_msr_common(vcpu, msr_index, data);
1140 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1142 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1145 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1148 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1150 case VCPU_EXREG_PDPTR:
1152 ept_save_pdptrs(vcpu);
1159 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1161 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1162 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1164 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1166 update_exception_bitmap(vcpu);
1169 static __init int cpu_has_kvm_support(void)
1171 return cpu_has_vmx();
1174 static __init int vmx_disabled_by_bios(void)
1178 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1179 return (msr & (FEATURE_CONTROL_LOCKED |
1180 FEATURE_CONTROL_VMXON_ENABLED))
1181 == FEATURE_CONTROL_LOCKED;
1182 /* locked but not enabled */
1185 static int hardware_enable(void *garbage)
1187 int cpu = raw_smp_processor_id();
1188 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1191 if (read_cr4() & X86_CR4_VMXE)
1194 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1195 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1196 if ((old & (FEATURE_CONTROL_LOCKED |
1197 FEATURE_CONTROL_VMXON_ENABLED))
1198 != (FEATURE_CONTROL_LOCKED |
1199 FEATURE_CONTROL_VMXON_ENABLED))
1200 /* enable and lock */
1201 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
1202 FEATURE_CONTROL_LOCKED |
1203 FEATURE_CONTROL_VMXON_ENABLED);
1204 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1205 asm volatile (ASM_VMX_VMXON_RAX
1206 : : "a"(&phys_addr), "m"(phys_addr)
1214 static void vmclear_local_vcpus(void)
1216 int cpu = raw_smp_processor_id();
1217 struct vcpu_vmx *vmx, *n;
1219 list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1225 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1228 static void kvm_cpu_vmxoff(void)
1230 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1231 write_cr4(read_cr4() & ~X86_CR4_VMXE);
1234 static void hardware_disable(void *garbage)
1236 vmclear_local_vcpus();
1240 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1241 u32 msr, u32 *result)
1243 u32 vmx_msr_low, vmx_msr_high;
1244 u32 ctl = ctl_min | ctl_opt;
1246 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1248 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1249 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
1251 /* Ensure minimum (required) set of control bits are supported. */
1259 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1261 u32 vmx_msr_low, vmx_msr_high;
1262 u32 min, opt, min2, opt2;
1263 u32 _pin_based_exec_control = 0;
1264 u32 _cpu_based_exec_control = 0;
1265 u32 _cpu_based_2nd_exec_control = 0;
1266 u32 _vmexit_control = 0;
1267 u32 _vmentry_control = 0;
1269 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1270 opt = PIN_BASED_VIRTUAL_NMIS;
1271 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1272 &_pin_based_exec_control) < 0)
1275 min = CPU_BASED_HLT_EXITING |
1276 #ifdef CONFIG_X86_64
1277 CPU_BASED_CR8_LOAD_EXITING |
1278 CPU_BASED_CR8_STORE_EXITING |
1280 CPU_BASED_CR3_LOAD_EXITING |
1281 CPU_BASED_CR3_STORE_EXITING |
1282 CPU_BASED_USE_IO_BITMAPS |
1283 CPU_BASED_MOV_DR_EXITING |
1284 CPU_BASED_USE_TSC_OFFSETING |
1285 CPU_BASED_MWAIT_EXITING |
1286 CPU_BASED_MONITOR_EXITING |
1287 CPU_BASED_INVLPG_EXITING;
1288 opt = CPU_BASED_TPR_SHADOW |
1289 CPU_BASED_USE_MSR_BITMAPS |
1290 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1291 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1292 &_cpu_based_exec_control) < 0)
1294 #ifdef CONFIG_X86_64
1295 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1296 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1297 ~CPU_BASED_CR8_STORE_EXITING;
1299 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1301 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1302 SECONDARY_EXEC_WBINVD_EXITING |
1303 SECONDARY_EXEC_ENABLE_VPID |
1304 SECONDARY_EXEC_ENABLE_EPT |
1305 SECONDARY_EXEC_UNRESTRICTED_GUEST |
1306 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
1307 SECONDARY_EXEC_RDTSCP;
1308 if (adjust_vmx_controls(min2, opt2,
1309 MSR_IA32_VMX_PROCBASED_CTLS2,
1310 &_cpu_based_2nd_exec_control) < 0)
1313 #ifndef CONFIG_X86_64
1314 if (!(_cpu_based_2nd_exec_control &
1315 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1316 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1318 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1319 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1321 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
1322 CPU_BASED_CR3_STORE_EXITING |
1323 CPU_BASED_INVLPG_EXITING);
1324 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1325 vmx_capability.ept, vmx_capability.vpid);
1329 #ifdef CONFIG_X86_64
1330 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1332 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1333 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1334 &_vmexit_control) < 0)
1338 opt = VM_ENTRY_LOAD_IA32_PAT;
1339 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1340 &_vmentry_control) < 0)
1343 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1345 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1346 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1349 #ifdef CONFIG_X86_64
1350 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1351 if (vmx_msr_high & (1u<<16))
1355 /* Require Write-Back (WB) memory type for VMCS accesses. */
1356 if (((vmx_msr_high >> 18) & 15) != 6)
1359 vmcs_conf->size = vmx_msr_high & 0x1fff;
1360 vmcs_conf->order = get_order(vmcs_config.size);
1361 vmcs_conf->revision_id = vmx_msr_low;
1363 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1364 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1365 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1366 vmcs_conf->vmexit_ctrl = _vmexit_control;
1367 vmcs_conf->vmentry_ctrl = _vmentry_control;
1372 static struct vmcs *alloc_vmcs_cpu(int cpu)
1374 int node = cpu_to_node(cpu);
1378 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
1381 vmcs = page_address(pages);
1382 memset(vmcs, 0, vmcs_config.size);
1383 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1387 static struct vmcs *alloc_vmcs(void)
1389 return alloc_vmcs_cpu(raw_smp_processor_id());
1392 static void free_vmcs(struct vmcs *vmcs)
1394 free_pages((unsigned long)vmcs, vmcs_config.order);
1397 static void free_kvm_area(void)
1401 for_each_possible_cpu(cpu) {
1402 free_vmcs(per_cpu(vmxarea, cpu));
1403 per_cpu(vmxarea, cpu) = NULL;
1407 static __init int alloc_kvm_area(void)
1411 for_each_possible_cpu(cpu) {
1414 vmcs = alloc_vmcs_cpu(cpu);
1420 per_cpu(vmxarea, cpu) = vmcs;
1425 static __init int hardware_setup(void)
1427 if (setup_vmcs_config(&vmcs_config) < 0)
1430 if (boot_cpu_has(X86_FEATURE_NX))
1431 kvm_enable_efer_bits(EFER_NX);
1433 if (!cpu_has_vmx_vpid())
1436 if (!cpu_has_vmx_ept()) {
1438 enable_unrestricted_guest = 0;
1441 if (!cpu_has_vmx_unrestricted_guest())
1442 enable_unrestricted_guest = 0;
1444 if (!cpu_has_vmx_flexpriority())
1445 flexpriority_enabled = 0;
1447 if (!cpu_has_vmx_tpr_shadow())
1448 kvm_x86_ops->update_cr8_intercept = NULL;
1450 if (enable_ept && !cpu_has_vmx_ept_2m_page())
1451 kvm_disable_largepages();
1453 if (!cpu_has_vmx_ple())
1456 return alloc_kvm_area();
1459 static __exit void hardware_unsetup(void)
1464 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1466 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1468 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1469 vmcs_write16(sf->selector, save->selector);
1470 vmcs_writel(sf->base, save->base);
1471 vmcs_write32(sf->limit, save->limit);
1472 vmcs_write32(sf->ar_bytes, save->ar);
1474 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1476 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1480 static void enter_pmode(struct kvm_vcpu *vcpu)
1482 unsigned long flags;
1483 struct vcpu_vmx *vmx = to_vmx(vcpu);
1485 vmx->emulation_required = 1;
1486 vmx->rmode.vm86_active = 0;
1488 vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
1489 vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
1490 vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
1492 flags = vmcs_readl(GUEST_RFLAGS);
1493 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1494 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1495 vmcs_writel(GUEST_RFLAGS, flags);
1497 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1498 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1500 update_exception_bitmap(vcpu);
1502 if (emulate_invalid_guest_state)
1505 fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
1506 fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
1507 fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
1508 fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
1510 vmcs_write16(GUEST_SS_SELECTOR, 0);
1511 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1513 vmcs_write16(GUEST_CS_SELECTOR,
1514 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1515 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1518 static gva_t rmode_tss_base(struct kvm *kvm)
1520 if (!kvm->arch.tss_addr) {
1521 struct kvm_memslots *slots;
1524 slots = rcu_dereference(kvm->memslots);
1525 base_gfn = kvm->memslots->memslots[0].base_gfn +
1526 kvm->memslots->memslots[0].npages - 3;
1527 return base_gfn << PAGE_SHIFT;
1529 return kvm->arch.tss_addr;
1532 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1534 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1536 save->selector = vmcs_read16(sf->selector);
1537 save->base = vmcs_readl(sf->base);
1538 save->limit = vmcs_read32(sf->limit);
1539 save->ar = vmcs_read32(sf->ar_bytes);
1540 vmcs_write16(sf->selector, save->base >> 4);
1541 vmcs_write32(sf->base, save->base & 0xfffff);
1542 vmcs_write32(sf->limit, 0xffff);
1543 vmcs_write32(sf->ar_bytes, 0xf3);
1546 static void enter_rmode(struct kvm_vcpu *vcpu)
1548 unsigned long flags;
1549 struct vcpu_vmx *vmx = to_vmx(vcpu);
1551 if (enable_unrestricted_guest)
1554 vmx->emulation_required = 1;
1555 vmx->rmode.vm86_active = 1;
1557 vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1558 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1560 vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1561 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1563 vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1564 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1566 flags = vmcs_readl(GUEST_RFLAGS);
1567 vmx->rmode.save_rflags = flags;
1569 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1571 vmcs_writel(GUEST_RFLAGS, flags);
1572 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1573 update_exception_bitmap(vcpu);
1575 if (emulate_invalid_guest_state)
1576 goto continue_rmode;
1578 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1579 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1580 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1582 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1583 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1584 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1585 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1586 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1588 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
1589 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
1590 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
1591 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
1594 kvm_mmu_reset_context(vcpu);
1595 init_rmode(vcpu->kvm);
1598 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1600 struct vcpu_vmx *vmx = to_vmx(vcpu);
1601 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1607 * Force kernel_gs_base reloading before EFER changes, as control
1608 * of this msr depends on is_long_mode().
1610 vmx_load_host_state(to_vmx(vcpu));
1611 vcpu->arch.efer = efer;
1612 if (efer & EFER_LMA) {
1613 vmcs_write32(VM_ENTRY_CONTROLS,
1614 vmcs_read32(VM_ENTRY_CONTROLS) |
1615 VM_ENTRY_IA32E_MODE);
1618 vmcs_write32(VM_ENTRY_CONTROLS,
1619 vmcs_read32(VM_ENTRY_CONTROLS) &
1620 ~VM_ENTRY_IA32E_MODE);
1622 msr->data = efer & ~EFER_LME;
1627 #ifdef CONFIG_X86_64
1629 static void enter_lmode(struct kvm_vcpu *vcpu)
1633 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1634 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1635 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1637 vmcs_write32(GUEST_TR_AR_BYTES,
1638 (guest_tr_ar & ~AR_TYPE_MASK)
1639 | AR_TYPE_BUSY_64_TSS);
1641 vcpu->arch.efer |= EFER_LMA;
1642 vmx_set_efer(vcpu, vcpu->arch.efer);
1645 static void exit_lmode(struct kvm_vcpu *vcpu)
1647 vcpu->arch.efer &= ~EFER_LMA;
1649 vmcs_write32(VM_ENTRY_CONTROLS,
1650 vmcs_read32(VM_ENTRY_CONTROLS)
1651 & ~VM_ENTRY_IA32E_MODE);
1656 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1658 vpid_sync_vcpu_all(to_vmx(vcpu));
1660 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1663 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1665 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
1667 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
1668 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
1671 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1673 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
1675 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
1676 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
1679 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1681 if (!test_bit(VCPU_EXREG_PDPTR,
1682 (unsigned long *)&vcpu->arch.regs_dirty))
1685 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1686 vmcs_write64(GUEST_PDPTR0, vcpu->arch.pdptrs[0]);
1687 vmcs_write64(GUEST_PDPTR1, vcpu->arch.pdptrs[1]);
1688 vmcs_write64(GUEST_PDPTR2, vcpu->arch.pdptrs[2]);
1689 vmcs_write64(GUEST_PDPTR3, vcpu->arch.pdptrs[3]);
1693 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
1695 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1696 vcpu->arch.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
1697 vcpu->arch.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
1698 vcpu->arch.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
1699 vcpu->arch.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
1702 __set_bit(VCPU_EXREG_PDPTR,
1703 (unsigned long *)&vcpu->arch.regs_avail);
1704 __set_bit(VCPU_EXREG_PDPTR,
1705 (unsigned long *)&vcpu->arch.regs_dirty);
1708 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1710 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1712 struct kvm_vcpu *vcpu)
1714 if (!(cr0 & X86_CR0_PG)) {
1715 /* From paging/starting to nonpaging */
1716 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1717 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1718 (CPU_BASED_CR3_LOAD_EXITING |
1719 CPU_BASED_CR3_STORE_EXITING));
1720 vcpu->arch.cr0 = cr0;
1721 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1722 } else if (!is_paging(vcpu)) {
1723 /* From nonpaging to paging */
1724 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1725 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1726 ~(CPU_BASED_CR3_LOAD_EXITING |
1727 CPU_BASED_CR3_STORE_EXITING));
1728 vcpu->arch.cr0 = cr0;
1729 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1732 if (!(cr0 & X86_CR0_WP))
1733 *hw_cr0 &= ~X86_CR0_WP;
1736 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1738 struct vcpu_vmx *vmx = to_vmx(vcpu);
1739 unsigned long hw_cr0;
1741 if (enable_unrestricted_guest)
1742 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
1743 | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
1745 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
1747 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
1750 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
1753 #ifdef CONFIG_X86_64
1754 if (vcpu->arch.efer & EFER_LME) {
1755 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1757 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1763 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1765 if (!vcpu->fpu_active)
1766 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
1768 vmcs_writel(CR0_READ_SHADOW, cr0);
1769 vmcs_writel(GUEST_CR0, hw_cr0);
1770 vcpu->arch.cr0 = cr0;
1773 static u64 construct_eptp(unsigned long root_hpa)
1777 /* TODO write the value reading from MSR */
1778 eptp = VMX_EPT_DEFAULT_MT |
1779 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1780 eptp |= (root_hpa & PAGE_MASK);
1785 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1787 unsigned long guest_cr3;
1792 eptp = construct_eptp(cr3);
1793 vmcs_write64(EPT_POINTER, eptp);
1794 guest_cr3 = is_paging(vcpu) ? vcpu->arch.cr3 :
1795 vcpu->kvm->arch.ept_identity_map_addr;
1796 ept_load_pdptrs(vcpu);
1799 vmx_flush_tlb(vcpu);
1800 vmcs_writel(GUEST_CR3, guest_cr3);
1803 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1805 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
1806 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
1808 vcpu->arch.cr4 = cr4;
1810 if (!is_paging(vcpu)) {
1811 hw_cr4 &= ~X86_CR4_PAE;
1812 hw_cr4 |= X86_CR4_PSE;
1813 } else if (!(cr4 & X86_CR4_PAE)) {
1814 hw_cr4 &= ~X86_CR4_PAE;
1818 vmcs_writel(CR4_READ_SHADOW, cr4);
1819 vmcs_writel(GUEST_CR4, hw_cr4);
1822 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1824 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1826 return vmcs_readl(sf->base);
1829 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1830 struct kvm_segment *var, int seg)
1832 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1835 var->base = vmcs_readl(sf->base);
1836 var->limit = vmcs_read32(sf->limit);
1837 var->selector = vmcs_read16(sf->selector);
1838 ar = vmcs_read32(sf->ar_bytes);
1839 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
1841 var->type = ar & 15;
1842 var->s = (ar >> 4) & 1;
1843 var->dpl = (ar >> 5) & 3;
1844 var->present = (ar >> 7) & 1;
1845 var->avl = (ar >> 12) & 1;
1846 var->l = (ar >> 13) & 1;
1847 var->db = (ar >> 14) & 1;
1848 var->g = (ar >> 15) & 1;
1849 var->unusable = (ar >> 16) & 1;
1852 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
1854 if (!is_protmode(vcpu))
1857 if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
1860 return vmcs_read16(GUEST_CS_SELECTOR) & 3;
1863 static u32 vmx_segment_access_rights(struct kvm_segment *var)
1870 ar = var->type & 15;
1871 ar |= (var->s & 1) << 4;
1872 ar |= (var->dpl & 3) << 5;
1873 ar |= (var->present & 1) << 7;
1874 ar |= (var->avl & 1) << 12;
1875 ar |= (var->l & 1) << 13;
1876 ar |= (var->db & 1) << 14;
1877 ar |= (var->g & 1) << 15;
1879 if (ar == 0) /* a 0 value means unusable */
1880 ar = AR_UNUSABLE_MASK;
1885 static void vmx_set_segment(struct kvm_vcpu *vcpu,
1886 struct kvm_segment *var, int seg)
1888 struct vcpu_vmx *vmx = to_vmx(vcpu);
1889 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1892 if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
1893 vmx->rmode.tr.selector = var->selector;
1894 vmx->rmode.tr.base = var->base;
1895 vmx->rmode.tr.limit = var->limit;
1896 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
1899 vmcs_writel(sf->base, var->base);
1900 vmcs_write32(sf->limit, var->limit);
1901 vmcs_write16(sf->selector, var->selector);
1902 if (vmx->rmode.vm86_active && var->s) {
1904 * Hack real-mode segments into vm86 compatibility.
1906 if (var->base == 0xffff0000 && var->selector == 0xf000)
1907 vmcs_writel(sf->base, 0xf0000);
1910 ar = vmx_segment_access_rights(var);
1913 * Fix the "Accessed" bit in AR field of segment registers for older
1915 * IA32 arch specifies that at the time of processor reset the
1916 * "Accessed" bit in the AR field of segment registers is 1. And qemu
1917 * is setting it to 0 in the usedland code. This causes invalid guest
1918 * state vmexit when "unrestricted guest" mode is turned on.
1919 * Fix for this setup issue in cpu_reset is being pushed in the qemu
1920 * tree. Newer qemu binaries with that qemu fix would not need this
1923 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
1924 ar |= 0x1; /* Accessed */
1926 vmcs_write32(sf->ar_bytes, ar);
1929 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1931 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1933 *db = (ar >> 14) & 1;
1934 *l = (ar >> 13) & 1;
1937 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1939 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1940 dt->base = vmcs_readl(GUEST_IDTR_BASE);
1943 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1945 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1946 vmcs_writel(GUEST_IDTR_BASE, dt->base);
1949 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1951 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1952 dt->base = vmcs_readl(GUEST_GDTR_BASE);
1955 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1957 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1958 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1961 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
1963 struct kvm_segment var;
1966 vmx_get_segment(vcpu, &var, seg);
1967 ar = vmx_segment_access_rights(&var);
1969 if (var.base != (var.selector << 4))
1971 if (var.limit != 0xffff)
1979 static bool code_segment_valid(struct kvm_vcpu *vcpu)
1981 struct kvm_segment cs;
1982 unsigned int cs_rpl;
1984 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
1985 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
1989 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
1993 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
1994 if (cs.dpl > cs_rpl)
1997 if (cs.dpl != cs_rpl)
2003 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
2007 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
2009 struct kvm_segment ss;
2010 unsigned int ss_rpl;
2012 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2013 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
2017 if (ss.type != 3 && ss.type != 7)
2021 if (ss.dpl != ss_rpl) /* DPL != RPL */
2029 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
2031 struct kvm_segment var;
2034 vmx_get_segment(vcpu, &var, seg);
2035 rpl = var.selector & SELECTOR_RPL_MASK;
2043 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
2044 if (var.dpl < rpl) /* DPL < RPL */
2048 /* TODO: Add other members to kvm_segment_field to allow checking for other access
2054 static bool tr_valid(struct kvm_vcpu *vcpu)
2056 struct kvm_segment tr;
2058 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
2062 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2064 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
2072 static bool ldtr_valid(struct kvm_vcpu *vcpu)
2074 struct kvm_segment ldtr;
2076 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
2080 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2090 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
2092 struct kvm_segment cs, ss;
2094 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2095 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2097 return ((cs.selector & SELECTOR_RPL_MASK) ==
2098 (ss.selector & SELECTOR_RPL_MASK));
2102 * Check if guest state is valid. Returns true if valid, false if
2104 * We assume that registers are always usable
2106 static bool guest_state_valid(struct kvm_vcpu *vcpu)
2108 /* real mode guest state checks */
2109 if (!is_protmode(vcpu)) {
2110 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
2112 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
2114 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
2116 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
2118 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
2120 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
2123 /* protected mode guest state checks */
2124 if (!cs_ss_rpl_check(vcpu))
2126 if (!code_segment_valid(vcpu))
2128 if (!stack_segment_valid(vcpu))
2130 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
2132 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
2134 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
2136 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2138 if (!tr_valid(vcpu))
2140 if (!ldtr_valid(vcpu))
2144 * - Add checks on RIP
2145 * - Add checks on RFLAGS
2151 static int init_rmode_tss(struct kvm *kvm)
2153 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
2158 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2161 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
2162 r = kvm_write_guest_page(kvm, fn++, &data,
2163 TSS_IOPB_BASE_OFFSET, sizeof(u16));
2166 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2169 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2173 r = kvm_write_guest_page(kvm, fn, &data,
2174 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2184 static int init_rmode_identity_map(struct kvm *kvm)
2187 pfn_t identity_map_pfn;
2192 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2193 printk(KERN_ERR "EPT: identity-mapping pagetable "
2194 "haven't been allocated!\n");
2197 if (likely(kvm->arch.ept_identity_pagetable_done))
2200 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
2201 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2204 /* Set up identity-mapping pagetable for EPT in real mode */
2205 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2206 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2207 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2208 r = kvm_write_guest_page(kvm, identity_map_pfn,
2209 &tmp, i * sizeof(tmp), sizeof(tmp));
2213 kvm->arch.ept_identity_pagetable_done = true;
2219 static void seg_setup(int seg)
2221 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2224 vmcs_write16(sf->selector, 0);
2225 vmcs_writel(sf->base, 0);
2226 vmcs_write32(sf->limit, 0xffff);
2227 if (enable_unrestricted_guest) {
2229 if (seg == VCPU_SREG_CS)
2230 ar |= 0x08; /* code segment */
2234 vmcs_write32(sf->ar_bytes, ar);
2237 static int alloc_apic_access_page(struct kvm *kvm)
2239 struct kvm_userspace_memory_region kvm_userspace_mem;
2242 mutex_lock(&kvm->slots_lock);
2243 if (kvm->arch.apic_access_page)
2245 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2246 kvm_userspace_mem.flags = 0;
2247 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2248 kvm_userspace_mem.memory_size = PAGE_SIZE;
2249 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2253 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2255 mutex_unlock(&kvm->slots_lock);
2259 static int alloc_identity_pagetable(struct kvm *kvm)
2261 struct kvm_userspace_memory_region kvm_userspace_mem;
2264 mutex_lock(&kvm->slots_lock);
2265 if (kvm->arch.ept_identity_pagetable)
2267 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2268 kvm_userspace_mem.flags = 0;
2269 kvm_userspace_mem.guest_phys_addr =
2270 kvm->arch.ept_identity_map_addr;
2271 kvm_userspace_mem.memory_size = PAGE_SIZE;
2272 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2276 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2277 kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
2279 mutex_unlock(&kvm->slots_lock);
2283 static void allocate_vpid(struct vcpu_vmx *vmx)
2290 spin_lock(&vmx_vpid_lock);
2291 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2292 if (vpid < VMX_NR_VPIDS) {
2294 __set_bit(vpid, vmx_vpid_bitmap);
2296 spin_unlock(&vmx_vpid_lock);
2299 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
2301 int f = sizeof(unsigned long);
2303 if (!cpu_has_vmx_msr_bitmap())
2307 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2308 * have the write-low and read-high bitmap offsets the wrong way round.
2309 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2311 if (msr <= 0x1fff) {
2312 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2313 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
2314 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2316 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2317 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
2321 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2324 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2325 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2329 * Sets up the vmcs for emulated real mode.
2331 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2333 u32 host_sysenter_cs, msr_low, msr_high;
2335 u64 host_pat, tsc_this, tsc_base;
2337 struct descriptor_table dt;
2339 unsigned long kvm_vmx_return;
2343 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2344 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
2346 if (cpu_has_vmx_msr_bitmap())
2347 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
2349 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2352 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2353 vmcs_config.pin_based_exec_ctrl);
2355 exec_control = vmcs_config.cpu_based_exec_ctrl;
2356 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2357 exec_control &= ~CPU_BASED_TPR_SHADOW;
2358 #ifdef CONFIG_X86_64
2359 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2360 CPU_BASED_CR8_LOAD_EXITING;
2364 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2365 CPU_BASED_CR3_LOAD_EXITING |
2366 CPU_BASED_INVLPG_EXITING;
2367 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2369 if (cpu_has_secondary_exec_ctrls()) {
2370 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2371 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2373 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2375 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2377 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2378 enable_unrestricted_guest = 0;
2380 if (!enable_unrestricted_guest)
2381 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2383 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
2384 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2388 vmcs_write32(PLE_GAP, ple_gap);
2389 vmcs_write32(PLE_WINDOW, ple_window);
2392 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2393 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2394 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
2396 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
2397 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
2398 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
2400 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
2401 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2402 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2403 vmcs_write16(HOST_FS_SELECTOR, kvm_read_fs()); /* 22.2.4 */
2404 vmcs_write16(HOST_GS_SELECTOR, kvm_read_gs()); /* 22.2.4 */
2405 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2406 #ifdef CONFIG_X86_64
2407 rdmsrl(MSR_FS_BASE, a);
2408 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2409 rdmsrl(MSR_GS_BASE, a);
2410 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2412 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2413 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2416 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
2419 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
2421 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2422 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2423 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2424 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2425 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2427 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2428 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2429 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2430 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
2431 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2432 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
2434 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2435 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2436 host_pat = msr_low | ((u64) msr_high << 32);
2437 vmcs_write64(HOST_IA32_PAT, host_pat);
2439 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2440 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2441 host_pat = msr_low | ((u64) msr_high << 32);
2442 /* Write the default value follow host pat */
2443 vmcs_write64(GUEST_IA32_PAT, host_pat);
2444 /* Keep arch.pat sync with GUEST_IA32_PAT */
2445 vmx->vcpu.arch.pat = host_pat;
2448 for (i = 0; i < NR_VMX_MSR; ++i) {
2449 u32 index = vmx_msr_index[i];
2450 u32 data_low, data_high;
2453 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2455 if (wrmsr_safe(index, data_low, data_high) < 0)
2457 vmx->guest_msrs[j].index = i;
2458 vmx->guest_msrs[j].data = 0;
2459 vmx->guest_msrs[j].mask = -1ull;
2463 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2465 /* 22.2.1, 20.8.1 */
2466 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2468 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2469 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
2471 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
2472 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
2474 tsc_base = vmx->vcpu.kvm->arch.vm_init_tsc;
2476 if (tsc_this < vmx->vcpu.kvm->arch.vm_init_tsc)
2477 tsc_base = tsc_this;
2479 guest_write_tsc(0, tsc_base);
2484 static int init_rmode(struct kvm *kvm)
2486 if (!init_rmode_tss(kvm))
2488 if (!init_rmode_identity_map(kvm))
2493 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2495 struct vcpu_vmx *vmx = to_vmx(vcpu);
2499 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2500 idx = srcu_read_lock(&vcpu->kvm->srcu);
2501 if (!init_rmode(vmx->vcpu.kvm)) {
2506 vmx->rmode.vm86_active = 0;
2508 vmx->soft_vnmi_blocked = 0;
2510 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2511 kvm_set_cr8(&vmx->vcpu, 0);
2512 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2513 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2514 msr |= MSR_IA32_APICBASE_BSP;
2515 kvm_set_apic_base(&vmx->vcpu, msr);
2517 fx_init(&vmx->vcpu);
2519 seg_setup(VCPU_SREG_CS);
2521 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2522 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
2524 if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
2525 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2526 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2528 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2529 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2532 seg_setup(VCPU_SREG_DS);
2533 seg_setup(VCPU_SREG_ES);
2534 seg_setup(VCPU_SREG_FS);
2535 seg_setup(VCPU_SREG_GS);
2536 seg_setup(VCPU_SREG_SS);
2538 vmcs_write16(GUEST_TR_SELECTOR, 0);
2539 vmcs_writel(GUEST_TR_BASE, 0);
2540 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2541 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2543 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2544 vmcs_writel(GUEST_LDTR_BASE, 0);
2545 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2546 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2548 vmcs_write32(GUEST_SYSENTER_CS, 0);
2549 vmcs_writel(GUEST_SYSENTER_ESP, 0);
2550 vmcs_writel(GUEST_SYSENTER_EIP, 0);
2552 vmcs_writel(GUEST_RFLAGS, 0x02);
2553 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2554 kvm_rip_write(vcpu, 0xfff0);
2556 kvm_rip_write(vcpu, 0);
2557 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2559 vmcs_writel(GUEST_DR7, 0x400);
2561 vmcs_writel(GUEST_GDTR_BASE, 0);
2562 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2564 vmcs_writel(GUEST_IDTR_BASE, 0);
2565 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2567 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
2568 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2569 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2571 /* Special registers */
2572 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2576 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
2578 if (cpu_has_vmx_tpr_shadow()) {
2579 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2580 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2581 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2582 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2583 vmcs_write32(TPR_THRESHOLD, 0);
2586 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2587 vmcs_write64(APIC_ACCESS_ADDR,
2588 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2591 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2593 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
2594 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
2595 vmx_set_cr4(&vmx->vcpu, 0);
2596 vmx_set_efer(&vmx->vcpu, 0);
2597 vmx_fpu_activate(&vmx->vcpu);
2598 update_exception_bitmap(&vmx->vcpu);
2600 vpid_sync_vcpu_all(vmx);
2604 /* HACK: Don't enable emulation on guest boot/reset */
2605 vmx->emulation_required = 0;
2608 srcu_read_unlock(&vcpu->kvm->srcu, idx);
2612 static void enable_irq_window(struct kvm_vcpu *vcpu)
2614 u32 cpu_based_vm_exec_control;
2616 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2617 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2618 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2621 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2623 u32 cpu_based_vm_exec_control;
2625 if (!cpu_has_virtual_nmis()) {
2626 enable_irq_window(vcpu);
2630 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2631 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2632 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2635 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
2637 struct vcpu_vmx *vmx = to_vmx(vcpu);
2639 int irq = vcpu->arch.interrupt.nr;
2641 trace_kvm_inj_virq(irq);
2643 ++vcpu->stat.irq_injections;
2644 if (vmx->rmode.vm86_active) {
2645 vmx->rmode.irq.pending = true;
2646 vmx->rmode.irq.vector = irq;
2647 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2648 if (vcpu->arch.interrupt.soft)
2649 vmx->rmode.irq.rip +=
2650 vmx->vcpu.arch.event_exit_inst_len;
2651 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2652 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
2653 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2654 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2657 intr = irq | INTR_INFO_VALID_MASK;
2658 if (vcpu->arch.interrupt.soft) {
2659 intr |= INTR_TYPE_SOFT_INTR;
2660 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2661 vmx->vcpu.arch.event_exit_inst_len);
2663 intr |= INTR_TYPE_EXT_INTR;
2664 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
2667 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2669 struct vcpu_vmx *vmx = to_vmx(vcpu);
2671 if (!cpu_has_virtual_nmis()) {
2673 * Tracking the NMI-blocked state in software is built upon
2674 * finding the next open IRQ window. This, in turn, depends on
2675 * well-behaving guests: They have to keep IRQs disabled at
2676 * least as long as the NMI handler runs. Otherwise we may
2677 * cause NMI nesting, maybe breaking the guest. But as this is
2678 * highly unlikely, we can live with the residual risk.
2680 vmx->soft_vnmi_blocked = 1;
2681 vmx->vnmi_blocked_time = 0;
2684 ++vcpu->stat.nmi_injections;
2685 if (vmx->rmode.vm86_active) {
2686 vmx->rmode.irq.pending = true;
2687 vmx->rmode.irq.vector = NMI_VECTOR;
2688 vmx->rmode.irq.rip = kvm_rip_read(vcpu);
2689 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2690 NMI_VECTOR | INTR_TYPE_SOFT_INTR |
2691 INTR_INFO_VALID_MASK);
2692 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
2693 kvm_rip_write(vcpu, vmx->rmode.irq.rip - 1);
2696 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2697 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2700 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
2702 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2705 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2706 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_NMI));
2709 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
2711 if (!cpu_has_virtual_nmis())
2712 return to_vmx(vcpu)->soft_vnmi_blocked;
2714 return !!(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2715 GUEST_INTR_STATE_NMI);
2718 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
2720 struct vcpu_vmx *vmx = to_vmx(vcpu);
2722 if (!cpu_has_virtual_nmis()) {
2723 if (vmx->soft_vnmi_blocked != masked) {
2724 vmx->soft_vnmi_blocked = masked;
2725 vmx->vnmi_blocked_time = 0;
2729 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
2730 GUEST_INTR_STATE_NMI);
2732 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
2733 GUEST_INTR_STATE_NMI);
2737 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2739 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2740 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2741 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
2744 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2747 struct kvm_userspace_memory_region tss_mem = {
2748 .slot = TSS_PRIVATE_MEMSLOT,
2749 .guest_phys_addr = addr,
2750 .memory_size = PAGE_SIZE * 3,
2754 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2757 kvm->arch.tss_addr = addr;
2761 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2762 int vec, u32 err_code)
2765 * Instruction with address size override prefix opcode 0x67
2766 * Cause the #SS fault with 0 error code in VM86 mode.
2768 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
2769 if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DONE)
2772 * Forward all other exceptions that are valid in real mode.
2773 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2774 * the required debugging infrastructure rework.
2778 if (vcpu->guest_debug &
2779 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
2781 kvm_queue_exception(vcpu, vec);
2785 * Update instruction length as we may reinject the exception
2786 * from user space while in guest debugging mode.
2788 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
2789 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2790 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
2801 kvm_queue_exception(vcpu, vec);
2808 * Trigger machine check on the host. We assume all the MSRs are already set up
2809 * by the CPU and that we still run on the same CPU as the MCE occurred on.
2810 * We pass a fake environment to the machine check handler because we want
2811 * the guest to be always treated like user space, no matter what context
2812 * it used internally.
2814 static void kvm_machine_check(void)
2816 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
2817 struct pt_regs regs = {
2818 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
2819 .flags = X86_EFLAGS_IF,
2822 do_machine_check(®s, 0);
2826 static int handle_machine_check(struct kvm_vcpu *vcpu)
2828 /* already handled by vcpu_run */
2832 static int handle_exception(struct kvm_vcpu *vcpu)
2834 struct vcpu_vmx *vmx = to_vmx(vcpu);
2835 struct kvm_run *kvm_run = vcpu->run;
2836 u32 intr_info, ex_no, error_code;
2837 unsigned long cr2, rip, dr6;
2839 enum emulation_result er;
2841 vect_info = vmx->idt_vectoring_info;
2842 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2844 if (is_machine_check(intr_info))
2845 return handle_machine_check(vcpu);
2847 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
2848 !is_page_fault(intr_info)) {
2849 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2850 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
2851 vcpu->run->internal.ndata = 2;
2852 vcpu->run->internal.data[0] = vect_info;
2853 vcpu->run->internal.data[1] = intr_info;
2857 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
2858 return 1; /* already handled by vmx_vcpu_run() */
2860 if (is_no_device(intr_info)) {
2861 vmx_fpu_activate(vcpu);
2865 if (is_invalid_opcode(intr_info)) {
2866 er = emulate_instruction(vcpu, 0, 0, EMULTYPE_TRAP_UD);
2867 if (er != EMULATE_DONE)
2868 kvm_queue_exception(vcpu, UD_VECTOR);
2873 rip = kvm_rip_read(vcpu);
2874 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
2875 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
2876 if (is_page_fault(intr_info)) {
2877 /* EPT won't cause page fault directly */
2880 cr2 = vmcs_readl(EXIT_QUALIFICATION);
2881 trace_kvm_page_fault(cr2, error_code);
2883 if (kvm_event_needs_reinjection(vcpu))
2884 kvm_mmu_unprotect_page_virt(vcpu, cr2);
2885 return kvm_mmu_page_fault(vcpu, cr2, error_code);
2888 if (vmx->rmode.vm86_active &&
2889 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
2891 if (vcpu->arch.halt_request) {
2892 vcpu->arch.halt_request = 0;
2893 return kvm_emulate_halt(vcpu);
2898 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
2901 dr6 = vmcs_readl(EXIT_QUALIFICATION);
2902 if (!(vcpu->guest_debug &
2903 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
2904 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
2905 kvm_queue_exception(vcpu, DB_VECTOR);
2908 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
2909 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
2913 * Update instruction length as we may reinject #BP from
2914 * user space while in guest debugging mode. Reading it for
2915 * #DB as well causes no harm, it is not used in that case.
2917 vmx->vcpu.arch.event_exit_inst_len =
2918 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2919 kvm_run->exit_reason = KVM_EXIT_DEBUG;
2920 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
2921 kvm_run->debug.arch.exception = ex_no;
2924 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
2925 kvm_run->ex.exception = ex_no;
2926 kvm_run->ex.error_code = error_code;
2932 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
2934 ++vcpu->stat.irq_exits;
2938 static int handle_triple_fault(struct kvm_vcpu *vcpu)
2940 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
2944 static int handle_io(struct kvm_vcpu *vcpu)
2946 unsigned long exit_qualification;
2947 int size, in, string;
2950 ++vcpu->stat.io_exits;
2951 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2952 string = (exit_qualification & 16) != 0;
2955 if (emulate_instruction(vcpu, 0, 0, 0) == EMULATE_DO_MMIO)
2960 size = (exit_qualification & 7) + 1;
2961 in = (exit_qualification & 8) != 0;
2962 port = exit_qualification >> 16;
2964 skip_emulated_instruction(vcpu);
2965 return kvm_emulate_pio(vcpu, in, size, port);
2969 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
2972 * Patch in the VMCALL instruction:
2974 hypercall[0] = 0x0f;
2975 hypercall[1] = 0x01;
2976 hypercall[2] = 0xc1;
2979 static int handle_cr(struct kvm_vcpu *vcpu)
2981 unsigned long exit_qualification, val;
2985 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2986 cr = exit_qualification & 15;
2987 reg = (exit_qualification >> 8) & 15;
2988 switch ((exit_qualification >> 4) & 3) {
2989 case 0: /* mov to cr */
2990 val = kvm_register_read(vcpu, reg);
2991 trace_kvm_cr_write(cr, val);
2994 kvm_set_cr0(vcpu, val);
2995 skip_emulated_instruction(vcpu);
2998 kvm_set_cr3(vcpu, val);
2999 skip_emulated_instruction(vcpu);
3002 kvm_set_cr4(vcpu, val);
3003 skip_emulated_instruction(vcpu);
3006 u8 cr8_prev = kvm_get_cr8(vcpu);
3007 u8 cr8 = kvm_register_read(vcpu, reg);
3008 kvm_set_cr8(vcpu, cr8);
3009 skip_emulated_instruction(vcpu);
3010 if (irqchip_in_kernel(vcpu->kvm))
3012 if (cr8_prev <= cr8)
3014 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
3020 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3021 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
3022 skip_emulated_instruction(vcpu);
3023 vmx_fpu_activate(vcpu);
3025 case 1: /*mov from cr*/
3028 kvm_register_write(vcpu, reg, vcpu->arch.cr3);
3029 trace_kvm_cr_read(cr, vcpu->arch.cr3);
3030 skip_emulated_instruction(vcpu);
3033 val = kvm_get_cr8(vcpu);
3034 kvm_register_write(vcpu, reg, val);
3035 trace_kvm_cr_read(cr, val);
3036 skip_emulated_instruction(vcpu);
3041 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
3042 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
3043 kvm_lmsw(vcpu, val);
3045 skip_emulated_instruction(vcpu);
3050 vcpu->run->exit_reason = 0;
3051 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
3052 (int)(exit_qualification >> 4) & 3, cr);
3056 static int check_dr_alias(struct kvm_vcpu *vcpu)
3058 if (kvm_read_cr4_bits(vcpu, X86_CR4_DE)) {
3059 kvm_queue_exception(vcpu, UD_VECTOR);
3065 static int handle_dr(struct kvm_vcpu *vcpu)
3067 unsigned long exit_qualification;
3071 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
3072 if (!kvm_require_cpl(vcpu, 0))
3074 dr = vmcs_readl(GUEST_DR7);
3077 * As the vm-exit takes precedence over the debug trap, we
3078 * need to emulate the latter, either for the host or the
3079 * guest debugging itself.
3081 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
3082 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
3083 vcpu->run->debug.arch.dr7 = dr;
3084 vcpu->run->debug.arch.pc =
3085 vmcs_readl(GUEST_CS_BASE) +
3086 vmcs_readl(GUEST_RIP);
3087 vcpu->run->debug.arch.exception = DB_VECTOR;
3088 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
3091 vcpu->arch.dr7 &= ~DR7_GD;
3092 vcpu->arch.dr6 |= DR6_BD;
3093 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3094 kvm_queue_exception(vcpu, DB_VECTOR);
3099 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3100 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
3101 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
3102 if (exit_qualification & TYPE_MOV_FROM_DR) {
3105 val = vcpu->arch.db[dr];
3108 if (check_dr_alias(vcpu) < 0)
3112 val = vcpu->arch.dr6;
3115 if (check_dr_alias(vcpu) < 0)
3119 val = vcpu->arch.dr7;
3122 kvm_register_write(vcpu, reg, val);
3124 val = vcpu->arch.regs[reg];
3127 vcpu->arch.db[dr] = val;
3128 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP))
3129 vcpu->arch.eff_db[dr] = val;
3132 if (check_dr_alias(vcpu) < 0)
3136 if (val & 0xffffffff00000000ULL) {
3137 kvm_inject_gp(vcpu, 0);
3140 vcpu->arch.dr6 = (val & DR6_VOLATILE) | DR6_FIXED_1;
3143 if (check_dr_alias(vcpu) < 0)
3147 if (val & 0xffffffff00000000ULL) {
3148 kvm_inject_gp(vcpu, 0);
3151 vcpu->arch.dr7 = (val & DR7_VOLATILE) | DR7_FIXED_1;
3152 if (!(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)) {
3153 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3154 vcpu->arch.switch_db_regs =
3155 (val & DR7_BP_EN_MASK);
3160 skip_emulated_instruction(vcpu);
3164 static int handle_cpuid(struct kvm_vcpu *vcpu)
3166 kvm_emulate_cpuid(vcpu);
3170 static int handle_rdmsr(struct kvm_vcpu *vcpu)
3172 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3175 if (vmx_get_msr(vcpu, ecx, &data)) {
3176 trace_kvm_msr_read_ex(ecx);
3177 kvm_inject_gp(vcpu, 0);
3181 trace_kvm_msr_read(ecx, data);
3183 /* FIXME: handling of bits 32:63 of rax, rdx */
3184 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
3185 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
3186 skip_emulated_instruction(vcpu);
3190 static int handle_wrmsr(struct kvm_vcpu *vcpu)
3192 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3193 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
3194 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3196 if (vmx_set_msr(vcpu, ecx, data) != 0) {
3197 trace_kvm_msr_write_ex(ecx, data);
3198 kvm_inject_gp(vcpu, 0);
3202 trace_kvm_msr_write(ecx, data);
3203 skip_emulated_instruction(vcpu);
3207 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
3212 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
3214 u32 cpu_based_vm_exec_control;
3216 /* clear pending irq */
3217 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3218 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3219 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3221 ++vcpu->stat.irq_window_exits;
3224 * If the user space waits to inject interrupts, exit as soon as
3227 if (!irqchip_in_kernel(vcpu->kvm) &&
3228 vcpu->run->request_interrupt_window &&
3229 !kvm_cpu_has_interrupt(vcpu)) {
3230 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3236 static int handle_halt(struct kvm_vcpu *vcpu)
3238 skip_emulated_instruction(vcpu);
3239 return kvm_emulate_halt(vcpu);
3242 static int handle_vmcall(struct kvm_vcpu *vcpu)
3244 skip_emulated_instruction(vcpu);
3245 kvm_emulate_hypercall(vcpu);
3249 static int handle_vmx_insn(struct kvm_vcpu *vcpu)
3251 kvm_queue_exception(vcpu, UD_VECTOR);
3255 static int handle_invlpg(struct kvm_vcpu *vcpu)
3257 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3259 kvm_mmu_invlpg(vcpu, exit_qualification);
3260 skip_emulated_instruction(vcpu);
3264 static int handle_wbinvd(struct kvm_vcpu *vcpu)
3266 skip_emulated_instruction(vcpu);
3267 /* TODO: Add support for VT-d/pass-through device */
3271 static int handle_apic_access(struct kvm_vcpu *vcpu)
3273 unsigned long exit_qualification;
3274 enum emulation_result er;
3275 unsigned long offset;
3277 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3278 offset = exit_qualification & 0xffful;
3280 er = emulate_instruction(vcpu, 0, 0, 0);
3282 if (er != EMULATE_DONE) {
3284 "Fail to handle apic access vmexit! Offset is 0x%lx\n",
3291 static int handle_task_switch(struct kvm_vcpu *vcpu)
3293 struct vcpu_vmx *vmx = to_vmx(vcpu);
3294 unsigned long exit_qualification;
3296 int reason, type, idt_v;
3298 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3299 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
3301 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3303 reason = (u32)exit_qualification >> 30;
3304 if (reason == TASK_SWITCH_GATE && idt_v) {
3306 case INTR_TYPE_NMI_INTR:
3307 vcpu->arch.nmi_injected = false;
3308 if (cpu_has_virtual_nmis())
3309 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3310 GUEST_INTR_STATE_NMI);
3312 case INTR_TYPE_EXT_INTR:
3313 case INTR_TYPE_SOFT_INTR:
3314 kvm_clear_interrupt_queue(vcpu);
3316 case INTR_TYPE_HARD_EXCEPTION:
3317 case INTR_TYPE_SOFT_EXCEPTION:
3318 kvm_clear_exception_queue(vcpu);
3324 tss_selector = exit_qualification;
3326 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3327 type != INTR_TYPE_EXT_INTR &&
3328 type != INTR_TYPE_NMI_INTR))
3329 skip_emulated_instruction(vcpu);
3331 if (!kvm_task_switch(vcpu, tss_selector, reason))
3334 /* clear all local breakpoint enable flags */
3335 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3338 * TODO: What about debug traps on tss switch?
3339 * Are we supposed to inject them and update dr6?
3345 static int handle_ept_violation(struct kvm_vcpu *vcpu)
3347 unsigned long exit_qualification;
3351 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3353 if (exit_qualification & (1 << 6)) {
3354 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3358 gla_validity = (exit_qualification >> 7) & 0x3;
3359 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3360 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3361 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3362 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3363 vmcs_readl(GUEST_LINEAR_ADDRESS));
3364 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3365 (long unsigned int)exit_qualification);
3366 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3367 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
3371 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3372 trace_kvm_page_fault(gpa, exit_qualification);
3373 return kvm_mmu_page_fault(vcpu, gpa & PAGE_MASK, 0);
3376 static u64 ept_rsvd_mask(u64 spte, int level)
3381 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
3382 mask |= (1ULL << i);
3385 /* bits 7:3 reserved */
3387 else if (level == 2) {
3388 if (spte & (1ULL << 7))
3389 /* 2MB ref, bits 20:12 reserved */
3392 /* bits 6:3 reserved */
3399 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
3402 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
3404 /* 010b (write-only) */
3405 WARN_ON((spte & 0x7) == 0x2);
3407 /* 110b (write/execute) */
3408 WARN_ON((spte & 0x7) == 0x6);
3410 /* 100b (execute-only) and value not supported by logical processor */
3411 if (!cpu_has_vmx_ept_execute_only())
3412 WARN_ON((spte & 0x7) == 0x4);
3416 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
3418 if (rsvd_bits != 0) {
3419 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
3420 __func__, rsvd_bits);
3424 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
3425 u64 ept_mem_type = (spte & 0x38) >> 3;
3427 if (ept_mem_type == 2 || ept_mem_type == 3 ||
3428 ept_mem_type == 7) {
3429 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
3430 __func__, ept_mem_type);
3437 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
3443 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3445 printk(KERN_ERR "EPT: Misconfiguration.\n");
3446 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
3448 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
3450 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
3451 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
3453 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3454 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
3459 static int handle_nmi_window(struct kvm_vcpu *vcpu)
3461 u32 cpu_based_vm_exec_control;
3463 /* clear pending NMI */
3464 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3465 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3466 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3467 ++vcpu->stat.nmi_window_exits;
3472 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
3474 struct vcpu_vmx *vmx = to_vmx(vcpu);
3475 enum emulation_result err = EMULATE_DONE;
3478 while (!guest_state_valid(vcpu)) {
3479 err = emulate_instruction(vcpu, 0, 0, 0);
3481 if (err == EMULATE_DO_MMIO) {
3486 if (err != EMULATE_DONE) {
3487 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3488 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3489 vcpu->run->internal.ndata = 0;
3494 if (signal_pending(current))
3500 vmx->emulation_required = 0;
3506 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
3507 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
3509 static int handle_pause(struct kvm_vcpu *vcpu)
3511 skip_emulated_instruction(vcpu);
3512 kvm_vcpu_on_spin(vcpu);
3517 static int handle_invalid_op(struct kvm_vcpu *vcpu)
3519 kvm_queue_exception(vcpu, UD_VECTOR);
3524 * The exit handlers return 1 if the exit was handled fully and guest execution
3525 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
3526 * to be done to userspace and return 0.
3528 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
3529 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
3530 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
3531 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
3532 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
3533 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
3534 [EXIT_REASON_CR_ACCESS] = handle_cr,
3535 [EXIT_REASON_DR_ACCESS] = handle_dr,
3536 [EXIT_REASON_CPUID] = handle_cpuid,
3537 [EXIT_REASON_MSR_READ] = handle_rdmsr,
3538 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
3539 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
3540 [EXIT_REASON_HLT] = handle_halt,
3541 [EXIT_REASON_INVLPG] = handle_invlpg,
3542 [EXIT_REASON_VMCALL] = handle_vmcall,
3543 [EXIT_REASON_VMCLEAR] = handle_vmx_insn,
3544 [EXIT_REASON_VMLAUNCH] = handle_vmx_insn,
3545 [EXIT_REASON_VMPTRLD] = handle_vmx_insn,
3546 [EXIT_REASON_VMPTRST] = handle_vmx_insn,
3547 [EXIT_REASON_VMREAD] = handle_vmx_insn,
3548 [EXIT_REASON_VMRESUME] = handle_vmx_insn,
3549 [EXIT_REASON_VMWRITE] = handle_vmx_insn,
3550 [EXIT_REASON_VMOFF] = handle_vmx_insn,
3551 [EXIT_REASON_VMON] = handle_vmx_insn,
3552 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
3553 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
3554 [EXIT_REASON_WBINVD] = handle_wbinvd,
3555 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
3556 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
3557 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
3558 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
3559 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
3560 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_invalid_op,
3561 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_invalid_op,
3564 static const int kvm_vmx_max_exit_handlers =
3565 ARRAY_SIZE(kvm_vmx_exit_handlers);
3568 * The guest has exited. See if we can fix it or if we need userspace
3571 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
3573 struct vcpu_vmx *vmx = to_vmx(vcpu);
3574 u32 exit_reason = vmx->exit_reason;
3575 u32 vectoring_info = vmx->idt_vectoring_info;
3577 trace_kvm_exit(exit_reason, kvm_rip_read(vcpu));
3579 /* If guest state is invalid, start emulating */
3580 if (vmx->emulation_required && emulate_invalid_guest_state)
3581 return handle_invalid_guest_state(vcpu);
3583 /* Access CR3 don't cause VMExit in paging mode, so we need
3584 * to sync with guest real CR3. */
3585 if (enable_ept && is_paging(vcpu))
3586 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3588 if (unlikely(vmx->fail)) {
3589 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3590 vcpu->run->fail_entry.hardware_entry_failure_reason
3591 = vmcs_read32(VM_INSTRUCTION_ERROR);
3595 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3596 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3597 exit_reason != EXIT_REASON_EPT_VIOLATION &&
3598 exit_reason != EXIT_REASON_TASK_SWITCH))
3599 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3600 "(0x%x) and exit reason is 0x%x\n",
3601 __func__, vectoring_info, exit_reason);
3603 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3604 if (vmx_interrupt_allowed(vcpu)) {
3605 vmx->soft_vnmi_blocked = 0;
3606 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3607 vcpu->arch.nmi_pending) {
3609 * This CPU don't support us in finding the end of an
3610 * NMI-blocked window if the guest runs with IRQs
3611 * disabled. So we pull the trigger after 1 s of
3612 * futile waiting, but inform the user about this.
3614 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3615 "state on VCPU %d after 1 s timeout\n",
3616 __func__, vcpu->vcpu_id);
3617 vmx->soft_vnmi_blocked = 0;
3621 if (exit_reason < kvm_vmx_max_exit_handlers
3622 && kvm_vmx_exit_handlers[exit_reason])
3623 return kvm_vmx_exit_handlers[exit_reason](vcpu);
3625 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3626 vcpu->run->hw.hardware_exit_reason = exit_reason;
3631 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3633 if (irr == -1 || tpr < irr) {
3634 vmcs_write32(TPR_THRESHOLD, 0);
3638 vmcs_write32(TPR_THRESHOLD, irr);
3641 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3644 u32 idt_vectoring_info = vmx->idt_vectoring_info;
3648 bool idtv_info_valid;
3650 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3652 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
3654 /* Handle machine checks before interrupts are enabled */
3655 if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3656 || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3657 && is_machine_check(exit_intr_info)))
3658 kvm_machine_check();
3660 /* We need to handle NMIs before interrupts are enabled */
3661 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3662 (exit_intr_info & INTR_INFO_VALID_MASK))
3665 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3667 if (cpu_has_virtual_nmis()) {
3668 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3669 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3671 * SDM 3: 27.7.1.2 (September 2008)
3672 * Re-set bit "block by NMI" before VM entry if vmexit caused by
3673 * a guest IRET fault.
3674 * SDM 3: 23.2.2 (September 2008)
3675 * Bit 12 is undefined in any of the following cases:
3676 * If the VM exit sets the valid bit in the IDT-vectoring
3677 * information field.
3678 * If the VM exit is due to a double fault.
3680 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3681 vector != DF_VECTOR && !idtv_info_valid)
3682 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3683 GUEST_INTR_STATE_NMI);
3684 } else if (unlikely(vmx->soft_vnmi_blocked))
3685 vmx->vnmi_blocked_time +=
3686 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3688 vmx->vcpu.arch.nmi_injected = false;
3689 kvm_clear_exception_queue(&vmx->vcpu);
3690 kvm_clear_interrupt_queue(&vmx->vcpu);
3692 if (!idtv_info_valid)
3695 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3696 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3699 case INTR_TYPE_NMI_INTR:
3700 vmx->vcpu.arch.nmi_injected = true;
3702 * SDM 3: 27.7.1.2 (September 2008)
3703 * Clear bit "block by NMI" before VM entry if a NMI
3706 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3707 GUEST_INTR_STATE_NMI);
3709 case INTR_TYPE_SOFT_EXCEPTION:
3710 vmx->vcpu.arch.event_exit_inst_len =
3711 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3713 case INTR_TYPE_HARD_EXCEPTION:
3714 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3715 u32 err = vmcs_read32(IDT_VECTORING_ERROR_CODE);
3716 kvm_queue_exception_e(&vmx->vcpu, vector, err);
3718 kvm_queue_exception(&vmx->vcpu, vector);
3720 case INTR_TYPE_SOFT_INTR:
3721 vmx->vcpu.arch.event_exit_inst_len =
3722 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3724 case INTR_TYPE_EXT_INTR:
3725 kvm_queue_interrupt(&vmx->vcpu, vector,
3726 type == INTR_TYPE_SOFT_INTR);
3734 * Failure to inject an interrupt should give us the information
3735 * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs
3736 * when fetching the interrupt redirection bitmap in the real-mode
3737 * tss, this doesn't happen. So we do it ourselves.
3739 static void fixup_rmode_irq(struct vcpu_vmx *vmx)
3741 vmx->rmode.irq.pending = 0;
3742 if (kvm_rip_read(&vmx->vcpu) + 1 != vmx->rmode.irq.rip)
3744 kvm_rip_write(&vmx->vcpu, vmx->rmode.irq.rip);
3745 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
3746 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
3747 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
3750 vmx->idt_vectoring_info =
3751 VECTORING_INFO_VALID_MASK
3752 | INTR_TYPE_EXT_INTR
3753 | vmx->rmode.irq.vector;
3756 #ifdef CONFIG_X86_64
3764 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
3766 struct vcpu_vmx *vmx = to_vmx(vcpu);
3768 /* Record the guest's net vcpu time for enforced NMI injections. */
3769 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3770 vmx->entry_time = ktime_get();
3772 /* Don't enter VMX if guest state is invalid, let the exit handler
3773 start emulation until we arrive back to a valid state */
3774 if (vmx->emulation_required && emulate_invalid_guest_state)
3777 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3778 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3779 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3780 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3782 /* When single-stepping over STI and MOV SS, we must clear the
3783 * corresponding interruptibility bits in the guest state. Otherwise
3784 * vmentry fails as it then expects bit 14 (BS) in pending debug
3785 * exceptions being set, but that's not correct for the guest debugging
3787 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3788 vmx_set_interrupt_shadow(vcpu, 0);
3791 * Loading guest fpu may have cleared host cr0.ts
3793 vmcs_writel(HOST_CR0, read_cr0());
3796 /* Store host registers */
3797 "push %%"R"dx; push %%"R"bp;"
3799 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3801 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
3802 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
3804 /* Reload cr2 if changed */
3805 "mov %c[cr2](%0), %%"R"ax \n\t"
3806 "mov %%cr2, %%"R"dx \n\t"
3807 "cmp %%"R"ax, %%"R"dx \n\t"
3809 "mov %%"R"ax, %%cr2 \n\t"
3811 /* Check if vmlaunch of vmresume is needed */
3812 "cmpl $0, %c[launched](%0) \n\t"
3813 /* Load guest registers. Don't clobber flags. */
3814 "mov %c[rax](%0), %%"R"ax \n\t"
3815 "mov %c[rbx](%0), %%"R"bx \n\t"
3816 "mov %c[rdx](%0), %%"R"dx \n\t"
3817 "mov %c[rsi](%0), %%"R"si \n\t"
3818 "mov %c[rdi](%0), %%"R"di \n\t"
3819 "mov %c[rbp](%0), %%"R"bp \n\t"
3820 #ifdef CONFIG_X86_64
3821 "mov %c[r8](%0), %%r8 \n\t"
3822 "mov %c[r9](%0), %%r9 \n\t"
3823 "mov %c[r10](%0), %%r10 \n\t"
3824 "mov %c[r11](%0), %%r11 \n\t"
3825 "mov %c[r12](%0), %%r12 \n\t"
3826 "mov %c[r13](%0), %%r13 \n\t"
3827 "mov %c[r14](%0), %%r14 \n\t"
3828 "mov %c[r15](%0), %%r15 \n\t"
3830 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
3832 /* Enter guest mode */
3833 "jne .Llaunched \n\t"
3834 __ex(ASM_VMX_VMLAUNCH) "\n\t"
3835 "jmp .Lkvm_vmx_return \n\t"
3836 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
3837 ".Lkvm_vmx_return: "
3838 /* Save guest registers, load host registers, keep flags */
3839 "xchg %0, (%%"R"sp) \n\t"
3840 "mov %%"R"ax, %c[rax](%0) \n\t"
3841 "mov %%"R"bx, %c[rbx](%0) \n\t"
3842 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
3843 "mov %%"R"dx, %c[rdx](%0) \n\t"
3844 "mov %%"R"si, %c[rsi](%0) \n\t"
3845 "mov %%"R"di, %c[rdi](%0) \n\t"
3846 "mov %%"R"bp, %c[rbp](%0) \n\t"
3847 #ifdef CONFIG_X86_64
3848 "mov %%r8, %c[r8](%0) \n\t"
3849 "mov %%r9, %c[r9](%0) \n\t"
3850 "mov %%r10, %c[r10](%0) \n\t"
3851 "mov %%r11, %c[r11](%0) \n\t"
3852 "mov %%r12, %c[r12](%0) \n\t"
3853 "mov %%r13, %c[r13](%0) \n\t"
3854 "mov %%r14, %c[r14](%0) \n\t"
3855 "mov %%r15, %c[r15](%0) \n\t"
3857 "mov %%cr2, %%"R"ax \n\t"
3858 "mov %%"R"ax, %c[cr2](%0) \n\t"
3860 "pop %%"R"bp; pop %%"R"bp; pop %%"R"dx \n\t"
3861 "setbe %c[fail](%0) \n\t"
3862 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
3863 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
3864 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
3865 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
3866 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
3867 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
3868 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
3869 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
3870 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
3871 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
3872 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
3873 #ifdef CONFIG_X86_64
3874 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
3875 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
3876 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
3877 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
3878 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
3879 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
3880 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
3881 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
3883 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
3885 , R"bx", R"di", R"si"
3886 #ifdef CONFIG_X86_64
3887 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
3891 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
3892 | (1 << VCPU_EXREG_PDPTR));
3893 vcpu->arch.regs_dirty = 0;
3895 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
3896 if (vmx->rmode.irq.pending)
3897 fixup_rmode_irq(vmx);
3899 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
3902 vmx_complete_interrupts(vmx);
3908 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
3910 struct vcpu_vmx *vmx = to_vmx(vcpu);
3914 free_vmcs(vmx->vmcs);
3919 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
3921 struct vcpu_vmx *vmx = to_vmx(vcpu);
3923 spin_lock(&vmx_vpid_lock);
3925 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3926 spin_unlock(&vmx_vpid_lock);
3927 vmx_free_vmcs(vcpu);
3928 kfree(vmx->guest_msrs);
3929 kvm_vcpu_uninit(vcpu);
3930 kmem_cache_free(kvm_vcpu_cache, vmx);
3933 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
3936 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
3940 return ERR_PTR(-ENOMEM);
3944 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
3948 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
3949 if (!vmx->guest_msrs) {
3954 vmx->vmcs = alloc_vmcs();
3958 vmcs_clear(vmx->vmcs);
3961 vmx_vcpu_load(&vmx->vcpu, cpu);
3962 err = vmx_vcpu_setup(vmx);
3963 vmx_vcpu_put(&vmx->vcpu);
3967 if (vm_need_virtualize_apic_accesses(kvm))
3968 if (alloc_apic_access_page(kvm) != 0)
3972 if (!kvm->arch.ept_identity_map_addr)
3973 kvm->arch.ept_identity_map_addr =
3974 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
3975 if (alloc_identity_pagetable(kvm) != 0)
3982 free_vmcs(vmx->vmcs);
3984 kfree(vmx->guest_msrs);
3986 kvm_vcpu_uninit(&vmx->vcpu);
3988 kmem_cache_free(kvm_vcpu_cache, vmx);
3989 return ERR_PTR(err);
3992 static void __init vmx_check_processor_compat(void *rtn)
3994 struct vmcs_config vmcs_conf;
3997 if (setup_vmcs_config(&vmcs_conf) < 0)
3999 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
4000 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
4001 smp_processor_id());
4006 static int get_ept_level(void)
4008 return VMX_EPT_DEFAULT_GAW + 1;
4011 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
4015 /* For VT-d and EPT combination
4016 * 1. MMIO: always map as UC
4018 * a. VT-d without snooping control feature: can't guarantee the
4019 * result, try to trust guest.
4020 * b. VT-d with snooping control feature: snooping control feature of
4021 * VT-d engine can guarantee the cache correctness. Just set it
4022 * to WB to keep consistent with host. So the same as item 3.
4023 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
4024 * consistent with host MTRR
4027 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
4028 else if (vcpu->kvm->arch.iommu_domain &&
4029 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
4030 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
4031 VMX_EPT_MT_EPTE_SHIFT;
4033 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
4039 #define _ER(x) { EXIT_REASON_##x, #x }
4041 static const struct trace_print_flags vmx_exit_reasons_str[] = {
4043 _ER(EXTERNAL_INTERRUPT),
4045 _ER(PENDING_INTERRUPT),
4065 _ER(IO_INSTRUCTION),
4068 _ER(MWAIT_INSTRUCTION),
4069 _ER(MONITOR_INSTRUCTION),
4070 _ER(PAUSE_INSTRUCTION),
4071 _ER(MCE_DURING_VMENTRY),
4072 _ER(TPR_BELOW_THRESHOLD),
4082 static int vmx_get_lpage_level(void)
4084 if (enable_ept && !cpu_has_vmx_ept_1g_page())
4085 return PT_DIRECTORY_LEVEL;
4087 /* For shadow and EPT supported 1GB page */
4088 return PT_PDPE_LEVEL;
4091 static inline u32 bit(int bitno)
4093 return 1 << (bitno & 31);
4096 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
4098 struct kvm_cpuid_entry2 *best;
4099 struct vcpu_vmx *vmx = to_vmx(vcpu);
4102 vmx->rdtscp_enabled = false;
4103 if (vmx_rdtscp_supported()) {
4104 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
4105 if (exec_control & SECONDARY_EXEC_RDTSCP) {
4106 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
4107 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
4108 vmx->rdtscp_enabled = true;
4110 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4111 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4118 static struct kvm_x86_ops vmx_x86_ops = {
4119 .cpu_has_kvm_support = cpu_has_kvm_support,
4120 .disabled_by_bios = vmx_disabled_by_bios,
4121 .hardware_setup = hardware_setup,
4122 .hardware_unsetup = hardware_unsetup,
4123 .check_processor_compatibility = vmx_check_processor_compat,
4124 .hardware_enable = hardware_enable,
4125 .hardware_disable = hardware_disable,
4126 .cpu_has_accelerated_tpr = report_flexpriority,
4128 .vcpu_create = vmx_create_vcpu,
4129 .vcpu_free = vmx_free_vcpu,
4130 .vcpu_reset = vmx_vcpu_reset,
4132 .prepare_guest_switch = vmx_save_host_state,
4133 .vcpu_load = vmx_vcpu_load,
4134 .vcpu_put = vmx_vcpu_put,
4136 .set_guest_debug = set_guest_debug,
4137 .get_msr = vmx_get_msr,
4138 .set_msr = vmx_set_msr,
4139 .get_segment_base = vmx_get_segment_base,
4140 .get_segment = vmx_get_segment,
4141 .set_segment = vmx_set_segment,
4142 .get_cpl = vmx_get_cpl,
4143 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
4144 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
4145 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
4146 .set_cr0 = vmx_set_cr0,
4147 .set_cr3 = vmx_set_cr3,
4148 .set_cr4 = vmx_set_cr4,
4149 .set_efer = vmx_set_efer,
4150 .get_idt = vmx_get_idt,
4151 .set_idt = vmx_set_idt,
4152 .get_gdt = vmx_get_gdt,
4153 .set_gdt = vmx_set_gdt,
4154 .cache_reg = vmx_cache_reg,
4155 .get_rflags = vmx_get_rflags,
4156 .set_rflags = vmx_set_rflags,
4157 .fpu_activate = vmx_fpu_activate,
4158 .fpu_deactivate = vmx_fpu_deactivate,
4160 .tlb_flush = vmx_flush_tlb,
4162 .run = vmx_vcpu_run,
4163 .handle_exit = vmx_handle_exit,
4164 .skip_emulated_instruction = skip_emulated_instruction,
4165 .set_interrupt_shadow = vmx_set_interrupt_shadow,
4166 .get_interrupt_shadow = vmx_get_interrupt_shadow,
4167 .patch_hypercall = vmx_patch_hypercall,
4168 .set_irq = vmx_inject_irq,
4169 .set_nmi = vmx_inject_nmi,
4170 .queue_exception = vmx_queue_exception,
4171 .interrupt_allowed = vmx_interrupt_allowed,
4172 .nmi_allowed = vmx_nmi_allowed,
4173 .get_nmi_mask = vmx_get_nmi_mask,
4174 .set_nmi_mask = vmx_set_nmi_mask,
4175 .enable_nmi_window = enable_nmi_window,
4176 .enable_irq_window = enable_irq_window,
4177 .update_cr8_intercept = update_cr8_intercept,
4179 .set_tss_addr = vmx_set_tss_addr,
4180 .get_tdp_level = get_ept_level,
4181 .get_mt_mask = vmx_get_mt_mask,
4183 .exit_reasons_str = vmx_exit_reasons_str,
4184 .get_lpage_level = vmx_get_lpage_level,
4186 .cpuid_update = vmx_cpuid_update,
4188 .rdtscp_supported = vmx_rdtscp_supported,
4191 static int __init vmx_init(void)
4195 rdmsrl_safe(MSR_EFER, &host_efer);
4197 for (i = 0; i < NR_VMX_MSR; ++i)
4198 kvm_define_shared_msr(i, vmx_msr_index[i]);
4200 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
4201 if (!vmx_io_bitmap_a)
4204 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
4205 if (!vmx_io_bitmap_b) {
4210 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
4211 if (!vmx_msr_bitmap_legacy) {
4216 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
4217 if (!vmx_msr_bitmap_longmode) {
4223 * Allow direct access to the PC debug port (it is often used for I/O
4224 * delays, but the vmexits simply slow things down).
4226 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
4227 clear_bit(0x80, vmx_io_bitmap_a);
4229 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
4231 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
4232 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
4234 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
4236 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
4240 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
4241 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
4242 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
4243 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
4244 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
4245 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
4248 bypass_guest_pf = 0;
4249 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK |
4250 VMX_EPT_WRITABLE_MASK);
4251 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4252 VMX_EPT_EXECUTABLE_MASK);
4257 if (bypass_guest_pf)
4258 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
4263 free_page((unsigned long)vmx_msr_bitmap_longmode);
4265 free_page((unsigned long)vmx_msr_bitmap_legacy);
4267 free_page((unsigned long)vmx_io_bitmap_b);
4269 free_page((unsigned long)vmx_io_bitmap_a);
4273 static void __exit vmx_exit(void)
4275 free_page((unsigned long)vmx_msr_bitmap_legacy);
4276 free_page((unsigned long)vmx_msr_bitmap_longmode);
4277 free_page((unsigned long)vmx_io_bitmap_b);
4278 free_page((unsigned long)vmx_io_bitmap_a);
4283 module_init(vmx_init)
4284 module_exit(vmx_exit)