Merge branch 'for-next'
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / arch / x86 / kvm / vmx.c
blobbf89ec2cfb823ceff3bcae33f00bf1cecfab10c5
1 /*
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 affiliates.
10 * Authors:
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.
19 #include "irq.h"
20 #include "mmu.h"
22 #include <linux/kvm_host.h>
23 #include <linux/module.h>
24 #include <linux/kernel.h>
25 #include <linux/mm.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"
33 #include "x86.h"
35 #include <asm/io.h>
36 #include <asm/desc.h>
37 #include <asm/vmx.h>
38 #include <asm/virtext.h>
39 #include <asm/mce.h>
40 #include <asm/i387.h>
41 #include <asm/xcr.h>
43 #include "trace.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 static int __read_mostly yield_on_hlt = 1;
73 module_param(yield_on_hlt, bool, S_IRUGO);
75 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST \
76 (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
77 #define KVM_GUEST_CR0_MASK \
78 (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
79 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST \
80 (X86_CR0_WP | X86_CR0_NE)
81 #define KVM_VM_CR0_ALWAYS_ON \
82 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
83 #define KVM_CR4_GUEST_OWNED_BITS \
84 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
85 | X86_CR4_OSXMMEXCPT)
87 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
88 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
90 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
93 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
94 * ple_gap: upper bound on the amount of time between two successive
95 * executions of PAUSE in a loop. Also indicate if ple enabled.
96 * According to test, this time is usually small than 41 cycles.
97 * ple_window: upper bound on the amount of time a guest is allowed to execute
98 * in a PAUSE loop. Tests indicate that most spinlocks are held for
99 * less than 2^12 cycles
100 * Time is measured based on a counter that runs at the same rate as the TSC,
101 * refer SDM volume 3b section 21.6.13 & 22.1.3.
103 #define KVM_VMX_DEFAULT_PLE_GAP 41
104 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
105 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
106 module_param(ple_gap, int, S_IRUGO);
108 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
109 module_param(ple_window, int, S_IRUGO);
111 #define NR_AUTOLOAD_MSRS 1
113 struct vmcs {
114 u32 revision_id;
115 u32 abort;
116 char data[0];
119 struct shared_msr_entry {
120 unsigned index;
121 u64 data;
122 u64 mask;
125 struct vcpu_vmx {
126 struct kvm_vcpu vcpu;
127 struct list_head local_vcpus_link;
128 unsigned long host_rsp;
129 int launched;
130 u8 fail;
131 u32 exit_intr_info;
132 u32 idt_vectoring_info;
133 struct shared_msr_entry *guest_msrs;
134 int nmsrs;
135 int save_nmsrs;
136 #ifdef CONFIG_X86_64
137 u64 msr_host_kernel_gs_base;
138 u64 msr_guest_kernel_gs_base;
139 #endif
140 struct vmcs *vmcs;
141 struct msr_autoload {
142 unsigned nr;
143 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
144 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
145 } msr_autoload;
146 struct {
147 int loaded;
148 u16 fs_sel, gs_sel, ldt_sel;
149 int gs_ldt_reload_needed;
150 int fs_reload_needed;
151 } host_state;
152 struct {
153 int vm86_active;
154 ulong save_rflags;
155 struct kvm_save_segment {
156 u16 selector;
157 unsigned long base;
158 u32 limit;
159 u32 ar;
160 } tr, es, ds, fs, gs;
161 } rmode;
162 int vpid;
163 bool emulation_required;
165 /* Support for vnmi-less CPUs */
166 int soft_vnmi_blocked;
167 ktime_t entry_time;
168 s64 vnmi_blocked_time;
169 u32 exit_reason;
171 bool rdtscp_enabled;
174 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
176 return container_of(vcpu, struct vcpu_vmx, vcpu);
179 static int init_rmode(struct kvm *kvm);
180 static u64 construct_eptp(unsigned long root_hpa);
181 static void kvm_cpu_vmxon(u64 addr);
182 static void kvm_cpu_vmxoff(void);
183 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
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 bool cpu_has_load_ia32_efer;
197 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
198 static DEFINE_SPINLOCK(vmx_vpid_lock);
200 static struct vmcs_config {
201 int size;
202 int order;
203 u32 revision_id;
204 u32 pin_based_exec_ctrl;
205 u32 cpu_based_exec_ctrl;
206 u32 cpu_based_2nd_exec_ctrl;
207 u32 vmexit_ctrl;
208 u32 vmentry_ctrl;
209 } vmcs_config;
211 static struct vmx_capability {
212 u32 ept;
213 u32 vpid;
214 } vmx_capability;
216 #define VMX_SEGMENT_FIELD(seg) \
217 [VCPU_SREG_##seg] = { \
218 .selector = GUEST_##seg##_SELECTOR, \
219 .base = GUEST_##seg##_BASE, \
220 .limit = GUEST_##seg##_LIMIT, \
221 .ar_bytes = GUEST_##seg##_AR_BYTES, \
224 static struct kvm_vmx_segment_field {
225 unsigned selector;
226 unsigned base;
227 unsigned limit;
228 unsigned ar_bytes;
229 } kvm_vmx_segment_fields[] = {
230 VMX_SEGMENT_FIELD(CS),
231 VMX_SEGMENT_FIELD(DS),
232 VMX_SEGMENT_FIELD(ES),
233 VMX_SEGMENT_FIELD(FS),
234 VMX_SEGMENT_FIELD(GS),
235 VMX_SEGMENT_FIELD(SS),
236 VMX_SEGMENT_FIELD(TR),
237 VMX_SEGMENT_FIELD(LDTR),
240 static u64 host_efer;
242 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
245 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
246 * away by decrementing the array size.
248 static const u32 vmx_msr_index[] = {
249 #ifdef CONFIG_X86_64
250 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
251 #endif
252 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
254 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
256 static inline bool is_page_fault(u32 intr_info)
258 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
259 INTR_INFO_VALID_MASK)) ==
260 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
263 static inline bool is_no_device(u32 intr_info)
265 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
266 INTR_INFO_VALID_MASK)) ==
267 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
270 static inline bool is_invalid_opcode(u32 intr_info)
272 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
273 INTR_INFO_VALID_MASK)) ==
274 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
277 static inline bool is_external_interrupt(u32 intr_info)
279 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
280 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
283 static inline bool is_machine_check(u32 intr_info)
285 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
286 INTR_INFO_VALID_MASK)) ==
287 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
290 static inline bool cpu_has_vmx_msr_bitmap(void)
292 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
295 static inline bool cpu_has_vmx_tpr_shadow(void)
297 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
300 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
302 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
305 static inline bool cpu_has_secondary_exec_ctrls(void)
307 return vmcs_config.cpu_based_exec_ctrl &
308 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
311 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
313 return vmcs_config.cpu_based_2nd_exec_ctrl &
314 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
317 static inline bool cpu_has_vmx_flexpriority(void)
319 return cpu_has_vmx_tpr_shadow() &&
320 cpu_has_vmx_virtualize_apic_accesses();
323 static inline bool cpu_has_vmx_ept_execute_only(void)
325 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
328 static inline bool cpu_has_vmx_eptp_uncacheable(void)
330 return vmx_capability.ept & VMX_EPTP_UC_BIT;
333 static inline bool cpu_has_vmx_eptp_writeback(void)
335 return vmx_capability.ept & VMX_EPTP_WB_BIT;
338 static inline bool cpu_has_vmx_ept_2m_page(void)
340 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
343 static inline bool cpu_has_vmx_ept_1g_page(void)
345 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
348 static inline bool cpu_has_vmx_ept_4levels(void)
350 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
353 static inline bool cpu_has_vmx_invept_individual_addr(void)
355 return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
358 static inline bool cpu_has_vmx_invept_context(void)
360 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
363 static inline bool cpu_has_vmx_invept_global(void)
365 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
368 static inline bool cpu_has_vmx_invvpid_single(void)
370 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
373 static inline bool cpu_has_vmx_invvpid_global(void)
375 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
378 static inline bool cpu_has_vmx_ept(void)
380 return vmcs_config.cpu_based_2nd_exec_ctrl &
381 SECONDARY_EXEC_ENABLE_EPT;
384 static inline bool cpu_has_vmx_unrestricted_guest(void)
386 return vmcs_config.cpu_based_2nd_exec_ctrl &
387 SECONDARY_EXEC_UNRESTRICTED_GUEST;
390 static inline bool cpu_has_vmx_ple(void)
392 return vmcs_config.cpu_based_2nd_exec_ctrl &
393 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
396 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
398 return flexpriority_enabled && irqchip_in_kernel(kvm);
401 static inline bool cpu_has_vmx_vpid(void)
403 return vmcs_config.cpu_based_2nd_exec_ctrl &
404 SECONDARY_EXEC_ENABLE_VPID;
407 static inline bool cpu_has_vmx_rdtscp(void)
409 return vmcs_config.cpu_based_2nd_exec_ctrl &
410 SECONDARY_EXEC_RDTSCP;
413 static inline bool cpu_has_virtual_nmis(void)
415 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
418 static inline bool cpu_has_vmx_wbinvd_exit(void)
420 return vmcs_config.cpu_based_2nd_exec_ctrl &
421 SECONDARY_EXEC_WBINVD_EXITING;
424 static inline bool report_flexpriority(void)
426 return flexpriority_enabled;
429 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
431 int i;
433 for (i = 0; i < vmx->nmsrs; ++i)
434 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
435 return i;
436 return -1;
439 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
441 struct {
442 u64 vpid : 16;
443 u64 rsvd : 48;
444 u64 gva;
445 } operand = { vpid, 0, gva };
447 asm volatile (__ex(ASM_VMX_INVVPID)
448 /* CF==1 or ZF==1 --> rc = -1 */
449 "; ja 1f ; ud2 ; 1:"
450 : : "a"(&operand), "c"(ext) : "cc", "memory");
453 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
455 struct {
456 u64 eptp, gpa;
457 } operand = {eptp, gpa};
459 asm volatile (__ex(ASM_VMX_INVEPT)
460 /* CF==1 or ZF==1 --> rc = -1 */
461 "; ja 1f ; ud2 ; 1:\n"
462 : : "a" (&operand), "c" (ext) : "cc", "memory");
465 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
467 int i;
469 i = __find_msr_index(vmx, msr);
470 if (i >= 0)
471 return &vmx->guest_msrs[i];
472 return NULL;
475 static void vmcs_clear(struct vmcs *vmcs)
477 u64 phys_addr = __pa(vmcs);
478 u8 error;
480 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
481 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
482 : "cc", "memory");
483 if (error)
484 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
485 vmcs, phys_addr);
488 static void vmcs_load(struct vmcs *vmcs)
490 u64 phys_addr = __pa(vmcs);
491 u8 error;
493 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
494 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
495 : "cc", "memory");
496 if (error)
497 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
498 vmcs, phys_addr);
501 static void __vcpu_clear(void *arg)
503 struct vcpu_vmx *vmx = arg;
504 int cpu = raw_smp_processor_id();
506 if (vmx->vcpu.cpu == cpu)
507 vmcs_clear(vmx->vmcs);
508 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
509 per_cpu(current_vmcs, cpu) = NULL;
510 list_del(&vmx->local_vcpus_link);
511 vmx->vcpu.cpu = -1;
512 vmx->launched = 0;
515 static void vcpu_clear(struct vcpu_vmx *vmx)
517 if (vmx->vcpu.cpu == -1)
518 return;
519 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 1);
522 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
524 if (vmx->vpid == 0)
525 return;
527 if (cpu_has_vmx_invvpid_single())
528 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
531 static inline void vpid_sync_vcpu_global(void)
533 if (cpu_has_vmx_invvpid_global())
534 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
537 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
539 if (cpu_has_vmx_invvpid_single())
540 vpid_sync_vcpu_single(vmx);
541 else
542 vpid_sync_vcpu_global();
545 static inline void ept_sync_global(void)
547 if (cpu_has_vmx_invept_global())
548 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
551 static inline void ept_sync_context(u64 eptp)
553 if (enable_ept) {
554 if (cpu_has_vmx_invept_context())
555 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
556 else
557 ept_sync_global();
561 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
563 if (enable_ept) {
564 if (cpu_has_vmx_invept_individual_addr())
565 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
566 eptp, gpa);
567 else
568 ept_sync_context(eptp);
572 static unsigned long vmcs_readl(unsigned long field)
574 unsigned long value = 0;
576 asm volatile (__ex(ASM_VMX_VMREAD_RDX_RAX)
577 : "+a"(value) : "d"(field) : "cc");
578 return value;
581 static u16 vmcs_read16(unsigned long field)
583 return vmcs_readl(field);
586 static u32 vmcs_read32(unsigned long field)
588 return vmcs_readl(field);
591 static u64 vmcs_read64(unsigned long field)
593 #ifdef CONFIG_X86_64
594 return vmcs_readl(field);
595 #else
596 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
597 #endif
600 static noinline void vmwrite_error(unsigned long field, unsigned long value)
602 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
603 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
604 dump_stack();
607 static void vmcs_writel(unsigned long field, unsigned long value)
609 u8 error;
611 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
612 : "=q"(error) : "a"(value), "d"(field) : "cc");
613 if (unlikely(error))
614 vmwrite_error(field, value);
617 static void vmcs_write16(unsigned long field, u16 value)
619 vmcs_writel(field, value);
622 static void vmcs_write32(unsigned long field, u32 value)
624 vmcs_writel(field, value);
627 static void vmcs_write64(unsigned long field, u64 value)
629 vmcs_writel(field, value);
630 #ifndef CONFIG_X86_64
631 asm volatile ("");
632 vmcs_writel(field+1, value >> 32);
633 #endif
636 static void vmcs_clear_bits(unsigned long field, u32 mask)
638 vmcs_writel(field, vmcs_readl(field) & ~mask);
641 static void vmcs_set_bits(unsigned long field, u32 mask)
643 vmcs_writel(field, vmcs_readl(field) | mask);
646 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
648 u32 eb;
650 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
651 (1u << NM_VECTOR) | (1u << DB_VECTOR);
652 if ((vcpu->guest_debug &
653 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
654 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
655 eb |= 1u << BP_VECTOR;
656 if (to_vmx(vcpu)->rmode.vm86_active)
657 eb = ~0;
658 if (enable_ept)
659 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
660 if (vcpu->fpu_active)
661 eb &= ~(1u << NM_VECTOR);
662 vmcs_write32(EXCEPTION_BITMAP, eb);
665 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
667 unsigned i;
668 struct msr_autoload *m = &vmx->msr_autoload;
670 if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
671 vmcs_clear_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
672 vmcs_clear_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
673 return;
676 for (i = 0; i < m->nr; ++i)
677 if (m->guest[i].index == msr)
678 break;
680 if (i == m->nr)
681 return;
682 --m->nr;
683 m->guest[i] = m->guest[m->nr];
684 m->host[i] = m->host[m->nr];
685 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
686 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
689 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
690 u64 guest_val, u64 host_val)
692 unsigned i;
693 struct msr_autoload *m = &vmx->msr_autoload;
695 if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
696 vmcs_write64(GUEST_IA32_EFER, guest_val);
697 vmcs_write64(HOST_IA32_EFER, host_val);
698 vmcs_set_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
699 vmcs_set_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
700 return;
703 for (i = 0; i < m->nr; ++i)
704 if (m->guest[i].index == msr)
705 break;
707 if (i == m->nr) {
708 ++m->nr;
709 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
710 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
713 m->guest[i].index = msr;
714 m->guest[i].value = guest_val;
715 m->host[i].index = msr;
716 m->host[i].value = host_val;
719 static void reload_tss(void)
722 * VT restores TR but not its size. Useless.
724 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
725 struct desc_struct *descs;
727 descs = (void *)gdt->address;
728 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
729 load_TR_desc();
732 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
734 u64 guest_efer;
735 u64 ignore_bits;
737 guest_efer = vmx->vcpu.arch.efer;
740 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
741 * outside long mode
743 ignore_bits = EFER_NX | EFER_SCE;
744 #ifdef CONFIG_X86_64
745 ignore_bits |= EFER_LMA | EFER_LME;
746 /* SCE is meaningful only in long mode on Intel */
747 if (guest_efer & EFER_LMA)
748 ignore_bits &= ~(u64)EFER_SCE;
749 #endif
750 guest_efer &= ~ignore_bits;
751 guest_efer |= host_efer & ignore_bits;
752 vmx->guest_msrs[efer_offset].data = guest_efer;
753 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
755 clear_atomic_switch_msr(vmx, MSR_EFER);
756 /* On ept, can't emulate nx, and must switch nx atomically */
757 if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
758 guest_efer = vmx->vcpu.arch.efer;
759 if (!(guest_efer & EFER_LMA))
760 guest_efer &= ~EFER_LME;
761 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
762 return false;
765 return true;
768 static unsigned long segment_base(u16 selector)
770 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
771 struct desc_struct *d;
772 unsigned long table_base;
773 unsigned long v;
775 if (!(selector & ~3))
776 return 0;
778 table_base = gdt->address;
780 if (selector & 4) { /* from ldt */
781 u16 ldt_selector = kvm_read_ldt();
783 if (!(ldt_selector & ~3))
784 return 0;
786 table_base = segment_base(ldt_selector);
788 d = (struct desc_struct *)(table_base + (selector & ~7));
789 v = get_desc_base(d);
790 #ifdef CONFIG_X86_64
791 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
792 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
793 #endif
794 return v;
797 static inline unsigned long kvm_read_tr_base(void)
799 u16 tr;
800 asm("str %0" : "=g"(tr));
801 return segment_base(tr);
804 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
806 struct vcpu_vmx *vmx = to_vmx(vcpu);
807 int i;
809 if (vmx->host_state.loaded)
810 return;
812 vmx->host_state.loaded = 1;
814 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
815 * allow segment selectors with cpl > 0 or ti == 1.
817 vmx->host_state.ldt_sel = kvm_read_ldt();
818 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
819 savesegment(fs, vmx->host_state.fs_sel);
820 if (!(vmx->host_state.fs_sel & 7)) {
821 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
822 vmx->host_state.fs_reload_needed = 0;
823 } else {
824 vmcs_write16(HOST_FS_SELECTOR, 0);
825 vmx->host_state.fs_reload_needed = 1;
827 savesegment(gs, vmx->host_state.gs_sel);
828 if (!(vmx->host_state.gs_sel & 7))
829 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
830 else {
831 vmcs_write16(HOST_GS_SELECTOR, 0);
832 vmx->host_state.gs_ldt_reload_needed = 1;
835 #ifdef CONFIG_X86_64
836 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
837 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
838 #else
839 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
840 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
841 #endif
843 #ifdef CONFIG_X86_64
844 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
845 if (is_long_mode(&vmx->vcpu))
846 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
847 #endif
848 for (i = 0; i < vmx->save_nmsrs; ++i)
849 kvm_set_shared_msr(vmx->guest_msrs[i].index,
850 vmx->guest_msrs[i].data,
851 vmx->guest_msrs[i].mask);
854 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
856 if (!vmx->host_state.loaded)
857 return;
859 ++vmx->vcpu.stat.host_state_reload;
860 vmx->host_state.loaded = 0;
861 #ifdef CONFIG_X86_64
862 if (is_long_mode(&vmx->vcpu))
863 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
864 #endif
865 if (vmx->host_state.gs_ldt_reload_needed) {
866 kvm_load_ldt(vmx->host_state.ldt_sel);
867 #ifdef CONFIG_X86_64
868 load_gs_index(vmx->host_state.gs_sel);
869 #else
870 loadsegment(gs, vmx->host_state.gs_sel);
871 #endif
873 if (vmx->host_state.fs_reload_needed)
874 loadsegment(fs, vmx->host_state.fs_sel);
875 reload_tss();
876 #ifdef CONFIG_X86_64
877 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
878 #endif
879 if (current_thread_info()->status & TS_USEDFPU)
880 clts();
881 load_gdt(&__get_cpu_var(host_gdt));
884 static void vmx_load_host_state(struct vcpu_vmx *vmx)
886 preempt_disable();
887 __vmx_load_host_state(vmx);
888 preempt_enable();
892 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
893 * vcpu mutex is already taken.
895 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
897 struct vcpu_vmx *vmx = to_vmx(vcpu);
898 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
900 if (!vmm_exclusive)
901 kvm_cpu_vmxon(phys_addr);
902 else if (vcpu->cpu != cpu)
903 vcpu_clear(vmx);
905 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
906 per_cpu(current_vmcs, cpu) = vmx->vmcs;
907 vmcs_load(vmx->vmcs);
910 if (vcpu->cpu != cpu) {
911 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
912 unsigned long sysenter_esp;
914 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
915 local_irq_disable();
916 list_add(&vmx->local_vcpus_link,
917 &per_cpu(vcpus_on_cpu, cpu));
918 local_irq_enable();
921 * Linux uses per-cpu TSS and GDT, so set these when switching
922 * processors.
924 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
925 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
927 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
928 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
932 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
934 __vmx_load_host_state(to_vmx(vcpu));
935 if (!vmm_exclusive) {
936 __vcpu_clear(to_vmx(vcpu));
937 kvm_cpu_vmxoff();
941 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
943 ulong cr0;
945 if (vcpu->fpu_active)
946 return;
947 vcpu->fpu_active = 1;
948 cr0 = vmcs_readl(GUEST_CR0);
949 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
950 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
951 vmcs_writel(GUEST_CR0, cr0);
952 update_exception_bitmap(vcpu);
953 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
954 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
957 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
959 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
961 vmx_decache_cr0_guest_bits(vcpu);
962 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
963 update_exception_bitmap(vcpu);
964 vcpu->arch.cr0_guest_owned_bits = 0;
965 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
966 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
969 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
971 unsigned long rflags, save_rflags;
973 rflags = vmcs_readl(GUEST_RFLAGS);
974 if (to_vmx(vcpu)->rmode.vm86_active) {
975 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
976 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
977 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
979 return rflags;
982 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
984 if (to_vmx(vcpu)->rmode.vm86_active) {
985 to_vmx(vcpu)->rmode.save_rflags = rflags;
986 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
988 vmcs_writel(GUEST_RFLAGS, rflags);
991 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
993 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
994 int ret = 0;
996 if (interruptibility & GUEST_INTR_STATE_STI)
997 ret |= KVM_X86_SHADOW_INT_STI;
998 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
999 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1001 return ret & mask;
1004 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1006 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1007 u32 interruptibility = interruptibility_old;
1009 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1011 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1012 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1013 else if (mask & KVM_X86_SHADOW_INT_STI)
1014 interruptibility |= GUEST_INTR_STATE_STI;
1016 if ((interruptibility != interruptibility_old))
1017 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1020 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1022 unsigned long rip;
1024 rip = kvm_rip_read(vcpu);
1025 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1026 kvm_rip_write(vcpu, rip);
1028 /* skipping an emulated instruction also counts */
1029 vmx_set_interrupt_shadow(vcpu, 0);
1032 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1034 /* Ensure that we clear the HLT state in the VMCS. We don't need to
1035 * explicitly skip the instruction because if the HLT state is set, then
1036 * the instruction is already executing and RIP has already been
1037 * advanced. */
1038 if (!yield_on_hlt &&
1039 vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1040 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1043 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1044 bool has_error_code, u32 error_code,
1045 bool reinject)
1047 struct vcpu_vmx *vmx = to_vmx(vcpu);
1048 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1050 if (has_error_code) {
1051 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1052 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1055 if (vmx->rmode.vm86_active) {
1056 if (kvm_inject_realmode_interrupt(vcpu, nr) != EMULATE_DONE)
1057 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1058 return;
1061 if (kvm_exception_is_soft(nr)) {
1062 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1063 vmx->vcpu.arch.event_exit_inst_len);
1064 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1065 } else
1066 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1068 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1069 vmx_clear_hlt(vcpu);
1072 static bool vmx_rdtscp_supported(void)
1074 return cpu_has_vmx_rdtscp();
1078 * Swap MSR entry in host/guest MSR entry array.
1080 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1082 struct shared_msr_entry tmp;
1084 tmp = vmx->guest_msrs[to];
1085 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1086 vmx->guest_msrs[from] = tmp;
1090 * Set up the vmcs to automatically save and restore system
1091 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1092 * mode, as fiddling with msrs is very expensive.
1094 static void setup_msrs(struct vcpu_vmx *vmx)
1096 int save_nmsrs, index;
1097 unsigned long *msr_bitmap;
1099 vmx_load_host_state(vmx);
1100 save_nmsrs = 0;
1101 #ifdef CONFIG_X86_64
1102 if (is_long_mode(&vmx->vcpu)) {
1103 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1104 if (index >= 0)
1105 move_msr_up(vmx, index, save_nmsrs++);
1106 index = __find_msr_index(vmx, MSR_LSTAR);
1107 if (index >= 0)
1108 move_msr_up(vmx, index, save_nmsrs++);
1109 index = __find_msr_index(vmx, MSR_CSTAR);
1110 if (index >= 0)
1111 move_msr_up(vmx, index, save_nmsrs++);
1112 index = __find_msr_index(vmx, MSR_TSC_AUX);
1113 if (index >= 0 && vmx->rdtscp_enabled)
1114 move_msr_up(vmx, index, save_nmsrs++);
1116 * MSR_STAR is only needed on long mode guests, and only
1117 * if efer.sce is enabled.
1119 index = __find_msr_index(vmx, MSR_STAR);
1120 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1121 move_msr_up(vmx, index, save_nmsrs++);
1123 #endif
1124 index = __find_msr_index(vmx, MSR_EFER);
1125 if (index >= 0 && update_transition_efer(vmx, index))
1126 move_msr_up(vmx, index, save_nmsrs++);
1128 vmx->save_nmsrs = save_nmsrs;
1130 if (cpu_has_vmx_msr_bitmap()) {
1131 if (is_long_mode(&vmx->vcpu))
1132 msr_bitmap = vmx_msr_bitmap_longmode;
1133 else
1134 msr_bitmap = vmx_msr_bitmap_legacy;
1136 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1141 * reads and returns guest's timestamp counter "register"
1142 * guest_tsc = host_tsc + tsc_offset -- 21.3
1144 static u64 guest_read_tsc(void)
1146 u64 host_tsc, tsc_offset;
1148 rdtscll(host_tsc);
1149 tsc_offset = vmcs_read64(TSC_OFFSET);
1150 return host_tsc + tsc_offset;
1154 * writes 'offset' into guest's timestamp counter offset register
1156 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1158 vmcs_write64(TSC_OFFSET, offset);
1161 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
1163 u64 offset = vmcs_read64(TSC_OFFSET);
1164 vmcs_write64(TSC_OFFSET, offset + adjustment);
1168 * Reads an msr value (of 'msr_index') into 'pdata'.
1169 * Returns 0 on success, non-0 otherwise.
1170 * Assumes vcpu_load() was already called.
1172 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1174 u64 data;
1175 struct shared_msr_entry *msr;
1177 if (!pdata) {
1178 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
1179 return -EINVAL;
1182 switch (msr_index) {
1183 #ifdef CONFIG_X86_64
1184 case MSR_FS_BASE:
1185 data = vmcs_readl(GUEST_FS_BASE);
1186 break;
1187 case MSR_GS_BASE:
1188 data = vmcs_readl(GUEST_GS_BASE);
1189 break;
1190 case MSR_KERNEL_GS_BASE:
1191 vmx_load_host_state(to_vmx(vcpu));
1192 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
1193 break;
1194 #endif
1195 case MSR_EFER:
1196 return kvm_get_msr_common(vcpu, msr_index, pdata);
1197 case MSR_IA32_TSC:
1198 data = guest_read_tsc();
1199 break;
1200 case MSR_IA32_SYSENTER_CS:
1201 data = vmcs_read32(GUEST_SYSENTER_CS);
1202 break;
1203 case MSR_IA32_SYSENTER_EIP:
1204 data = vmcs_readl(GUEST_SYSENTER_EIP);
1205 break;
1206 case MSR_IA32_SYSENTER_ESP:
1207 data = vmcs_readl(GUEST_SYSENTER_ESP);
1208 break;
1209 case MSR_TSC_AUX:
1210 if (!to_vmx(vcpu)->rdtscp_enabled)
1211 return 1;
1212 /* Otherwise falls through */
1213 default:
1214 vmx_load_host_state(to_vmx(vcpu));
1215 msr = find_msr_entry(to_vmx(vcpu), msr_index);
1216 if (msr) {
1217 vmx_load_host_state(to_vmx(vcpu));
1218 data = msr->data;
1219 break;
1221 return kvm_get_msr_common(vcpu, msr_index, pdata);
1224 *pdata = data;
1225 return 0;
1229 * Writes msr value into into the appropriate "register".
1230 * Returns 0 on success, non-0 otherwise.
1231 * Assumes vcpu_load() was already called.
1233 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1235 struct vcpu_vmx *vmx = to_vmx(vcpu);
1236 struct shared_msr_entry *msr;
1237 int ret = 0;
1239 switch (msr_index) {
1240 case MSR_EFER:
1241 vmx_load_host_state(vmx);
1242 ret = kvm_set_msr_common(vcpu, msr_index, data);
1243 break;
1244 #ifdef CONFIG_X86_64
1245 case MSR_FS_BASE:
1246 vmcs_writel(GUEST_FS_BASE, data);
1247 break;
1248 case MSR_GS_BASE:
1249 vmcs_writel(GUEST_GS_BASE, data);
1250 break;
1251 case MSR_KERNEL_GS_BASE:
1252 vmx_load_host_state(vmx);
1253 vmx->msr_guest_kernel_gs_base = data;
1254 break;
1255 #endif
1256 case MSR_IA32_SYSENTER_CS:
1257 vmcs_write32(GUEST_SYSENTER_CS, data);
1258 break;
1259 case MSR_IA32_SYSENTER_EIP:
1260 vmcs_writel(GUEST_SYSENTER_EIP, data);
1261 break;
1262 case MSR_IA32_SYSENTER_ESP:
1263 vmcs_writel(GUEST_SYSENTER_ESP, data);
1264 break;
1265 case MSR_IA32_TSC:
1266 kvm_write_tsc(vcpu, data);
1267 break;
1268 case MSR_IA32_CR_PAT:
1269 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
1270 vmcs_write64(GUEST_IA32_PAT, data);
1271 vcpu->arch.pat = data;
1272 break;
1274 ret = kvm_set_msr_common(vcpu, msr_index, data);
1275 break;
1276 case MSR_TSC_AUX:
1277 if (!vmx->rdtscp_enabled)
1278 return 1;
1279 /* Check reserved bit, higher 32 bits should be zero */
1280 if ((data >> 32) != 0)
1281 return 1;
1282 /* Otherwise falls through */
1283 default:
1284 msr = find_msr_entry(vmx, msr_index);
1285 if (msr) {
1286 vmx_load_host_state(vmx);
1287 msr->data = data;
1288 break;
1290 ret = kvm_set_msr_common(vcpu, msr_index, data);
1293 return ret;
1296 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
1298 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
1299 switch (reg) {
1300 case VCPU_REGS_RSP:
1301 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
1302 break;
1303 case VCPU_REGS_RIP:
1304 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
1305 break;
1306 case VCPU_EXREG_PDPTR:
1307 if (enable_ept)
1308 ept_save_pdptrs(vcpu);
1309 break;
1310 default:
1311 break;
1315 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
1317 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
1318 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
1319 else
1320 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
1322 update_exception_bitmap(vcpu);
1325 static __init int cpu_has_kvm_support(void)
1327 return cpu_has_vmx();
1330 static __init int vmx_disabled_by_bios(void)
1332 u64 msr;
1334 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
1335 if (msr & FEATURE_CONTROL_LOCKED) {
1336 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
1337 && tboot_enabled())
1338 return 1;
1339 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
1340 && !tboot_enabled()) {
1341 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
1342 " activate TXT before enabling KVM\n");
1343 return 1;
1347 return 0;
1348 /* locked but not enabled */
1351 static void kvm_cpu_vmxon(u64 addr)
1353 asm volatile (ASM_VMX_VMXON_RAX
1354 : : "a"(&addr), "m"(addr)
1355 : "memory", "cc");
1358 static int hardware_enable(void *garbage)
1360 int cpu = raw_smp_processor_id();
1361 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1362 u64 old, test_bits;
1364 if (read_cr4() & X86_CR4_VMXE)
1365 return -EBUSY;
1367 INIT_LIST_HEAD(&per_cpu(vcpus_on_cpu, cpu));
1368 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
1370 test_bits = FEATURE_CONTROL_LOCKED;
1371 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
1372 if (tboot_enabled())
1373 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
1375 if ((old & test_bits) != test_bits) {
1376 /* enable and lock */
1377 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
1379 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
1381 if (vmm_exclusive) {
1382 kvm_cpu_vmxon(phys_addr);
1383 ept_sync_global();
1386 store_gdt(&__get_cpu_var(host_gdt));
1388 return 0;
1391 static void vmclear_local_vcpus(void)
1393 int cpu = raw_smp_processor_id();
1394 struct vcpu_vmx *vmx, *n;
1396 list_for_each_entry_safe(vmx, n, &per_cpu(vcpus_on_cpu, cpu),
1397 local_vcpus_link)
1398 __vcpu_clear(vmx);
1402 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
1403 * tricks.
1405 static void kvm_cpu_vmxoff(void)
1407 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
1410 static void hardware_disable(void *garbage)
1412 if (vmm_exclusive) {
1413 vmclear_local_vcpus();
1414 kvm_cpu_vmxoff();
1416 write_cr4(read_cr4() & ~X86_CR4_VMXE);
1419 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
1420 u32 msr, u32 *result)
1422 u32 vmx_msr_low, vmx_msr_high;
1423 u32 ctl = ctl_min | ctl_opt;
1425 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1427 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
1428 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
1430 /* Ensure minimum (required) set of control bits are supported. */
1431 if (ctl_min & ~ctl)
1432 return -EIO;
1434 *result = ctl;
1435 return 0;
1438 static __init bool allow_1_setting(u32 msr, u32 ctl)
1440 u32 vmx_msr_low, vmx_msr_high;
1442 rdmsr(msr, vmx_msr_low, vmx_msr_high);
1443 return vmx_msr_high & ctl;
1446 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
1448 u32 vmx_msr_low, vmx_msr_high;
1449 u32 min, opt, min2, opt2;
1450 u32 _pin_based_exec_control = 0;
1451 u32 _cpu_based_exec_control = 0;
1452 u32 _cpu_based_2nd_exec_control = 0;
1453 u32 _vmexit_control = 0;
1454 u32 _vmentry_control = 0;
1456 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
1457 opt = PIN_BASED_VIRTUAL_NMIS;
1458 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
1459 &_pin_based_exec_control) < 0)
1460 return -EIO;
1462 min =
1463 #ifdef CONFIG_X86_64
1464 CPU_BASED_CR8_LOAD_EXITING |
1465 CPU_BASED_CR8_STORE_EXITING |
1466 #endif
1467 CPU_BASED_CR3_LOAD_EXITING |
1468 CPU_BASED_CR3_STORE_EXITING |
1469 CPU_BASED_USE_IO_BITMAPS |
1470 CPU_BASED_MOV_DR_EXITING |
1471 CPU_BASED_USE_TSC_OFFSETING |
1472 CPU_BASED_MWAIT_EXITING |
1473 CPU_BASED_MONITOR_EXITING |
1474 CPU_BASED_INVLPG_EXITING;
1476 if (yield_on_hlt)
1477 min |= CPU_BASED_HLT_EXITING;
1479 opt = CPU_BASED_TPR_SHADOW |
1480 CPU_BASED_USE_MSR_BITMAPS |
1481 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1482 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
1483 &_cpu_based_exec_control) < 0)
1484 return -EIO;
1485 #ifdef CONFIG_X86_64
1486 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
1487 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
1488 ~CPU_BASED_CR8_STORE_EXITING;
1489 #endif
1490 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
1491 min2 = 0;
1492 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
1493 SECONDARY_EXEC_WBINVD_EXITING |
1494 SECONDARY_EXEC_ENABLE_VPID |
1495 SECONDARY_EXEC_ENABLE_EPT |
1496 SECONDARY_EXEC_UNRESTRICTED_GUEST |
1497 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
1498 SECONDARY_EXEC_RDTSCP;
1499 if (adjust_vmx_controls(min2, opt2,
1500 MSR_IA32_VMX_PROCBASED_CTLS2,
1501 &_cpu_based_2nd_exec_control) < 0)
1502 return -EIO;
1504 #ifndef CONFIG_X86_64
1505 if (!(_cpu_based_2nd_exec_control &
1506 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
1507 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
1508 #endif
1509 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
1510 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
1511 enabled */
1512 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
1513 CPU_BASED_CR3_STORE_EXITING |
1514 CPU_BASED_INVLPG_EXITING);
1515 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
1516 vmx_capability.ept, vmx_capability.vpid);
1519 min = 0;
1520 #ifdef CONFIG_X86_64
1521 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
1522 #endif
1523 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
1524 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
1525 &_vmexit_control) < 0)
1526 return -EIO;
1528 min = 0;
1529 opt = VM_ENTRY_LOAD_IA32_PAT;
1530 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1531 &_vmentry_control) < 0)
1532 return -EIO;
1534 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1536 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1537 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1538 return -EIO;
1540 #ifdef CONFIG_X86_64
1541 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1542 if (vmx_msr_high & (1u<<16))
1543 return -EIO;
1544 #endif
1546 /* Require Write-Back (WB) memory type for VMCS accesses. */
1547 if (((vmx_msr_high >> 18) & 15) != 6)
1548 return -EIO;
1550 vmcs_conf->size = vmx_msr_high & 0x1fff;
1551 vmcs_conf->order = get_order(vmcs_config.size);
1552 vmcs_conf->revision_id = vmx_msr_low;
1554 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1555 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1556 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1557 vmcs_conf->vmexit_ctrl = _vmexit_control;
1558 vmcs_conf->vmentry_ctrl = _vmentry_control;
1560 cpu_has_load_ia32_efer =
1561 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
1562 VM_ENTRY_LOAD_IA32_EFER)
1563 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
1564 VM_EXIT_LOAD_IA32_EFER);
1566 return 0;
1569 static struct vmcs *alloc_vmcs_cpu(int cpu)
1571 int node = cpu_to_node(cpu);
1572 struct page *pages;
1573 struct vmcs *vmcs;
1575 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
1576 if (!pages)
1577 return NULL;
1578 vmcs = page_address(pages);
1579 memset(vmcs, 0, vmcs_config.size);
1580 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1581 return vmcs;
1584 static struct vmcs *alloc_vmcs(void)
1586 return alloc_vmcs_cpu(raw_smp_processor_id());
1589 static void free_vmcs(struct vmcs *vmcs)
1591 free_pages((unsigned long)vmcs, vmcs_config.order);
1594 static void free_kvm_area(void)
1596 int cpu;
1598 for_each_possible_cpu(cpu) {
1599 free_vmcs(per_cpu(vmxarea, cpu));
1600 per_cpu(vmxarea, cpu) = NULL;
1604 static __init int alloc_kvm_area(void)
1606 int cpu;
1608 for_each_possible_cpu(cpu) {
1609 struct vmcs *vmcs;
1611 vmcs = alloc_vmcs_cpu(cpu);
1612 if (!vmcs) {
1613 free_kvm_area();
1614 return -ENOMEM;
1617 per_cpu(vmxarea, cpu) = vmcs;
1619 return 0;
1622 static __init int hardware_setup(void)
1624 if (setup_vmcs_config(&vmcs_config) < 0)
1625 return -EIO;
1627 if (boot_cpu_has(X86_FEATURE_NX))
1628 kvm_enable_efer_bits(EFER_NX);
1630 if (!cpu_has_vmx_vpid())
1631 enable_vpid = 0;
1633 if (!cpu_has_vmx_ept() ||
1634 !cpu_has_vmx_ept_4levels()) {
1635 enable_ept = 0;
1636 enable_unrestricted_guest = 0;
1639 if (!cpu_has_vmx_unrestricted_guest())
1640 enable_unrestricted_guest = 0;
1642 if (!cpu_has_vmx_flexpriority())
1643 flexpriority_enabled = 0;
1645 if (!cpu_has_vmx_tpr_shadow())
1646 kvm_x86_ops->update_cr8_intercept = NULL;
1648 if (enable_ept && !cpu_has_vmx_ept_2m_page())
1649 kvm_disable_largepages();
1651 if (!cpu_has_vmx_ple())
1652 ple_gap = 0;
1654 return alloc_kvm_area();
1657 static __exit void hardware_unsetup(void)
1659 free_kvm_area();
1662 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1664 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1666 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1667 vmcs_write16(sf->selector, save->selector);
1668 vmcs_writel(sf->base, save->base);
1669 vmcs_write32(sf->limit, save->limit);
1670 vmcs_write32(sf->ar_bytes, save->ar);
1671 } else {
1672 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1673 << AR_DPL_SHIFT;
1674 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1678 static void enter_pmode(struct kvm_vcpu *vcpu)
1680 unsigned long flags;
1681 struct vcpu_vmx *vmx = to_vmx(vcpu);
1683 vmx->emulation_required = 1;
1684 vmx->rmode.vm86_active = 0;
1686 vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
1687 vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
1688 vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
1690 flags = vmcs_readl(GUEST_RFLAGS);
1691 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1692 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1693 vmcs_writel(GUEST_RFLAGS, flags);
1695 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1696 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1698 update_exception_bitmap(vcpu);
1700 if (emulate_invalid_guest_state)
1701 return;
1703 fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
1704 fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
1705 fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
1706 fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
1708 vmcs_write16(GUEST_SS_SELECTOR, 0);
1709 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1711 vmcs_write16(GUEST_CS_SELECTOR,
1712 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1713 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1716 static gva_t rmode_tss_base(struct kvm *kvm)
1718 if (!kvm->arch.tss_addr) {
1719 struct kvm_memslots *slots;
1720 gfn_t base_gfn;
1722 slots = kvm_memslots(kvm);
1723 base_gfn = slots->memslots[0].base_gfn +
1724 kvm->memslots->memslots[0].npages - 3;
1725 return base_gfn << PAGE_SHIFT;
1727 return kvm->arch.tss_addr;
1730 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1732 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1734 save->selector = vmcs_read16(sf->selector);
1735 save->base = vmcs_readl(sf->base);
1736 save->limit = vmcs_read32(sf->limit);
1737 save->ar = vmcs_read32(sf->ar_bytes);
1738 vmcs_write16(sf->selector, save->base >> 4);
1739 vmcs_write32(sf->base, save->base & 0xffff0);
1740 vmcs_write32(sf->limit, 0xffff);
1741 vmcs_write32(sf->ar_bytes, 0xf3);
1742 if (save->base & 0xf)
1743 printk_once(KERN_WARNING "kvm: segment base is not paragraph"
1744 " aligned when entering protected mode (seg=%d)",
1745 seg);
1748 static void enter_rmode(struct kvm_vcpu *vcpu)
1750 unsigned long flags;
1751 struct vcpu_vmx *vmx = to_vmx(vcpu);
1753 if (enable_unrestricted_guest)
1754 return;
1756 vmx->emulation_required = 1;
1757 vmx->rmode.vm86_active = 1;
1759 vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1760 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1762 vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1763 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1765 vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1766 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1768 flags = vmcs_readl(GUEST_RFLAGS);
1769 vmx->rmode.save_rflags = flags;
1771 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1773 vmcs_writel(GUEST_RFLAGS, flags);
1774 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1775 update_exception_bitmap(vcpu);
1777 if (emulate_invalid_guest_state)
1778 goto continue_rmode;
1780 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1781 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1782 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1784 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1785 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1786 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1787 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1788 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1790 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
1791 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
1792 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
1793 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
1795 continue_rmode:
1796 kvm_mmu_reset_context(vcpu);
1797 init_rmode(vcpu->kvm);
1800 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1802 struct vcpu_vmx *vmx = to_vmx(vcpu);
1803 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1805 if (!msr)
1806 return;
1809 * Force kernel_gs_base reloading before EFER changes, as control
1810 * of this msr depends on is_long_mode().
1812 vmx_load_host_state(to_vmx(vcpu));
1813 vcpu->arch.efer = efer;
1814 if (efer & EFER_LMA) {
1815 vmcs_write32(VM_ENTRY_CONTROLS,
1816 vmcs_read32(VM_ENTRY_CONTROLS) |
1817 VM_ENTRY_IA32E_MODE);
1818 msr->data = efer;
1819 } else {
1820 vmcs_write32(VM_ENTRY_CONTROLS,
1821 vmcs_read32(VM_ENTRY_CONTROLS) &
1822 ~VM_ENTRY_IA32E_MODE);
1824 msr->data = efer & ~EFER_LME;
1826 setup_msrs(vmx);
1829 #ifdef CONFIG_X86_64
1831 static void enter_lmode(struct kvm_vcpu *vcpu)
1833 u32 guest_tr_ar;
1835 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1836 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1837 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1838 __func__);
1839 vmcs_write32(GUEST_TR_AR_BYTES,
1840 (guest_tr_ar & ~AR_TYPE_MASK)
1841 | AR_TYPE_BUSY_64_TSS);
1843 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
1846 static void exit_lmode(struct kvm_vcpu *vcpu)
1848 vmcs_write32(VM_ENTRY_CONTROLS,
1849 vmcs_read32(VM_ENTRY_CONTROLS)
1850 & ~VM_ENTRY_IA32E_MODE);
1851 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
1854 #endif
1856 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
1858 vpid_sync_context(to_vmx(vcpu));
1859 if (enable_ept) {
1860 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
1861 return;
1862 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
1866 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
1868 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
1870 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
1871 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
1874 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
1876 if (enable_ept && is_paging(vcpu))
1877 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
1878 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
1881 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1883 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
1885 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
1886 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
1889 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
1891 if (!test_bit(VCPU_EXREG_PDPTR,
1892 (unsigned long *)&vcpu->arch.regs_dirty))
1893 return;
1895 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1896 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
1897 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
1898 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
1899 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
1903 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
1905 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
1906 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
1907 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
1908 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
1909 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
1912 __set_bit(VCPU_EXREG_PDPTR,
1913 (unsigned long *)&vcpu->arch.regs_avail);
1914 __set_bit(VCPU_EXREG_PDPTR,
1915 (unsigned long *)&vcpu->arch.regs_dirty);
1918 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
1920 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
1921 unsigned long cr0,
1922 struct kvm_vcpu *vcpu)
1924 vmx_decache_cr3(vcpu);
1925 if (!(cr0 & X86_CR0_PG)) {
1926 /* From paging/starting to nonpaging */
1927 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1928 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
1929 (CPU_BASED_CR3_LOAD_EXITING |
1930 CPU_BASED_CR3_STORE_EXITING));
1931 vcpu->arch.cr0 = cr0;
1932 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1933 } else if (!is_paging(vcpu)) {
1934 /* From nonpaging to paging */
1935 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
1936 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
1937 ~(CPU_BASED_CR3_LOAD_EXITING |
1938 CPU_BASED_CR3_STORE_EXITING));
1939 vcpu->arch.cr0 = cr0;
1940 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
1943 if (!(cr0 & X86_CR0_WP))
1944 *hw_cr0 &= ~X86_CR0_WP;
1947 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1949 struct vcpu_vmx *vmx = to_vmx(vcpu);
1950 unsigned long hw_cr0;
1952 if (enable_unrestricted_guest)
1953 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
1954 | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
1955 else
1956 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
1958 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
1959 enter_pmode(vcpu);
1961 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
1962 enter_rmode(vcpu);
1964 #ifdef CONFIG_X86_64
1965 if (vcpu->arch.efer & EFER_LME) {
1966 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1967 enter_lmode(vcpu);
1968 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1969 exit_lmode(vcpu);
1971 #endif
1973 if (enable_ept)
1974 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
1976 if (!vcpu->fpu_active)
1977 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
1979 vmcs_writel(CR0_READ_SHADOW, cr0);
1980 vmcs_writel(GUEST_CR0, hw_cr0);
1981 vcpu->arch.cr0 = cr0;
1984 static u64 construct_eptp(unsigned long root_hpa)
1986 u64 eptp;
1988 /* TODO write the value reading from MSR */
1989 eptp = VMX_EPT_DEFAULT_MT |
1990 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
1991 eptp |= (root_hpa & PAGE_MASK);
1993 return eptp;
1996 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1998 unsigned long guest_cr3;
1999 u64 eptp;
2001 guest_cr3 = cr3;
2002 if (enable_ept) {
2003 eptp = construct_eptp(cr3);
2004 vmcs_write64(EPT_POINTER, eptp);
2005 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
2006 vcpu->kvm->arch.ept_identity_map_addr;
2007 ept_load_pdptrs(vcpu);
2010 vmx_flush_tlb(vcpu);
2011 vmcs_writel(GUEST_CR3, guest_cr3);
2014 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2016 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
2017 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
2019 vcpu->arch.cr4 = cr4;
2020 if (enable_ept) {
2021 if (!is_paging(vcpu)) {
2022 hw_cr4 &= ~X86_CR4_PAE;
2023 hw_cr4 |= X86_CR4_PSE;
2024 } else if (!(cr4 & X86_CR4_PAE)) {
2025 hw_cr4 &= ~X86_CR4_PAE;
2029 vmcs_writel(CR4_READ_SHADOW, cr4);
2030 vmcs_writel(GUEST_CR4, hw_cr4);
2033 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2035 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2037 return vmcs_readl(sf->base);
2040 static void vmx_get_segment(struct kvm_vcpu *vcpu,
2041 struct kvm_segment *var, int seg)
2043 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2044 u32 ar;
2046 var->base = vmcs_readl(sf->base);
2047 var->limit = vmcs_read32(sf->limit);
2048 var->selector = vmcs_read16(sf->selector);
2049 ar = vmcs_read32(sf->ar_bytes);
2050 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
2051 ar = 0;
2052 var->type = ar & 15;
2053 var->s = (ar >> 4) & 1;
2054 var->dpl = (ar >> 5) & 3;
2055 var->present = (ar >> 7) & 1;
2056 var->avl = (ar >> 12) & 1;
2057 var->l = (ar >> 13) & 1;
2058 var->db = (ar >> 14) & 1;
2059 var->g = (ar >> 15) & 1;
2060 var->unusable = (ar >> 16) & 1;
2063 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
2065 if (!is_protmode(vcpu))
2066 return 0;
2068 if (vmx_get_rflags(vcpu) & X86_EFLAGS_VM) /* if virtual 8086 */
2069 return 3;
2071 return vmcs_read16(GUEST_CS_SELECTOR) & 3;
2074 static u32 vmx_segment_access_rights(struct kvm_segment *var)
2076 u32 ar;
2078 if (var->unusable)
2079 ar = 1 << 16;
2080 else {
2081 ar = var->type & 15;
2082 ar |= (var->s & 1) << 4;
2083 ar |= (var->dpl & 3) << 5;
2084 ar |= (var->present & 1) << 7;
2085 ar |= (var->avl & 1) << 12;
2086 ar |= (var->l & 1) << 13;
2087 ar |= (var->db & 1) << 14;
2088 ar |= (var->g & 1) << 15;
2090 if (ar == 0) /* a 0 value means unusable */
2091 ar = AR_UNUSABLE_MASK;
2093 return ar;
2096 static void vmx_set_segment(struct kvm_vcpu *vcpu,
2097 struct kvm_segment *var, int seg)
2099 struct vcpu_vmx *vmx = to_vmx(vcpu);
2100 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2101 u32 ar;
2103 if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
2104 vmx->rmode.tr.selector = var->selector;
2105 vmx->rmode.tr.base = var->base;
2106 vmx->rmode.tr.limit = var->limit;
2107 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
2108 return;
2110 vmcs_writel(sf->base, var->base);
2111 vmcs_write32(sf->limit, var->limit);
2112 vmcs_write16(sf->selector, var->selector);
2113 if (vmx->rmode.vm86_active && var->s) {
2115 * Hack real-mode segments into vm86 compatibility.
2117 if (var->base == 0xffff0000 && var->selector == 0xf000)
2118 vmcs_writel(sf->base, 0xf0000);
2119 ar = 0xf3;
2120 } else
2121 ar = vmx_segment_access_rights(var);
2124 * Fix the "Accessed" bit in AR field of segment registers for older
2125 * qemu binaries.
2126 * IA32 arch specifies that at the time of processor reset the
2127 * "Accessed" bit in the AR field of segment registers is 1. And qemu
2128 * is setting it to 0 in the usedland code. This causes invalid guest
2129 * state vmexit when "unrestricted guest" mode is turned on.
2130 * Fix for this setup issue in cpu_reset is being pushed in the qemu
2131 * tree. Newer qemu binaries with that qemu fix would not need this
2132 * kvm hack.
2134 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
2135 ar |= 0x1; /* Accessed */
2137 vmcs_write32(sf->ar_bytes, ar);
2140 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2142 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
2144 *db = (ar >> 14) & 1;
2145 *l = (ar >> 13) & 1;
2148 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2150 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
2151 dt->address = vmcs_readl(GUEST_IDTR_BASE);
2154 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2156 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
2157 vmcs_writel(GUEST_IDTR_BASE, dt->address);
2160 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2162 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
2163 dt->address = vmcs_readl(GUEST_GDTR_BASE);
2166 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2168 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
2169 vmcs_writel(GUEST_GDTR_BASE, dt->address);
2172 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
2174 struct kvm_segment var;
2175 u32 ar;
2177 vmx_get_segment(vcpu, &var, seg);
2178 ar = vmx_segment_access_rights(&var);
2180 if (var.base != (var.selector << 4))
2181 return false;
2182 if (var.limit != 0xffff)
2183 return false;
2184 if (ar != 0xf3)
2185 return false;
2187 return true;
2190 static bool code_segment_valid(struct kvm_vcpu *vcpu)
2192 struct kvm_segment cs;
2193 unsigned int cs_rpl;
2195 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2196 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
2198 if (cs.unusable)
2199 return false;
2200 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
2201 return false;
2202 if (!cs.s)
2203 return false;
2204 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
2205 if (cs.dpl > cs_rpl)
2206 return false;
2207 } else {
2208 if (cs.dpl != cs_rpl)
2209 return false;
2211 if (!cs.present)
2212 return false;
2214 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
2215 return true;
2218 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
2220 struct kvm_segment ss;
2221 unsigned int ss_rpl;
2223 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2224 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
2226 if (ss.unusable)
2227 return true;
2228 if (ss.type != 3 && ss.type != 7)
2229 return false;
2230 if (!ss.s)
2231 return false;
2232 if (ss.dpl != ss_rpl) /* DPL != RPL */
2233 return false;
2234 if (!ss.present)
2235 return false;
2237 return true;
2240 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
2242 struct kvm_segment var;
2243 unsigned int rpl;
2245 vmx_get_segment(vcpu, &var, seg);
2246 rpl = var.selector & SELECTOR_RPL_MASK;
2248 if (var.unusable)
2249 return true;
2250 if (!var.s)
2251 return false;
2252 if (!var.present)
2253 return false;
2254 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
2255 if (var.dpl < rpl) /* DPL < RPL */
2256 return false;
2259 /* TODO: Add other members to kvm_segment_field to allow checking for other access
2260 * rights flags
2262 return true;
2265 static bool tr_valid(struct kvm_vcpu *vcpu)
2267 struct kvm_segment tr;
2269 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
2271 if (tr.unusable)
2272 return false;
2273 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2274 return false;
2275 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
2276 return false;
2277 if (!tr.present)
2278 return false;
2280 return true;
2283 static bool ldtr_valid(struct kvm_vcpu *vcpu)
2285 struct kvm_segment ldtr;
2287 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
2289 if (ldtr.unusable)
2290 return true;
2291 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
2292 return false;
2293 if (ldtr.type != 2)
2294 return false;
2295 if (!ldtr.present)
2296 return false;
2298 return true;
2301 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
2303 struct kvm_segment cs, ss;
2305 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
2306 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
2308 return ((cs.selector & SELECTOR_RPL_MASK) ==
2309 (ss.selector & SELECTOR_RPL_MASK));
2313 * Check if guest state is valid. Returns true if valid, false if
2314 * not.
2315 * We assume that registers are always usable
2317 static bool guest_state_valid(struct kvm_vcpu *vcpu)
2319 /* real mode guest state checks */
2320 if (!is_protmode(vcpu)) {
2321 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
2322 return false;
2323 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
2324 return false;
2325 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
2326 return false;
2327 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
2328 return false;
2329 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
2330 return false;
2331 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
2332 return false;
2333 } else {
2334 /* protected mode guest state checks */
2335 if (!cs_ss_rpl_check(vcpu))
2336 return false;
2337 if (!code_segment_valid(vcpu))
2338 return false;
2339 if (!stack_segment_valid(vcpu))
2340 return false;
2341 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
2342 return false;
2343 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
2344 return false;
2345 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
2346 return false;
2347 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
2348 return false;
2349 if (!tr_valid(vcpu))
2350 return false;
2351 if (!ldtr_valid(vcpu))
2352 return false;
2354 /* TODO:
2355 * - Add checks on RIP
2356 * - Add checks on RFLAGS
2359 return true;
2362 static int init_rmode_tss(struct kvm *kvm)
2364 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
2365 u16 data = 0;
2366 int ret = 0;
2367 int r;
2369 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2370 if (r < 0)
2371 goto out;
2372 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
2373 r = kvm_write_guest_page(kvm, fn++, &data,
2374 TSS_IOPB_BASE_OFFSET, sizeof(u16));
2375 if (r < 0)
2376 goto out;
2377 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
2378 if (r < 0)
2379 goto out;
2380 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
2381 if (r < 0)
2382 goto out;
2383 data = ~0;
2384 r = kvm_write_guest_page(kvm, fn, &data,
2385 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
2386 sizeof(u8));
2387 if (r < 0)
2388 goto out;
2390 ret = 1;
2391 out:
2392 return ret;
2395 static int init_rmode_identity_map(struct kvm *kvm)
2397 int i, r, ret;
2398 pfn_t identity_map_pfn;
2399 u32 tmp;
2401 if (!enable_ept)
2402 return 1;
2403 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
2404 printk(KERN_ERR "EPT: identity-mapping pagetable "
2405 "haven't been allocated!\n");
2406 return 0;
2408 if (likely(kvm->arch.ept_identity_pagetable_done))
2409 return 1;
2410 ret = 0;
2411 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
2412 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
2413 if (r < 0)
2414 goto out;
2415 /* Set up identity-mapping pagetable for EPT in real mode */
2416 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
2417 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
2418 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
2419 r = kvm_write_guest_page(kvm, identity_map_pfn,
2420 &tmp, i * sizeof(tmp), sizeof(tmp));
2421 if (r < 0)
2422 goto out;
2424 kvm->arch.ept_identity_pagetable_done = true;
2425 ret = 1;
2426 out:
2427 return ret;
2430 static void seg_setup(int seg)
2432 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2433 unsigned int ar;
2435 vmcs_write16(sf->selector, 0);
2436 vmcs_writel(sf->base, 0);
2437 vmcs_write32(sf->limit, 0xffff);
2438 if (enable_unrestricted_guest) {
2439 ar = 0x93;
2440 if (seg == VCPU_SREG_CS)
2441 ar |= 0x08; /* code segment */
2442 } else
2443 ar = 0xf3;
2445 vmcs_write32(sf->ar_bytes, ar);
2448 static int alloc_apic_access_page(struct kvm *kvm)
2450 struct kvm_userspace_memory_region kvm_userspace_mem;
2451 int r = 0;
2453 mutex_lock(&kvm->slots_lock);
2454 if (kvm->arch.apic_access_page)
2455 goto out;
2456 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
2457 kvm_userspace_mem.flags = 0;
2458 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
2459 kvm_userspace_mem.memory_size = PAGE_SIZE;
2460 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2461 if (r)
2462 goto out;
2464 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
2465 out:
2466 mutex_unlock(&kvm->slots_lock);
2467 return r;
2470 static int alloc_identity_pagetable(struct kvm *kvm)
2472 struct kvm_userspace_memory_region kvm_userspace_mem;
2473 int r = 0;
2475 mutex_lock(&kvm->slots_lock);
2476 if (kvm->arch.ept_identity_pagetable)
2477 goto out;
2478 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
2479 kvm_userspace_mem.flags = 0;
2480 kvm_userspace_mem.guest_phys_addr =
2481 kvm->arch.ept_identity_map_addr;
2482 kvm_userspace_mem.memory_size = PAGE_SIZE;
2483 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
2484 if (r)
2485 goto out;
2487 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
2488 kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
2489 out:
2490 mutex_unlock(&kvm->slots_lock);
2491 return r;
2494 static void allocate_vpid(struct vcpu_vmx *vmx)
2496 int vpid;
2498 vmx->vpid = 0;
2499 if (!enable_vpid)
2500 return;
2501 spin_lock(&vmx_vpid_lock);
2502 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
2503 if (vpid < VMX_NR_VPIDS) {
2504 vmx->vpid = vpid;
2505 __set_bit(vpid, vmx_vpid_bitmap);
2507 spin_unlock(&vmx_vpid_lock);
2510 static void free_vpid(struct vcpu_vmx *vmx)
2512 if (!enable_vpid)
2513 return;
2514 spin_lock(&vmx_vpid_lock);
2515 if (vmx->vpid != 0)
2516 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
2517 spin_unlock(&vmx_vpid_lock);
2520 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
2522 int f = sizeof(unsigned long);
2524 if (!cpu_has_vmx_msr_bitmap())
2525 return;
2528 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
2529 * have the write-low and read-high bitmap offsets the wrong way round.
2530 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
2532 if (msr <= 0x1fff) {
2533 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
2534 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
2535 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
2536 msr &= 0x1fff;
2537 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
2538 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
2542 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
2544 if (!longmode_only)
2545 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
2546 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
2550 * Sets up the vmcs for emulated real mode.
2552 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
2554 u32 host_sysenter_cs, msr_low, msr_high;
2555 u32 junk;
2556 u64 host_pat;
2557 unsigned long a;
2558 struct desc_ptr dt;
2559 int i;
2560 unsigned long kvm_vmx_return;
2561 u32 exec_control;
2563 /* I/O */
2564 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
2565 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
2567 if (cpu_has_vmx_msr_bitmap())
2568 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
2570 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
2572 /* Control */
2573 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
2574 vmcs_config.pin_based_exec_ctrl);
2576 exec_control = vmcs_config.cpu_based_exec_ctrl;
2577 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
2578 exec_control &= ~CPU_BASED_TPR_SHADOW;
2579 #ifdef CONFIG_X86_64
2580 exec_control |= CPU_BASED_CR8_STORE_EXITING |
2581 CPU_BASED_CR8_LOAD_EXITING;
2582 #endif
2584 if (!enable_ept)
2585 exec_control |= CPU_BASED_CR3_STORE_EXITING |
2586 CPU_BASED_CR3_LOAD_EXITING |
2587 CPU_BASED_INVLPG_EXITING;
2588 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
2590 if (cpu_has_secondary_exec_ctrls()) {
2591 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
2592 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2593 exec_control &=
2594 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2595 if (vmx->vpid == 0)
2596 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
2597 if (!enable_ept) {
2598 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
2599 enable_unrestricted_guest = 0;
2601 if (!enable_unrestricted_guest)
2602 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
2603 if (!ple_gap)
2604 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
2605 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
2608 if (ple_gap) {
2609 vmcs_write32(PLE_GAP, ple_gap);
2610 vmcs_write32(PLE_WINDOW, ple_window);
2613 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
2614 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
2615 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
2617 vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS); /* 22.2.3 */
2618 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
2619 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
2621 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
2622 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2623 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2624 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
2625 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
2626 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
2627 #ifdef CONFIG_X86_64
2628 rdmsrl(MSR_FS_BASE, a);
2629 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
2630 rdmsrl(MSR_GS_BASE, a);
2631 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
2632 #else
2633 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
2634 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
2635 #endif
2637 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
2639 native_store_idt(&dt);
2640 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
2642 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
2643 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
2644 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
2645 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
2646 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
2647 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
2648 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
2650 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
2651 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
2652 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
2653 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
2654 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
2655 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
2657 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
2658 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2659 host_pat = msr_low | ((u64) msr_high << 32);
2660 vmcs_write64(HOST_IA32_PAT, host_pat);
2662 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2663 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
2664 host_pat = msr_low | ((u64) msr_high << 32);
2665 /* Write the default value follow host pat */
2666 vmcs_write64(GUEST_IA32_PAT, host_pat);
2667 /* Keep arch.pat sync with GUEST_IA32_PAT */
2668 vmx->vcpu.arch.pat = host_pat;
2671 for (i = 0; i < NR_VMX_MSR; ++i) {
2672 u32 index = vmx_msr_index[i];
2673 u32 data_low, data_high;
2674 int j = vmx->nmsrs;
2676 if (rdmsr_safe(index, &data_low, &data_high) < 0)
2677 continue;
2678 if (wrmsr_safe(index, data_low, data_high) < 0)
2679 continue;
2680 vmx->guest_msrs[j].index = i;
2681 vmx->guest_msrs[j].data = 0;
2682 vmx->guest_msrs[j].mask = -1ull;
2683 ++vmx->nmsrs;
2686 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
2688 /* 22.2.1, 20.8.1 */
2689 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
2691 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
2692 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
2693 if (enable_ept)
2694 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
2695 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
2697 kvm_write_tsc(&vmx->vcpu, 0);
2699 return 0;
2702 static int init_rmode(struct kvm *kvm)
2704 int idx, ret = 0;
2706 idx = srcu_read_lock(&kvm->srcu);
2707 if (!init_rmode_tss(kvm))
2708 goto exit;
2709 if (!init_rmode_identity_map(kvm))
2710 goto exit;
2712 ret = 1;
2713 exit:
2714 srcu_read_unlock(&kvm->srcu, idx);
2715 return ret;
2718 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
2720 struct vcpu_vmx *vmx = to_vmx(vcpu);
2721 u64 msr;
2722 int ret;
2724 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
2725 if (!init_rmode(vmx->vcpu.kvm)) {
2726 ret = -ENOMEM;
2727 goto out;
2730 vmx->rmode.vm86_active = 0;
2732 vmx->soft_vnmi_blocked = 0;
2734 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
2735 kvm_set_cr8(&vmx->vcpu, 0);
2736 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
2737 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2738 msr |= MSR_IA32_APICBASE_BSP;
2739 kvm_set_apic_base(&vmx->vcpu, msr);
2741 ret = fx_init(&vmx->vcpu);
2742 if (ret != 0)
2743 goto out;
2745 seg_setup(VCPU_SREG_CS);
2747 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
2748 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
2750 if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
2751 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
2752 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
2753 } else {
2754 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
2755 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
2758 seg_setup(VCPU_SREG_DS);
2759 seg_setup(VCPU_SREG_ES);
2760 seg_setup(VCPU_SREG_FS);
2761 seg_setup(VCPU_SREG_GS);
2762 seg_setup(VCPU_SREG_SS);
2764 vmcs_write16(GUEST_TR_SELECTOR, 0);
2765 vmcs_writel(GUEST_TR_BASE, 0);
2766 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
2767 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2769 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
2770 vmcs_writel(GUEST_LDTR_BASE, 0);
2771 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
2772 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
2774 vmcs_write32(GUEST_SYSENTER_CS, 0);
2775 vmcs_writel(GUEST_SYSENTER_ESP, 0);
2776 vmcs_writel(GUEST_SYSENTER_EIP, 0);
2778 vmcs_writel(GUEST_RFLAGS, 0x02);
2779 if (kvm_vcpu_is_bsp(&vmx->vcpu))
2780 kvm_rip_write(vcpu, 0xfff0);
2781 else
2782 kvm_rip_write(vcpu, 0);
2783 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
2785 vmcs_writel(GUEST_DR7, 0x400);
2787 vmcs_writel(GUEST_GDTR_BASE, 0);
2788 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
2790 vmcs_writel(GUEST_IDTR_BASE, 0);
2791 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
2793 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
2794 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
2795 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
2797 /* Special registers */
2798 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
2800 setup_msrs(vmx);
2802 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
2804 if (cpu_has_vmx_tpr_shadow()) {
2805 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
2806 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
2807 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
2808 page_to_phys(vmx->vcpu.arch.apic->regs_page));
2809 vmcs_write32(TPR_THRESHOLD, 0);
2812 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
2813 vmcs_write64(APIC_ACCESS_ADDR,
2814 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
2816 if (vmx->vpid != 0)
2817 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
2819 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
2820 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
2821 vmx_set_cr4(&vmx->vcpu, 0);
2822 vmx_set_efer(&vmx->vcpu, 0);
2823 vmx_fpu_activate(&vmx->vcpu);
2824 update_exception_bitmap(&vmx->vcpu);
2826 vpid_sync_context(vmx);
2828 ret = 0;
2830 /* HACK: Don't enable emulation on guest boot/reset */
2831 vmx->emulation_required = 0;
2833 out:
2834 return ret;
2837 static void enable_irq_window(struct kvm_vcpu *vcpu)
2839 u32 cpu_based_vm_exec_control;
2841 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2842 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2843 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2846 static void enable_nmi_window(struct kvm_vcpu *vcpu)
2848 u32 cpu_based_vm_exec_control;
2850 if (!cpu_has_virtual_nmis()) {
2851 enable_irq_window(vcpu);
2852 return;
2855 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
2856 enable_irq_window(vcpu);
2857 return;
2859 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2860 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
2861 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2864 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
2866 struct vcpu_vmx *vmx = to_vmx(vcpu);
2867 uint32_t intr;
2868 int irq = vcpu->arch.interrupt.nr;
2870 trace_kvm_inj_virq(irq);
2872 ++vcpu->stat.irq_injections;
2873 if (vmx->rmode.vm86_active) {
2874 if (kvm_inject_realmode_interrupt(vcpu, irq) != EMULATE_DONE)
2875 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2876 return;
2878 intr = irq | INTR_INFO_VALID_MASK;
2879 if (vcpu->arch.interrupt.soft) {
2880 intr |= INTR_TYPE_SOFT_INTR;
2881 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2882 vmx->vcpu.arch.event_exit_inst_len);
2883 } else
2884 intr |= INTR_TYPE_EXT_INTR;
2885 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
2886 vmx_clear_hlt(vcpu);
2889 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
2891 struct vcpu_vmx *vmx = to_vmx(vcpu);
2893 if (!cpu_has_virtual_nmis()) {
2895 * Tracking the NMI-blocked state in software is built upon
2896 * finding the next open IRQ window. This, in turn, depends on
2897 * well-behaving guests: They have to keep IRQs disabled at
2898 * least as long as the NMI handler runs. Otherwise we may
2899 * cause NMI nesting, maybe breaking the guest. But as this is
2900 * highly unlikely, we can live with the residual risk.
2902 vmx->soft_vnmi_blocked = 1;
2903 vmx->vnmi_blocked_time = 0;
2906 ++vcpu->stat.nmi_injections;
2907 if (vmx->rmode.vm86_active) {
2908 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR) != EMULATE_DONE)
2909 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2910 return;
2912 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
2913 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
2914 vmx_clear_hlt(vcpu);
2917 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
2919 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
2920 return 0;
2922 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2923 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
2924 | GUEST_INTR_STATE_NMI));
2927 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
2929 if (!cpu_has_virtual_nmis())
2930 return to_vmx(vcpu)->soft_vnmi_blocked;
2931 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
2934 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
2936 struct vcpu_vmx *vmx = to_vmx(vcpu);
2938 if (!cpu_has_virtual_nmis()) {
2939 if (vmx->soft_vnmi_blocked != masked) {
2940 vmx->soft_vnmi_blocked = masked;
2941 vmx->vnmi_blocked_time = 0;
2943 } else {
2944 if (masked)
2945 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
2946 GUEST_INTR_STATE_NMI);
2947 else
2948 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
2949 GUEST_INTR_STATE_NMI);
2953 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
2955 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2956 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
2957 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
2960 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
2962 int ret;
2963 struct kvm_userspace_memory_region tss_mem = {
2964 .slot = TSS_PRIVATE_MEMSLOT,
2965 .guest_phys_addr = addr,
2966 .memory_size = PAGE_SIZE * 3,
2967 .flags = 0,
2970 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
2971 if (ret)
2972 return ret;
2973 kvm->arch.tss_addr = addr;
2974 return 0;
2977 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
2978 int vec, u32 err_code)
2981 * Instruction with address size override prefix opcode 0x67
2982 * Cause the #SS fault with 0 error code in VM86 mode.
2984 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
2985 if (emulate_instruction(vcpu, 0) == EMULATE_DONE)
2986 return 1;
2988 * Forward all other exceptions that are valid in real mode.
2989 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
2990 * the required debugging infrastructure rework.
2992 switch (vec) {
2993 case DB_VECTOR:
2994 if (vcpu->guest_debug &
2995 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
2996 return 0;
2997 kvm_queue_exception(vcpu, vec);
2998 return 1;
2999 case BP_VECTOR:
3001 * Update instruction length as we may reinject the exception
3002 * from user space while in guest debugging mode.
3004 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
3005 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3006 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
3007 return 0;
3008 /* fall through */
3009 case DE_VECTOR:
3010 case OF_VECTOR:
3011 case BR_VECTOR:
3012 case UD_VECTOR:
3013 case DF_VECTOR:
3014 case SS_VECTOR:
3015 case GP_VECTOR:
3016 case MF_VECTOR:
3017 kvm_queue_exception(vcpu, vec);
3018 return 1;
3020 return 0;
3024 * Trigger machine check on the host. We assume all the MSRs are already set up
3025 * by the CPU and that we still run on the same CPU as the MCE occurred on.
3026 * We pass a fake environment to the machine check handler because we want
3027 * the guest to be always treated like user space, no matter what context
3028 * it used internally.
3030 static void kvm_machine_check(void)
3032 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
3033 struct pt_regs regs = {
3034 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
3035 .flags = X86_EFLAGS_IF,
3038 do_machine_check(&regs, 0);
3039 #endif
3042 static int handle_machine_check(struct kvm_vcpu *vcpu)
3044 /* already handled by vcpu_run */
3045 return 1;
3048 static int handle_exception(struct kvm_vcpu *vcpu)
3050 struct vcpu_vmx *vmx = to_vmx(vcpu);
3051 struct kvm_run *kvm_run = vcpu->run;
3052 u32 intr_info, ex_no, error_code;
3053 unsigned long cr2, rip, dr6;
3054 u32 vect_info;
3055 enum emulation_result er;
3057 vect_info = vmx->idt_vectoring_info;
3058 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
3060 if (is_machine_check(intr_info))
3061 return handle_machine_check(vcpu);
3063 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
3064 !is_page_fault(intr_info)) {
3065 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3066 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
3067 vcpu->run->internal.ndata = 2;
3068 vcpu->run->internal.data[0] = vect_info;
3069 vcpu->run->internal.data[1] = intr_info;
3070 return 0;
3073 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
3074 return 1; /* already handled by vmx_vcpu_run() */
3076 if (is_no_device(intr_info)) {
3077 vmx_fpu_activate(vcpu);
3078 return 1;
3081 if (is_invalid_opcode(intr_info)) {
3082 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
3083 if (er != EMULATE_DONE)
3084 kvm_queue_exception(vcpu, UD_VECTOR);
3085 return 1;
3088 error_code = 0;
3089 rip = kvm_rip_read(vcpu);
3090 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
3091 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
3092 if (is_page_fault(intr_info)) {
3093 /* EPT won't cause page fault directly */
3094 if (enable_ept)
3095 BUG();
3096 cr2 = vmcs_readl(EXIT_QUALIFICATION);
3097 trace_kvm_page_fault(cr2, error_code);
3099 if (kvm_event_needs_reinjection(vcpu))
3100 kvm_mmu_unprotect_page_virt(vcpu, cr2);
3101 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
3104 if (vmx->rmode.vm86_active &&
3105 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
3106 error_code)) {
3107 if (vcpu->arch.halt_request) {
3108 vcpu->arch.halt_request = 0;
3109 return kvm_emulate_halt(vcpu);
3111 return 1;
3114 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
3115 switch (ex_no) {
3116 case DB_VECTOR:
3117 dr6 = vmcs_readl(EXIT_QUALIFICATION);
3118 if (!(vcpu->guest_debug &
3119 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
3120 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
3121 kvm_queue_exception(vcpu, DB_VECTOR);
3122 return 1;
3124 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
3125 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
3126 /* fall through */
3127 case BP_VECTOR:
3129 * Update instruction length as we may reinject #BP from
3130 * user space while in guest debugging mode. Reading it for
3131 * #DB as well causes no harm, it is not used in that case.
3133 vmx->vcpu.arch.event_exit_inst_len =
3134 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3135 kvm_run->exit_reason = KVM_EXIT_DEBUG;
3136 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
3137 kvm_run->debug.arch.exception = ex_no;
3138 break;
3139 default:
3140 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
3141 kvm_run->ex.exception = ex_no;
3142 kvm_run->ex.error_code = error_code;
3143 break;
3145 return 0;
3148 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
3150 ++vcpu->stat.irq_exits;
3151 return 1;
3154 static int handle_triple_fault(struct kvm_vcpu *vcpu)
3156 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
3157 return 0;
3160 static int handle_io(struct kvm_vcpu *vcpu)
3162 unsigned long exit_qualification;
3163 int size, in, string;
3164 unsigned port;
3166 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3167 string = (exit_qualification & 16) != 0;
3168 in = (exit_qualification & 8) != 0;
3170 ++vcpu->stat.io_exits;
3172 if (string || in)
3173 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
3175 port = exit_qualification >> 16;
3176 size = (exit_qualification & 7) + 1;
3177 skip_emulated_instruction(vcpu);
3179 return kvm_fast_pio_out(vcpu, size, port);
3182 static void
3183 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
3186 * Patch in the VMCALL instruction:
3188 hypercall[0] = 0x0f;
3189 hypercall[1] = 0x01;
3190 hypercall[2] = 0xc1;
3193 static int handle_cr(struct kvm_vcpu *vcpu)
3195 unsigned long exit_qualification, val;
3196 int cr;
3197 int reg;
3198 int err;
3200 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3201 cr = exit_qualification & 15;
3202 reg = (exit_qualification >> 8) & 15;
3203 switch ((exit_qualification >> 4) & 3) {
3204 case 0: /* mov to cr */
3205 val = kvm_register_read(vcpu, reg);
3206 trace_kvm_cr_write(cr, val);
3207 switch (cr) {
3208 case 0:
3209 err = kvm_set_cr0(vcpu, val);
3210 kvm_complete_insn_gp(vcpu, err);
3211 return 1;
3212 case 3:
3213 err = kvm_set_cr3(vcpu, val);
3214 kvm_complete_insn_gp(vcpu, err);
3215 return 1;
3216 case 4:
3217 err = kvm_set_cr4(vcpu, val);
3218 kvm_complete_insn_gp(vcpu, err);
3219 return 1;
3220 case 8: {
3221 u8 cr8_prev = kvm_get_cr8(vcpu);
3222 u8 cr8 = kvm_register_read(vcpu, reg);
3223 err = kvm_set_cr8(vcpu, cr8);
3224 kvm_complete_insn_gp(vcpu, err);
3225 if (irqchip_in_kernel(vcpu->kvm))
3226 return 1;
3227 if (cr8_prev <= cr8)
3228 return 1;
3229 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
3230 return 0;
3233 break;
3234 case 2: /* clts */
3235 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
3236 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
3237 skip_emulated_instruction(vcpu);
3238 vmx_fpu_activate(vcpu);
3239 return 1;
3240 case 1: /*mov from cr*/
3241 switch (cr) {
3242 case 3:
3243 val = kvm_read_cr3(vcpu);
3244 kvm_register_write(vcpu, reg, val);
3245 trace_kvm_cr_read(cr, val);
3246 skip_emulated_instruction(vcpu);
3247 return 1;
3248 case 8:
3249 val = kvm_get_cr8(vcpu);
3250 kvm_register_write(vcpu, reg, val);
3251 trace_kvm_cr_read(cr, val);
3252 skip_emulated_instruction(vcpu);
3253 return 1;
3255 break;
3256 case 3: /* lmsw */
3257 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
3258 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
3259 kvm_lmsw(vcpu, val);
3261 skip_emulated_instruction(vcpu);
3262 return 1;
3263 default:
3264 break;
3266 vcpu->run->exit_reason = 0;
3267 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
3268 (int)(exit_qualification >> 4) & 3, cr);
3269 return 0;
3272 static int handle_dr(struct kvm_vcpu *vcpu)
3274 unsigned long exit_qualification;
3275 int dr, reg;
3277 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
3278 if (!kvm_require_cpl(vcpu, 0))
3279 return 1;
3280 dr = vmcs_readl(GUEST_DR7);
3281 if (dr & DR7_GD) {
3283 * As the vm-exit takes precedence over the debug trap, we
3284 * need to emulate the latter, either for the host or the
3285 * guest debugging itself.
3287 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
3288 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
3289 vcpu->run->debug.arch.dr7 = dr;
3290 vcpu->run->debug.arch.pc =
3291 vmcs_readl(GUEST_CS_BASE) +
3292 vmcs_readl(GUEST_RIP);
3293 vcpu->run->debug.arch.exception = DB_VECTOR;
3294 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
3295 return 0;
3296 } else {
3297 vcpu->arch.dr7 &= ~DR7_GD;
3298 vcpu->arch.dr6 |= DR6_BD;
3299 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
3300 kvm_queue_exception(vcpu, DB_VECTOR);
3301 return 1;
3305 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3306 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
3307 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
3308 if (exit_qualification & TYPE_MOV_FROM_DR) {
3309 unsigned long val;
3310 if (!kvm_get_dr(vcpu, dr, &val))
3311 kvm_register_write(vcpu, reg, val);
3312 } else
3313 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
3314 skip_emulated_instruction(vcpu);
3315 return 1;
3318 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
3320 vmcs_writel(GUEST_DR7, val);
3323 static int handle_cpuid(struct kvm_vcpu *vcpu)
3325 kvm_emulate_cpuid(vcpu);
3326 return 1;
3329 static int handle_rdmsr(struct kvm_vcpu *vcpu)
3331 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3332 u64 data;
3334 if (vmx_get_msr(vcpu, ecx, &data)) {
3335 trace_kvm_msr_read_ex(ecx);
3336 kvm_inject_gp(vcpu, 0);
3337 return 1;
3340 trace_kvm_msr_read(ecx, data);
3342 /* FIXME: handling of bits 32:63 of rax, rdx */
3343 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
3344 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
3345 skip_emulated_instruction(vcpu);
3346 return 1;
3349 static int handle_wrmsr(struct kvm_vcpu *vcpu)
3351 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
3352 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
3353 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
3355 if (vmx_set_msr(vcpu, ecx, data) != 0) {
3356 trace_kvm_msr_write_ex(ecx, data);
3357 kvm_inject_gp(vcpu, 0);
3358 return 1;
3361 trace_kvm_msr_write(ecx, data);
3362 skip_emulated_instruction(vcpu);
3363 return 1;
3366 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
3368 kvm_make_request(KVM_REQ_EVENT, vcpu);
3369 return 1;
3372 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
3374 u32 cpu_based_vm_exec_control;
3376 /* clear pending irq */
3377 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3378 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
3379 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3381 kvm_make_request(KVM_REQ_EVENT, vcpu);
3383 ++vcpu->stat.irq_window_exits;
3386 * If the user space waits to inject interrupts, exit as soon as
3387 * possible
3389 if (!irqchip_in_kernel(vcpu->kvm) &&
3390 vcpu->run->request_interrupt_window &&
3391 !kvm_cpu_has_interrupt(vcpu)) {
3392 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
3393 return 0;
3395 return 1;
3398 static int handle_halt(struct kvm_vcpu *vcpu)
3400 skip_emulated_instruction(vcpu);
3401 return kvm_emulate_halt(vcpu);
3404 static int handle_vmcall(struct kvm_vcpu *vcpu)
3406 skip_emulated_instruction(vcpu);
3407 kvm_emulate_hypercall(vcpu);
3408 return 1;
3411 static int handle_vmx_insn(struct kvm_vcpu *vcpu)
3413 kvm_queue_exception(vcpu, UD_VECTOR);
3414 return 1;
3417 static int handle_invd(struct kvm_vcpu *vcpu)
3419 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
3422 static int handle_invlpg(struct kvm_vcpu *vcpu)
3424 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3426 kvm_mmu_invlpg(vcpu, exit_qualification);
3427 skip_emulated_instruction(vcpu);
3428 return 1;
3431 static int handle_wbinvd(struct kvm_vcpu *vcpu)
3433 skip_emulated_instruction(vcpu);
3434 kvm_emulate_wbinvd(vcpu);
3435 return 1;
3438 static int handle_xsetbv(struct kvm_vcpu *vcpu)
3440 u64 new_bv = kvm_read_edx_eax(vcpu);
3441 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
3443 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
3444 skip_emulated_instruction(vcpu);
3445 return 1;
3448 static int handle_apic_access(struct kvm_vcpu *vcpu)
3450 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
3453 static int handle_task_switch(struct kvm_vcpu *vcpu)
3455 struct vcpu_vmx *vmx = to_vmx(vcpu);
3456 unsigned long exit_qualification;
3457 bool has_error_code = false;
3458 u32 error_code = 0;
3459 u16 tss_selector;
3460 int reason, type, idt_v;
3462 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
3463 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
3465 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3467 reason = (u32)exit_qualification >> 30;
3468 if (reason == TASK_SWITCH_GATE && idt_v) {
3469 switch (type) {
3470 case INTR_TYPE_NMI_INTR:
3471 vcpu->arch.nmi_injected = false;
3472 if (cpu_has_virtual_nmis())
3473 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3474 GUEST_INTR_STATE_NMI);
3475 break;
3476 case INTR_TYPE_EXT_INTR:
3477 case INTR_TYPE_SOFT_INTR:
3478 kvm_clear_interrupt_queue(vcpu);
3479 break;
3480 case INTR_TYPE_HARD_EXCEPTION:
3481 if (vmx->idt_vectoring_info &
3482 VECTORING_INFO_DELIVER_CODE_MASK) {
3483 has_error_code = true;
3484 error_code =
3485 vmcs_read32(IDT_VECTORING_ERROR_CODE);
3487 /* fall through */
3488 case INTR_TYPE_SOFT_EXCEPTION:
3489 kvm_clear_exception_queue(vcpu);
3490 break;
3491 default:
3492 break;
3495 tss_selector = exit_qualification;
3497 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
3498 type != INTR_TYPE_EXT_INTR &&
3499 type != INTR_TYPE_NMI_INTR))
3500 skip_emulated_instruction(vcpu);
3502 if (kvm_task_switch(vcpu, tss_selector, reason,
3503 has_error_code, error_code) == EMULATE_FAIL) {
3504 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3505 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
3506 vcpu->run->internal.ndata = 0;
3507 return 0;
3510 /* clear all local breakpoint enable flags */
3511 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
3514 * TODO: What about debug traps on tss switch?
3515 * Are we supposed to inject them and update dr6?
3518 return 1;
3521 static int handle_ept_violation(struct kvm_vcpu *vcpu)
3523 unsigned long exit_qualification;
3524 gpa_t gpa;
3525 int gla_validity;
3527 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
3529 if (exit_qualification & (1 << 6)) {
3530 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
3531 return -EINVAL;
3534 gla_validity = (exit_qualification >> 7) & 0x3;
3535 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
3536 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
3537 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
3538 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
3539 vmcs_readl(GUEST_LINEAR_ADDRESS));
3540 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
3541 (long unsigned int)exit_qualification);
3542 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3543 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
3544 return 0;
3547 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3548 trace_kvm_page_fault(gpa, exit_qualification);
3549 return kvm_mmu_page_fault(vcpu, gpa, exit_qualification & 0x3, NULL, 0);
3552 static u64 ept_rsvd_mask(u64 spte, int level)
3554 int i;
3555 u64 mask = 0;
3557 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
3558 mask |= (1ULL << i);
3560 if (level > 2)
3561 /* bits 7:3 reserved */
3562 mask |= 0xf8;
3563 else if (level == 2) {
3564 if (spte & (1ULL << 7))
3565 /* 2MB ref, bits 20:12 reserved */
3566 mask |= 0x1ff000;
3567 else
3568 /* bits 6:3 reserved */
3569 mask |= 0x78;
3572 return mask;
3575 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
3576 int level)
3578 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
3580 /* 010b (write-only) */
3581 WARN_ON((spte & 0x7) == 0x2);
3583 /* 110b (write/execute) */
3584 WARN_ON((spte & 0x7) == 0x6);
3586 /* 100b (execute-only) and value not supported by logical processor */
3587 if (!cpu_has_vmx_ept_execute_only())
3588 WARN_ON((spte & 0x7) == 0x4);
3590 /* not 000b */
3591 if ((spte & 0x7)) {
3592 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
3594 if (rsvd_bits != 0) {
3595 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
3596 __func__, rsvd_bits);
3597 WARN_ON(1);
3600 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
3601 u64 ept_mem_type = (spte & 0x38) >> 3;
3603 if (ept_mem_type == 2 || ept_mem_type == 3 ||
3604 ept_mem_type == 7) {
3605 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
3606 __func__, ept_mem_type);
3607 WARN_ON(1);
3613 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
3615 u64 sptes[4];
3616 int nr_sptes, i;
3617 gpa_t gpa;
3619 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
3621 printk(KERN_ERR "EPT: Misconfiguration.\n");
3622 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
3624 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
3626 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
3627 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
3629 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3630 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
3632 return 0;
3635 static int handle_nmi_window(struct kvm_vcpu *vcpu)
3637 u32 cpu_based_vm_exec_control;
3639 /* clear pending NMI */
3640 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3641 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
3642 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3643 ++vcpu->stat.nmi_window_exits;
3644 kvm_make_request(KVM_REQ_EVENT, vcpu);
3646 return 1;
3649 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
3651 struct vcpu_vmx *vmx = to_vmx(vcpu);
3652 enum emulation_result err = EMULATE_DONE;
3653 int ret = 1;
3654 u32 cpu_exec_ctrl;
3655 bool intr_window_requested;
3657 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3658 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
3660 while (!guest_state_valid(vcpu)) {
3661 if (intr_window_requested
3662 && (kvm_get_rflags(&vmx->vcpu) & X86_EFLAGS_IF))
3663 return handle_interrupt_window(&vmx->vcpu);
3665 err = emulate_instruction(vcpu, 0);
3667 if (err == EMULATE_DO_MMIO) {
3668 ret = 0;
3669 goto out;
3672 if (err != EMULATE_DONE)
3673 return 0;
3675 if (signal_pending(current))
3676 goto out;
3677 if (need_resched())
3678 schedule();
3681 vmx->emulation_required = 0;
3682 out:
3683 return ret;
3687 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
3688 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
3690 static int handle_pause(struct kvm_vcpu *vcpu)
3692 skip_emulated_instruction(vcpu);
3693 kvm_vcpu_on_spin(vcpu);
3695 return 1;
3698 static int handle_invalid_op(struct kvm_vcpu *vcpu)
3700 kvm_queue_exception(vcpu, UD_VECTOR);
3701 return 1;
3705 * The exit handlers return 1 if the exit was handled fully and guest execution
3706 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
3707 * to be done to userspace and return 0.
3709 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
3710 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
3711 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
3712 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
3713 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
3714 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
3715 [EXIT_REASON_CR_ACCESS] = handle_cr,
3716 [EXIT_REASON_DR_ACCESS] = handle_dr,
3717 [EXIT_REASON_CPUID] = handle_cpuid,
3718 [EXIT_REASON_MSR_READ] = handle_rdmsr,
3719 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
3720 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
3721 [EXIT_REASON_HLT] = handle_halt,
3722 [EXIT_REASON_INVD] = handle_invd,
3723 [EXIT_REASON_INVLPG] = handle_invlpg,
3724 [EXIT_REASON_VMCALL] = handle_vmcall,
3725 [EXIT_REASON_VMCLEAR] = handle_vmx_insn,
3726 [EXIT_REASON_VMLAUNCH] = handle_vmx_insn,
3727 [EXIT_REASON_VMPTRLD] = handle_vmx_insn,
3728 [EXIT_REASON_VMPTRST] = handle_vmx_insn,
3729 [EXIT_REASON_VMREAD] = handle_vmx_insn,
3730 [EXIT_REASON_VMRESUME] = handle_vmx_insn,
3731 [EXIT_REASON_VMWRITE] = handle_vmx_insn,
3732 [EXIT_REASON_VMOFF] = handle_vmx_insn,
3733 [EXIT_REASON_VMON] = handle_vmx_insn,
3734 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
3735 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
3736 [EXIT_REASON_WBINVD] = handle_wbinvd,
3737 [EXIT_REASON_XSETBV] = handle_xsetbv,
3738 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
3739 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
3740 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
3741 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
3742 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
3743 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_invalid_op,
3744 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_invalid_op,
3747 static const int kvm_vmx_max_exit_handlers =
3748 ARRAY_SIZE(kvm_vmx_exit_handlers);
3750 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
3752 *info1 = vmcs_readl(EXIT_QUALIFICATION);
3753 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
3757 * The guest has exited. See if we can fix it or if we need userspace
3758 * assistance.
3760 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
3762 struct vcpu_vmx *vmx = to_vmx(vcpu);
3763 u32 exit_reason = vmx->exit_reason;
3764 u32 vectoring_info = vmx->idt_vectoring_info;
3766 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
3768 /* If guest state is invalid, start emulating */
3769 if (vmx->emulation_required && emulate_invalid_guest_state)
3770 return handle_invalid_guest_state(vcpu);
3772 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
3773 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3774 vcpu->run->fail_entry.hardware_entry_failure_reason
3775 = exit_reason;
3776 return 0;
3779 if (unlikely(vmx->fail)) {
3780 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
3781 vcpu->run->fail_entry.hardware_entry_failure_reason
3782 = vmcs_read32(VM_INSTRUCTION_ERROR);
3783 return 0;
3786 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
3787 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
3788 exit_reason != EXIT_REASON_EPT_VIOLATION &&
3789 exit_reason != EXIT_REASON_TASK_SWITCH))
3790 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
3791 "(0x%x) and exit reason is 0x%x\n",
3792 __func__, vectoring_info, exit_reason);
3794 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
3795 if (vmx_interrupt_allowed(vcpu)) {
3796 vmx->soft_vnmi_blocked = 0;
3797 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
3798 vcpu->arch.nmi_pending) {
3800 * This CPU don't support us in finding the end of an
3801 * NMI-blocked window if the guest runs with IRQs
3802 * disabled. So we pull the trigger after 1 s of
3803 * futile waiting, but inform the user about this.
3805 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
3806 "state on VCPU %d after 1 s timeout\n",
3807 __func__, vcpu->vcpu_id);
3808 vmx->soft_vnmi_blocked = 0;
3812 if (exit_reason < kvm_vmx_max_exit_handlers
3813 && kvm_vmx_exit_handlers[exit_reason])
3814 return kvm_vmx_exit_handlers[exit_reason](vcpu);
3815 else {
3816 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
3817 vcpu->run->hw.hardware_exit_reason = exit_reason;
3819 return 0;
3822 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
3824 if (irr == -1 || tpr < irr) {
3825 vmcs_write32(TPR_THRESHOLD, 0);
3826 return;
3829 vmcs_write32(TPR_THRESHOLD, irr);
3832 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
3834 u32 exit_intr_info = vmx->exit_intr_info;
3836 /* Handle machine checks before interrupts are enabled */
3837 if ((vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY)
3838 || (vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI
3839 && is_machine_check(exit_intr_info)))
3840 kvm_machine_check();
3842 /* We need to handle NMIs before interrupts are enabled */
3843 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
3844 (exit_intr_info & INTR_INFO_VALID_MASK)) {
3845 kvm_before_handle_nmi(&vmx->vcpu);
3846 asm("int $2");
3847 kvm_after_handle_nmi(&vmx->vcpu);
3851 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
3853 u32 exit_intr_info = vmx->exit_intr_info;
3854 bool unblock_nmi;
3855 u8 vector;
3856 bool idtv_info_valid;
3858 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3860 if (cpu_has_virtual_nmis()) {
3861 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
3862 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
3864 * SDM 3: 27.7.1.2 (September 2008)
3865 * Re-set bit "block by NMI" before VM entry if vmexit caused by
3866 * a guest IRET fault.
3867 * SDM 3: 23.2.2 (September 2008)
3868 * Bit 12 is undefined in any of the following cases:
3869 * If the VM exit sets the valid bit in the IDT-vectoring
3870 * information field.
3871 * If the VM exit is due to a double fault.
3873 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
3874 vector != DF_VECTOR && !idtv_info_valid)
3875 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3876 GUEST_INTR_STATE_NMI);
3877 } else if (unlikely(vmx->soft_vnmi_blocked))
3878 vmx->vnmi_blocked_time +=
3879 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
3882 static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
3883 u32 idt_vectoring_info,
3884 int instr_len_field,
3885 int error_code_field)
3887 u8 vector;
3888 int type;
3889 bool idtv_info_valid;
3891 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
3893 vmx->vcpu.arch.nmi_injected = false;
3894 kvm_clear_exception_queue(&vmx->vcpu);
3895 kvm_clear_interrupt_queue(&vmx->vcpu);
3897 if (!idtv_info_valid)
3898 return;
3900 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
3902 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
3903 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
3905 switch (type) {
3906 case INTR_TYPE_NMI_INTR:
3907 vmx->vcpu.arch.nmi_injected = true;
3909 * SDM 3: 27.7.1.2 (September 2008)
3910 * Clear bit "block by NMI" before VM entry if a NMI
3911 * delivery faulted.
3913 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3914 GUEST_INTR_STATE_NMI);
3915 break;
3916 case INTR_TYPE_SOFT_EXCEPTION:
3917 vmx->vcpu.arch.event_exit_inst_len =
3918 vmcs_read32(instr_len_field);
3919 /* fall through */
3920 case INTR_TYPE_HARD_EXCEPTION:
3921 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
3922 u32 err = vmcs_read32(error_code_field);
3923 kvm_queue_exception_e(&vmx->vcpu, vector, err);
3924 } else
3925 kvm_queue_exception(&vmx->vcpu, vector);
3926 break;
3927 case INTR_TYPE_SOFT_INTR:
3928 vmx->vcpu.arch.event_exit_inst_len =
3929 vmcs_read32(instr_len_field);
3930 /* fall through */
3931 case INTR_TYPE_EXT_INTR:
3932 kvm_queue_interrupt(&vmx->vcpu, vector,
3933 type == INTR_TYPE_SOFT_INTR);
3934 break;
3935 default:
3936 break;
3940 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
3942 __vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
3943 VM_EXIT_INSTRUCTION_LEN,
3944 IDT_VECTORING_ERROR_CODE);
3947 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
3949 __vmx_complete_interrupts(to_vmx(vcpu),
3950 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
3951 VM_ENTRY_INSTRUCTION_LEN,
3952 VM_ENTRY_EXCEPTION_ERROR_CODE);
3954 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
3957 #ifdef CONFIG_X86_64
3958 #define R "r"
3959 #define Q "q"
3960 #else
3961 #define R "e"
3962 #define Q "l"
3963 #endif
3965 static void vmx_vcpu_run(struct kvm_vcpu *vcpu)
3967 struct vcpu_vmx *vmx = to_vmx(vcpu);
3969 /* Record the guest's net vcpu time for enforced NMI injections. */
3970 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
3971 vmx->entry_time = ktime_get();
3973 /* Don't enter VMX if guest state is invalid, let the exit handler
3974 start emulation until we arrive back to a valid state */
3975 if (vmx->emulation_required && emulate_invalid_guest_state)
3976 return;
3978 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
3979 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
3980 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
3981 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
3983 /* When single-stepping over STI and MOV SS, we must clear the
3984 * corresponding interruptibility bits in the guest state. Otherwise
3985 * vmentry fails as it then expects bit 14 (BS) in pending debug
3986 * exceptions being set, but that's not correct for the guest debugging
3987 * case. */
3988 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
3989 vmx_set_interrupt_shadow(vcpu, 0);
3991 asm(
3992 /* Store host registers */
3993 "push %%"R"dx; push %%"R"bp;"
3994 "push %%"R"cx \n\t"
3995 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
3996 "je 1f \n\t"
3997 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
3998 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
3999 "1: \n\t"
4000 /* Reload cr2 if changed */
4001 "mov %c[cr2](%0), %%"R"ax \n\t"
4002 "mov %%cr2, %%"R"dx \n\t"
4003 "cmp %%"R"ax, %%"R"dx \n\t"
4004 "je 2f \n\t"
4005 "mov %%"R"ax, %%cr2 \n\t"
4006 "2: \n\t"
4007 /* Check if vmlaunch of vmresume is needed */
4008 "cmpl $0, %c[launched](%0) \n\t"
4009 /* Load guest registers. Don't clobber flags. */
4010 "mov %c[rax](%0), %%"R"ax \n\t"
4011 "mov %c[rbx](%0), %%"R"bx \n\t"
4012 "mov %c[rdx](%0), %%"R"dx \n\t"
4013 "mov %c[rsi](%0), %%"R"si \n\t"
4014 "mov %c[rdi](%0), %%"R"di \n\t"
4015 "mov %c[rbp](%0), %%"R"bp \n\t"
4016 #ifdef CONFIG_X86_64
4017 "mov %c[r8](%0), %%r8 \n\t"
4018 "mov %c[r9](%0), %%r9 \n\t"
4019 "mov %c[r10](%0), %%r10 \n\t"
4020 "mov %c[r11](%0), %%r11 \n\t"
4021 "mov %c[r12](%0), %%r12 \n\t"
4022 "mov %c[r13](%0), %%r13 \n\t"
4023 "mov %c[r14](%0), %%r14 \n\t"
4024 "mov %c[r15](%0), %%r15 \n\t"
4025 #endif
4026 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
4028 /* Enter guest mode */
4029 "jne .Llaunched \n\t"
4030 __ex(ASM_VMX_VMLAUNCH) "\n\t"
4031 "jmp .Lkvm_vmx_return \n\t"
4032 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
4033 ".Lkvm_vmx_return: "
4034 /* Save guest registers, load host registers, keep flags */
4035 "xchg %0, (%%"R"sp) \n\t"
4036 "mov %%"R"ax, %c[rax](%0) \n\t"
4037 "mov %%"R"bx, %c[rbx](%0) \n\t"
4038 "push"Q" (%%"R"sp); pop"Q" %c[rcx](%0) \n\t"
4039 "mov %%"R"dx, %c[rdx](%0) \n\t"
4040 "mov %%"R"si, %c[rsi](%0) \n\t"
4041 "mov %%"R"di, %c[rdi](%0) \n\t"
4042 "mov %%"R"bp, %c[rbp](%0) \n\t"
4043 #ifdef CONFIG_X86_64
4044 "mov %%r8, %c[r8](%0) \n\t"
4045 "mov %%r9, %c[r9](%0) \n\t"
4046 "mov %%r10, %c[r10](%0) \n\t"
4047 "mov %%r11, %c[r11](%0) \n\t"
4048 "mov %%r12, %c[r12](%0) \n\t"
4049 "mov %%r13, %c[r13](%0) \n\t"
4050 "mov %%r14, %c[r14](%0) \n\t"
4051 "mov %%r15, %c[r15](%0) \n\t"
4052 #endif
4053 "mov %%cr2, %%"R"ax \n\t"
4054 "mov %%"R"ax, %c[cr2](%0) \n\t"
4056 "pop %%"R"bp; pop %%"R"bp; pop %%"R"dx \n\t"
4057 "setbe %c[fail](%0) \n\t"
4058 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
4059 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
4060 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
4061 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
4062 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
4063 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
4064 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
4065 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
4066 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
4067 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
4068 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
4069 #ifdef CONFIG_X86_64
4070 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
4071 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
4072 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
4073 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
4074 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
4075 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
4076 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
4077 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
4078 #endif
4079 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
4080 : "cc", "memory"
4081 , R"ax", R"bx", R"di", R"si"
4082 #ifdef CONFIG_X86_64
4083 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
4084 #endif
4087 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
4088 | (1 << VCPU_EXREG_PDPTR)
4089 | (1 << VCPU_EXREG_CR3));
4090 vcpu->arch.regs_dirty = 0;
4092 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
4094 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
4095 vmx->launched = 1;
4097 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
4098 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
4100 vmx_complete_atomic_exit(vmx);
4101 vmx_recover_nmi_blocking(vmx);
4102 vmx_complete_interrupts(vmx);
4105 #undef R
4106 #undef Q
4108 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
4110 struct vcpu_vmx *vmx = to_vmx(vcpu);
4112 if (vmx->vmcs) {
4113 vcpu_clear(vmx);
4114 free_vmcs(vmx->vmcs);
4115 vmx->vmcs = NULL;
4119 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
4121 struct vcpu_vmx *vmx = to_vmx(vcpu);
4123 free_vpid(vmx);
4124 vmx_free_vmcs(vcpu);
4125 kfree(vmx->guest_msrs);
4126 kvm_vcpu_uninit(vcpu);
4127 kmem_cache_free(kvm_vcpu_cache, vmx);
4130 static inline void vmcs_init(struct vmcs *vmcs)
4132 u64 phys_addr = __pa(per_cpu(vmxarea, raw_smp_processor_id()));
4134 if (!vmm_exclusive)
4135 kvm_cpu_vmxon(phys_addr);
4137 vmcs_clear(vmcs);
4139 if (!vmm_exclusive)
4140 kvm_cpu_vmxoff();
4143 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
4145 int err;
4146 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
4147 int cpu;
4149 if (!vmx)
4150 return ERR_PTR(-ENOMEM);
4152 allocate_vpid(vmx);
4154 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
4155 if (err)
4156 goto free_vcpu;
4158 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
4159 if (!vmx->guest_msrs) {
4160 err = -ENOMEM;
4161 goto uninit_vcpu;
4164 vmx->vmcs = alloc_vmcs();
4165 if (!vmx->vmcs)
4166 goto free_msrs;
4168 vmcs_init(vmx->vmcs);
4170 cpu = get_cpu();
4171 vmx_vcpu_load(&vmx->vcpu, cpu);
4172 vmx->vcpu.cpu = cpu;
4173 err = vmx_vcpu_setup(vmx);
4174 vmx_vcpu_put(&vmx->vcpu);
4175 put_cpu();
4176 if (err)
4177 goto free_vmcs;
4178 if (vm_need_virtualize_apic_accesses(kvm))
4179 if (alloc_apic_access_page(kvm) != 0)
4180 goto free_vmcs;
4182 if (enable_ept) {
4183 if (!kvm->arch.ept_identity_map_addr)
4184 kvm->arch.ept_identity_map_addr =
4185 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
4186 if (alloc_identity_pagetable(kvm) != 0)
4187 goto free_vmcs;
4190 return &vmx->vcpu;
4192 free_vmcs:
4193 free_vmcs(vmx->vmcs);
4194 free_msrs:
4195 kfree(vmx->guest_msrs);
4196 uninit_vcpu:
4197 kvm_vcpu_uninit(&vmx->vcpu);
4198 free_vcpu:
4199 free_vpid(vmx);
4200 kmem_cache_free(kvm_vcpu_cache, vmx);
4201 return ERR_PTR(err);
4204 static void __init vmx_check_processor_compat(void *rtn)
4206 struct vmcs_config vmcs_conf;
4208 *(int *)rtn = 0;
4209 if (setup_vmcs_config(&vmcs_conf) < 0)
4210 *(int *)rtn = -EIO;
4211 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
4212 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
4213 smp_processor_id());
4214 *(int *)rtn = -EIO;
4218 static int get_ept_level(void)
4220 return VMX_EPT_DEFAULT_GAW + 1;
4223 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
4225 u64 ret;
4227 /* For VT-d and EPT combination
4228 * 1. MMIO: always map as UC
4229 * 2. EPT with VT-d:
4230 * a. VT-d without snooping control feature: can't guarantee the
4231 * result, try to trust guest.
4232 * b. VT-d with snooping control feature: snooping control feature of
4233 * VT-d engine can guarantee the cache correctness. Just set it
4234 * to WB to keep consistent with host. So the same as item 3.
4235 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
4236 * consistent with host MTRR
4238 if (is_mmio)
4239 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
4240 else if (vcpu->kvm->arch.iommu_domain &&
4241 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
4242 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
4243 VMX_EPT_MT_EPTE_SHIFT;
4244 else
4245 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
4246 | VMX_EPT_IPAT_BIT;
4248 return ret;
4251 #define _ER(x) { EXIT_REASON_##x, #x }
4253 static const struct trace_print_flags vmx_exit_reasons_str[] = {
4254 _ER(EXCEPTION_NMI),
4255 _ER(EXTERNAL_INTERRUPT),
4256 _ER(TRIPLE_FAULT),
4257 _ER(PENDING_INTERRUPT),
4258 _ER(NMI_WINDOW),
4259 _ER(TASK_SWITCH),
4260 _ER(CPUID),
4261 _ER(HLT),
4262 _ER(INVLPG),
4263 _ER(RDPMC),
4264 _ER(RDTSC),
4265 _ER(VMCALL),
4266 _ER(VMCLEAR),
4267 _ER(VMLAUNCH),
4268 _ER(VMPTRLD),
4269 _ER(VMPTRST),
4270 _ER(VMREAD),
4271 _ER(VMRESUME),
4272 _ER(VMWRITE),
4273 _ER(VMOFF),
4274 _ER(VMON),
4275 _ER(CR_ACCESS),
4276 _ER(DR_ACCESS),
4277 _ER(IO_INSTRUCTION),
4278 _ER(MSR_READ),
4279 _ER(MSR_WRITE),
4280 _ER(MWAIT_INSTRUCTION),
4281 _ER(MONITOR_INSTRUCTION),
4282 _ER(PAUSE_INSTRUCTION),
4283 _ER(MCE_DURING_VMENTRY),
4284 _ER(TPR_BELOW_THRESHOLD),
4285 _ER(APIC_ACCESS),
4286 _ER(EPT_VIOLATION),
4287 _ER(EPT_MISCONFIG),
4288 _ER(WBINVD),
4289 { -1, NULL }
4292 #undef _ER
4294 static int vmx_get_lpage_level(void)
4296 if (enable_ept && !cpu_has_vmx_ept_1g_page())
4297 return PT_DIRECTORY_LEVEL;
4298 else
4299 /* For shadow and EPT supported 1GB page */
4300 return PT_PDPE_LEVEL;
4303 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
4305 struct kvm_cpuid_entry2 *best;
4306 struct vcpu_vmx *vmx = to_vmx(vcpu);
4307 u32 exec_control;
4309 vmx->rdtscp_enabled = false;
4310 if (vmx_rdtscp_supported()) {
4311 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
4312 if (exec_control & SECONDARY_EXEC_RDTSCP) {
4313 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
4314 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
4315 vmx->rdtscp_enabled = true;
4316 else {
4317 exec_control &= ~SECONDARY_EXEC_RDTSCP;
4318 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4319 exec_control);
4325 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
4329 static struct kvm_x86_ops vmx_x86_ops = {
4330 .cpu_has_kvm_support = cpu_has_kvm_support,
4331 .disabled_by_bios = vmx_disabled_by_bios,
4332 .hardware_setup = hardware_setup,
4333 .hardware_unsetup = hardware_unsetup,
4334 .check_processor_compatibility = vmx_check_processor_compat,
4335 .hardware_enable = hardware_enable,
4336 .hardware_disable = hardware_disable,
4337 .cpu_has_accelerated_tpr = report_flexpriority,
4339 .vcpu_create = vmx_create_vcpu,
4340 .vcpu_free = vmx_free_vcpu,
4341 .vcpu_reset = vmx_vcpu_reset,
4343 .prepare_guest_switch = vmx_save_host_state,
4344 .vcpu_load = vmx_vcpu_load,
4345 .vcpu_put = vmx_vcpu_put,
4347 .set_guest_debug = set_guest_debug,
4348 .get_msr = vmx_get_msr,
4349 .set_msr = vmx_set_msr,
4350 .get_segment_base = vmx_get_segment_base,
4351 .get_segment = vmx_get_segment,
4352 .set_segment = vmx_set_segment,
4353 .get_cpl = vmx_get_cpl,
4354 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
4355 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
4356 .decache_cr3 = vmx_decache_cr3,
4357 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
4358 .set_cr0 = vmx_set_cr0,
4359 .set_cr3 = vmx_set_cr3,
4360 .set_cr4 = vmx_set_cr4,
4361 .set_efer = vmx_set_efer,
4362 .get_idt = vmx_get_idt,
4363 .set_idt = vmx_set_idt,
4364 .get_gdt = vmx_get_gdt,
4365 .set_gdt = vmx_set_gdt,
4366 .set_dr7 = vmx_set_dr7,
4367 .cache_reg = vmx_cache_reg,
4368 .get_rflags = vmx_get_rflags,
4369 .set_rflags = vmx_set_rflags,
4370 .fpu_activate = vmx_fpu_activate,
4371 .fpu_deactivate = vmx_fpu_deactivate,
4373 .tlb_flush = vmx_flush_tlb,
4375 .run = vmx_vcpu_run,
4376 .handle_exit = vmx_handle_exit,
4377 .skip_emulated_instruction = skip_emulated_instruction,
4378 .set_interrupt_shadow = vmx_set_interrupt_shadow,
4379 .get_interrupt_shadow = vmx_get_interrupt_shadow,
4380 .patch_hypercall = vmx_patch_hypercall,
4381 .set_irq = vmx_inject_irq,
4382 .set_nmi = vmx_inject_nmi,
4383 .queue_exception = vmx_queue_exception,
4384 .cancel_injection = vmx_cancel_injection,
4385 .interrupt_allowed = vmx_interrupt_allowed,
4386 .nmi_allowed = vmx_nmi_allowed,
4387 .get_nmi_mask = vmx_get_nmi_mask,
4388 .set_nmi_mask = vmx_set_nmi_mask,
4389 .enable_nmi_window = enable_nmi_window,
4390 .enable_irq_window = enable_irq_window,
4391 .update_cr8_intercept = update_cr8_intercept,
4393 .set_tss_addr = vmx_set_tss_addr,
4394 .get_tdp_level = get_ept_level,
4395 .get_mt_mask = vmx_get_mt_mask,
4397 .get_exit_info = vmx_get_exit_info,
4398 .exit_reasons_str = vmx_exit_reasons_str,
4400 .get_lpage_level = vmx_get_lpage_level,
4402 .cpuid_update = vmx_cpuid_update,
4404 .rdtscp_supported = vmx_rdtscp_supported,
4406 .set_supported_cpuid = vmx_set_supported_cpuid,
4408 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
4410 .write_tsc_offset = vmx_write_tsc_offset,
4411 .adjust_tsc_offset = vmx_adjust_tsc_offset,
4413 .set_tdp_cr3 = vmx_set_cr3,
4416 static int __init vmx_init(void)
4418 int r, i;
4420 rdmsrl_safe(MSR_EFER, &host_efer);
4422 for (i = 0; i < NR_VMX_MSR; ++i)
4423 kvm_define_shared_msr(i, vmx_msr_index[i]);
4425 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
4426 if (!vmx_io_bitmap_a)
4427 return -ENOMEM;
4429 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
4430 if (!vmx_io_bitmap_b) {
4431 r = -ENOMEM;
4432 goto out;
4435 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
4436 if (!vmx_msr_bitmap_legacy) {
4437 r = -ENOMEM;
4438 goto out1;
4441 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
4442 if (!vmx_msr_bitmap_longmode) {
4443 r = -ENOMEM;
4444 goto out2;
4448 * Allow direct access to the PC debug port (it is often used for I/O
4449 * delays, but the vmexits simply slow things down).
4451 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
4452 clear_bit(0x80, vmx_io_bitmap_a);
4454 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
4456 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
4457 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
4459 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
4461 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
4462 __alignof__(struct vcpu_vmx), THIS_MODULE);
4463 if (r)
4464 goto out3;
4466 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
4467 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
4468 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
4469 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
4470 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
4471 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
4473 if (enable_ept) {
4474 bypass_guest_pf = 0;
4475 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
4476 VMX_EPT_EXECUTABLE_MASK);
4477 kvm_enable_tdp();
4478 } else
4479 kvm_disable_tdp();
4481 if (bypass_guest_pf)
4482 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
4484 return 0;
4486 out3:
4487 free_page((unsigned long)vmx_msr_bitmap_longmode);
4488 out2:
4489 free_page((unsigned long)vmx_msr_bitmap_legacy);
4490 out1:
4491 free_page((unsigned long)vmx_io_bitmap_b);
4492 out:
4493 free_page((unsigned long)vmx_io_bitmap_a);
4494 return r;
4497 static void __exit vmx_exit(void)
4499 free_page((unsigned long)vmx_msr_bitmap_legacy);
4500 free_page((unsigned long)vmx_msr_bitmap_longmode);
4501 free_page((unsigned long)vmx_io_bitmap_b);
4502 free_page((unsigned long)vmx_io_bitmap_a);
4504 kvm_exit();
4507 module_init(vmx_init)
4508 module_exit(vmx_exit)