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.
8 * Copyright 2010 Red Hat, Inc. and/or its affilates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
23 #include <linux/module.h>
24 #include <linux/kernel.h>
26 #include <linux/highmem.h>
27 #include <linux/sched.h>
28 #include <linux/moduleparam.h>
29 #include <linux/ftrace_event.h>
30 #include <linux/slab.h>
31 #include <linux/tboot.h>
32 #include "kvm_cache_regs.h"
38 #include <asm/virtext.h>
45 #define __ex(x) __kvm_handle_fault_on_reboot(x)
47 MODULE_AUTHOR("Qumranet");
48 MODULE_LICENSE("GPL");
50 static int __read_mostly bypass_guest_pf
= 1;
51 module_param(bypass_guest_pf
, bool, S_IRUGO
);
53 static int __read_mostly enable_vpid
= 1;
54 module_param_named(vpid
, enable_vpid
, bool, 0444);
56 static int __read_mostly flexpriority_enabled
= 1;
57 module_param_named(flexpriority
, flexpriority_enabled
, bool, S_IRUGO
);
59 static int __read_mostly enable_ept
= 1;
60 module_param_named(ept
, enable_ept
, bool, S_IRUGO
);
62 static int __read_mostly enable_unrestricted_guest
= 1;
63 module_param_named(unrestricted_guest
,
64 enable_unrestricted_guest
, bool, S_IRUGO
);
66 static int __read_mostly emulate_invalid_guest_state
= 0;
67 module_param(emulate_invalid_guest_state
, bool, S_IRUGO
);
69 static int __read_mostly vmm_exclusive
= 1;
70 module_param(vmm_exclusive
, bool, S_IRUGO
);
72 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST \
73 (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
74 #define KVM_GUEST_CR0_MASK \
75 (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
76 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST \
77 (X86_CR0_WP | X86_CR0_NE)
78 #define KVM_VM_CR0_ALWAYS_ON \
79 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
80 #define KVM_CR4_GUEST_OWNED_BITS \
81 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
84 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
85 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
87 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
90 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
91 * ple_gap: upper bound on the amount of time between two successive
92 * executions of PAUSE in a loop. Also indicate if ple enabled.
93 * According to test, this time is usually small than 41 cycles.
94 * ple_window: upper bound on the amount of time a guest is allowed to execute
95 * in a PAUSE loop. Tests indicate that most spinlocks are held for
96 * less than 2^12 cycles
97 * Time is measured based on a counter that runs at the same rate as the TSC,
98 * refer SDM volume 3b section 21.6.13 & 22.1.3.
100 #define KVM_VMX_DEFAULT_PLE_GAP 41
101 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
102 static int ple_gap
= KVM_VMX_DEFAULT_PLE_GAP
;
103 module_param(ple_gap
, int, S_IRUGO
);
105 static int ple_window
= KVM_VMX_DEFAULT_PLE_WINDOW
;
106 module_param(ple_window
, int, S_IRUGO
);
108 #define NR_AUTOLOAD_MSRS 1
116 struct shared_msr_entry
{
123 struct kvm_vcpu vcpu
;
124 struct list_head local_vcpus_link
;
125 unsigned long host_rsp
;
128 u32 idt_vectoring_info
;
129 struct shared_msr_entry
*guest_msrs
;
133 u64 msr_host_kernel_gs_base
;
134 u64 msr_guest_kernel_gs_base
;
137 struct msr_autoload
{
139 struct vmx_msr_entry guest
[NR_AUTOLOAD_MSRS
];
140 struct vmx_msr_entry host
[NR_AUTOLOAD_MSRS
];
144 u16 fs_sel
, gs_sel
, ldt_sel
;
145 int gs_ldt_reload_needed
;
146 int fs_reload_needed
;
151 struct kvm_save_segment
{
156 } tr
, es
, ds
, fs
, gs
;
164 bool emulation_required
;
166 /* Support for vnmi-less CPUs */
167 int soft_vnmi_blocked
;
169 s64 vnmi_blocked_time
;
175 static inline struct vcpu_vmx
*to_vmx(struct kvm_vcpu
*vcpu
)
177 return container_of(vcpu
, struct vcpu_vmx
, vcpu
);
180 static int init_rmode(struct kvm
*kvm
);
181 static u64
construct_eptp(unsigned long root_hpa
);
182 static void kvm_cpu_vmxon(u64 addr
);
183 static void kvm_cpu_vmxoff(void);
185 static DEFINE_PER_CPU(struct vmcs
*, vmxarea
);
186 static DEFINE_PER_CPU(struct vmcs
*, current_vmcs
);
187 static DEFINE_PER_CPU(struct list_head
, vcpus_on_cpu
);
188 static DEFINE_PER_CPU(struct desc_ptr
, host_gdt
);
190 static unsigned long *vmx_io_bitmap_a
;
191 static unsigned long *vmx_io_bitmap_b
;
192 static unsigned long *vmx_msr_bitmap_legacy
;
193 static unsigned long *vmx_msr_bitmap_longmode
;
195 static DECLARE_BITMAP(vmx_vpid_bitmap
, VMX_NR_VPIDS
);
196 static DEFINE_SPINLOCK(vmx_vpid_lock
);
198 static struct vmcs_config
{
202 u32 pin_based_exec_ctrl
;
203 u32 cpu_based_exec_ctrl
;
204 u32 cpu_based_2nd_exec_ctrl
;
209 static struct vmx_capability
{
214 #define VMX_SEGMENT_FIELD(seg) \
215 [VCPU_SREG_##seg] = { \
216 .selector = GUEST_##seg##_SELECTOR, \
217 .base = GUEST_##seg##_BASE, \
218 .limit = GUEST_##seg##_LIMIT, \
219 .ar_bytes = GUEST_##seg##_AR_BYTES, \
222 static struct kvm_vmx_segment_field
{
227 } kvm_vmx_segment_fields
[] = {
228 VMX_SEGMENT_FIELD(CS
),
229 VMX_SEGMENT_FIELD(DS
),
230 VMX_SEGMENT_FIELD(ES
),
231 VMX_SEGMENT_FIELD(FS
),
232 VMX_SEGMENT_FIELD(GS
),
233 VMX_SEGMENT_FIELD(SS
),
234 VMX_SEGMENT_FIELD(TR
),
235 VMX_SEGMENT_FIELD(LDTR
),
238 static u64 host_efer
;
240 static void ept_save_pdptrs(struct kvm_vcpu
*vcpu
);
243 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
244 * away by decrementing the array size.
246 static const u32 vmx_msr_index
[] = {
248 MSR_SYSCALL_MASK
, MSR_LSTAR
, MSR_CSTAR
,
250 MSR_EFER
, MSR_TSC_AUX
, MSR_STAR
,
252 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
254 static inline bool is_page_fault(u32 intr_info
)
256 return (intr_info
& (INTR_INFO_INTR_TYPE_MASK
| INTR_INFO_VECTOR_MASK
|
257 INTR_INFO_VALID_MASK
)) ==
258 (INTR_TYPE_HARD_EXCEPTION
| PF_VECTOR
| INTR_INFO_VALID_MASK
);
261 static inline bool is_no_device(u32 intr_info
)
263 return (intr_info
& (INTR_INFO_INTR_TYPE_MASK
| INTR_INFO_VECTOR_MASK
|
264 INTR_INFO_VALID_MASK
)) ==
265 (INTR_TYPE_HARD_EXCEPTION
| NM_VECTOR
| INTR_INFO_VALID_MASK
);
268 static inline bool is_invalid_opcode(u32 intr_info
)
270 return (intr_info
& (INTR_INFO_INTR_TYPE_MASK
| INTR_INFO_VECTOR_MASK
|
271 INTR_INFO_VALID_MASK
)) ==
272 (INTR_TYPE_HARD_EXCEPTION
| UD_VECTOR
| INTR_INFO_VALID_MASK
);
275 static inline bool is_external_interrupt(u32 intr_info
)
277 return (intr_info
& (INTR_INFO_INTR_TYPE_MASK
| INTR_INFO_VALID_MASK
))
278 == (INTR_TYPE_EXT_INTR
| INTR_INFO_VALID_MASK
);
281 static inline bool is_machine_check(u32 intr_info
)
283 return (intr_info
& (INTR_INFO_INTR_TYPE_MASK
| INTR_INFO_VECTOR_MASK
|
284 INTR_INFO_VALID_MASK
)) ==
285 (INTR_TYPE_HARD_EXCEPTION
| MC_VECTOR
| INTR_INFO_VALID_MASK
);
288 static inline bool cpu_has_vmx_msr_bitmap(void)
290 return vmcs_config
.cpu_based_exec_ctrl
& CPU_BASED_USE_MSR_BITMAPS
;
293 static inline bool cpu_has_vmx_tpr_shadow(void)
295 return vmcs_config
.cpu_based_exec_ctrl
& CPU_BASED_TPR_SHADOW
;
298 static inline bool vm_need_tpr_shadow(struct kvm
*kvm
)
300 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm
));
303 static inline bool cpu_has_secondary_exec_ctrls(void)
305 return vmcs_config
.cpu_based_exec_ctrl
&
306 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS
;
309 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
311 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
312 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
;
315 static inline bool cpu_has_vmx_flexpriority(void)
317 return cpu_has_vmx_tpr_shadow() &&
318 cpu_has_vmx_virtualize_apic_accesses();
321 static inline bool cpu_has_vmx_ept_execute_only(void)
323 return vmx_capability
.ept
& VMX_EPT_EXECUTE_ONLY_BIT
;
326 static inline bool cpu_has_vmx_eptp_uncacheable(void)
328 return vmx_capability
.ept
& VMX_EPTP_UC_BIT
;
331 static inline bool cpu_has_vmx_eptp_writeback(void)
333 return vmx_capability
.ept
& VMX_EPTP_WB_BIT
;
336 static inline bool cpu_has_vmx_ept_2m_page(void)
338 return vmx_capability
.ept
& VMX_EPT_2MB_PAGE_BIT
;
341 static inline bool cpu_has_vmx_ept_1g_page(void)
343 return vmx_capability
.ept
& VMX_EPT_1GB_PAGE_BIT
;
346 static inline bool cpu_has_vmx_ept_4levels(void)
348 return vmx_capability
.ept
& VMX_EPT_PAGE_WALK_4_BIT
;
351 static inline bool cpu_has_vmx_invept_individual_addr(void)
353 return vmx_capability
.ept
& VMX_EPT_EXTENT_INDIVIDUAL_BIT
;
356 static inline bool cpu_has_vmx_invept_context(void)
358 return vmx_capability
.ept
& VMX_EPT_EXTENT_CONTEXT_BIT
;
361 static inline bool cpu_has_vmx_invept_global(void)
363 return vmx_capability
.ept
& VMX_EPT_EXTENT_GLOBAL_BIT
;
366 static inline bool cpu_has_vmx_invvpid_single(void)
368 return vmx_capability
.vpid
& VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT
;
371 static inline bool cpu_has_vmx_invvpid_global(void)
373 return vmx_capability
.vpid
& VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT
;
376 static inline bool cpu_has_vmx_ept(void)
378 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
379 SECONDARY_EXEC_ENABLE_EPT
;
382 static inline bool cpu_has_vmx_unrestricted_guest(void)
384 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
385 SECONDARY_EXEC_UNRESTRICTED_GUEST
;
388 static inline bool cpu_has_vmx_ple(void)
390 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
391 SECONDARY_EXEC_PAUSE_LOOP_EXITING
;
394 static inline bool vm_need_virtualize_apic_accesses(struct kvm
*kvm
)
396 return flexpriority_enabled
&& irqchip_in_kernel(kvm
);
399 static inline bool cpu_has_vmx_vpid(void)
401 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
402 SECONDARY_EXEC_ENABLE_VPID
;
405 static inline bool cpu_has_vmx_rdtscp(void)
407 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
408 SECONDARY_EXEC_RDTSCP
;
411 static inline bool cpu_has_virtual_nmis(void)
413 return vmcs_config
.pin_based_exec_ctrl
& PIN_BASED_VIRTUAL_NMIS
;
416 static inline bool cpu_has_vmx_wbinvd_exit(void)
418 return vmcs_config
.cpu_based_2nd_exec_ctrl
&
419 SECONDARY_EXEC_WBINVD_EXITING
;
422 static inline bool report_flexpriority(void)
424 return flexpriority_enabled
;
427 static int __find_msr_index(struct vcpu_vmx
*vmx
, u32 msr
)
431 for (i
= 0; i
< vmx
->nmsrs
; ++i
)
432 if (vmx_msr_index
[vmx
->guest_msrs
[i
].index
] == msr
)
437 static inline void __invvpid(int ext
, u16 vpid
, gva_t gva
)
443 } operand
= { vpid
, 0, gva
};
445 asm volatile (__ex(ASM_VMX_INVVPID
)
446 /* CF==1 or ZF==1 --> rc = -1 */
448 : : "a"(&operand
), "c"(ext
) : "cc", "memory");
451 static inline void __invept(int ext
, u64 eptp
, gpa_t gpa
)
455 } operand
= {eptp
, gpa
};
457 asm volatile (__ex(ASM_VMX_INVEPT
)
458 /* CF==1 or ZF==1 --> rc = -1 */
459 "; ja 1f ; ud2 ; 1:\n"
460 : : "a" (&operand
), "c" (ext
) : "cc", "memory");
463 static struct shared_msr_entry
*find_msr_entry(struct vcpu_vmx
*vmx
, u32 msr
)
467 i
= __find_msr_index(vmx
, msr
);
469 return &vmx
->guest_msrs
[i
];
473 static void vmcs_clear(struct vmcs
*vmcs
)
475 u64 phys_addr
= __pa(vmcs
);
478 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX
) "; setna %0"
479 : "=g"(error
) : "a"(&phys_addr
), "m"(phys_addr
)
482 printk(KERN_ERR
"kvm: vmclear fail: %p/%llx\n",
486 static void vmcs_load(struct vmcs
*vmcs
)
488 u64 phys_addr
= __pa(vmcs
);
491 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX
) "; setna %0"
492 : "=g"(error
) : "a"(&phys_addr
), "m"(phys_addr
)
495 printk(KERN_ERR
"kvm: vmptrld %p/%llx fail\n",
499 static void __vcpu_clear(void *arg
)
501 struct vcpu_vmx
*vmx
= arg
;
502 int cpu
= raw_smp_processor_id();
504 if (vmx
->vcpu
.cpu
== cpu
)
505 vmcs_clear(vmx
->vmcs
);
506 if (per_cpu(current_vmcs
, cpu
) == vmx
->vmcs
)
507 per_cpu(current_vmcs
, cpu
) = NULL
;
508 rdtscll(vmx
->vcpu
.arch
.host_tsc
);
509 list_del(&vmx
->local_vcpus_link
);
514 static void vcpu_clear(struct vcpu_vmx
*vmx
)
516 if (vmx
->vcpu
.cpu
== -1)
518 smp_call_function_single(vmx
->vcpu
.cpu
, __vcpu_clear
, vmx
, 1);
521 static inline void vpid_sync_vcpu_single(struct vcpu_vmx
*vmx
)
526 if (cpu_has_vmx_invvpid_single())
527 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT
, vmx
->vpid
, 0);
530 static inline void vpid_sync_vcpu_global(void)
532 if (cpu_has_vmx_invvpid_global())
533 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT
, 0, 0);
536 static inline void vpid_sync_context(struct vcpu_vmx
*vmx
)
538 if (cpu_has_vmx_invvpid_single())
539 vpid_sync_vcpu_single(vmx
);
541 vpid_sync_vcpu_global();
544 static inline void ept_sync_global(void)
546 if (cpu_has_vmx_invept_global())
547 __invept(VMX_EPT_EXTENT_GLOBAL
, 0, 0);
550 static inline void ept_sync_context(u64 eptp
)
553 if (cpu_has_vmx_invept_context())
554 __invept(VMX_EPT_EXTENT_CONTEXT
, eptp
, 0);
560 static inline void ept_sync_individual_addr(u64 eptp
, gpa_t gpa
)
563 if (cpu_has_vmx_invept_individual_addr())
564 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR
,
567 ept_sync_context(eptp
);
571 static unsigned long vmcs_readl(unsigned long field
)
575 asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX
)
576 : "=a"(value
) : "d"(field
) : "cc");
580 static u16
vmcs_read16(unsigned long field
)
582 return vmcs_readl(field
);
585 static u32
vmcs_read32(unsigned long field
)
587 return vmcs_readl(field
);
590 static u64
vmcs_read64(unsigned long field
)
593 return vmcs_readl(field
);
595 return vmcs_readl(field
) | ((u64
)vmcs_readl(field
+1) << 32);
599 static noinline
void vmwrite_error(unsigned long field
, unsigned long value
)
601 printk(KERN_ERR
"vmwrite error: reg %lx value %lx (err %d)\n",
602 field
, value
, vmcs_read32(VM_INSTRUCTION_ERROR
));
606 static void vmcs_writel(unsigned long field
, unsigned long value
)
610 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX
) "; setna %0"
611 : "=q"(error
) : "a"(value
), "d"(field
) : "cc");
613 vmwrite_error(field
, value
);
616 static void vmcs_write16(unsigned long field
, u16 value
)
618 vmcs_writel(field
, value
);
621 static void vmcs_write32(unsigned long field
, u32 value
)
623 vmcs_writel(field
, value
);
626 static void vmcs_write64(unsigned long field
, u64 value
)
628 vmcs_writel(field
, value
);
629 #ifndef CONFIG_X86_64
631 vmcs_writel(field
+1, value
>> 32);
635 static void vmcs_clear_bits(unsigned long field
, u32 mask
)
637 vmcs_writel(field
, vmcs_readl(field
) & ~mask
);
640 static void vmcs_set_bits(unsigned long field
, u32 mask
)
642 vmcs_writel(field
, vmcs_readl(field
) | mask
);
645 static void update_exception_bitmap(struct kvm_vcpu
*vcpu
)
649 eb
= (1u << PF_VECTOR
) | (1u << UD_VECTOR
) | (1u << MC_VECTOR
) |
650 (1u << NM_VECTOR
) | (1u << DB_VECTOR
);
651 if ((vcpu
->guest_debug
&
652 (KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_SW_BP
)) ==
653 (KVM_GUESTDBG_ENABLE
| KVM_GUESTDBG_USE_SW_BP
))
654 eb
|= 1u << BP_VECTOR
;
655 if (to_vmx(vcpu
)->rmode
.vm86_active
)
658 eb
&= ~(1u << PF_VECTOR
); /* bypass_guest_pf = 0 */
659 if (vcpu
->fpu_active
)
660 eb
&= ~(1u << NM_VECTOR
);
661 vmcs_write32(EXCEPTION_BITMAP
, eb
);
664 static void clear_atomic_switch_msr(struct vcpu_vmx
*vmx
, unsigned msr
)
667 struct msr_autoload
*m
= &vmx
->msr_autoload
;
669 for (i
= 0; i
< m
->nr
; ++i
)
670 if (m
->guest
[i
].index
== msr
)
676 m
->guest
[i
] = m
->guest
[m
->nr
];
677 m
->host
[i
] = m
->host
[m
->nr
];
678 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT
, m
->nr
);
679 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT
, m
->nr
);
682 static void add_atomic_switch_msr(struct vcpu_vmx
*vmx
, unsigned msr
,
683 u64 guest_val
, u64 host_val
)
686 struct msr_autoload
*m
= &vmx
->msr_autoload
;
688 for (i
= 0; i
< m
->nr
; ++i
)
689 if (m
->guest
[i
].index
== msr
)
694 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT
, m
->nr
);
695 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT
, m
->nr
);
698 m
->guest
[i
].index
= msr
;
699 m
->guest
[i
].value
= guest_val
;
700 m
->host
[i
].index
= msr
;
701 m
->host
[i
].value
= host_val
;
704 static void reload_tss(void)
707 * VT restores TR but not its size. Useless.
709 struct desc_ptr
*gdt
= &__get_cpu_var(host_gdt
);
710 struct desc_struct
*descs
;
712 descs
= (void *)gdt
->address
;
713 descs
[GDT_ENTRY_TSS
].type
= 9; /* available TSS */
717 static bool update_transition_efer(struct vcpu_vmx
*vmx
, int efer_offset
)
722 guest_efer
= vmx
->vcpu
.arch
.efer
;
725 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
728 ignore_bits
= EFER_NX
| EFER_SCE
;
730 ignore_bits
|= EFER_LMA
| EFER_LME
;
731 /* SCE is meaningful only in long mode on Intel */
732 if (guest_efer
& EFER_LMA
)
733 ignore_bits
&= ~(u64
)EFER_SCE
;
735 guest_efer
&= ~ignore_bits
;
736 guest_efer
|= host_efer
& ignore_bits
;
737 vmx
->guest_msrs
[efer_offset
].data
= guest_efer
;
738 vmx
->guest_msrs
[efer_offset
].mask
= ~ignore_bits
;
740 clear_atomic_switch_msr(vmx
, MSR_EFER
);
741 /* On ept, can't emulate nx, and must switch nx atomically */
742 if (enable_ept
&& ((vmx
->vcpu
.arch
.efer
^ host_efer
) & EFER_NX
)) {
743 guest_efer
= vmx
->vcpu
.arch
.efer
;
744 if (!(guest_efer
& EFER_LMA
))
745 guest_efer
&= ~EFER_LME
;
746 add_atomic_switch_msr(vmx
, MSR_EFER
, guest_efer
, host_efer
);
753 static unsigned long segment_base(u16 selector
)
755 struct desc_ptr
*gdt
= &__get_cpu_var(host_gdt
);
756 struct desc_struct
*d
;
757 unsigned long table_base
;
760 if (!(selector
& ~3))
763 table_base
= gdt
->address
;
765 if (selector
& 4) { /* from ldt */
766 u16 ldt_selector
= kvm_read_ldt();
768 if (!(ldt_selector
& ~3))
771 table_base
= segment_base(ldt_selector
);
773 d
= (struct desc_struct
*)(table_base
+ (selector
& ~7));
774 v
= get_desc_base(d
);
776 if (d
->s
== 0 && (d
->type
== 2 || d
->type
== 9 || d
->type
== 11))
777 v
|= ((unsigned long)((struct ldttss_desc64
*)d
)->base3
) << 32;
782 static inline unsigned long kvm_read_tr_base(void)
785 asm("str %0" : "=g"(tr
));
786 return segment_base(tr
);
789 static void vmx_save_host_state(struct kvm_vcpu
*vcpu
)
791 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
794 if (vmx
->host_state
.loaded
)
797 vmx
->host_state
.loaded
= 1;
799 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
800 * allow segment selectors with cpl > 0 or ti == 1.
802 vmx
->host_state
.ldt_sel
= kvm_read_ldt();
803 vmx
->host_state
.gs_ldt_reload_needed
= vmx
->host_state
.ldt_sel
;
804 savesegment(fs
, vmx
->host_state
.fs_sel
);
805 if (!(vmx
->host_state
.fs_sel
& 7)) {
806 vmcs_write16(HOST_FS_SELECTOR
, vmx
->host_state
.fs_sel
);
807 vmx
->host_state
.fs_reload_needed
= 0;
809 vmcs_write16(HOST_FS_SELECTOR
, 0);
810 vmx
->host_state
.fs_reload_needed
= 1;
812 savesegment(gs
, vmx
->host_state
.gs_sel
);
813 if (!(vmx
->host_state
.gs_sel
& 7))
814 vmcs_write16(HOST_GS_SELECTOR
, vmx
->host_state
.gs_sel
);
816 vmcs_write16(HOST_GS_SELECTOR
, 0);
817 vmx
->host_state
.gs_ldt_reload_needed
= 1;
821 vmcs_writel(HOST_FS_BASE
, read_msr(MSR_FS_BASE
));
822 vmcs_writel(HOST_GS_BASE
, read_msr(MSR_GS_BASE
));
824 vmcs_writel(HOST_FS_BASE
, segment_base(vmx
->host_state
.fs_sel
));
825 vmcs_writel(HOST_GS_BASE
, segment_base(vmx
->host_state
.gs_sel
));
829 if (is_long_mode(&vmx
->vcpu
)) {
830 rdmsrl(MSR_KERNEL_GS_BASE
, vmx
->msr_host_kernel_gs_base
);
831 wrmsrl(MSR_KERNEL_GS_BASE
, vmx
->msr_guest_kernel_gs_base
);
834 for (i
= 0; i
< vmx
->save_nmsrs
; ++i
)
835 kvm_set_shared_msr(vmx
->guest_msrs
[i
].index
,
836 vmx
->guest_msrs
[i
].data
,
837 vmx
->guest_msrs
[i
].mask
);
840 static void __vmx_load_host_state(struct vcpu_vmx
*vmx
)
842 if (!vmx
->host_state
.loaded
)
845 ++vmx
->vcpu
.stat
.host_state_reload
;
846 vmx
->host_state
.loaded
= 0;
847 if (vmx
->host_state
.fs_reload_needed
)
848 loadsegment(fs
, vmx
->host_state
.fs_sel
);
849 if (vmx
->host_state
.gs_ldt_reload_needed
) {
850 kvm_load_ldt(vmx
->host_state
.ldt_sel
);
852 load_gs_index(vmx
->host_state
.gs_sel
);
853 wrmsrl(MSR_KERNEL_GS_BASE
, current
->thread
.gs
);
855 loadsegment(gs
, vmx
->host_state
.gs_sel
);
860 if (is_long_mode(&vmx
->vcpu
)) {
861 rdmsrl(MSR_KERNEL_GS_BASE
, vmx
->msr_guest_kernel_gs_base
);
862 wrmsrl(MSR_KERNEL_GS_BASE
, vmx
->msr_host_kernel_gs_base
);
865 if (current_thread_info()->status
& TS_USEDFPU
)
867 load_gdt(&__get_cpu_var(host_gdt
));
870 static void vmx_load_host_state(struct vcpu_vmx
*vmx
)
873 __vmx_load_host_state(vmx
);
878 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
879 * vcpu mutex is already taken.
881 static void vmx_vcpu_load(struct kvm_vcpu
*vcpu
, int cpu
)
883 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
884 u64 tsc_this
, delta
, new_offset
;
885 u64 phys_addr
= __pa(per_cpu(vmxarea
, cpu
));
888 kvm_cpu_vmxon(phys_addr
);
889 else if (vcpu
->cpu
!= cpu
)
892 if (per_cpu(current_vmcs
, cpu
) != vmx
->vmcs
) {
893 per_cpu(current_vmcs
, cpu
) = vmx
->vmcs
;
894 vmcs_load(vmx
->vmcs
);
897 if (vcpu
->cpu
!= cpu
) {
898 struct desc_ptr
*gdt
= &__get_cpu_var(host_gdt
);
899 unsigned long sysenter_esp
;
901 kvm_migrate_timers(vcpu
);
902 kvm_make_request(KVM_REQ_TLB_FLUSH
, vcpu
);
904 list_add(&vmx
->local_vcpus_link
,
905 &per_cpu(vcpus_on_cpu
, cpu
));
910 * Linux uses per-cpu TSS and GDT, so set these when switching
913 vmcs_writel(HOST_TR_BASE
, kvm_read_tr_base()); /* 22.2.4 */
914 vmcs_writel(HOST_GDTR_BASE
, gdt
->address
); /* 22.2.4 */
916 rdmsrl(MSR_IA32_SYSENTER_ESP
, sysenter_esp
);
917 vmcs_writel(HOST_IA32_SYSENTER_ESP
, sysenter_esp
); /* 22.2.3 */
920 * Make sure the time stamp counter is monotonous.
923 if (tsc_this
< vcpu
->arch
.host_tsc
) {
924 delta
= vcpu
->arch
.host_tsc
- tsc_this
;
925 new_offset
= vmcs_read64(TSC_OFFSET
) + delta
;
926 vmcs_write64(TSC_OFFSET
, new_offset
);
931 static void vmx_vcpu_put(struct kvm_vcpu
*vcpu
)
933 __vmx_load_host_state(to_vmx(vcpu
));
934 if (!vmm_exclusive
) {
935 __vcpu_clear(to_vmx(vcpu
));
940 static void vmx_fpu_activate(struct kvm_vcpu
*vcpu
)
944 if (vcpu
->fpu_active
)
946 vcpu
->fpu_active
= 1;
947 cr0
= vmcs_readl(GUEST_CR0
);
948 cr0
&= ~(X86_CR0_TS
| X86_CR0_MP
);
949 cr0
|= kvm_read_cr0_bits(vcpu
, X86_CR0_TS
| X86_CR0_MP
);
950 vmcs_writel(GUEST_CR0
, cr0
);
951 update_exception_bitmap(vcpu
);
952 vcpu
->arch
.cr0_guest_owned_bits
= X86_CR0_TS
;
953 vmcs_writel(CR0_GUEST_HOST_MASK
, ~vcpu
->arch
.cr0_guest_owned_bits
);
956 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu
*vcpu
);
958 static void vmx_fpu_deactivate(struct kvm_vcpu
*vcpu
)
960 vmx_decache_cr0_guest_bits(vcpu
);
961 vmcs_set_bits(GUEST_CR0
, X86_CR0_TS
| X86_CR0_MP
);
962 update_exception_bitmap(vcpu
);
963 vcpu
->arch
.cr0_guest_owned_bits
= 0;
964 vmcs_writel(CR0_GUEST_HOST_MASK
, ~vcpu
->arch
.cr0_guest_owned_bits
);
965 vmcs_writel(CR0_READ_SHADOW
, vcpu
->arch
.cr0
);
968 static unsigned long vmx_get_rflags(struct kvm_vcpu
*vcpu
)
970 unsigned long rflags
, save_rflags
;
972 rflags
= vmcs_readl(GUEST_RFLAGS
);
973 if (to_vmx(vcpu
)->rmode
.vm86_active
) {
974 rflags
&= RMODE_GUEST_OWNED_EFLAGS_BITS
;
975 save_rflags
= to_vmx(vcpu
)->rmode
.save_rflags
;
976 rflags
|= save_rflags
& ~RMODE_GUEST_OWNED_EFLAGS_BITS
;
981 static void vmx_set_rflags(struct kvm_vcpu
*vcpu
, unsigned long rflags
)
983 if (to_vmx(vcpu
)->rmode
.vm86_active
) {
984 to_vmx(vcpu
)->rmode
.save_rflags
= rflags
;
985 rflags
|= X86_EFLAGS_IOPL
| X86_EFLAGS_VM
;
987 vmcs_writel(GUEST_RFLAGS
, rflags
);
990 static u32
vmx_get_interrupt_shadow(struct kvm_vcpu
*vcpu
, int mask
)
992 u32 interruptibility
= vmcs_read32(GUEST_INTERRUPTIBILITY_INFO
);
995 if (interruptibility
& GUEST_INTR_STATE_STI
)
996 ret
|= KVM_X86_SHADOW_INT_STI
;
997 if (interruptibility
& GUEST_INTR_STATE_MOV_SS
)
998 ret
|= KVM_X86_SHADOW_INT_MOV_SS
;
1003 static void vmx_set_interrupt_shadow(struct kvm_vcpu
*vcpu
, int mask
)
1005 u32 interruptibility_old
= vmcs_read32(GUEST_INTERRUPTIBILITY_INFO
);
1006 u32 interruptibility
= interruptibility_old
;
1008 interruptibility
&= ~(GUEST_INTR_STATE_STI
| GUEST_INTR_STATE_MOV_SS
);
1010 if (mask
& KVM_X86_SHADOW_INT_MOV_SS
)
1011 interruptibility
|= GUEST_INTR_STATE_MOV_SS
;
1012 else if (mask
& KVM_X86_SHADOW_INT_STI
)
1013 interruptibility
|= GUEST_INTR_STATE_STI
;
1015 if ((interruptibility
!= interruptibility_old
))
1016 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO
, interruptibility
);
1019 static void skip_emulated_instruction(struct kvm_vcpu
*vcpu
)
1023 rip
= kvm_rip_read(vcpu
);
1024 rip
+= vmcs_read32(VM_EXIT_INSTRUCTION_LEN
);
1025 kvm_rip_write(vcpu
, rip
);
1027 /* skipping an emulated instruction also counts */
1028 vmx_set_interrupt_shadow(vcpu
, 0);
1031 static void vmx_queue_exception(struct kvm_vcpu
*vcpu
, unsigned nr
,
1032 bool has_error_code
, u32 error_code
,
1035 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
1036 u32 intr_info
= nr
| INTR_INFO_VALID_MASK
;
1038 if (has_error_code
) {
1039 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE
, error_code
);
1040 intr_info
|= INTR_INFO_DELIVER_CODE_MASK
;
1043 if (vmx
->rmode
.vm86_active
) {
1044 vmx
->rmode
.irq
.pending
= true;
1045 vmx
->rmode
.irq
.vector
= nr
;
1046 vmx
->rmode
.irq
.rip
= kvm_rip_read(vcpu
);
1047 if (kvm_exception_is_soft(nr
))
1048 vmx
->rmode
.irq
.rip
+=
1049 vmx
->vcpu
.arch
.event_exit_inst_len
;
1050 intr_info
|= INTR_TYPE_SOFT_INTR
;
1051 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD
, intr_info
);
1052 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN
, 1);
1053 kvm_rip_write(vcpu
, vmx
->rmode
.irq
.rip
- 1);
1057 if (kvm_exception_is_soft(nr
)) {
1058 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN
,
1059 vmx
->vcpu
.arch
.event_exit_inst_len
);
1060 intr_info
|= INTR_TYPE_SOFT_EXCEPTION
;
1062 intr_info
|= INTR_TYPE_HARD_EXCEPTION
;
1064 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD
, intr_info
);
1067 static bool vmx_rdtscp_supported(void)
1069 return cpu_has_vmx_rdtscp();
1073 * Swap MSR entry in host/guest MSR entry array.
1075 static void move_msr_up(struct vcpu_vmx
*vmx
, int from
, int to
)
1077 struct shared_msr_entry tmp
;
1079 tmp
= vmx
->guest_msrs
[to
];
1080 vmx
->guest_msrs
[to
] = vmx
->guest_msrs
[from
];
1081 vmx
->guest_msrs
[from
] = tmp
;
1085 * Set up the vmcs to automatically save and restore system
1086 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1087 * mode, as fiddling with msrs is very expensive.
1089 static void setup_msrs(struct vcpu_vmx
*vmx
)
1091 int save_nmsrs
, index
;
1092 unsigned long *msr_bitmap
;
1094 vmx_load_host_state(vmx
);
1096 #ifdef CONFIG_X86_64
1097 if (is_long_mode(&vmx
->vcpu
)) {
1098 index
= __find_msr_index(vmx
, MSR_SYSCALL_MASK
);
1100 move_msr_up(vmx
, index
, save_nmsrs
++);
1101 index
= __find_msr_index(vmx
, MSR_LSTAR
);
1103 move_msr_up(vmx
, index
, save_nmsrs
++);
1104 index
= __find_msr_index(vmx
, MSR_CSTAR
);
1106 move_msr_up(vmx
, index
, save_nmsrs
++);
1107 index
= __find_msr_index(vmx
, MSR_TSC_AUX
);
1108 if (index
>= 0 && vmx
->rdtscp_enabled
)
1109 move_msr_up(vmx
, index
, save_nmsrs
++);
1111 * MSR_STAR is only needed on long mode guests, and only
1112 * if efer.sce is enabled.
1114 index
= __find_msr_index(vmx
, MSR_STAR
);
1115 if ((index
>= 0) && (vmx
->vcpu
.arch
.efer
& EFER_SCE
))
1116 move_msr_up(vmx
, index
, save_nmsrs
++);
1119 index
= __find_msr_index(vmx
, MSR_EFER
);
1120 if (index
>= 0 && update_transition_efer(vmx
, index
))
1121 move_msr_up(vmx
, index
, save_nmsrs
++);
1123 vmx
->save_nmsrs
= save_nmsrs
;
1125 if (cpu_has_vmx_msr_bitmap()) {
1126 if (is_long_mode(&vmx
->vcpu
))
1127 msr_bitmap
= vmx_msr_bitmap_longmode
;
1129 msr_bitmap
= vmx_msr_bitmap_legacy
;
1131 vmcs_write64(MSR_BITMAP
, __pa(msr_bitmap
));
1136 * reads and returns guest's timestamp counter "register"
1137 * guest_tsc = host_tsc + tsc_offset -- 21.3
1139 static u64
guest_read_tsc(void)
1141 u64 host_tsc
, tsc_offset
;
1144 tsc_offset
= vmcs_read64(TSC_OFFSET
);
1145 return host_tsc
+ tsc_offset
;
1149 * writes 'offset' into guest's timestamp counter offset register
1151 static void vmx_write_tsc_offset(struct kvm_vcpu
*vcpu
, u64 offset
)
1153 vmcs_write64(TSC_OFFSET
, offset
);
1157 * Reads an msr value (of 'msr_index') into 'pdata'.
1158 * Returns 0 on success, non-0 otherwise.
1159 * Assumes vcpu_load() was already called.
1161 static int vmx_get_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64
*pdata
)
1164 struct shared_msr_entry
*msr
;
1167 printk(KERN_ERR
"BUG: get_msr called with NULL pdata\n");
1171 switch (msr_index
) {
1172 #ifdef CONFIG_X86_64
1174 data
= vmcs_readl(GUEST_FS_BASE
);
1177 data
= vmcs_readl(GUEST_GS_BASE
);
1179 case MSR_KERNEL_GS_BASE
:
1180 vmx_load_host_state(to_vmx(vcpu
));
1181 data
= to_vmx(vcpu
)->msr_guest_kernel_gs_base
;
1185 return kvm_get_msr_common(vcpu
, msr_index
, pdata
);
1187 data
= guest_read_tsc();
1189 case MSR_IA32_SYSENTER_CS
:
1190 data
= vmcs_read32(GUEST_SYSENTER_CS
);
1192 case MSR_IA32_SYSENTER_EIP
:
1193 data
= vmcs_readl(GUEST_SYSENTER_EIP
);
1195 case MSR_IA32_SYSENTER_ESP
:
1196 data
= vmcs_readl(GUEST_SYSENTER_ESP
);
1199 if (!to_vmx(vcpu
)->rdtscp_enabled
)
1201 /* Otherwise falls through */
1203 vmx_load_host_state(to_vmx(vcpu
));
1204 msr
= find_msr_entry(to_vmx(vcpu
), msr_index
);
1206 vmx_load_host_state(to_vmx(vcpu
));
1210 return kvm_get_msr_common(vcpu
, msr_index
, pdata
);
1218 * Writes msr value into into the appropriate "register".
1219 * Returns 0 on success, non-0 otherwise.
1220 * Assumes vcpu_load() was already called.
1222 static int vmx_set_msr(struct kvm_vcpu
*vcpu
, u32 msr_index
, u64 data
)
1224 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
1225 struct shared_msr_entry
*msr
;
1228 switch (msr_index
) {
1230 vmx_load_host_state(vmx
);
1231 ret
= kvm_set_msr_common(vcpu
, msr_index
, data
);
1233 #ifdef CONFIG_X86_64
1235 vmcs_writel(GUEST_FS_BASE
, data
);
1238 vmcs_writel(GUEST_GS_BASE
, data
);
1240 case MSR_KERNEL_GS_BASE
:
1241 vmx_load_host_state(vmx
);
1242 vmx
->msr_guest_kernel_gs_base
= data
;
1245 case MSR_IA32_SYSENTER_CS
:
1246 vmcs_write32(GUEST_SYSENTER_CS
, data
);
1248 case MSR_IA32_SYSENTER_EIP
:
1249 vmcs_writel(GUEST_SYSENTER_EIP
, data
);
1251 case MSR_IA32_SYSENTER_ESP
:
1252 vmcs_writel(GUEST_SYSENTER_ESP
, data
);
1255 kvm_write_tsc(vcpu
, data
);
1257 case MSR_IA32_CR_PAT
:
1258 if (vmcs_config
.vmentry_ctrl
& VM_ENTRY_LOAD_IA32_PAT
) {
1259 vmcs_write64(GUEST_IA32_PAT
, data
);
1260 vcpu
->arch
.pat
= data
;
1263 ret
= kvm_set_msr_common(vcpu
, msr_index
, data
);
1266 if (!vmx
->rdtscp_enabled
)
1268 /* Check reserved bit, higher 32 bits should be zero */
1269 if ((data
>> 32) != 0)
1271 /* Otherwise falls through */
1273 msr
= find_msr_entry(vmx
, msr_index
);
1275 vmx_load_host_state(vmx
);
1279 ret
= kvm_set_msr_common(vcpu
, msr_index
, data
);
1285 static void vmx_cache_reg(struct kvm_vcpu
*vcpu
, enum kvm_reg reg
)
1287 __set_bit(reg
, (unsigned long *)&vcpu
->arch
.regs_avail
);
1290 vcpu
->arch
.regs
[VCPU_REGS_RSP
] = vmcs_readl(GUEST_RSP
);
1293 vcpu
->arch
.regs
[VCPU_REGS_RIP
] = vmcs_readl(GUEST_RIP
);
1295 case VCPU_EXREG_PDPTR
:
1297 ept_save_pdptrs(vcpu
);
1304 static void set_guest_debug(struct kvm_vcpu
*vcpu
, struct kvm_guest_debug
*dbg
)
1306 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
)
1307 vmcs_writel(GUEST_DR7
, dbg
->arch
.debugreg
[7]);
1309 vmcs_writel(GUEST_DR7
, vcpu
->arch
.dr7
);
1311 update_exception_bitmap(vcpu
);
1314 static __init
int cpu_has_kvm_support(void)
1316 return cpu_has_vmx();
1319 static __init
int vmx_disabled_by_bios(void)
1323 rdmsrl(MSR_IA32_FEATURE_CONTROL
, msr
);
1324 if (msr
& FEATURE_CONTROL_LOCKED
) {
1325 if (!(msr
& FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX
)
1328 if (!(msr
& FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX
)
1329 && !tboot_enabled())
1334 /* locked but not enabled */
1337 static void kvm_cpu_vmxon(u64 addr
)
1339 asm volatile (ASM_VMX_VMXON_RAX
1340 : : "a"(&addr
), "m"(addr
)
1344 static int hardware_enable(void *garbage
)
1346 int cpu
= raw_smp_processor_id();
1347 u64 phys_addr
= __pa(per_cpu(vmxarea
, cpu
));
1350 if (read_cr4() & X86_CR4_VMXE
)
1353 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu
, cpu
));
1354 rdmsrl(MSR_IA32_FEATURE_CONTROL
, old
);
1356 test_bits
= FEATURE_CONTROL_LOCKED
;
1357 test_bits
|= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX
;
1358 if (tboot_enabled())
1359 test_bits
|= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX
;
1361 if ((old
& test_bits
) != test_bits
) {
1362 /* enable and lock */
1363 wrmsrl(MSR_IA32_FEATURE_CONTROL
, old
| test_bits
);
1365 write_cr4(read_cr4() | X86_CR4_VMXE
); /* FIXME: not cpu hotplug safe */
1367 if (vmm_exclusive
) {
1368 kvm_cpu_vmxon(phys_addr
);
1372 store_gdt(&__get_cpu_var(host_gdt
));
1377 static void vmclear_local_vcpus(void)
1379 int cpu
= raw_smp_processor_id();
1380 struct vcpu_vmx
*vmx
, *n
;
1382 list_for_each_entry_safe(vmx
, n
, &per_cpu(vcpus_on_cpu
, cpu
),
1388 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1391 static void kvm_cpu_vmxoff(void)
1393 asm volatile (__ex(ASM_VMX_VMXOFF
) : : : "cc");
1396 static void hardware_disable(void *garbage
)
1398 if (vmm_exclusive
) {
1399 vmclear_local_vcpus();
1402 write_cr4(read_cr4() & ~X86_CR4_VMXE
);
1405 static __init
int adjust_vmx_controls(u32 ctl_min
, u32 ctl_opt
,
1406 u32 msr
, u32
*result
)
1408 u32 vmx_msr_low
, vmx_msr_high
;
1409 u32 ctl
= ctl_min
| ctl_opt
;
1411 rdmsr(msr
, vmx_msr_low
, vmx_msr_high
);
1413 ctl
&= vmx_msr_high
; /* bit == 0 in high word ==> must be zero */
1414 ctl
|= vmx_msr_low
; /* bit == 1 in low word ==> must be one */
1416 /* Ensure minimum (required) set of control bits are supported. */
1424 static __init
int setup_vmcs_config(struct vmcs_config
*vmcs_conf
)
1426 u32 vmx_msr_low
, vmx_msr_high
;
1427 u32 min
, opt
, min2
, opt2
;
1428 u32 _pin_based_exec_control
= 0;
1429 u32 _cpu_based_exec_control
= 0;
1430 u32 _cpu_based_2nd_exec_control
= 0;
1431 u32 _vmexit_control
= 0;
1432 u32 _vmentry_control
= 0;
1434 min
= PIN_BASED_EXT_INTR_MASK
| PIN_BASED_NMI_EXITING
;
1435 opt
= PIN_BASED_VIRTUAL_NMIS
;
1436 if (adjust_vmx_controls(min
, opt
, MSR_IA32_VMX_PINBASED_CTLS
,
1437 &_pin_based_exec_control
) < 0)
1440 min
= CPU_BASED_HLT_EXITING
|
1441 #ifdef CONFIG_X86_64
1442 CPU_BASED_CR8_LOAD_EXITING
|
1443 CPU_BASED_CR8_STORE_EXITING
|
1445 CPU_BASED_CR3_LOAD_EXITING
|
1446 CPU_BASED_CR3_STORE_EXITING
|
1447 CPU_BASED_USE_IO_BITMAPS
|
1448 CPU_BASED_MOV_DR_EXITING
|
1449 CPU_BASED_USE_TSC_OFFSETING
|
1450 CPU_BASED_MWAIT_EXITING
|
1451 CPU_BASED_MONITOR_EXITING
|
1452 CPU_BASED_INVLPG_EXITING
;
1453 opt
= CPU_BASED_TPR_SHADOW
|
1454 CPU_BASED_USE_MSR_BITMAPS
|
1455 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS
;
1456 if (adjust_vmx_controls(min
, opt
, MSR_IA32_VMX_PROCBASED_CTLS
,
1457 &_cpu_based_exec_control
) < 0)
1459 #ifdef CONFIG_X86_64
1460 if ((_cpu_based_exec_control
& CPU_BASED_TPR_SHADOW
))
1461 _cpu_based_exec_control
&= ~CPU_BASED_CR8_LOAD_EXITING
&
1462 ~CPU_BASED_CR8_STORE_EXITING
;
1464 if (_cpu_based_exec_control
& CPU_BASED_ACTIVATE_SECONDARY_CONTROLS
) {
1466 opt2
= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
|
1467 SECONDARY_EXEC_WBINVD_EXITING
|
1468 SECONDARY_EXEC_ENABLE_VPID
|
1469 SECONDARY_EXEC_ENABLE_EPT
|
1470 SECONDARY_EXEC_UNRESTRICTED_GUEST
|
1471 SECONDARY_EXEC_PAUSE_LOOP_EXITING
|
1472 SECONDARY_EXEC_RDTSCP
;
1473 if (adjust_vmx_controls(min2
, opt2
,
1474 MSR_IA32_VMX_PROCBASED_CTLS2
,
1475 &_cpu_based_2nd_exec_control
) < 0)
1478 #ifndef CONFIG_X86_64
1479 if (!(_cpu_based_2nd_exec_control
&
1480 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
))
1481 _cpu_based_exec_control
&= ~CPU_BASED_TPR_SHADOW
;
1483 if (_cpu_based_2nd_exec_control
& SECONDARY_EXEC_ENABLE_EPT
) {
1484 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1486 _cpu_based_exec_control
&= ~(CPU_BASED_CR3_LOAD_EXITING
|
1487 CPU_BASED_CR3_STORE_EXITING
|
1488 CPU_BASED_INVLPG_EXITING
);
1489 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP
,
1490 vmx_capability
.ept
, vmx_capability
.vpid
);
1494 #ifdef CONFIG_X86_64
1495 min
|= VM_EXIT_HOST_ADDR_SPACE_SIZE
;
1497 opt
= VM_EXIT_SAVE_IA32_PAT
| VM_EXIT_LOAD_IA32_PAT
;
1498 if (adjust_vmx_controls(min
, opt
, MSR_IA32_VMX_EXIT_CTLS
,
1499 &_vmexit_control
) < 0)
1503 opt
= VM_ENTRY_LOAD_IA32_PAT
;
1504 if (adjust_vmx_controls(min
, opt
, MSR_IA32_VMX_ENTRY_CTLS
,
1505 &_vmentry_control
) < 0)
1508 rdmsr(MSR_IA32_VMX_BASIC
, vmx_msr_low
, vmx_msr_high
);
1510 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1511 if ((vmx_msr_high
& 0x1fff) > PAGE_SIZE
)
1514 #ifdef CONFIG_X86_64
1515 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1516 if (vmx_msr_high
& (1u<<16))
1520 /* Require Write-Back (WB) memory type for VMCS accesses. */
1521 if (((vmx_msr_high
>> 18) & 15) != 6)
1524 vmcs_conf
->size
= vmx_msr_high
& 0x1fff;
1525 vmcs_conf
->order
= get_order(vmcs_config
.size
);
1526 vmcs_conf
->revision_id
= vmx_msr_low
;
1528 vmcs_conf
->pin_based_exec_ctrl
= _pin_based_exec_control
;
1529 vmcs_conf
->cpu_based_exec_ctrl
= _cpu_based_exec_control
;
1530 vmcs_conf
->cpu_based_2nd_exec_ctrl
= _cpu_based_2nd_exec_control
;
1531 vmcs_conf
->vmexit_ctrl
= _vmexit_control
;
1532 vmcs_conf
->vmentry_ctrl
= _vmentry_control
;
1537 static struct vmcs
*alloc_vmcs_cpu(int cpu
)
1539 int node
= cpu_to_node(cpu
);
1543 pages
= alloc_pages_exact_node(node
, GFP_KERNEL
, vmcs_config
.order
);
1546 vmcs
= page_address(pages
);
1547 memset(vmcs
, 0, vmcs_config
.size
);
1548 vmcs
->revision_id
= vmcs_config
.revision_id
; /* vmcs revision id */
1552 static struct vmcs
*alloc_vmcs(void)
1554 return alloc_vmcs_cpu(raw_smp_processor_id());
1557 static void free_vmcs(struct vmcs
*vmcs
)
1559 free_pages((unsigned long)vmcs
, vmcs_config
.order
);
1562 static void free_kvm_area(void)
1566 for_each_possible_cpu(cpu
) {
1567 free_vmcs(per_cpu(vmxarea
, cpu
));
1568 per_cpu(vmxarea
, cpu
) = NULL
;
1572 static __init
int alloc_kvm_area(void)
1576 for_each_possible_cpu(cpu
) {
1579 vmcs
= alloc_vmcs_cpu(cpu
);
1585 per_cpu(vmxarea
, cpu
) = vmcs
;
1590 static __init
int hardware_setup(void)
1592 if (setup_vmcs_config(&vmcs_config
) < 0)
1595 if (boot_cpu_has(X86_FEATURE_NX
))
1596 kvm_enable_efer_bits(EFER_NX
);
1598 if (!cpu_has_vmx_vpid())
1601 if (!cpu_has_vmx_ept() ||
1602 !cpu_has_vmx_ept_4levels()) {
1604 enable_unrestricted_guest
= 0;
1607 if (!cpu_has_vmx_unrestricted_guest())
1608 enable_unrestricted_guest
= 0;
1610 if (!cpu_has_vmx_flexpriority())
1611 flexpriority_enabled
= 0;
1613 if (!cpu_has_vmx_tpr_shadow())
1614 kvm_x86_ops
->update_cr8_intercept
= NULL
;
1616 if (enable_ept
&& !cpu_has_vmx_ept_2m_page())
1617 kvm_disable_largepages();
1619 if (!cpu_has_vmx_ple())
1622 return alloc_kvm_area();
1625 static __exit
void hardware_unsetup(void)
1630 static void fix_pmode_dataseg(int seg
, struct kvm_save_segment
*save
)
1632 struct kvm_vmx_segment_field
*sf
= &kvm_vmx_segment_fields
[seg
];
1634 if (vmcs_readl(sf
->base
) == save
->base
&& (save
->base
& AR_S_MASK
)) {
1635 vmcs_write16(sf
->selector
, save
->selector
);
1636 vmcs_writel(sf
->base
, save
->base
);
1637 vmcs_write32(sf
->limit
, save
->limit
);
1638 vmcs_write32(sf
->ar_bytes
, save
->ar
);
1640 u32 dpl
= (vmcs_read16(sf
->selector
) & SELECTOR_RPL_MASK
)
1642 vmcs_write32(sf
->ar_bytes
, 0x93 | dpl
);
1646 static void enter_pmode(struct kvm_vcpu
*vcpu
)
1648 unsigned long flags
;
1649 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
1651 vmx
->emulation_required
= 1;
1652 vmx
->rmode
.vm86_active
= 0;
1654 vmcs_writel(GUEST_TR_BASE
, vmx
->rmode
.tr
.base
);
1655 vmcs_write32(GUEST_TR_LIMIT
, vmx
->rmode
.tr
.limit
);
1656 vmcs_write32(GUEST_TR_AR_BYTES
, vmx
->rmode
.tr
.ar
);
1658 flags
= vmcs_readl(GUEST_RFLAGS
);
1659 flags
&= RMODE_GUEST_OWNED_EFLAGS_BITS
;
1660 flags
|= vmx
->rmode
.save_rflags
& ~RMODE_GUEST_OWNED_EFLAGS_BITS
;
1661 vmcs_writel(GUEST_RFLAGS
, flags
);
1663 vmcs_writel(GUEST_CR4
, (vmcs_readl(GUEST_CR4
) & ~X86_CR4_VME
) |
1664 (vmcs_readl(CR4_READ_SHADOW
) & X86_CR4_VME
));
1666 update_exception_bitmap(vcpu
);
1668 if (emulate_invalid_guest_state
)
1671 fix_pmode_dataseg(VCPU_SREG_ES
, &vmx
->rmode
.es
);
1672 fix_pmode_dataseg(VCPU_SREG_DS
, &vmx
->rmode
.ds
);
1673 fix_pmode_dataseg(VCPU_SREG_GS
, &vmx
->rmode
.gs
);
1674 fix_pmode_dataseg(VCPU_SREG_FS
, &vmx
->rmode
.fs
);
1676 vmcs_write16(GUEST_SS_SELECTOR
, 0);
1677 vmcs_write32(GUEST_SS_AR_BYTES
, 0x93);
1679 vmcs_write16(GUEST_CS_SELECTOR
,
1680 vmcs_read16(GUEST_CS_SELECTOR
) & ~SELECTOR_RPL_MASK
);
1681 vmcs_write32(GUEST_CS_AR_BYTES
, 0x9b);
1684 static gva_t
rmode_tss_base(struct kvm
*kvm
)
1686 if (!kvm
->arch
.tss_addr
) {
1687 struct kvm_memslots
*slots
;
1690 slots
= kvm_memslots(kvm
);
1691 base_gfn
= slots
->memslots
[0].base_gfn
+
1692 kvm
->memslots
->memslots
[0].npages
- 3;
1693 return base_gfn
<< PAGE_SHIFT
;
1695 return kvm
->arch
.tss_addr
;
1698 static void fix_rmode_seg(int seg
, struct kvm_save_segment
*save
)
1700 struct kvm_vmx_segment_field
*sf
= &kvm_vmx_segment_fields
[seg
];
1702 save
->selector
= vmcs_read16(sf
->selector
);
1703 save
->base
= vmcs_readl(sf
->base
);
1704 save
->limit
= vmcs_read32(sf
->limit
);
1705 save
->ar
= vmcs_read32(sf
->ar_bytes
);
1706 vmcs_write16(sf
->selector
, save
->base
>> 4);
1707 vmcs_write32(sf
->base
, save
->base
& 0xfffff);
1708 vmcs_write32(sf
->limit
, 0xffff);
1709 vmcs_write32(sf
->ar_bytes
, 0xf3);
1712 static void enter_rmode(struct kvm_vcpu
*vcpu
)
1714 unsigned long flags
;
1715 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
1717 if (enable_unrestricted_guest
)
1720 vmx
->emulation_required
= 1;
1721 vmx
->rmode
.vm86_active
= 1;
1723 vmx
->rmode
.tr
.base
= vmcs_readl(GUEST_TR_BASE
);
1724 vmcs_writel(GUEST_TR_BASE
, rmode_tss_base(vcpu
->kvm
));
1726 vmx
->rmode
.tr
.limit
= vmcs_read32(GUEST_TR_LIMIT
);
1727 vmcs_write32(GUEST_TR_LIMIT
, RMODE_TSS_SIZE
- 1);
1729 vmx
->rmode
.tr
.ar
= vmcs_read32(GUEST_TR_AR_BYTES
);
1730 vmcs_write32(GUEST_TR_AR_BYTES
, 0x008b);
1732 flags
= vmcs_readl(GUEST_RFLAGS
);
1733 vmx
->rmode
.save_rflags
= flags
;
1735 flags
|= X86_EFLAGS_IOPL
| X86_EFLAGS_VM
;
1737 vmcs_writel(GUEST_RFLAGS
, flags
);
1738 vmcs_writel(GUEST_CR4
, vmcs_readl(GUEST_CR4
) | X86_CR4_VME
);
1739 update_exception_bitmap(vcpu
);
1741 if (emulate_invalid_guest_state
)
1742 goto continue_rmode
;
1744 vmcs_write16(GUEST_SS_SELECTOR
, vmcs_readl(GUEST_SS_BASE
) >> 4);
1745 vmcs_write32(GUEST_SS_LIMIT
, 0xffff);
1746 vmcs_write32(GUEST_SS_AR_BYTES
, 0xf3);
1748 vmcs_write32(GUEST_CS_AR_BYTES
, 0xf3);
1749 vmcs_write32(GUEST_CS_LIMIT
, 0xffff);
1750 if (vmcs_readl(GUEST_CS_BASE
) == 0xffff0000)
1751 vmcs_writel(GUEST_CS_BASE
, 0xf0000);
1752 vmcs_write16(GUEST_CS_SELECTOR
, vmcs_readl(GUEST_CS_BASE
) >> 4);
1754 fix_rmode_seg(VCPU_SREG_ES
, &vmx
->rmode
.es
);
1755 fix_rmode_seg(VCPU_SREG_DS
, &vmx
->rmode
.ds
);
1756 fix_rmode_seg(VCPU_SREG_GS
, &vmx
->rmode
.gs
);
1757 fix_rmode_seg(VCPU_SREG_FS
, &vmx
->rmode
.fs
);
1760 kvm_mmu_reset_context(vcpu
);
1761 init_rmode(vcpu
->kvm
);
1764 static void vmx_set_efer(struct kvm_vcpu
*vcpu
, u64 efer
)
1766 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
1767 struct shared_msr_entry
*msr
= find_msr_entry(vmx
, MSR_EFER
);
1773 * Force kernel_gs_base reloading before EFER changes, as control
1774 * of this msr depends on is_long_mode().
1776 vmx_load_host_state(to_vmx(vcpu
));
1777 vcpu
->arch
.efer
= efer
;
1778 if (efer
& EFER_LMA
) {
1779 vmcs_write32(VM_ENTRY_CONTROLS
,
1780 vmcs_read32(VM_ENTRY_CONTROLS
) |
1781 VM_ENTRY_IA32E_MODE
);
1784 vmcs_write32(VM_ENTRY_CONTROLS
,
1785 vmcs_read32(VM_ENTRY_CONTROLS
) &
1786 ~VM_ENTRY_IA32E_MODE
);
1788 msr
->data
= efer
& ~EFER_LME
;
1793 #ifdef CONFIG_X86_64
1795 static void enter_lmode(struct kvm_vcpu
*vcpu
)
1799 guest_tr_ar
= vmcs_read32(GUEST_TR_AR_BYTES
);
1800 if ((guest_tr_ar
& AR_TYPE_MASK
) != AR_TYPE_BUSY_64_TSS
) {
1801 printk(KERN_DEBUG
"%s: tss fixup for long mode. \n",
1803 vmcs_write32(GUEST_TR_AR_BYTES
,
1804 (guest_tr_ar
& ~AR_TYPE_MASK
)
1805 | AR_TYPE_BUSY_64_TSS
);
1807 vmx_set_efer(vcpu
, vcpu
->arch
.efer
| EFER_LMA
);
1810 static void exit_lmode(struct kvm_vcpu
*vcpu
)
1812 vmcs_write32(VM_ENTRY_CONTROLS
,
1813 vmcs_read32(VM_ENTRY_CONTROLS
)
1814 & ~VM_ENTRY_IA32E_MODE
);
1815 vmx_set_efer(vcpu
, vcpu
->arch
.efer
& ~EFER_LMA
);
1820 static void vmx_flush_tlb(struct kvm_vcpu
*vcpu
)
1822 vpid_sync_context(to_vmx(vcpu
));
1824 if (!VALID_PAGE(vcpu
->arch
.mmu
.root_hpa
))
1826 ept_sync_context(construct_eptp(vcpu
->arch
.mmu
.root_hpa
));
1830 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu
*vcpu
)
1832 ulong cr0_guest_owned_bits
= vcpu
->arch
.cr0_guest_owned_bits
;
1834 vcpu
->arch
.cr0
&= ~cr0_guest_owned_bits
;
1835 vcpu
->arch
.cr0
|= vmcs_readl(GUEST_CR0
) & cr0_guest_owned_bits
;
1838 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu
*vcpu
)
1840 ulong cr4_guest_owned_bits
= vcpu
->arch
.cr4_guest_owned_bits
;
1842 vcpu
->arch
.cr4
&= ~cr4_guest_owned_bits
;
1843 vcpu
->arch
.cr4
|= vmcs_readl(GUEST_CR4
) & cr4_guest_owned_bits
;
1846 static void ept_load_pdptrs(struct kvm_vcpu
*vcpu
)
1848 if (!test_bit(VCPU_EXREG_PDPTR
,
1849 (unsigned long *)&vcpu
->arch
.regs_dirty
))
1852 if (is_paging(vcpu
) && is_pae(vcpu
) && !is_long_mode(vcpu
)) {
1853 vmcs_write64(GUEST_PDPTR0
, vcpu
->arch
.pdptrs
[0]);
1854 vmcs_write64(GUEST_PDPTR1
, vcpu
->arch
.pdptrs
[1]);
1855 vmcs_write64(GUEST_PDPTR2
, vcpu
->arch
.pdptrs
[2]);
1856 vmcs_write64(GUEST_PDPTR3
, vcpu
->arch
.pdptrs
[3]);
1860 static void ept_save_pdptrs(struct kvm_vcpu
*vcpu
)
1862 if (is_paging(vcpu
) && is_pae(vcpu
) && !is_long_mode(vcpu
)) {
1863 vcpu
->arch
.pdptrs
[0] = vmcs_read64(GUEST_PDPTR0
);
1864 vcpu
->arch
.pdptrs
[1] = vmcs_read64(GUEST_PDPTR1
);
1865 vcpu
->arch
.pdptrs
[2] = vmcs_read64(GUEST_PDPTR2
);
1866 vcpu
->arch
.pdptrs
[3] = vmcs_read64(GUEST_PDPTR3
);
1869 __set_bit(VCPU_EXREG_PDPTR
,
1870 (unsigned long *)&vcpu
->arch
.regs_avail
);
1871 __set_bit(VCPU_EXREG_PDPTR
,
1872 (unsigned long *)&vcpu
->arch
.regs_dirty
);
1875 static void vmx_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
);
1877 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0
,
1879 struct kvm_vcpu
*vcpu
)
1881 if (!(cr0
& X86_CR0_PG
)) {
1882 /* From paging/starting to nonpaging */
1883 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL
,
1884 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL
) |
1885 (CPU_BASED_CR3_LOAD_EXITING
|
1886 CPU_BASED_CR3_STORE_EXITING
));
1887 vcpu
->arch
.cr0
= cr0
;
1888 vmx_set_cr4(vcpu
, kvm_read_cr4(vcpu
));
1889 } else if (!is_paging(vcpu
)) {
1890 /* From nonpaging to paging */
1891 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL
,
1892 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL
) &
1893 ~(CPU_BASED_CR3_LOAD_EXITING
|
1894 CPU_BASED_CR3_STORE_EXITING
));
1895 vcpu
->arch
.cr0
= cr0
;
1896 vmx_set_cr4(vcpu
, kvm_read_cr4(vcpu
));
1899 if (!(cr0
& X86_CR0_WP
))
1900 *hw_cr0
&= ~X86_CR0_WP
;
1903 static void vmx_set_cr0(struct kvm_vcpu
*vcpu
, unsigned long cr0
)
1905 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
1906 unsigned long hw_cr0
;
1908 if (enable_unrestricted_guest
)
1909 hw_cr0
= (cr0
& ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST
)
1910 | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST
;
1912 hw_cr0
= (cr0
& ~KVM_GUEST_CR0_MASK
) | KVM_VM_CR0_ALWAYS_ON
;
1914 if (vmx
->rmode
.vm86_active
&& (cr0
& X86_CR0_PE
))
1917 if (!vmx
->rmode
.vm86_active
&& !(cr0
& X86_CR0_PE
))
1920 #ifdef CONFIG_X86_64
1921 if (vcpu
->arch
.efer
& EFER_LME
) {
1922 if (!is_paging(vcpu
) && (cr0
& X86_CR0_PG
))
1924 if (is_paging(vcpu
) && !(cr0
& X86_CR0_PG
))
1930 ept_update_paging_mode_cr0(&hw_cr0
, cr0
, vcpu
);
1932 if (!vcpu
->fpu_active
)
1933 hw_cr0
|= X86_CR0_TS
| X86_CR0_MP
;
1935 vmcs_writel(CR0_READ_SHADOW
, cr0
);
1936 vmcs_writel(GUEST_CR0
, hw_cr0
);
1937 vcpu
->arch
.cr0
= cr0
;
1940 static u64
construct_eptp(unsigned long root_hpa
)
1944 /* TODO write the value reading from MSR */
1945 eptp
= VMX_EPT_DEFAULT_MT
|
1946 VMX_EPT_DEFAULT_GAW
<< VMX_EPT_GAW_EPTP_SHIFT
;
1947 eptp
|= (root_hpa
& PAGE_MASK
);
1952 static void vmx_set_cr3(struct kvm_vcpu
*vcpu
, unsigned long cr3
)
1954 unsigned long guest_cr3
;
1959 eptp
= construct_eptp(cr3
);
1960 vmcs_write64(EPT_POINTER
, eptp
);
1961 guest_cr3
= is_paging(vcpu
) ? vcpu
->arch
.cr3
:
1962 vcpu
->kvm
->arch
.ept_identity_map_addr
;
1963 ept_load_pdptrs(vcpu
);
1966 vmx_flush_tlb(vcpu
);
1967 vmcs_writel(GUEST_CR3
, guest_cr3
);
1970 static void vmx_set_cr4(struct kvm_vcpu
*vcpu
, unsigned long cr4
)
1972 unsigned long hw_cr4
= cr4
| (to_vmx(vcpu
)->rmode
.vm86_active
?
1973 KVM_RMODE_VM_CR4_ALWAYS_ON
: KVM_PMODE_VM_CR4_ALWAYS_ON
);
1975 vcpu
->arch
.cr4
= cr4
;
1977 if (!is_paging(vcpu
)) {
1978 hw_cr4
&= ~X86_CR4_PAE
;
1979 hw_cr4
|= X86_CR4_PSE
;
1980 } else if (!(cr4
& X86_CR4_PAE
)) {
1981 hw_cr4
&= ~X86_CR4_PAE
;
1985 vmcs_writel(CR4_READ_SHADOW
, cr4
);
1986 vmcs_writel(GUEST_CR4
, hw_cr4
);
1989 static u64
vmx_get_segment_base(struct kvm_vcpu
*vcpu
, int seg
)
1991 struct kvm_vmx_segment_field
*sf
= &kvm_vmx_segment_fields
[seg
];
1993 return vmcs_readl(sf
->base
);
1996 static void vmx_get_segment(struct kvm_vcpu
*vcpu
,
1997 struct kvm_segment
*var
, int seg
)
1999 struct kvm_vmx_segment_field
*sf
= &kvm_vmx_segment_fields
[seg
];
2002 var
->base
= vmcs_readl(sf
->base
);
2003 var
->limit
= vmcs_read32(sf
->limit
);
2004 var
->selector
= vmcs_read16(sf
->selector
);
2005 ar
= vmcs_read32(sf
->ar_bytes
);
2006 if ((ar
& AR_UNUSABLE_MASK
) && !emulate_invalid_guest_state
)
2008 var
->type
= ar
& 15;
2009 var
->s
= (ar
>> 4) & 1;
2010 var
->dpl
= (ar
>> 5) & 3;
2011 var
->present
= (ar
>> 7) & 1;
2012 var
->avl
= (ar
>> 12) & 1;
2013 var
->l
= (ar
>> 13) & 1;
2014 var
->db
= (ar
>> 14) & 1;
2015 var
->g
= (ar
>> 15) & 1;
2016 var
->unusable
= (ar
>> 16) & 1;
2019 static int vmx_get_cpl(struct kvm_vcpu
*vcpu
)
2021 if (!is_protmode(vcpu
))
2024 if (vmx_get_rflags(vcpu
) & X86_EFLAGS_VM
) /* if virtual 8086 */
2027 return vmcs_read16(GUEST_CS_SELECTOR
) & 3;
2030 static u32
vmx_segment_access_rights(struct kvm_segment
*var
)
2037 ar
= var
->type
& 15;
2038 ar
|= (var
->s
& 1) << 4;
2039 ar
|= (var
->dpl
& 3) << 5;
2040 ar
|= (var
->present
& 1) << 7;
2041 ar
|= (var
->avl
& 1) << 12;
2042 ar
|= (var
->l
& 1) << 13;
2043 ar
|= (var
->db
& 1) << 14;
2044 ar
|= (var
->g
& 1) << 15;
2046 if (ar
== 0) /* a 0 value means unusable */
2047 ar
= AR_UNUSABLE_MASK
;
2052 static void vmx_set_segment(struct kvm_vcpu
*vcpu
,
2053 struct kvm_segment
*var
, int seg
)
2055 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
2056 struct kvm_vmx_segment_field
*sf
= &kvm_vmx_segment_fields
[seg
];
2059 if (vmx
->rmode
.vm86_active
&& seg
== VCPU_SREG_TR
) {
2060 vmx
->rmode
.tr
.selector
= var
->selector
;
2061 vmx
->rmode
.tr
.base
= var
->base
;
2062 vmx
->rmode
.tr
.limit
= var
->limit
;
2063 vmx
->rmode
.tr
.ar
= vmx_segment_access_rights(var
);
2066 vmcs_writel(sf
->base
, var
->base
);
2067 vmcs_write32(sf
->limit
, var
->limit
);
2068 vmcs_write16(sf
->selector
, var
->selector
);
2069 if (vmx
->rmode
.vm86_active
&& var
->s
) {
2071 * Hack real-mode segments into vm86 compatibility.
2073 if (var
->base
== 0xffff0000 && var
->selector
== 0xf000)
2074 vmcs_writel(sf
->base
, 0xf0000);
2077 ar
= vmx_segment_access_rights(var
);
2080 * Fix the "Accessed" bit in AR field of segment registers for older
2082 * IA32 arch specifies that at the time of processor reset the
2083 * "Accessed" bit in the AR field of segment registers is 1. And qemu
2084 * is setting it to 0 in the usedland code. This causes invalid guest
2085 * state vmexit when "unrestricted guest" mode is turned on.
2086 * Fix for this setup issue in cpu_reset is being pushed in the qemu
2087 * tree. Newer qemu binaries with that qemu fix would not need this
2090 if (enable_unrestricted_guest
&& (seg
!= VCPU_SREG_LDTR
))
2091 ar
|= 0x1; /* Accessed */
2093 vmcs_write32(sf
->ar_bytes
, ar
);
2096 static void vmx_get_cs_db_l_bits(struct kvm_vcpu
*vcpu
, int *db
, int *l
)
2098 u32 ar
= vmcs_read32(GUEST_CS_AR_BYTES
);
2100 *db
= (ar
>> 14) & 1;
2101 *l
= (ar
>> 13) & 1;
2104 static void vmx_get_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
2106 dt
->size
= vmcs_read32(GUEST_IDTR_LIMIT
);
2107 dt
->address
= vmcs_readl(GUEST_IDTR_BASE
);
2110 static void vmx_set_idt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
2112 vmcs_write32(GUEST_IDTR_LIMIT
, dt
->size
);
2113 vmcs_writel(GUEST_IDTR_BASE
, dt
->address
);
2116 static void vmx_get_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
2118 dt
->size
= vmcs_read32(GUEST_GDTR_LIMIT
);
2119 dt
->address
= vmcs_readl(GUEST_GDTR_BASE
);
2122 static void vmx_set_gdt(struct kvm_vcpu
*vcpu
, struct desc_ptr
*dt
)
2124 vmcs_write32(GUEST_GDTR_LIMIT
, dt
->size
);
2125 vmcs_writel(GUEST_GDTR_BASE
, dt
->address
);
2128 static bool rmode_segment_valid(struct kvm_vcpu
*vcpu
, int seg
)
2130 struct kvm_segment var
;
2133 vmx_get_segment(vcpu
, &var
, seg
);
2134 ar
= vmx_segment_access_rights(&var
);
2136 if (var
.base
!= (var
.selector
<< 4))
2138 if (var
.limit
!= 0xffff)
2146 static bool code_segment_valid(struct kvm_vcpu
*vcpu
)
2148 struct kvm_segment cs
;
2149 unsigned int cs_rpl
;
2151 vmx_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
2152 cs_rpl
= cs
.selector
& SELECTOR_RPL_MASK
;
2156 if (~cs
.type
& (AR_TYPE_CODE_MASK
|AR_TYPE_ACCESSES_MASK
))
2160 if (cs
.type
& AR_TYPE_WRITEABLE_MASK
) {
2161 if (cs
.dpl
> cs_rpl
)
2164 if (cs
.dpl
!= cs_rpl
)
2170 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
2174 static bool stack_segment_valid(struct kvm_vcpu
*vcpu
)
2176 struct kvm_segment ss
;
2177 unsigned int ss_rpl
;
2179 vmx_get_segment(vcpu
, &ss
, VCPU_SREG_SS
);
2180 ss_rpl
= ss
.selector
& SELECTOR_RPL_MASK
;
2184 if (ss
.type
!= 3 && ss
.type
!= 7)
2188 if (ss
.dpl
!= ss_rpl
) /* DPL != RPL */
2196 static bool data_segment_valid(struct kvm_vcpu
*vcpu
, int seg
)
2198 struct kvm_segment var
;
2201 vmx_get_segment(vcpu
, &var
, seg
);
2202 rpl
= var
.selector
& SELECTOR_RPL_MASK
;
2210 if (~var
.type
& (AR_TYPE_CODE_MASK
|AR_TYPE_WRITEABLE_MASK
)) {
2211 if (var
.dpl
< rpl
) /* DPL < RPL */
2215 /* TODO: Add other members to kvm_segment_field to allow checking for other access
2221 static bool tr_valid(struct kvm_vcpu
*vcpu
)
2223 struct kvm_segment tr
;
2225 vmx_get_segment(vcpu
, &tr
, VCPU_SREG_TR
);
2229 if (tr
.selector
& SELECTOR_TI_MASK
) /* TI = 1 */
2231 if (tr
.type
!= 3 && tr
.type
!= 11) /* TODO: Check if guest is in IA32e mode */
2239 static bool ldtr_valid(struct kvm_vcpu
*vcpu
)
2241 struct kvm_segment ldtr
;
2243 vmx_get_segment(vcpu
, &ldtr
, VCPU_SREG_LDTR
);
2247 if (ldtr
.selector
& SELECTOR_TI_MASK
) /* TI = 1 */
2257 static bool cs_ss_rpl_check(struct kvm_vcpu
*vcpu
)
2259 struct kvm_segment cs
, ss
;
2261 vmx_get_segment(vcpu
, &cs
, VCPU_SREG_CS
);
2262 vmx_get_segment(vcpu
, &ss
, VCPU_SREG_SS
);
2264 return ((cs
.selector
& SELECTOR_RPL_MASK
) ==
2265 (ss
.selector
& SELECTOR_RPL_MASK
));
2269 * Check if guest state is valid. Returns true if valid, false if
2271 * We assume that registers are always usable
2273 static bool guest_state_valid(struct kvm_vcpu
*vcpu
)
2275 /* real mode guest state checks */
2276 if (!is_protmode(vcpu
)) {
2277 if (!rmode_segment_valid(vcpu
, VCPU_SREG_CS
))
2279 if (!rmode_segment_valid(vcpu
, VCPU_SREG_SS
))
2281 if (!rmode_segment_valid(vcpu
, VCPU_SREG_DS
))
2283 if (!rmode_segment_valid(vcpu
, VCPU_SREG_ES
))
2285 if (!rmode_segment_valid(vcpu
, VCPU_SREG_FS
))
2287 if (!rmode_segment_valid(vcpu
, VCPU_SREG_GS
))
2290 /* protected mode guest state checks */
2291 if (!cs_ss_rpl_check(vcpu
))
2293 if (!code_segment_valid(vcpu
))
2295 if (!stack_segment_valid(vcpu
))
2297 if (!data_segment_valid(vcpu
, VCPU_SREG_DS
))
2299 if (!data_segment_valid(vcpu
, VCPU_SREG_ES
))
2301 if (!data_segment_valid(vcpu
, VCPU_SREG_FS
))
2303 if (!data_segment_valid(vcpu
, VCPU_SREG_GS
))
2305 if (!tr_valid(vcpu
))
2307 if (!ldtr_valid(vcpu
))
2311 * - Add checks on RIP
2312 * - Add checks on RFLAGS
2318 static int init_rmode_tss(struct kvm
*kvm
)
2320 gfn_t fn
= rmode_tss_base(kvm
) >> PAGE_SHIFT
;
2325 r
= kvm_clear_guest_page(kvm
, fn
, 0, PAGE_SIZE
);
2328 data
= TSS_BASE_SIZE
+ TSS_REDIRECTION_SIZE
;
2329 r
= kvm_write_guest_page(kvm
, fn
++, &data
,
2330 TSS_IOPB_BASE_OFFSET
, sizeof(u16
));
2333 r
= kvm_clear_guest_page(kvm
, fn
++, 0, PAGE_SIZE
);
2336 r
= kvm_clear_guest_page(kvm
, fn
, 0, PAGE_SIZE
);
2340 r
= kvm_write_guest_page(kvm
, fn
, &data
,
2341 RMODE_TSS_SIZE
- 2 * PAGE_SIZE
- 1,
2351 static int init_rmode_identity_map(struct kvm
*kvm
)
2354 pfn_t identity_map_pfn
;
2359 if (unlikely(!kvm
->arch
.ept_identity_pagetable
)) {
2360 printk(KERN_ERR
"EPT: identity-mapping pagetable "
2361 "haven't been allocated!\n");
2364 if (likely(kvm
->arch
.ept_identity_pagetable_done
))
2367 identity_map_pfn
= kvm
->arch
.ept_identity_map_addr
>> PAGE_SHIFT
;
2368 r
= kvm_clear_guest_page(kvm
, identity_map_pfn
, 0, PAGE_SIZE
);
2371 /* Set up identity-mapping pagetable for EPT in real mode */
2372 for (i
= 0; i
< PT32_ENT_PER_PAGE
; i
++) {
2373 tmp
= (i
<< 22) + (_PAGE_PRESENT
| _PAGE_RW
| _PAGE_USER
|
2374 _PAGE_ACCESSED
| _PAGE_DIRTY
| _PAGE_PSE
);
2375 r
= kvm_write_guest_page(kvm
, identity_map_pfn
,
2376 &tmp
, i
* sizeof(tmp
), sizeof(tmp
));
2380 kvm
->arch
.ept_identity_pagetable_done
= true;
2386 static void seg_setup(int seg
)
2388 struct kvm_vmx_segment_field
*sf
= &kvm_vmx_segment_fields
[seg
];
2391 vmcs_write16(sf
->selector
, 0);
2392 vmcs_writel(sf
->base
, 0);
2393 vmcs_write32(sf
->limit
, 0xffff);
2394 if (enable_unrestricted_guest
) {
2396 if (seg
== VCPU_SREG_CS
)
2397 ar
|= 0x08; /* code segment */
2401 vmcs_write32(sf
->ar_bytes
, ar
);
2404 static int alloc_apic_access_page(struct kvm
*kvm
)
2406 struct kvm_userspace_memory_region kvm_userspace_mem
;
2409 mutex_lock(&kvm
->slots_lock
);
2410 if (kvm
->arch
.apic_access_page
)
2412 kvm_userspace_mem
.slot
= APIC_ACCESS_PAGE_PRIVATE_MEMSLOT
;
2413 kvm_userspace_mem
.flags
= 0;
2414 kvm_userspace_mem
.guest_phys_addr
= 0xfee00000ULL
;
2415 kvm_userspace_mem
.memory_size
= PAGE_SIZE
;
2416 r
= __kvm_set_memory_region(kvm
, &kvm_userspace_mem
, 0);
2420 kvm
->arch
.apic_access_page
= gfn_to_page(kvm
, 0xfee00);
2422 mutex_unlock(&kvm
->slots_lock
);
2426 static int alloc_identity_pagetable(struct kvm
*kvm
)
2428 struct kvm_userspace_memory_region kvm_userspace_mem
;
2431 mutex_lock(&kvm
->slots_lock
);
2432 if (kvm
->arch
.ept_identity_pagetable
)
2434 kvm_userspace_mem
.slot
= IDENTITY_PAGETABLE_PRIVATE_MEMSLOT
;
2435 kvm_userspace_mem
.flags
= 0;
2436 kvm_userspace_mem
.guest_phys_addr
=
2437 kvm
->arch
.ept_identity_map_addr
;
2438 kvm_userspace_mem
.memory_size
= PAGE_SIZE
;
2439 r
= __kvm_set_memory_region(kvm
, &kvm_userspace_mem
, 0);
2443 kvm
->arch
.ept_identity_pagetable
= gfn_to_page(kvm
,
2444 kvm
->arch
.ept_identity_map_addr
>> PAGE_SHIFT
);
2446 mutex_unlock(&kvm
->slots_lock
);
2450 static void allocate_vpid(struct vcpu_vmx
*vmx
)
2457 spin_lock(&vmx_vpid_lock
);
2458 vpid
= find_first_zero_bit(vmx_vpid_bitmap
, VMX_NR_VPIDS
);
2459 if (vpid
< VMX_NR_VPIDS
) {
2461 __set_bit(vpid
, vmx_vpid_bitmap
);
2463 spin_unlock(&vmx_vpid_lock
);
2466 static void free_vpid(struct vcpu_vmx
*vmx
)
2470 spin_lock(&vmx_vpid_lock
);
2472 __clear_bit(vmx
->vpid
, vmx_vpid_bitmap
);
2473 spin_unlock(&vmx_vpid_lock
);
2476 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap
, u32 msr
)
2478 int f
= sizeof(unsigned long);
2480 if (!cpu_has_vmx_msr_bitmap())
2484 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2485 * have the write-low and read-high bitmap offsets the wrong way round.
2486 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2488 if (msr
<= 0x1fff) {
2489 __clear_bit(msr
, msr_bitmap
+ 0x000 / f
); /* read-low */
2490 __clear_bit(msr
, msr_bitmap
+ 0x800 / f
); /* write-low */
2491 } else if ((msr
>= 0xc0000000) && (msr
<= 0xc0001fff)) {
2493 __clear_bit(msr
, msr_bitmap
+ 0x400 / f
); /* read-high */
2494 __clear_bit(msr
, msr_bitmap
+ 0xc00 / f
); /* write-high */
2498 static void vmx_disable_intercept_for_msr(u32 msr
, bool longmode_only
)
2501 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy
, msr
);
2502 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode
, msr
);
2506 * Sets up the vmcs for emulated real mode.
2508 static int vmx_vcpu_setup(struct vcpu_vmx
*vmx
)
2510 u32 host_sysenter_cs
, msr_low
, msr_high
;
2516 unsigned long kvm_vmx_return
;
2520 vmcs_write64(IO_BITMAP_A
, __pa(vmx_io_bitmap_a
));
2521 vmcs_write64(IO_BITMAP_B
, __pa(vmx_io_bitmap_b
));
2523 if (cpu_has_vmx_msr_bitmap())
2524 vmcs_write64(MSR_BITMAP
, __pa(vmx_msr_bitmap_legacy
));
2526 vmcs_write64(VMCS_LINK_POINTER
, -1ull); /* 22.3.1.5 */
2529 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL
,
2530 vmcs_config
.pin_based_exec_ctrl
);
2532 exec_control
= vmcs_config
.cpu_based_exec_ctrl
;
2533 if (!vm_need_tpr_shadow(vmx
->vcpu
.kvm
)) {
2534 exec_control
&= ~CPU_BASED_TPR_SHADOW
;
2535 #ifdef CONFIG_X86_64
2536 exec_control
|= CPU_BASED_CR8_STORE_EXITING
|
2537 CPU_BASED_CR8_LOAD_EXITING
;
2541 exec_control
|= CPU_BASED_CR3_STORE_EXITING
|
2542 CPU_BASED_CR3_LOAD_EXITING
|
2543 CPU_BASED_INVLPG_EXITING
;
2544 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL
, exec_control
);
2546 if (cpu_has_secondary_exec_ctrls()) {
2547 exec_control
= vmcs_config
.cpu_based_2nd_exec_ctrl
;
2548 if (!vm_need_virtualize_apic_accesses(vmx
->vcpu
.kvm
))
2550 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES
;
2552 exec_control
&= ~SECONDARY_EXEC_ENABLE_VPID
;
2554 exec_control
&= ~SECONDARY_EXEC_ENABLE_EPT
;
2555 enable_unrestricted_guest
= 0;
2557 if (!enable_unrestricted_guest
)
2558 exec_control
&= ~SECONDARY_EXEC_UNRESTRICTED_GUEST
;
2560 exec_control
&= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING
;
2561 vmcs_write32(SECONDARY_VM_EXEC_CONTROL
, exec_control
);
2565 vmcs_write32(PLE_GAP
, ple_gap
);
2566 vmcs_write32(PLE_WINDOW
, ple_window
);
2569 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK
, !!bypass_guest_pf
);
2570 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH
, !!bypass_guest_pf
);
2571 vmcs_write32(CR3_TARGET_COUNT
, 0); /* 22.2.1 */
2573 vmcs_writel(HOST_CR0
, read_cr0() | X86_CR0_TS
); /* 22.2.3 */
2574 vmcs_writel(HOST_CR4
, read_cr4()); /* 22.2.3, 22.2.5 */
2575 vmcs_writel(HOST_CR3
, read_cr3()); /* 22.2.3 FIXME: shadow tables */
2577 vmcs_write16(HOST_CS_SELECTOR
, __KERNEL_CS
); /* 22.2.4 */
2578 vmcs_write16(HOST_DS_SELECTOR
, __KERNEL_DS
); /* 22.2.4 */
2579 vmcs_write16(HOST_ES_SELECTOR
, __KERNEL_DS
); /* 22.2.4 */
2580 vmcs_write16(HOST_FS_SELECTOR
, 0); /* 22.2.4 */
2581 vmcs_write16(HOST_GS_SELECTOR
, 0); /* 22.2.4 */
2582 vmcs_write16(HOST_SS_SELECTOR
, __KERNEL_DS
); /* 22.2.4 */
2583 #ifdef CONFIG_X86_64
2584 rdmsrl(MSR_FS_BASE
, a
);
2585 vmcs_writel(HOST_FS_BASE
, a
); /* 22.2.4 */
2586 rdmsrl(MSR_GS_BASE
, a
);
2587 vmcs_writel(HOST_GS_BASE
, a
); /* 22.2.4 */
2589 vmcs_writel(HOST_FS_BASE
, 0); /* 22.2.4 */
2590 vmcs_writel(HOST_GS_BASE
, 0); /* 22.2.4 */
2593 vmcs_write16(HOST_TR_SELECTOR
, GDT_ENTRY_TSS
*8); /* 22.2.4 */
2595 native_store_idt(&dt
);
2596 vmcs_writel(HOST_IDTR_BASE
, dt
.address
); /* 22.2.4 */
2598 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return
));
2599 vmcs_writel(HOST_RIP
, kvm_vmx_return
); /* 22.2.5 */
2600 vmcs_write32(VM_EXIT_MSR_STORE_COUNT
, 0);
2601 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT
, 0);
2602 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR
, __pa(vmx
->msr_autoload
.host
));
2603 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT
, 0);
2604 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR
, __pa(vmx
->msr_autoload
.guest
));
2606 rdmsr(MSR_IA32_SYSENTER_CS
, host_sysenter_cs
, junk
);
2607 vmcs_write32(HOST_IA32_SYSENTER_CS
, host_sysenter_cs
);
2608 rdmsrl(MSR_IA32_SYSENTER_ESP
, a
);
2609 vmcs_writel(HOST_IA32_SYSENTER_ESP
, a
); /* 22.2.3 */
2610 rdmsrl(MSR_IA32_SYSENTER_EIP
, a
);
2611 vmcs_writel(HOST_IA32_SYSENTER_EIP
, a
); /* 22.2.3 */
2613 if (vmcs_config
.vmexit_ctrl
& VM_EXIT_LOAD_IA32_PAT
) {
2614 rdmsr(MSR_IA32_CR_PAT
, msr_low
, msr_high
);
2615 host_pat
= msr_low
| ((u64
) msr_high
<< 32);
2616 vmcs_write64(HOST_IA32_PAT
, host_pat
);
2618 if (vmcs_config
.vmentry_ctrl
& VM_ENTRY_LOAD_IA32_PAT
) {
2619 rdmsr(MSR_IA32_CR_PAT
, msr_low
, msr_high
);
2620 host_pat
= msr_low
| ((u64
) msr_high
<< 32);
2621 /* Write the default value follow host pat */
2622 vmcs_write64(GUEST_IA32_PAT
, host_pat
);
2623 /* Keep arch.pat sync with GUEST_IA32_PAT */
2624 vmx
->vcpu
.arch
.pat
= host_pat
;
2627 for (i
= 0; i
< NR_VMX_MSR
; ++i
) {
2628 u32 index
= vmx_msr_index
[i
];
2629 u32 data_low
, data_high
;
2632 if (rdmsr_safe(index
, &data_low
, &data_high
) < 0)
2634 if (wrmsr_safe(index
, data_low
, data_high
) < 0)
2636 vmx
->guest_msrs
[j
].index
= i
;
2637 vmx
->guest_msrs
[j
].data
= 0;
2638 vmx
->guest_msrs
[j
].mask
= -1ull;
2642 vmcs_write32(VM_EXIT_CONTROLS
, vmcs_config
.vmexit_ctrl
);
2644 /* 22.2.1, 20.8.1 */
2645 vmcs_write32(VM_ENTRY_CONTROLS
, vmcs_config
.vmentry_ctrl
);
2647 vmcs_writel(CR0_GUEST_HOST_MASK
, ~0UL);
2648 vmx
->vcpu
.arch
.cr4_guest_owned_bits
= KVM_CR4_GUEST_OWNED_BITS
;
2650 vmx
->vcpu
.arch
.cr4_guest_owned_bits
|= X86_CR4_PGE
;
2651 vmcs_writel(CR4_GUEST_HOST_MASK
, ~vmx
->vcpu
.arch
.cr4_guest_owned_bits
);
2653 kvm_write_tsc(&vmx
->vcpu
, 0);
2658 static int init_rmode(struct kvm
*kvm
)
2662 idx
= srcu_read_lock(&kvm
->srcu
);
2663 if (!init_rmode_tss(kvm
))
2665 if (!init_rmode_identity_map(kvm
))
2670 srcu_read_unlock(&kvm
->srcu
, idx
);
2674 static int vmx_vcpu_reset(struct kvm_vcpu
*vcpu
)
2676 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
2680 vcpu
->arch
.regs_avail
= ~((1 << VCPU_REGS_RIP
) | (1 << VCPU_REGS_RSP
));
2681 if (!init_rmode(vmx
->vcpu
.kvm
)) {
2686 vmx
->rmode
.vm86_active
= 0;
2688 vmx
->soft_vnmi_blocked
= 0;
2690 vmx
->vcpu
.arch
.regs
[VCPU_REGS_RDX
] = get_rdx_init_val();
2691 kvm_set_cr8(&vmx
->vcpu
, 0);
2692 msr
= 0xfee00000 | MSR_IA32_APICBASE_ENABLE
;
2693 if (kvm_vcpu_is_bsp(&vmx
->vcpu
))
2694 msr
|= MSR_IA32_APICBASE_BSP
;
2695 kvm_set_apic_base(&vmx
->vcpu
, msr
);
2697 ret
= fx_init(&vmx
->vcpu
);
2701 seg_setup(VCPU_SREG_CS
);
2703 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2704 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
2706 if (kvm_vcpu_is_bsp(&vmx
->vcpu
)) {
2707 vmcs_write16(GUEST_CS_SELECTOR
, 0xf000);
2708 vmcs_writel(GUEST_CS_BASE
, 0x000f0000);
2710 vmcs_write16(GUEST_CS_SELECTOR
, vmx
->vcpu
.arch
.sipi_vector
<< 8);
2711 vmcs_writel(GUEST_CS_BASE
, vmx
->vcpu
.arch
.sipi_vector
<< 12);
2714 seg_setup(VCPU_SREG_DS
);
2715 seg_setup(VCPU_SREG_ES
);
2716 seg_setup(VCPU_SREG_FS
);
2717 seg_setup(VCPU_SREG_GS
);
2718 seg_setup(VCPU_SREG_SS
);
2720 vmcs_write16(GUEST_TR_SELECTOR
, 0);
2721 vmcs_writel(GUEST_TR_BASE
, 0);
2722 vmcs_write32(GUEST_TR_LIMIT
, 0xffff);
2723 vmcs_write32(GUEST_TR_AR_BYTES
, 0x008b);
2725 vmcs_write16(GUEST_LDTR_SELECTOR
, 0);
2726 vmcs_writel(GUEST_LDTR_BASE
, 0);
2727 vmcs_write32(GUEST_LDTR_LIMIT
, 0xffff);
2728 vmcs_write32(GUEST_LDTR_AR_BYTES
, 0x00082);
2730 vmcs_write32(GUEST_SYSENTER_CS
, 0);
2731 vmcs_writel(GUEST_SYSENTER_ESP
, 0);
2732 vmcs_writel(GUEST_SYSENTER_EIP
, 0);
2734 vmcs_writel(GUEST_RFLAGS
, 0x02);
2735 if (kvm_vcpu_is_bsp(&vmx
->vcpu
))
2736 kvm_rip_write(vcpu
, 0xfff0);
2738 kvm_rip_write(vcpu
, 0);
2739 kvm_register_write(vcpu
, VCPU_REGS_RSP
, 0);
2741 vmcs_writel(GUEST_DR7
, 0x400);
2743 vmcs_writel(GUEST_GDTR_BASE
, 0);
2744 vmcs_write32(GUEST_GDTR_LIMIT
, 0xffff);
2746 vmcs_writel(GUEST_IDTR_BASE
, 0);
2747 vmcs_write32(GUEST_IDTR_LIMIT
, 0xffff);
2749 vmcs_write32(GUEST_ACTIVITY_STATE
, 0);
2750 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO
, 0);
2751 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS
, 0);
2753 /* Special registers */
2754 vmcs_write64(GUEST_IA32_DEBUGCTL
, 0);
2758 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD
, 0); /* 22.2.1 */
2760 if (cpu_has_vmx_tpr_shadow()) {
2761 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR
, 0);
2762 if (vm_need_tpr_shadow(vmx
->vcpu
.kvm
))
2763 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR
,
2764 page_to_phys(vmx
->vcpu
.arch
.apic
->regs_page
));
2765 vmcs_write32(TPR_THRESHOLD
, 0);
2768 if (vm_need_virtualize_apic_accesses(vmx
->vcpu
.kvm
))
2769 vmcs_write64(APIC_ACCESS_ADDR
,
2770 page_to_phys(vmx
->vcpu
.kvm
->arch
.apic_access_page
));
2773 vmcs_write16(VIRTUAL_PROCESSOR_ID
, vmx
->vpid
);
2775 vmx
->vcpu
.arch
.cr0
= X86_CR0_NW
| X86_CR0_CD
| X86_CR0_ET
;
2776 vmx_set_cr0(&vmx
->vcpu
, kvm_read_cr0(vcpu
)); /* enter rmode */
2777 vmx_set_cr4(&vmx
->vcpu
, 0);
2778 vmx_set_efer(&vmx
->vcpu
, 0);
2779 vmx_fpu_activate(&vmx
->vcpu
);
2780 update_exception_bitmap(&vmx
->vcpu
);
2782 vpid_sync_context(vmx
);
2786 /* HACK: Don't enable emulation on guest boot/reset */
2787 vmx
->emulation_required
= 0;
2793 static void enable_irq_window(struct kvm_vcpu
*vcpu
)
2795 u32 cpu_based_vm_exec_control
;
2797 cpu_based_vm_exec_control
= vmcs_read32(CPU_BASED_VM_EXEC_CONTROL
);
2798 cpu_based_vm_exec_control
|= CPU_BASED_VIRTUAL_INTR_PENDING
;
2799 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL
, cpu_based_vm_exec_control
);
2802 static void enable_nmi_window(struct kvm_vcpu
*vcpu
)
2804 u32 cpu_based_vm_exec_control
;
2806 if (!cpu_has_virtual_nmis()) {
2807 enable_irq_window(vcpu
);
2811 cpu_based_vm_exec_control
= vmcs_read32(CPU_BASED_VM_EXEC_CONTROL
);
2812 cpu_based_vm_exec_control
|= CPU_BASED_VIRTUAL_NMI_PENDING
;
2813 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL
, cpu_based_vm_exec_control
);
2816 static void vmx_inject_irq(struct kvm_vcpu
*vcpu
)
2818 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
2820 int irq
= vcpu
->arch
.interrupt
.nr
;
2822 trace_kvm_inj_virq(irq
);
2824 ++vcpu
->stat
.irq_injections
;
2825 if (vmx
->rmode
.vm86_active
) {
2826 vmx
->rmode
.irq
.pending
= true;
2827 vmx
->rmode
.irq
.vector
= irq
;
2828 vmx
->rmode
.irq
.rip
= kvm_rip_read(vcpu
);
2829 if (vcpu
->arch
.interrupt
.soft
)
2830 vmx
->rmode
.irq
.rip
+=
2831 vmx
->vcpu
.arch
.event_exit_inst_len
;
2832 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD
,
2833 irq
| INTR_TYPE_SOFT_INTR
| INTR_INFO_VALID_MASK
);
2834 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN
, 1);
2835 kvm_rip_write(vcpu
, vmx
->rmode
.irq
.rip
- 1);
2838 intr
= irq
| INTR_INFO_VALID_MASK
;
2839 if (vcpu
->arch
.interrupt
.soft
) {
2840 intr
|= INTR_TYPE_SOFT_INTR
;
2841 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN
,
2842 vmx
->vcpu
.arch
.event_exit_inst_len
);
2844 intr
|= INTR_TYPE_EXT_INTR
;
2845 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD
, intr
);
2848 static void vmx_inject_nmi(struct kvm_vcpu
*vcpu
)
2850 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
2852 if (!cpu_has_virtual_nmis()) {
2854 * Tracking the NMI-blocked state in software is built upon
2855 * finding the next open IRQ window. This, in turn, depends on
2856 * well-behaving guests: They have to keep IRQs disabled at
2857 * least as long as the NMI handler runs. Otherwise we may
2858 * cause NMI nesting, maybe breaking the guest. But as this is
2859 * highly unlikely, we can live with the residual risk.
2861 vmx
->soft_vnmi_blocked
= 1;
2862 vmx
->vnmi_blocked_time
= 0;
2865 ++vcpu
->stat
.nmi_injections
;
2866 if (vmx
->rmode
.vm86_active
) {
2867 vmx
->rmode
.irq
.pending
= true;
2868 vmx
->rmode
.irq
.vector
= NMI_VECTOR
;
2869 vmx
->rmode
.irq
.rip
= kvm_rip_read(vcpu
);
2870 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD
,
2871 NMI_VECTOR
| INTR_TYPE_SOFT_INTR
|
2872 INTR_INFO_VALID_MASK
);
2873 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN
, 1);
2874 kvm_rip_write(vcpu
, vmx
->rmode
.irq
.rip
- 1);
2877 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD
,
2878 INTR_TYPE_NMI_INTR
| INTR_INFO_VALID_MASK
| NMI_VECTOR
);
2881 static int vmx_nmi_allowed(struct kvm_vcpu
*vcpu
)
2883 if (!cpu_has_virtual_nmis() && to_vmx(vcpu
)->soft_vnmi_blocked
)
2886 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO
) &
2887 (GUEST_INTR_STATE_MOV_SS
| GUEST_INTR_STATE_NMI
));
2890 static bool vmx_get_nmi_mask(struct kvm_vcpu
*vcpu
)
2892 if (!cpu_has_virtual_nmis())
2893 return to_vmx(vcpu
)->soft_vnmi_blocked
;
2894 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO
) & GUEST_INTR_STATE_NMI
;
2897 static void vmx_set_nmi_mask(struct kvm_vcpu
*vcpu
, bool masked
)
2899 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
2901 if (!cpu_has_virtual_nmis()) {
2902 if (vmx
->soft_vnmi_blocked
!= masked
) {
2903 vmx
->soft_vnmi_blocked
= masked
;
2904 vmx
->vnmi_blocked_time
= 0;
2908 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO
,
2909 GUEST_INTR_STATE_NMI
);
2911 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO
,
2912 GUEST_INTR_STATE_NMI
);
2916 static int vmx_interrupt_allowed(struct kvm_vcpu
*vcpu
)
2918 return (vmcs_readl(GUEST_RFLAGS
) & X86_EFLAGS_IF
) &&
2919 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO
) &
2920 (GUEST_INTR_STATE_STI
| GUEST_INTR_STATE_MOV_SS
));
2923 static int vmx_set_tss_addr(struct kvm
*kvm
, unsigned int addr
)
2926 struct kvm_userspace_memory_region tss_mem
= {
2927 .slot
= TSS_PRIVATE_MEMSLOT
,
2928 .guest_phys_addr
= addr
,
2929 .memory_size
= PAGE_SIZE
* 3,
2933 ret
= kvm_set_memory_region(kvm
, &tss_mem
, 0);
2936 kvm
->arch
.tss_addr
= addr
;
2940 static int handle_rmode_exception(struct kvm_vcpu
*vcpu
,
2941 int vec
, u32 err_code
)
2944 * Instruction with address size override prefix opcode 0x67
2945 * Cause the #SS fault with 0 error code in VM86 mode.
2947 if (((vec
== GP_VECTOR
) || (vec
== SS_VECTOR
)) && err_code
== 0)
2948 if (emulate_instruction(vcpu
, 0, 0, 0) == EMULATE_DONE
)
2951 * Forward all other exceptions that are valid in real mode.
2952 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2953 * the required debugging infrastructure rework.
2957 if (vcpu
->guest_debug
&
2958 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
))
2960 kvm_queue_exception(vcpu
, vec
);
2964 * Update instruction length as we may reinject the exception
2965 * from user space while in guest debugging mode.
2967 to_vmx(vcpu
)->vcpu
.arch
.event_exit_inst_len
=
2968 vmcs_read32(VM_EXIT_INSTRUCTION_LEN
);
2969 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_SW_BP
)
2980 kvm_queue_exception(vcpu
, vec
);
2987 * Trigger machine check on the host. We assume all the MSRs are already set up
2988 * by the CPU and that we still run on the same CPU as the MCE occurred on.
2989 * We pass a fake environment to the machine check handler because we want
2990 * the guest to be always treated like user space, no matter what context
2991 * it used internally.
2993 static void kvm_machine_check(void)
2995 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
2996 struct pt_regs regs
= {
2997 .cs
= 3, /* Fake ring 3 no matter what the guest ran on */
2998 .flags
= X86_EFLAGS_IF
,
3001 do_machine_check(®s
, 0);
3005 static int handle_machine_check(struct kvm_vcpu
*vcpu
)
3007 /* already handled by vcpu_run */
3011 static int handle_exception(struct kvm_vcpu
*vcpu
)
3013 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
3014 struct kvm_run
*kvm_run
= vcpu
->run
;
3015 u32 intr_info
, ex_no
, error_code
;
3016 unsigned long cr2
, rip
, dr6
;
3018 enum emulation_result er
;
3020 vect_info
= vmx
->idt_vectoring_info
;
3021 intr_info
= vmcs_read32(VM_EXIT_INTR_INFO
);
3023 if (is_machine_check(intr_info
))
3024 return handle_machine_check(vcpu
);
3026 if ((vect_info
& VECTORING_INFO_VALID_MASK
) &&
3027 !is_page_fault(intr_info
)) {
3028 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
3029 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_SIMUL_EX
;
3030 vcpu
->run
->internal
.ndata
= 2;
3031 vcpu
->run
->internal
.data
[0] = vect_info
;
3032 vcpu
->run
->internal
.data
[1] = intr_info
;
3036 if ((intr_info
& INTR_INFO_INTR_TYPE_MASK
) == INTR_TYPE_NMI_INTR
)
3037 return 1; /* already handled by vmx_vcpu_run() */
3039 if (is_no_device(intr_info
)) {
3040 vmx_fpu_activate(vcpu
);
3044 if (is_invalid_opcode(intr_info
)) {
3045 er
= emulate_instruction(vcpu
, 0, 0, EMULTYPE_TRAP_UD
);
3046 if (er
!= EMULATE_DONE
)
3047 kvm_queue_exception(vcpu
, UD_VECTOR
);
3052 rip
= kvm_rip_read(vcpu
);
3053 if (intr_info
& INTR_INFO_DELIVER_CODE_MASK
)
3054 error_code
= vmcs_read32(VM_EXIT_INTR_ERROR_CODE
);
3055 if (is_page_fault(intr_info
)) {
3056 /* EPT won't cause page fault directly */
3059 cr2
= vmcs_readl(EXIT_QUALIFICATION
);
3060 trace_kvm_page_fault(cr2
, error_code
);
3062 if (kvm_event_needs_reinjection(vcpu
))
3063 kvm_mmu_unprotect_page_virt(vcpu
, cr2
);
3064 return kvm_mmu_page_fault(vcpu
, cr2
, error_code
);
3067 if (vmx
->rmode
.vm86_active
&&
3068 handle_rmode_exception(vcpu
, intr_info
& INTR_INFO_VECTOR_MASK
,
3070 if (vcpu
->arch
.halt_request
) {
3071 vcpu
->arch
.halt_request
= 0;
3072 return kvm_emulate_halt(vcpu
);
3077 ex_no
= intr_info
& INTR_INFO_VECTOR_MASK
;
3080 dr6
= vmcs_readl(EXIT_QUALIFICATION
);
3081 if (!(vcpu
->guest_debug
&
3082 (KVM_GUESTDBG_SINGLESTEP
| KVM_GUESTDBG_USE_HW_BP
))) {
3083 vcpu
->arch
.dr6
= dr6
| DR6_FIXED_1
;
3084 kvm_queue_exception(vcpu
, DB_VECTOR
);
3087 kvm_run
->debug
.arch
.dr6
= dr6
| DR6_FIXED_1
;
3088 kvm_run
->debug
.arch
.dr7
= vmcs_readl(GUEST_DR7
);
3092 * Update instruction length as we may reinject #BP from
3093 * user space while in guest debugging mode. Reading it for
3094 * #DB as well causes no harm, it is not used in that case.
3096 vmx
->vcpu
.arch
.event_exit_inst_len
=
3097 vmcs_read32(VM_EXIT_INSTRUCTION_LEN
);
3098 kvm_run
->exit_reason
= KVM_EXIT_DEBUG
;
3099 kvm_run
->debug
.arch
.pc
= vmcs_readl(GUEST_CS_BASE
) + rip
;
3100 kvm_run
->debug
.arch
.exception
= ex_no
;
3103 kvm_run
->exit_reason
= KVM_EXIT_EXCEPTION
;
3104 kvm_run
->ex
.exception
= ex_no
;
3105 kvm_run
->ex
.error_code
= error_code
;
3111 static int handle_external_interrupt(struct kvm_vcpu
*vcpu
)
3113 ++vcpu
->stat
.irq_exits
;
3117 static int handle_triple_fault(struct kvm_vcpu
*vcpu
)
3119 vcpu
->run
->exit_reason
= KVM_EXIT_SHUTDOWN
;
3123 static int handle_io(struct kvm_vcpu
*vcpu
)
3125 unsigned long exit_qualification
;
3126 int size
, in
, string
;
3129 exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
3130 string
= (exit_qualification
& 16) != 0;
3131 in
= (exit_qualification
& 8) != 0;
3133 ++vcpu
->stat
.io_exits
;
3136 return emulate_instruction(vcpu
, 0, 0, 0) == EMULATE_DONE
;
3138 port
= exit_qualification
>> 16;
3139 size
= (exit_qualification
& 7) + 1;
3140 skip_emulated_instruction(vcpu
);
3142 return kvm_fast_pio_out(vcpu
, size
, port
);
3146 vmx_patch_hypercall(struct kvm_vcpu
*vcpu
, unsigned char *hypercall
)
3149 * Patch in the VMCALL instruction:
3151 hypercall
[0] = 0x0f;
3152 hypercall
[1] = 0x01;
3153 hypercall
[2] = 0xc1;
3156 static void complete_insn_gp(struct kvm_vcpu
*vcpu
, int err
)
3159 kvm_inject_gp(vcpu
, 0);
3161 skip_emulated_instruction(vcpu
);
3164 static int handle_cr(struct kvm_vcpu
*vcpu
)
3166 unsigned long exit_qualification
, val
;
3171 exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
3172 cr
= exit_qualification
& 15;
3173 reg
= (exit_qualification
>> 8) & 15;
3174 switch ((exit_qualification
>> 4) & 3) {
3175 case 0: /* mov to cr */
3176 val
= kvm_register_read(vcpu
, reg
);
3177 trace_kvm_cr_write(cr
, val
);
3180 err
= kvm_set_cr0(vcpu
, val
);
3181 complete_insn_gp(vcpu
, err
);
3184 err
= kvm_set_cr3(vcpu
, val
);
3185 complete_insn_gp(vcpu
, err
);
3188 err
= kvm_set_cr4(vcpu
, val
);
3189 complete_insn_gp(vcpu
, err
);
3192 u8 cr8_prev
= kvm_get_cr8(vcpu
);
3193 u8 cr8
= kvm_register_read(vcpu
, reg
);
3194 kvm_set_cr8(vcpu
, cr8
);
3195 skip_emulated_instruction(vcpu
);
3196 if (irqchip_in_kernel(vcpu
->kvm
))
3198 if (cr8_prev
<= cr8
)
3200 vcpu
->run
->exit_reason
= KVM_EXIT_SET_TPR
;
3206 vmx_set_cr0(vcpu
, kvm_read_cr0_bits(vcpu
, ~X86_CR0_TS
));
3207 trace_kvm_cr_write(0, kvm_read_cr0(vcpu
));
3208 skip_emulated_instruction(vcpu
);
3209 vmx_fpu_activate(vcpu
);
3211 case 1: /*mov from cr*/
3214 kvm_register_write(vcpu
, reg
, vcpu
->arch
.cr3
);
3215 trace_kvm_cr_read(cr
, vcpu
->arch
.cr3
);
3216 skip_emulated_instruction(vcpu
);
3219 val
= kvm_get_cr8(vcpu
);
3220 kvm_register_write(vcpu
, reg
, val
);
3221 trace_kvm_cr_read(cr
, val
);
3222 skip_emulated_instruction(vcpu
);
3227 val
= (exit_qualification
>> LMSW_SOURCE_DATA_SHIFT
) & 0x0f;
3228 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu
) & ~0xful
) | val
);
3229 kvm_lmsw(vcpu
, val
);
3231 skip_emulated_instruction(vcpu
);
3236 vcpu
->run
->exit_reason
= 0;
3237 pr_unimpl(vcpu
, "unhandled control register: op %d cr %d\n",
3238 (int)(exit_qualification
>> 4) & 3, cr
);
3242 static int handle_dr(struct kvm_vcpu
*vcpu
)
3244 unsigned long exit_qualification
;
3247 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
3248 if (!kvm_require_cpl(vcpu
, 0))
3250 dr
= vmcs_readl(GUEST_DR7
);
3253 * As the vm-exit takes precedence over the debug trap, we
3254 * need to emulate the latter, either for the host or the
3255 * guest debugging itself.
3257 if (vcpu
->guest_debug
& KVM_GUESTDBG_USE_HW_BP
) {
3258 vcpu
->run
->debug
.arch
.dr6
= vcpu
->arch
.dr6
;
3259 vcpu
->run
->debug
.arch
.dr7
= dr
;
3260 vcpu
->run
->debug
.arch
.pc
=
3261 vmcs_readl(GUEST_CS_BASE
) +
3262 vmcs_readl(GUEST_RIP
);
3263 vcpu
->run
->debug
.arch
.exception
= DB_VECTOR
;
3264 vcpu
->run
->exit_reason
= KVM_EXIT_DEBUG
;
3267 vcpu
->arch
.dr7
&= ~DR7_GD
;
3268 vcpu
->arch
.dr6
|= DR6_BD
;
3269 vmcs_writel(GUEST_DR7
, vcpu
->arch
.dr7
);
3270 kvm_queue_exception(vcpu
, DB_VECTOR
);
3275 exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
3276 dr
= exit_qualification
& DEBUG_REG_ACCESS_NUM
;
3277 reg
= DEBUG_REG_ACCESS_REG(exit_qualification
);
3278 if (exit_qualification
& TYPE_MOV_FROM_DR
) {
3280 if (!kvm_get_dr(vcpu
, dr
, &val
))
3281 kvm_register_write(vcpu
, reg
, val
);
3283 kvm_set_dr(vcpu
, dr
, vcpu
->arch
.regs
[reg
]);
3284 skip_emulated_instruction(vcpu
);
3288 static void vmx_set_dr7(struct kvm_vcpu
*vcpu
, unsigned long val
)
3290 vmcs_writel(GUEST_DR7
, val
);
3293 static int handle_cpuid(struct kvm_vcpu
*vcpu
)
3295 kvm_emulate_cpuid(vcpu
);
3299 static int handle_rdmsr(struct kvm_vcpu
*vcpu
)
3301 u32 ecx
= vcpu
->arch
.regs
[VCPU_REGS_RCX
];
3304 if (vmx_get_msr(vcpu
, ecx
, &data
)) {
3305 trace_kvm_msr_read_ex(ecx
);
3306 kvm_inject_gp(vcpu
, 0);
3310 trace_kvm_msr_read(ecx
, data
);
3312 /* FIXME: handling of bits 32:63 of rax, rdx */
3313 vcpu
->arch
.regs
[VCPU_REGS_RAX
] = data
& -1u;
3314 vcpu
->arch
.regs
[VCPU_REGS_RDX
] = (data
>> 32) & -1u;
3315 skip_emulated_instruction(vcpu
);
3319 static int handle_wrmsr(struct kvm_vcpu
*vcpu
)
3321 u32 ecx
= vcpu
->arch
.regs
[VCPU_REGS_RCX
];
3322 u64 data
= (vcpu
->arch
.regs
[VCPU_REGS_RAX
] & -1u)
3323 | ((u64
)(vcpu
->arch
.regs
[VCPU_REGS_RDX
] & -1u) << 32);
3325 if (vmx_set_msr(vcpu
, ecx
, data
) != 0) {
3326 trace_kvm_msr_write_ex(ecx
, data
);
3327 kvm_inject_gp(vcpu
, 0);
3331 trace_kvm_msr_write(ecx
, data
);
3332 skip_emulated_instruction(vcpu
);
3336 static int handle_tpr_below_threshold(struct kvm_vcpu
*vcpu
)
3341 static int handle_interrupt_window(struct kvm_vcpu
*vcpu
)
3343 u32 cpu_based_vm_exec_control
;
3345 /* clear pending irq */
3346 cpu_based_vm_exec_control
= vmcs_read32(CPU_BASED_VM_EXEC_CONTROL
);
3347 cpu_based_vm_exec_control
&= ~CPU_BASED_VIRTUAL_INTR_PENDING
;
3348 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL
, cpu_based_vm_exec_control
);
3350 ++vcpu
->stat
.irq_window_exits
;
3353 * If the user space waits to inject interrupts, exit as soon as
3356 if (!irqchip_in_kernel(vcpu
->kvm
) &&
3357 vcpu
->run
->request_interrupt_window
&&
3358 !kvm_cpu_has_interrupt(vcpu
)) {
3359 vcpu
->run
->exit_reason
= KVM_EXIT_IRQ_WINDOW_OPEN
;
3365 static int handle_halt(struct kvm_vcpu
*vcpu
)
3367 skip_emulated_instruction(vcpu
);
3368 return kvm_emulate_halt(vcpu
);
3371 static int handle_vmcall(struct kvm_vcpu
*vcpu
)
3373 skip_emulated_instruction(vcpu
);
3374 kvm_emulate_hypercall(vcpu
);
3378 static int handle_vmx_insn(struct kvm_vcpu
*vcpu
)
3380 kvm_queue_exception(vcpu
, UD_VECTOR
);
3384 static int handle_invlpg(struct kvm_vcpu
*vcpu
)
3386 unsigned long exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
3388 kvm_mmu_invlpg(vcpu
, exit_qualification
);
3389 skip_emulated_instruction(vcpu
);
3393 static int handle_wbinvd(struct kvm_vcpu
*vcpu
)
3395 skip_emulated_instruction(vcpu
);
3396 kvm_emulate_wbinvd(vcpu
);
3400 static int handle_xsetbv(struct kvm_vcpu
*vcpu
)
3402 u64 new_bv
= kvm_read_edx_eax(vcpu
);
3403 u32 index
= kvm_register_read(vcpu
, VCPU_REGS_RCX
);
3405 if (kvm_set_xcr(vcpu
, index
, new_bv
) == 0)
3406 skip_emulated_instruction(vcpu
);
3410 static int handle_apic_access(struct kvm_vcpu
*vcpu
)
3412 return emulate_instruction(vcpu
, 0, 0, 0) == EMULATE_DONE
;
3415 static int handle_task_switch(struct kvm_vcpu
*vcpu
)
3417 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
3418 unsigned long exit_qualification
;
3419 bool has_error_code
= false;
3422 int reason
, type
, idt_v
;
3424 idt_v
= (vmx
->idt_vectoring_info
& VECTORING_INFO_VALID_MASK
);
3425 type
= (vmx
->idt_vectoring_info
& VECTORING_INFO_TYPE_MASK
);
3427 exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
3429 reason
= (u32
)exit_qualification
>> 30;
3430 if (reason
== TASK_SWITCH_GATE
&& idt_v
) {
3432 case INTR_TYPE_NMI_INTR
:
3433 vcpu
->arch
.nmi_injected
= false;
3434 if (cpu_has_virtual_nmis())
3435 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO
,
3436 GUEST_INTR_STATE_NMI
);
3438 case INTR_TYPE_EXT_INTR
:
3439 case INTR_TYPE_SOFT_INTR
:
3440 kvm_clear_interrupt_queue(vcpu
);
3442 case INTR_TYPE_HARD_EXCEPTION
:
3443 if (vmx
->idt_vectoring_info
&
3444 VECTORING_INFO_DELIVER_CODE_MASK
) {
3445 has_error_code
= true;
3447 vmcs_read32(IDT_VECTORING_ERROR_CODE
);
3450 case INTR_TYPE_SOFT_EXCEPTION
:
3451 kvm_clear_exception_queue(vcpu
);
3457 tss_selector
= exit_qualification
;
3459 if (!idt_v
|| (type
!= INTR_TYPE_HARD_EXCEPTION
&&
3460 type
!= INTR_TYPE_EXT_INTR
&&
3461 type
!= INTR_TYPE_NMI_INTR
))
3462 skip_emulated_instruction(vcpu
);
3464 if (kvm_task_switch(vcpu
, tss_selector
, reason
,
3465 has_error_code
, error_code
) == EMULATE_FAIL
) {
3466 vcpu
->run
->exit_reason
= KVM_EXIT_INTERNAL_ERROR
;
3467 vcpu
->run
->internal
.suberror
= KVM_INTERNAL_ERROR_EMULATION
;
3468 vcpu
->run
->internal
.ndata
= 0;
3472 /* clear all local breakpoint enable flags */
3473 vmcs_writel(GUEST_DR7
, vmcs_readl(GUEST_DR7
) & ~55);
3476 * TODO: What about debug traps on tss switch?
3477 * Are we supposed to inject them and update dr6?
3483 static int handle_ept_violation(struct kvm_vcpu
*vcpu
)
3485 unsigned long exit_qualification
;
3489 exit_qualification
= vmcs_readl(EXIT_QUALIFICATION
);
3491 if (exit_qualification
& (1 << 6)) {
3492 printk(KERN_ERR
"EPT: GPA exceeds GAW!\n");
3496 gla_validity
= (exit_qualification
>> 7) & 0x3;
3497 if (gla_validity
!= 0x3 && gla_validity
!= 0x1 && gla_validity
!= 0) {
3498 printk(KERN_ERR
"EPT: Handling EPT violation failed!\n");
3499 printk(KERN_ERR
"EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3500 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS
),
3501 vmcs_readl(GUEST_LINEAR_ADDRESS
));
3502 printk(KERN_ERR
"EPT: Exit qualification is 0x%lx\n",
3503 (long unsigned int)exit_qualification
);
3504 vcpu
->run
->exit_reason
= KVM_EXIT_UNKNOWN
;
3505 vcpu
->run
->hw
.hardware_exit_reason
= EXIT_REASON_EPT_VIOLATION
;
3509 gpa
= vmcs_read64(GUEST_PHYSICAL_ADDRESS
);
3510 trace_kvm_page_fault(gpa
, exit_qualification
);
3511 return kvm_mmu_page_fault(vcpu
, gpa
& PAGE_MASK
, 0);
3514 static u64
ept_rsvd_mask(u64 spte
, int level
)
3519 for (i
= 51; i
> boot_cpu_data
.x86_phys_bits
; i
--)
3520 mask
|= (1ULL << i
);
3523 /* bits 7:3 reserved */
3525 else if (level
== 2) {
3526 if (spte
& (1ULL << 7))
3527 /* 2MB ref, bits 20:12 reserved */
3530 /* bits 6:3 reserved */
3537 static void ept_misconfig_inspect_spte(struct kvm_vcpu
*vcpu
, u64 spte
,
3540 printk(KERN_ERR
"%s: spte 0x%llx level %d\n", __func__
, spte
, level
);
3542 /* 010b (write-only) */
3543 WARN_ON((spte
& 0x7) == 0x2);
3545 /* 110b (write/execute) */
3546 WARN_ON((spte
& 0x7) == 0x6);
3548 /* 100b (execute-only) and value not supported by logical processor */
3549 if (!cpu_has_vmx_ept_execute_only())
3550 WARN_ON((spte
& 0x7) == 0x4);
3554 u64 rsvd_bits
= spte
& ept_rsvd_mask(spte
, level
);
3556 if (rsvd_bits
!= 0) {
3557 printk(KERN_ERR
"%s: rsvd_bits = 0x%llx\n",
3558 __func__
, rsvd_bits
);
3562 if (level
== 1 || (level
== 2 && (spte
& (1ULL << 7)))) {
3563 u64 ept_mem_type
= (spte
& 0x38) >> 3;
3565 if (ept_mem_type
== 2 || ept_mem_type
== 3 ||
3566 ept_mem_type
== 7) {
3567 printk(KERN_ERR
"%s: ept_mem_type=0x%llx\n",
3568 __func__
, ept_mem_type
);
3575 static int handle_ept_misconfig(struct kvm_vcpu
*vcpu
)
3581 gpa
= vmcs_read64(GUEST_PHYSICAL_ADDRESS
);
3583 printk(KERN_ERR
"EPT: Misconfiguration.\n");
3584 printk(KERN_ERR
"EPT: GPA: 0x%llx\n", gpa
);
3586 nr_sptes
= kvm_mmu_get_spte_hierarchy(vcpu
, gpa
, sptes
);
3588 for (i
= PT64_ROOT_LEVEL
; i
> PT64_ROOT_LEVEL
- nr_sptes
; --i
)
3589 ept_misconfig_inspect_spte(vcpu
, sptes
[i
-1], i
);
3591 vcpu
->run
->exit_reason
= KVM_EXIT_UNKNOWN
;
3592 vcpu
->run
->hw
.hardware_exit_reason
= EXIT_REASON_EPT_MISCONFIG
;
3597 static int handle_nmi_window(struct kvm_vcpu
*vcpu
)
3599 u32 cpu_based_vm_exec_control
;
3601 /* clear pending NMI */
3602 cpu_based_vm_exec_control
= vmcs_read32(CPU_BASED_VM_EXEC_CONTROL
);
3603 cpu_based_vm_exec_control
&= ~CPU_BASED_VIRTUAL_NMI_PENDING
;
3604 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL
, cpu_based_vm_exec_control
);
3605 ++vcpu
->stat
.nmi_window_exits
;
3610 static int handle_invalid_guest_state(struct kvm_vcpu
*vcpu
)
3612 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
3613 enum emulation_result err
= EMULATE_DONE
;
3616 while (!guest_state_valid(vcpu
)) {
3617 err
= emulate_instruction(vcpu
, 0, 0, 0);
3619 if (err
== EMULATE_DO_MMIO
) {
3624 if (err
!= EMULATE_DONE
)
3627 if (signal_pending(current
))
3633 vmx
->emulation_required
= 0;
3639 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
3640 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
3642 static int handle_pause(struct kvm_vcpu
*vcpu
)
3644 skip_emulated_instruction(vcpu
);
3645 kvm_vcpu_on_spin(vcpu
);
3650 static int handle_invalid_op(struct kvm_vcpu
*vcpu
)
3652 kvm_queue_exception(vcpu
, UD_VECTOR
);
3657 * The exit handlers return 1 if the exit was handled fully and guest execution
3658 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
3659 * to be done to userspace and return 0.
3661 static int (*kvm_vmx_exit_handlers
[])(struct kvm_vcpu
*vcpu
) = {
3662 [EXIT_REASON_EXCEPTION_NMI
] = handle_exception
,
3663 [EXIT_REASON_EXTERNAL_INTERRUPT
] = handle_external_interrupt
,
3664 [EXIT_REASON_TRIPLE_FAULT
] = handle_triple_fault
,
3665 [EXIT_REASON_NMI_WINDOW
] = handle_nmi_window
,
3666 [EXIT_REASON_IO_INSTRUCTION
] = handle_io
,
3667 [EXIT_REASON_CR_ACCESS
] = handle_cr
,
3668 [EXIT_REASON_DR_ACCESS
] = handle_dr
,
3669 [EXIT_REASON_CPUID
] = handle_cpuid
,
3670 [EXIT_REASON_MSR_READ
] = handle_rdmsr
,
3671 [EXIT_REASON_MSR_WRITE
] = handle_wrmsr
,
3672 [EXIT_REASON_PENDING_INTERRUPT
] = handle_interrupt_window
,
3673 [EXIT_REASON_HLT
] = handle_halt
,
3674 [EXIT_REASON_INVLPG
] = handle_invlpg
,
3675 [EXIT_REASON_VMCALL
] = handle_vmcall
,
3676 [EXIT_REASON_VMCLEAR
] = handle_vmx_insn
,
3677 [EXIT_REASON_VMLAUNCH
] = handle_vmx_insn
,
3678 [EXIT_REASON_VMPTRLD
] = handle_vmx_insn
,
3679 [EXIT_REASON_VMPTRST
] = handle_vmx_insn
,
3680 [EXIT_REASON_VMREAD
] = handle_vmx_insn
,
3681 [EXIT_REASON_VMRESUME
] = handle_vmx_insn
,
3682 [EXIT_REASON_VMWRITE
] = handle_vmx_insn
,
3683 [EXIT_REASON_VMOFF
] = handle_vmx_insn
,
3684 [EXIT_REASON_VMON
] = handle_vmx_insn
,
3685 [EXIT_REASON_TPR_BELOW_THRESHOLD
] = handle_tpr_below_threshold
,
3686 [EXIT_REASON_APIC_ACCESS
] = handle_apic_access
,
3687 [EXIT_REASON_WBINVD
] = handle_wbinvd
,
3688 [EXIT_REASON_XSETBV
] = handle_xsetbv
,
3689 [EXIT_REASON_TASK_SWITCH
] = handle_task_switch
,
3690 [EXIT_REASON_MCE_DURING_VMENTRY
] = handle_machine_check
,
3691 [EXIT_REASON_EPT_VIOLATION
] = handle_ept_violation
,
3692 [EXIT_REASON_EPT_MISCONFIG
] = handle_ept_misconfig
,
3693 [EXIT_REASON_PAUSE_INSTRUCTION
] = handle_pause
,
3694 [EXIT_REASON_MWAIT_INSTRUCTION
] = handle_invalid_op
,
3695 [EXIT_REASON_MONITOR_INSTRUCTION
] = handle_invalid_op
,
3698 static const int kvm_vmx_max_exit_handlers
=
3699 ARRAY_SIZE(kvm_vmx_exit_handlers
);
3702 * The guest has exited. See if we can fix it or if we need userspace
3705 static int vmx_handle_exit(struct kvm_vcpu
*vcpu
)
3707 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
3708 u32 exit_reason
= vmx
->exit_reason
;
3709 u32 vectoring_info
= vmx
->idt_vectoring_info
;
3711 trace_kvm_exit(exit_reason
, vcpu
);
3713 /* If guest state is invalid, start emulating */
3714 if (vmx
->emulation_required
&& emulate_invalid_guest_state
)
3715 return handle_invalid_guest_state(vcpu
);
3717 /* Access CR3 don't cause VMExit in paging mode, so we need
3718 * to sync with guest real CR3. */
3719 if (enable_ept
&& is_paging(vcpu
))
3720 vcpu
->arch
.cr3
= vmcs_readl(GUEST_CR3
);
3722 if (exit_reason
& VMX_EXIT_REASONS_FAILED_VMENTRY
) {
3723 vcpu
->run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
3724 vcpu
->run
->fail_entry
.hardware_entry_failure_reason
3729 if (unlikely(vmx
->fail
)) {
3730 vcpu
->run
->exit_reason
= KVM_EXIT_FAIL_ENTRY
;
3731 vcpu
->run
->fail_entry
.hardware_entry_failure_reason
3732 = vmcs_read32(VM_INSTRUCTION_ERROR
);
3736 if ((vectoring_info
& VECTORING_INFO_VALID_MASK
) &&
3737 (exit_reason
!= EXIT_REASON_EXCEPTION_NMI
&&
3738 exit_reason
!= EXIT_REASON_EPT_VIOLATION
&&
3739 exit_reason
!= EXIT_REASON_TASK_SWITCH
))
3740 printk(KERN_WARNING
"%s: unexpected, valid vectoring info "
3741 "(0x%x) and exit reason is 0x%x\n",
3742 __func__
, vectoring_info
, exit_reason
);
3744 if (unlikely(!cpu_has_virtual_nmis() && vmx
->soft_vnmi_blocked
)) {
3745 if (vmx_interrupt_allowed(vcpu
)) {
3746 vmx
->soft_vnmi_blocked
= 0;
3747 } else if (vmx
->vnmi_blocked_time
> 1000000000LL &&
3748 vcpu
->arch
.nmi_pending
) {
3750 * This CPU don't support us in finding the end of an
3751 * NMI-blocked window if the guest runs with IRQs
3752 * disabled. So we pull the trigger after 1 s of
3753 * futile waiting, but inform the user about this.
3755 printk(KERN_WARNING
"%s: Breaking out of NMI-blocked "
3756 "state on VCPU %d after 1 s timeout\n",
3757 __func__
, vcpu
->vcpu_id
);
3758 vmx
->soft_vnmi_blocked
= 0;
3762 if (exit_reason
< kvm_vmx_max_exit_handlers
3763 && kvm_vmx_exit_handlers
[exit_reason
])
3764 return kvm_vmx_exit_handlers
[exit_reason
](vcpu
);
3766 vcpu
->run
->exit_reason
= KVM_EXIT_UNKNOWN
;
3767 vcpu
->run
->hw
.hardware_exit_reason
= exit_reason
;
3772 static void update_cr8_intercept(struct kvm_vcpu
*vcpu
, int tpr
, int irr
)
3774 if (irr
== -1 || tpr
< irr
) {
3775 vmcs_write32(TPR_THRESHOLD
, 0);
3779 vmcs_write32(TPR_THRESHOLD
, irr
);
3782 static void vmx_complete_interrupts(struct vcpu_vmx
*vmx
)
3785 u32 idt_vectoring_info
= vmx
->idt_vectoring_info
;
3789 bool idtv_info_valid
;
3791 exit_intr_info
= vmcs_read32(VM_EXIT_INTR_INFO
);
3793 vmx
->exit_reason
= vmcs_read32(VM_EXIT_REASON
);
3795 /* Handle machine checks before interrupts are enabled */
3796 if ((vmx
->exit_reason
== EXIT_REASON_MCE_DURING_VMENTRY
)
3797 || (vmx
->exit_reason
== EXIT_REASON_EXCEPTION_NMI
3798 && is_machine_check(exit_intr_info
)))
3799 kvm_machine_check();
3801 /* We need to handle NMIs before interrupts are enabled */
3802 if ((exit_intr_info
& INTR_INFO_INTR_TYPE_MASK
) == INTR_TYPE_NMI_INTR
&&
3803 (exit_intr_info
& INTR_INFO_VALID_MASK
)) {
3804 kvm_before_handle_nmi(&vmx
->vcpu
);
3806 kvm_after_handle_nmi(&vmx
->vcpu
);
3809 idtv_info_valid
= idt_vectoring_info
& VECTORING_INFO_VALID_MASK
;
3811 if (cpu_has_virtual_nmis()) {
3812 unblock_nmi
= (exit_intr_info
& INTR_INFO_UNBLOCK_NMI
) != 0;
3813 vector
= exit_intr_info
& INTR_INFO_VECTOR_MASK
;
3815 * SDM 3: 27.7.1.2 (September 2008)
3816 * Re-set bit "block by NMI" before VM entry if vmexit caused by
3817 * a guest IRET fault.
3818 * SDM 3: 23.2.2 (September 2008)
3819 * Bit 12 is undefined in any of the following cases:
3820 * If the VM exit sets the valid bit in the IDT-vectoring
3821 * information field.
3822 * If the VM exit is due to a double fault.
3824 if ((exit_intr_info
& INTR_INFO_VALID_MASK
) && unblock_nmi
&&
3825 vector
!= DF_VECTOR
&& !idtv_info_valid
)
3826 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO
,
3827 GUEST_INTR_STATE_NMI
);
3828 } else if (unlikely(vmx
->soft_vnmi_blocked
))
3829 vmx
->vnmi_blocked_time
+=
3830 ktime_to_ns(ktime_sub(ktime_get(), vmx
->entry_time
));
3832 vmx
->vcpu
.arch
.nmi_injected
= false;
3833 kvm_clear_exception_queue(&vmx
->vcpu
);
3834 kvm_clear_interrupt_queue(&vmx
->vcpu
);
3836 if (!idtv_info_valid
)
3839 vector
= idt_vectoring_info
& VECTORING_INFO_VECTOR_MASK
;
3840 type
= idt_vectoring_info
& VECTORING_INFO_TYPE_MASK
;
3843 case INTR_TYPE_NMI_INTR
:
3844 vmx
->vcpu
.arch
.nmi_injected
= true;
3846 * SDM 3: 27.7.1.2 (September 2008)
3847 * Clear bit "block by NMI" before VM entry if a NMI
3850 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO
,
3851 GUEST_INTR_STATE_NMI
);
3853 case INTR_TYPE_SOFT_EXCEPTION
:
3854 vmx
->vcpu
.arch
.event_exit_inst_len
=
3855 vmcs_read32(VM_EXIT_INSTRUCTION_LEN
);
3857 case INTR_TYPE_HARD_EXCEPTION
:
3858 if (idt_vectoring_info
& VECTORING_INFO_DELIVER_CODE_MASK
) {
3859 u32 err
= vmcs_read32(IDT_VECTORING_ERROR_CODE
);
3860 kvm_queue_exception_e(&vmx
->vcpu
, vector
, err
);
3862 kvm_queue_exception(&vmx
->vcpu
, vector
);
3864 case INTR_TYPE_SOFT_INTR
:
3865 vmx
->vcpu
.arch
.event_exit_inst_len
=
3866 vmcs_read32(VM_EXIT_INSTRUCTION_LEN
);
3868 case INTR_TYPE_EXT_INTR
:
3869 kvm_queue_interrupt(&vmx
->vcpu
, vector
,
3870 type
== INTR_TYPE_SOFT_INTR
);
3878 * Failure to inject an interrupt should give us the information
3879 * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs
3880 * when fetching the interrupt redirection bitmap in the real-mode
3881 * tss, this doesn't happen. So we do it ourselves.
3883 static void fixup_rmode_irq(struct vcpu_vmx
*vmx
)
3885 vmx
->rmode
.irq
.pending
= 0;
3886 if (kvm_rip_read(&vmx
->vcpu
) + 1 != vmx
->rmode
.irq
.rip
)
3888 kvm_rip_write(&vmx
->vcpu
, vmx
->rmode
.irq
.rip
);
3889 if (vmx
->idt_vectoring_info
& VECTORING_INFO_VALID_MASK
) {
3890 vmx
->idt_vectoring_info
&= ~VECTORING_INFO_TYPE_MASK
;
3891 vmx
->idt_vectoring_info
|= INTR_TYPE_EXT_INTR
;
3894 vmx
->idt_vectoring_info
=
3895 VECTORING_INFO_VALID_MASK
3896 | INTR_TYPE_EXT_INTR
3897 | vmx
->rmode
.irq
.vector
;
3900 #ifdef CONFIG_X86_64
3908 static void vmx_vcpu_run(struct kvm_vcpu
*vcpu
)
3910 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
3912 /* Record the guest's net vcpu time for enforced NMI injections. */
3913 if (unlikely(!cpu_has_virtual_nmis() && vmx
->soft_vnmi_blocked
))
3914 vmx
->entry_time
= ktime_get();
3916 /* Don't enter VMX if guest state is invalid, let the exit handler
3917 start emulation until we arrive back to a valid state */
3918 if (vmx
->emulation_required
&& emulate_invalid_guest_state
)
3921 if (test_bit(VCPU_REGS_RSP
, (unsigned long *)&vcpu
->arch
.regs_dirty
))
3922 vmcs_writel(GUEST_RSP
, vcpu
->arch
.regs
[VCPU_REGS_RSP
]);
3923 if (test_bit(VCPU_REGS_RIP
, (unsigned long *)&vcpu
->arch
.regs_dirty
))
3924 vmcs_writel(GUEST_RIP
, vcpu
->arch
.regs
[VCPU_REGS_RIP
]);
3926 /* When single-stepping over STI and MOV SS, we must clear the
3927 * corresponding interruptibility bits in the guest state. Otherwise
3928 * vmentry fails as it then expects bit 14 (BS) in pending debug
3929 * exceptions being set, but that's not correct for the guest debugging
3931 if (vcpu
->guest_debug
& KVM_GUESTDBG_SINGLESTEP
)
3932 vmx_set_interrupt_shadow(vcpu
, 0);
3935 /* Store host registers */
3936 "push %%"R
"dx; push %%"R
"bp;"
3938 "cmp %%"R
"sp, %c[host_rsp](%0) \n\t"
3940 "mov %%"R
"sp, %c[host_rsp](%0) \n\t"
3941 __ex(ASM_VMX_VMWRITE_RSP_RDX
) "\n\t"
3943 /* Reload cr2 if changed */
3944 "mov %c[cr2](%0), %%"R
"ax \n\t"
3945 "mov %%cr2, %%"R
"dx \n\t"
3946 "cmp %%"R
"ax, %%"R
"dx \n\t"
3948 "mov %%"R
"ax, %%cr2 \n\t"
3950 /* Check if vmlaunch of vmresume is needed */
3951 "cmpl $0, %c[launched](%0) \n\t"
3952 /* Load guest registers. Don't clobber flags. */
3953 "mov %c[rax](%0), %%"R
"ax \n\t"
3954 "mov %c[rbx](%0), %%"R
"bx \n\t"
3955 "mov %c[rdx](%0), %%"R
"dx \n\t"
3956 "mov %c[rsi](%0), %%"R
"si \n\t"
3957 "mov %c[rdi](%0), %%"R
"di \n\t"
3958 "mov %c[rbp](%0), %%"R
"bp \n\t"
3959 #ifdef CONFIG_X86_64
3960 "mov %c[r8](%0), %%r8 \n\t"
3961 "mov %c[r9](%0), %%r9 \n\t"
3962 "mov %c[r10](%0), %%r10 \n\t"
3963 "mov %c[r11](%0), %%r11 \n\t"
3964 "mov %c[r12](%0), %%r12 \n\t"
3965 "mov %c[r13](%0), %%r13 \n\t"
3966 "mov %c[r14](%0), %%r14 \n\t"
3967 "mov %c[r15](%0), %%r15 \n\t"
3969 "mov %c[rcx](%0), %%"R
"cx \n\t" /* kills %0 (ecx) */
3971 /* Enter guest mode */
3972 "jne .Llaunched \n\t"
3973 __ex(ASM_VMX_VMLAUNCH
) "\n\t"
3974 "jmp .Lkvm_vmx_return \n\t"
3975 ".Llaunched: " __ex(ASM_VMX_VMRESUME
) "\n\t"
3976 ".Lkvm_vmx_return: "
3977 /* Save guest registers, load host registers, keep flags */
3978 "xchg %0, (%%"R
"sp) \n\t"
3979 "mov %%"R
"ax, %c[rax](%0) \n\t"
3980 "mov %%"R
"bx, %c[rbx](%0) \n\t"
3981 "push"Q
" (%%"R
"sp); pop"Q
" %c[rcx](%0) \n\t"
3982 "mov %%"R
"dx, %c[rdx](%0) \n\t"
3983 "mov %%"R
"si, %c[rsi](%0) \n\t"
3984 "mov %%"R
"di, %c[rdi](%0) \n\t"
3985 "mov %%"R
"bp, %c[rbp](%0) \n\t"
3986 #ifdef CONFIG_X86_64
3987 "mov %%r8, %c[r8](%0) \n\t"
3988 "mov %%r9, %c[r9](%0) \n\t"
3989 "mov %%r10, %c[r10](%0) \n\t"
3990 "mov %%r11, %c[r11](%0) \n\t"
3991 "mov %%r12, %c[r12](%0) \n\t"
3992 "mov %%r13, %c[r13](%0) \n\t"
3993 "mov %%r14, %c[r14](%0) \n\t"
3994 "mov %%r15, %c[r15](%0) \n\t"
3996 "mov %%cr2, %%"R
"ax \n\t"
3997 "mov %%"R
"ax, %c[cr2](%0) \n\t"
3999 "pop %%"R
"bp; pop %%"R
"bp; pop %%"R
"dx \n\t"
4000 "setbe %c[fail](%0) \n\t"
4001 : : "c"(vmx
), "d"((unsigned long)HOST_RSP
),
4002 [launched
]"i"(offsetof(struct vcpu_vmx
, launched
)),
4003 [fail
]"i"(offsetof(struct vcpu_vmx
, fail
)),
4004 [host_rsp
]"i"(offsetof(struct vcpu_vmx
, host_rsp
)),
4005 [rax
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_RAX
])),
4006 [rbx
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_RBX
])),
4007 [rcx
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_RCX
])),
4008 [rdx
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_RDX
])),
4009 [rsi
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_RSI
])),
4010 [rdi
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_RDI
])),
4011 [rbp
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_RBP
])),
4012 #ifdef CONFIG_X86_64
4013 [r8
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R8
])),
4014 [r9
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R9
])),
4015 [r10
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R10
])),
4016 [r11
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R11
])),
4017 [r12
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R12
])),
4018 [r13
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R13
])),
4019 [r14
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R14
])),
4020 [r15
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.regs
[VCPU_REGS_R15
])),
4022 [cr2
]"i"(offsetof(struct vcpu_vmx
, vcpu
.arch
.cr2
))
4024 , R
"bx", R
"di", R
"si"
4025 #ifdef CONFIG_X86_64
4026 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
4030 vcpu
->arch
.regs_avail
= ~((1 << VCPU_REGS_RIP
) | (1 << VCPU_REGS_RSP
)
4031 | (1 << VCPU_EXREG_PDPTR
));
4032 vcpu
->arch
.regs_dirty
= 0;
4034 vmx
->idt_vectoring_info
= vmcs_read32(IDT_VECTORING_INFO_FIELD
);
4035 if (vmx
->rmode
.irq
.pending
)
4036 fixup_rmode_irq(vmx
);
4038 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS
));
4041 vmx_complete_interrupts(vmx
);
4047 static void vmx_free_vmcs(struct kvm_vcpu
*vcpu
)
4049 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
4053 free_vmcs(vmx
->vmcs
);
4058 static void vmx_free_vcpu(struct kvm_vcpu
*vcpu
)
4060 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
4063 vmx_free_vmcs(vcpu
);
4064 kfree(vmx
->guest_msrs
);
4065 kvm_vcpu_uninit(vcpu
);
4066 kmem_cache_free(kvm_vcpu_cache
, vmx
);
4069 static inline void vmcs_init(struct vmcs
*vmcs
)
4071 u64 phys_addr
= __pa(per_cpu(vmxarea
, raw_smp_processor_id()));
4074 kvm_cpu_vmxon(phys_addr
);
4082 static struct kvm_vcpu
*vmx_create_vcpu(struct kvm
*kvm
, unsigned int id
)
4085 struct vcpu_vmx
*vmx
= kmem_cache_zalloc(kvm_vcpu_cache
, GFP_KERNEL
);
4089 return ERR_PTR(-ENOMEM
);
4093 err
= kvm_vcpu_init(&vmx
->vcpu
, kvm
, id
);
4097 vmx
->guest_msrs
= kmalloc(PAGE_SIZE
, GFP_KERNEL
);
4098 if (!vmx
->guest_msrs
) {
4103 vmx
->vmcs
= alloc_vmcs();
4107 vmcs_init(vmx
->vmcs
);
4110 vmx_vcpu_load(&vmx
->vcpu
, cpu
);
4111 err
= vmx_vcpu_setup(vmx
);
4112 vmx_vcpu_put(&vmx
->vcpu
);
4116 if (vm_need_virtualize_apic_accesses(kvm
))
4117 if (alloc_apic_access_page(kvm
) != 0)
4121 if (!kvm
->arch
.ept_identity_map_addr
)
4122 kvm
->arch
.ept_identity_map_addr
=
4123 VMX_EPT_IDENTITY_PAGETABLE_ADDR
;
4124 if (alloc_identity_pagetable(kvm
) != 0)
4131 free_vmcs(vmx
->vmcs
);
4133 kfree(vmx
->guest_msrs
);
4135 kvm_vcpu_uninit(&vmx
->vcpu
);
4138 kmem_cache_free(kvm_vcpu_cache
, vmx
);
4139 return ERR_PTR(err
);
4142 static void __init
vmx_check_processor_compat(void *rtn
)
4144 struct vmcs_config vmcs_conf
;
4147 if (setup_vmcs_config(&vmcs_conf
) < 0)
4149 if (memcmp(&vmcs_config
, &vmcs_conf
, sizeof(struct vmcs_config
)) != 0) {
4150 printk(KERN_ERR
"kvm: CPU %d feature inconsistency!\n",
4151 smp_processor_id());
4156 static int get_ept_level(void)
4158 return VMX_EPT_DEFAULT_GAW
+ 1;
4161 static u64
vmx_get_mt_mask(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool is_mmio
)
4165 /* For VT-d and EPT combination
4166 * 1. MMIO: always map as UC
4168 * a. VT-d without snooping control feature: can't guarantee the
4169 * result, try to trust guest.
4170 * b. VT-d with snooping control feature: snooping control feature of
4171 * VT-d engine can guarantee the cache correctness. Just set it
4172 * to WB to keep consistent with host. So the same as item 3.
4173 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
4174 * consistent with host MTRR
4177 ret
= MTRR_TYPE_UNCACHABLE
<< VMX_EPT_MT_EPTE_SHIFT
;
4178 else if (vcpu
->kvm
->arch
.iommu_domain
&&
4179 !(vcpu
->kvm
->arch
.iommu_flags
& KVM_IOMMU_CACHE_COHERENCY
))
4180 ret
= kvm_get_guest_memory_type(vcpu
, gfn
) <<
4181 VMX_EPT_MT_EPTE_SHIFT
;
4183 ret
= (MTRR_TYPE_WRBACK
<< VMX_EPT_MT_EPTE_SHIFT
)
4189 #define _ER(x) { EXIT_REASON_##x, #x }
4191 static const struct trace_print_flags vmx_exit_reasons_str
[] = {
4193 _ER(EXTERNAL_INTERRUPT
),
4195 _ER(PENDING_INTERRUPT
),
4215 _ER(IO_INSTRUCTION
),
4218 _ER(MWAIT_INSTRUCTION
),
4219 _ER(MONITOR_INSTRUCTION
),
4220 _ER(PAUSE_INSTRUCTION
),
4221 _ER(MCE_DURING_VMENTRY
),
4222 _ER(TPR_BELOW_THRESHOLD
),
4232 static int vmx_get_lpage_level(void)
4234 if (enable_ept
&& !cpu_has_vmx_ept_1g_page())
4235 return PT_DIRECTORY_LEVEL
;
4237 /* For shadow and EPT supported 1GB page */
4238 return PT_PDPE_LEVEL
;
4241 static inline u32
bit(int bitno
)
4243 return 1 << (bitno
& 31);
4246 static void vmx_cpuid_update(struct kvm_vcpu
*vcpu
)
4248 struct kvm_cpuid_entry2
*best
;
4249 struct vcpu_vmx
*vmx
= to_vmx(vcpu
);
4252 vmx
->rdtscp_enabled
= false;
4253 if (vmx_rdtscp_supported()) {
4254 exec_control
= vmcs_read32(SECONDARY_VM_EXEC_CONTROL
);
4255 if (exec_control
& SECONDARY_EXEC_RDTSCP
) {
4256 best
= kvm_find_cpuid_entry(vcpu
, 0x80000001, 0);
4257 if (best
&& (best
->edx
& bit(X86_FEATURE_RDTSCP
)))
4258 vmx
->rdtscp_enabled
= true;
4260 exec_control
&= ~SECONDARY_EXEC_RDTSCP
;
4261 vmcs_write32(SECONDARY_VM_EXEC_CONTROL
,
4268 static void vmx_set_supported_cpuid(u32 func
, struct kvm_cpuid_entry2
*entry
)
4272 static struct kvm_x86_ops vmx_x86_ops
= {
4273 .cpu_has_kvm_support
= cpu_has_kvm_support
,
4274 .disabled_by_bios
= vmx_disabled_by_bios
,
4275 .hardware_setup
= hardware_setup
,
4276 .hardware_unsetup
= hardware_unsetup
,
4277 .check_processor_compatibility
= vmx_check_processor_compat
,
4278 .hardware_enable
= hardware_enable
,
4279 .hardware_disable
= hardware_disable
,
4280 .cpu_has_accelerated_tpr
= report_flexpriority
,
4282 .vcpu_create
= vmx_create_vcpu
,
4283 .vcpu_free
= vmx_free_vcpu
,
4284 .vcpu_reset
= vmx_vcpu_reset
,
4286 .prepare_guest_switch
= vmx_save_host_state
,
4287 .vcpu_load
= vmx_vcpu_load
,
4288 .vcpu_put
= vmx_vcpu_put
,
4290 .set_guest_debug
= set_guest_debug
,
4291 .get_msr
= vmx_get_msr
,
4292 .set_msr
= vmx_set_msr
,
4293 .get_segment_base
= vmx_get_segment_base
,
4294 .get_segment
= vmx_get_segment
,
4295 .set_segment
= vmx_set_segment
,
4296 .get_cpl
= vmx_get_cpl
,
4297 .get_cs_db_l_bits
= vmx_get_cs_db_l_bits
,
4298 .decache_cr0_guest_bits
= vmx_decache_cr0_guest_bits
,
4299 .decache_cr4_guest_bits
= vmx_decache_cr4_guest_bits
,
4300 .set_cr0
= vmx_set_cr0
,
4301 .set_cr3
= vmx_set_cr3
,
4302 .set_cr4
= vmx_set_cr4
,
4303 .set_efer
= vmx_set_efer
,
4304 .get_idt
= vmx_get_idt
,
4305 .set_idt
= vmx_set_idt
,
4306 .get_gdt
= vmx_get_gdt
,
4307 .set_gdt
= vmx_set_gdt
,
4308 .set_dr7
= vmx_set_dr7
,
4309 .cache_reg
= vmx_cache_reg
,
4310 .get_rflags
= vmx_get_rflags
,
4311 .set_rflags
= vmx_set_rflags
,
4312 .fpu_activate
= vmx_fpu_activate
,
4313 .fpu_deactivate
= vmx_fpu_deactivate
,
4315 .tlb_flush
= vmx_flush_tlb
,
4317 .run
= vmx_vcpu_run
,
4318 .handle_exit
= vmx_handle_exit
,
4319 .skip_emulated_instruction
= skip_emulated_instruction
,
4320 .set_interrupt_shadow
= vmx_set_interrupt_shadow
,
4321 .get_interrupt_shadow
= vmx_get_interrupt_shadow
,
4322 .patch_hypercall
= vmx_patch_hypercall
,
4323 .set_irq
= vmx_inject_irq
,
4324 .set_nmi
= vmx_inject_nmi
,
4325 .queue_exception
= vmx_queue_exception
,
4326 .interrupt_allowed
= vmx_interrupt_allowed
,
4327 .nmi_allowed
= vmx_nmi_allowed
,
4328 .get_nmi_mask
= vmx_get_nmi_mask
,
4329 .set_nmi_mask
= vmx_set_nmi_mask
,
4330 .enable_nmi_window
= enable_nmi_window
,
4331 .enable_irq_window
= enable_irq_window
,
4332 .update_cr8_intercept
= update_cr8_intercept
,
4334 .set_tss_addr
= vmx_set_tss_addr
,
4335 .get_tdp_level
= get_ept_level
,
4336 .get_mt_mask
= vmx_get_mt_mask
,
4338 .exit_reasons_str
= vmx_exit_reasons_str
,
4339 .get_lpage_level
= vmx_get_lpage_level
,
4341 .cpuid_update
= vmx_cpuid_update
,
4343 .rdtscp_supported
= vmx_rdtscp_supported
,
4345 .set_supported_cpuid
= vmx_set_supported_cpuid
,
4347 .has_wbinvd_exit
= cpu_has_vmx_wbinvd_exit
,
4349 .write_tsc_offset
= vmx_write_tsc_offset
,
4352 static int __init
vmx_init(void)
4356 rdmsrl_safe(MSR_EFER
, &host_efer
);
4358 for (i
= 0; i
< NR_VMX_MSR
; ++i
)
4359 kvm_define_shared_msr(i
, vmx_msr_index
[i
]);
4361 vmx_io_bitmap_a
= (unsigned long *)__get_free_page(GFP_KERNEL
);
4362 if (!vmx_io_bitmap_a
)
4365 vmx_io_bitmap_b
= (unsigned long *)__get_free_page(GFP_KERNEL
);
4366 if (!vmx_io_bitmap_b
) {
4371 vmx_msr_bitmap_legacy
= (unsigned long *)__get_free_page(GFP_KERNEL
);
4372 if (!vmx_msr_bitmap_legacy
) {
4377 vmx_msr_bitmap_longmode
= (unsigned long *)__get_free_page(GFP_KERNEL
);
4378 if (!vmx_msr_bitmap_longmode
) {
4384 * Allow direct access to the PC debug port (it is often used for I/O
4385 * delays, but the vmexits simply slow things down).
4387 memset(vmx_io_bitmap_a
, 0xff, PAGE_SIZE
);
4388 clear_bit(0x80, vmx_io_bitmap_a
);
4390 memset(vmx_io_bitmap_b
, 0xff, PAGE_SIZE
);
4392 memset(vmx_msr_bitmap_legacy
, 0xff, PAGE_SIZE
);
4393 memset(vmx_msr_bitmap_longmode
, 0xff, PAGE_SIZE
);
4395 set_bit(0, vmx_vpid_bitmap
); /* 0 is reserved for host */
4397 r
= kvm_init(&vmx_x86_ops
, sizeof(struct vcpu_vmx
),
4398 __alignof__(struct vcpu_vmx
), THIS_MODULE
);
4402 vmx_disable_intercept_for_msr(MSR_FS_BASE
, false);
4403 vmx_disable_intercept_for_msr(MSR_GS_BASE
, false);
4404 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE
, true);
4405 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS
, false);
4406 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP
, false);
4407 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP
, false);
4410 bypass_guest_pf
= 0;
4411 kvm_mmu_set_base_ptes(VMX_EPT_READABLE_MASK
|
4412 VMX_EPT_WRITABLE_MASK
);
4413 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4414 VMX_EPT_EXECUTABLE_MASK
);
4419 if (bypass_guest_pf
)
4420 kvm_mmu_set_nonpresent_ptes(~0xffeull
, 0ull);
4425 free_page((unsigned long)vmx_msr_bitmap_longmode
);
4427 free_page((unsigned long)vmx_msr_bitmap_legacy
);
4429 free_page((unsigned long)vmx_io_bitmap_b
);
4431 free_page((unsigned long)vmx_io_bitmap_a
);
4435 static void __exit
vmx_exit(void)
4437 free_page((unsigned long)vmx_msr_bitmap_legacy
);
4438 free_page((unsigned long)vmx_msr_bitmap_longmode
);
4439 free_page((unsigned long)vmx_io_bitmap_b
);
4440 free_page((unsigned long)vmx_io_bitmap_a
);
4445 module_init(vmx_init
)
4446 module_exit(vmx_exit
)