2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
10 * Avi Kivity <avi@qumranet.com>
11 * Yaniv Kamay <yaniv@qumranet.com>
13 * This work is licensed under the terms of the GNU GPL, version 2. See
14 * the COPYING file in the top-level directory.
20 #include "x86_emulate.h"
23 #include "segment_descriptor.h"
26 #include <linux/module.h>
27 #include <linux/kernel.h>
29 #include <linux/highmem.h>
30 #include <linux/sched.h>
31 #include <linux/moduleparam.h>
36 MODULE_AUTHOR("Qumranet");
37 MODULE_LICENSE("GPL");
39 static int bypass_guest_pf = 1;
40 module_param(bypass_guest_pf, bool, 0);
52 u32 idt_vectoring_info;
53 struct kvm_msr_entry *guest_msrs;
54 struct kvm_msr_entry *host_msrs;
59 int msr_offset_kernel_gs_base;
64 u16 fs_sel, gs_sel, ldt_sel;
65 int gs_ldt_reload_needed;
67 int guest_efer_loaded;
78 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
80 return container_of(vcpu, struct vcpu_vmx, vcpu);
83 static int init_rmode_tss(struct kvm *kvm);
85 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
86 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
88 static struct page *vmx_io_bitmap_a;
89 static struct page *vmx_io_bitmap_b;
91 static struct vmcs_config {
95 u32 pin_based_exec_ctrl;
96 u32 cpu_based_exec_ctrl;
97 u32 cpu_based_2nd_exec_ctrl;
102 #define VMX_SEGMENT_FIELD(seg) \
103 [VCPU_SREG_##seg] = { \
104 .selector = GUEST_##seg##_SELECTOR, \
105 .base = GUEST_##seg##_BASE, \
106 .limit = GUEST_##seg##_LIMIT, \
107 .ar_bytes = GUEST_##seg##_AR_BYTES, \
110 static struct kvm_vmx_segment_field {
115 } kvm_vmx_segment_fields[] = {
116 VMX_SEGMENT_FIELD(CS),
117 VMX_SEGMENT_FIELD(DS),
118 VMX_SEGMENT_FIELD(ES),
119 VMX_SEGMENT_FIELD(FS),
120 VMX_SEGMENT_FIELD(GS),
121 VMX_SEGMENT_FIELD(SS),
122 VMX_SEGMENT_FIELD(TR),
123 VMX_SEGMENT_FIELD(LDTR),
127 * Keep MSR_K6_STAR at the end, as setup_msrs() will try to optimize it
128 * away by decrementing the array size.
130 static const u32 vmx_msr_index[] = {
132 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR, MSR_KERNEL_GS_BASE,
134 MSR_EFER, MSR_K6_STAR,
136 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
138 static void load_msrs(struct kvm_msr_entry *e, int n)
142 for (i = 0; i < n; ++i)
143 wrmsrl(e[i].index, e[i].data);
146 static void save_msrs(struct kvm_msr_entry *e, int n)
150 for (i = 0; i < n; ++i)
151 rdmsrl(e[i].index, e[i].data);
154 static inline int is_page_fault(u32 intr_info)
156 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
157 INTR_INFO_VALID_MASK)) ==
158 (INTR_TYPE_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
161 static inline int is_no_device(u32 intr_info)
163 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
164 INTR_INFO_VALID_MASK)) ==
165 (INTR_TYPE_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
168 static inline int is_invalid_opcode(u32 intr_info)
170 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
171 INTR_INFO_VALID_MASK)) ==
172 (INTR_TYPE_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
175 static inline int is_external_interrupt(u32 intr_info)
177 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
178 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
181 static inline int cpu_has_vmx_tpr_shadow(void)
183 return (vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW);
186 static inline int vm_need_tpr_shadow(struct kvm *kvm)
188 return ((cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm)));
191 static inline int cpu_has_secondary_exec_ctrls(void)
193 return (vmcs_config.cpu_based_exec_ctrl &
194 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS);
197 static inline int cpu_has_vmx_virtualize_apic_accesses(void)
199 return (vmcs_config.cpu_based_2nd_exec_ctrl &
200 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
203 static inline int vm_need_virtualize_apic_accesses(struct kvm *kvm)
205 return ((cpu_has_vmx_virtualize_apic_accesses()) &&
206 (irqchip_in_kernel(kvm)));
209 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
213 for (i = 0; i < vmx->nmsrs; ++i)
214 if (vmx->guest_msrs[i].index == msr)
219 static struct kvm_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
223 i = __find_msr_index(vmx, msr);
225 return &vmx->guest_msrs[i];
229 static void vmcs_clear(struct vmcs *vmcs)
231 u64 phys_addr = __pa(vmcs);
234 asm volatile (ASM_VMX_VMCLEAR_RAX "; setna %0"
235 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
238 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
242 static void __vcpu_clear(void *arg)
244 struct vcpu_vmx *vmx = arg;
245 int cpu = raw_smp_processor_id();
247 if (vmx->vcpu.cpu == cpu)
248 vmcs_clear(vmx->vmcs);
249 if (per_cpu(current_vmcs, cpu) == vmx->vmcs)
250 per_cpu(current_vmcs, cpu) = NULL;
251 rdtscll(vmx->vcpu.arch.host_tsc);
254 static void vcpu_clear(struct vcpu_vmx *vmx)
256 if (vmx->vcpu.cpu == -1)
258 smp_call_function_single(vmx->vcpu.cpu, __vcpu_clear, vmx, 0, 1);
262 static unsigned long vmcs_readl(unsigned long field)
266 asm volatile (ASM_VMX_VMREAD_RDX_RAX
267 : "=a"(value) : "d"(field) : "cc");
271 static u16 vmcs_read16(unsigned long field)
273 return vmcs_readl(field);
276 static u32 vmcs_read32(unsigned long field)
278 return vmcs_readl(field);
281 static u64 vmcs_read64(unsigned long field)
284 return vmcs_readl(field);
286 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
290 static noinline void vmwrite_error(unsigned long field, unsigned long value)
292 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
293 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
297 static void vmcs_writel(unsigned long field, unsigned long value)
301 asm volatile (ASM_VMX_VMWRITE_RAX_RDX "; setna %0"
302 : "=q"(error) : "a"(value), "d"(field) : "cc");
304 vmwrite_error(field, value);
307 static void vmcs_write16(unsigned long field, u16 value)
309 vmcs_writel(field, value);
312 static void vmcs_write32(unsigned long field, u32 value)
314 vmcs_writel(field, value);
317 static void vmcs_write64(unsigned long field, u64 value)
320 vmcs_writel(field, value);
322 vmcs_writel(field, value);
324 vmcs_writel(field+1, value >> 32);
328 static void vmcs_clear_bits(unsigned long field, u32 mask)
330 vmcs_writel(field, vmcs_readl(field) & ~mask);
333 static void vmcs_set_bits(unsigned long field, u32 mask)
335 vmcs_writel(field, vmcs_readl(field) | mask);
338 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
342 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR);
343 if (!vcpu->fpu_active)
344 eb |= 1u << NM_VECTOR;
345 if (vcpu->guest_debug.enabled)
347 if (vcpu->arch.rmode.active)
349 vmcs_write32(EXCEPTION_BITMAP, eb);
352 static void reload_tss(void)
354 #ifndef CONFIG_X86_64
357 * VT restores TR but not its size. Useless.
359 struct descriptor_table gdt;
360 struct segment_descriptor *descs;
363 descs = (void *)gdt.base;
364 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
369 static void load_transition_efer(struct vcpu_vmx *vmx)
371 int efer_offset = vmx->msr_offset_efer;
372 u64 host_efer = vmx->host_msrs[efer_offset].data;
373 u64 guest_efer = vmx->guest_msrs[efer_offset].data;
379 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
382 ignore_bits = EFER_NX | EFER_SCE;
384 ignore_bits |= EFER_LMA | EFER_LME;
385 /* SCE is meaningful only in long mode on Intel */
386 if (guest_efer & EFER_LMA)
387 ignore_bits &= ~(u64)EFER_SCE;
389 if ((guest_efer & ~ignore_bits) == (host_efer & ~ignore_bits))
392 vmx->host_state.guest_efer_loaded = 1;
393 guest_efer &= ~ignore_bits;
394 guest_efer |= host_efer & ignore_bits;
395 wrmsrl(MSR_EFER, guest_efer);
396 vmx->vcpu.stat.efer_reload++;
399 static void reload_host_efer(struct vcpu_vmx *vmx)
401 if (vmx->host_state.guest_efer_loaded) {
402 vmx->host_state.guest_efer_loaded = 0;
403 load_msrs(vmx->host_msrs + vmx->msr_offset_efer, 1);
407 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
409 struct vcpu_vmx *vmx = to_vmx(vcpu);
411 if (vmx->host_state.loaded)
414 vmx->host_state.loaded = 1;
416 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
417 * allow segment selectors with cpl > 0 or ti == 1.
419 vmx->host_state.ldt_sel = read_ldt();
420 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
421 vmx->host_state.fs_sel = read_fs();
422 if (!(vmx->host_state.fs_sel & 7)) {
423 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
424 vmx->host_state.fs_reload_needed = 0;
426 vmcs_write16(HOST_FS_SELECTOR, 0);
427 vmx->host_state.fs_reload_needed = 1;
429 vmx->host_state.gs_sel = read_gs();
430 if (!(vmx->host_state.gs_sel & 7))
431 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
433 vmcs_write16(HOST_GS_SELECTOR, 0);
434 vmx->host_state.gs_ldt_reload_needed = 1;
438 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
439 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
441 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
442 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
446 if (is_long_mode(&vmx->vcpu))
447 save_msrs(vmx->host_msrs +
448 vmx->msr_offset_kernel_gs_base, 1);
451 load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
452 load_transition_efer(vmx);
455 static void vmx_load_host_state(struct vcpu_vmx *vmx)
459 if (!vmx->host_state.loaded)
462 ++vmx->vcpu.stat.host_state_reload;
463 vmx->host_state.loaded = 0;
464 if (vmx->host_state.fs_reload_needed)
465 load_fs(vmx->host_state.fs_sel);
466 if (vmx->host_state.gs_ldt_reload_needed) {
467 load_ldt(vmx->host_state.ldt_sel);
469 * If we have to reload gs, we must take care to
470 * preserve our gs base.
472 local_irq_save(flags);
473 load_gs(vmx->host_state.gs_sel);
475 wrmsrl(MSR_GS_BASE, vmcs_readl(HOST_GS_BASE));
477 local_irq_restore(flags);
480 save_msrs(vmx->guest_msrs, vmx->save_nmsrs);
481 load_msrs(vmx->host_msrs, vmx->save_nmsrs);
482 reload_host_efer(vmx);
486 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
487 * vcpu mutex is already taken.
489 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
491 struct vcpu_vmx *vmx = to_vmx(vcpu);
492 u64 phys_addr = __pa(vmx->vmcs);
495 if (vcpu->cpu != cpu) {
497 kvm_migrate_apic_timer(vcpu);
500 if (per_cpu(current_vmcs, cpu) != vmx->vmcs) {
503 per_cpu(current_vmcs, cpu) = vmx->vmcs;
504 asm volatile (ASM_VMX_VMPTRLD_RAX "; setna %0"
505 : "=g"(error) : "a"(&phys_addr), "m"(phys_addr)
508 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
509 vmx->vmcs, phys_addr);
512 if (vcpu->cpu != cpu) {
513 struct descriptor_table dt;
514 unsigned long sysenter_esp;
518 * Linux uses per-cpu TSS and GDT, so set these when switching
521 vmcs_writel(HOST_TR_BASE, read_tr_base()); /* 22.2.4 */
523 vmcs_writel(HOST_GDTR_BASE, dt.base); /* 22.2.4 */
525 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
526 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
529 * Make sure the time stamp counter is monotonous.
532 delta = vcpu->arch.host_tsc - tsc_this;
533 vmcs_write64(TSC_OFFSET, vmcs_read64(TSC_OFFSET) + delta);
537 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
539 vmx_load_host_state(to_vmx(vcpu));
542 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
544 if (vcpu->fpu_active)
546 vcpu->fpu_active = 1;
547 vmcs_clear_bits(GUEST_CR0, X86_CR0_TS);
548 if (vcpu->arch.cr0 & X86_CR0_TS)
549 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
550 update_exception_bitmap(vcpu);
553 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
555 if (!vcpu->fpu_active)
557 vcpu->fpu_active = 0;
558 vmcs_set_bits(GUEST_CR0, X86_CR0_TS);
559 update_exception_bitmap(vcpu);
562 static void vmx_vcpu_decache(struct kvm_vcpu *vcpu)
564 vcpu_clear(to_vmx(vcpu));
567 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
569 return vmcs_readl(GUEST_RFLAGS);
572 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
574 if (vcpu->arch.rmode.active)
575 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
576 vmcs_writel(GUEST_RFLAGS, rflags);
579 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
582 u32 interruptibility;
584 rip = vmcs_readl(GUEST_RIP);
585 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
586 vmcs_writel(GUEST_RIP, rip);
589 * We emulated an instruction, so temporary interrupt blocking
590 * should be removed, if set.
592 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
593 if (interruptibility & 3)
594 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
595 interruptibility & ~3);
596 vcpu->arch.interrupt_window_open = 1;
599 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
600 bool has_error_code, u32 error_code)
602 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
603 nr | INTR_TYPE_EXCEPTION
604 | (has_error_code ? INTR_INFO_DELIEVER_CODE_MASK : 0)
605 | INTR_INFO_VALID_MASK);
607 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
610 static bool vmx_exception_injected(struct kvm_vcpu *vcpu)
612 struct vcpu_vmx *vmx = to_vmx(vcpu);
614 return !(vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
618 * Swap MSR entry in host/guest MSR entry array.
621 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
623 struct kvm_msr_entry tmp;
625 tmp = vmx->guest_msrs[to];
626 vmx->guest_msrs[to] = vmx->guest_msrs[from];
627 vmx->guest_msrs[from] = tmp;
628 tmp = vmx->host_msrs[to];
629 vmx->host_msrs[to] = vmx->host_msrs[from];
630 vmx->host_msrs[from] = tmp;
635 * Set up the vmcs to automatically save and restore system
636 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
637 * mode, as fiddling with msrs is very expensive.
639 static void setup_msrs(struct vcpu_vmx *vmx)
645 if (is_long_mode(&vmx->vcpu)) {
648 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
650 move_msr_up(vmx, index, save_nmsrs++);
651 index = __find_msr_index(vmx, MSR_LSTAR);
653 move_msr_up(vmx, index, save_nmsrs++);
654 index = __find_msr_index(vmx, MSR_CSTAR);
656 move_msr_up(vmx, index, save_nmsrs++);
657 index = __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
659 move_msr_up(vmx, index, save_nmsrs++);
661 * MSR_K6_STAR is only needed on long mode guests, and only
662 * if efer.sce is enabled.
664 index = __find_msr_index(vmx, MSR_K6_STAR);
665 if ((index >= 0) && (vmx->vcpu.arch.shadow_efer & EFER_SCE))
666 move_msr_up(vmx, index, save_nmsrs++);
669 vmx->save_nmsrs = save_nmsrs;
672 vmx->msr_offset_kernel_gs_base =
673 __find_msr_index(vmx, MSR_KERNEL_GS_BASE);
675 vmx->msr_offset_efer = __find_msr_index(vmx, MSR_EFER);
679 * reads and returns guest's timestamp counter "register"
680 * guest_tsc = host_tsc + tsc_offset -- 21.3
682 static u64 guest_read_tsc(void)
684 u64 host_tsc, tsc_offset;
687 tsc_offset = vmcs_read64(TSC_OFFSET);
688 return host_tsc + tsc_offset;
692 * writes 'guest_tsc' into guest's timestamp counter "register"
693 * guest_tsc = host_tsc + tsc_offset ==> tsc_offset = guest_tsc - host_tsc
695 static void guest_write_tsc(u64 guest_tsc)
700 vmcs_write64(TSC_OFFSET, guest_tsc - host_tsc);
704 * Reads an msr value (of 'msr_index') into 'pdata'.
705 * Returns 0 on success, non-0 otherwise.
706 * Assumes vcpu_load() was already called.
708 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
711 struct kvm_msr_entry *msr;
714 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
721 data = vmcs_readl(GUEST_FS_BASE);
724 data = vmcs_readl(GUEST_GS_BASE);
727 return kvm_get_msr_common(vcpu, msr_index, pdata);
729 case MSR_IA32_TIME_STAMP_COUNTER:
730 data = guest_read_tsc();
732 case MSR_IA32_SYSENTER_CS:
733 data = vmcs_read32(GUEST_SYSENTER_CS);
735 case MSR_IA32_SYSENTER_EIP:
736 data = vmcs_readl(GUEST_SYSENTER_EIP);
738 case MSR_IA32_SYSENTER_ESP:
739 data = vmcs_readl(GUEST_SYSENTER_ESP);
742 msr = find_msr_entry(to_vmx(vcpu), msr_index);
747 return kvm_get_msr_common(vcpu, msr_index, pdata);
755 * Writes msr value into into the appropriate "register".
756 * Returns 0 on success, non-0 otherwise.
757 * Assumes vcpu_load() was already called.
759 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
761 struct vcpu_vmx *vmx = to_vmx(vcpu);
762 struct kvm_msr_entry *msr;
768 ret = kvm_set_msr_common(vcpu, msr_index, data);
769 if (vmx->host_state.loaded) {
770 reload_host_efer(vmx);
771 load_transition_efer(vmx);
775 vmcs_writel(GUEST_FS_BASE, data);
778 vmcs_writel(GUEST_GS_BASE, data);
781 case MSR_IA32_SYSENTER_CS:
782 vmcs_write32(GUEST_SYSENTER_CS, data);
784 case MSR_IA32_SYSENTER_EIP:
785 vmcs_writel(GUEST_SYSENTER_EIP, data);
787 case MSR_IA32_SYSENTER_ESP:
788 vmcs_writel(GUEST_SYSENTER_ESP, data);
790 case MSR_IA32_TIME_STAMP_COUNTER:
791 guest_write_tsc(data);
794 msr = find_msr_entry(vmx, msr_index);
797 if (vmx->host_state.loaded)
798 load_msrs(vmx->guest_msrs, vmx->save_nmsrs);
801 ret = kvm_set_msr_common(vcpu, msr_index, data);
808 * Sync the rsp and rip registers into the vcpu structure. This allows
809 * registers to be accessed by indexing vcpu->arch.regs.
811 static void vcpu_load_rsp_rip(struct kvm_vcpu *vcpu)
813 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
814 vcpu->arch.rip = vmcs_readl(GUEST_RIP);
818 * Syncs rsp and rip back into the vmcs. Should be called after possible
821 static void vcpu_put_rsp_rip(struct kvm_vcpu *vcpu)
823 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
824 vmcs_writel(GUEST_RIP, vcpu->arch.rip);
827 static int set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_debug_guest *dbg)
829 unsigned long dr7 = 0x400;
832 old_singlestep = vcpu->guest_debug.singlestep;
834 vcpu->guest_debug.enabled = dbg->enabled;
835 if (vcpu->guest_debug.enabled) {
838 dr7 |= 0x200; /* exact */
839 for (i = 0; i < 4; ++i) {
840 if (!dbg->breakpoints[i].enabled)
842 vcpu->guest_debug.bp[i] = dbg->breakpoints[i].address;
843 dr7 |= 2 << (i*2); /* global enable */
844 dr7 |= 0 << (i*4+16); /* execution breakpoint */
847 vcpu->guest_debug.singlestep = dbg->singlestep;
849 vcpu->guest_debug.singlestep = 0;
851 if (old_singlestep && !vcpu->guest_debug.singlestep) {
854 flags = vmcs_readl(GUEST_RFLAGS);
855 flags &= ~(X86_EFLAGS_TF | X86_EFLAGS_RF);
856 vmcs_writel(GUEST_RFLAGS, flags);
859 update_exception_bitmap(vcpu);
860 vmcs_writel(GUEST_DR7, dr7);
865 static int vmx_get_irq(struct kvm_vcpu *vcpu)
867 struct vcpu_vmx *vmx = to_vmx(vcpu);
870 idtv_info_field = vmx->idt_vectoring_info;
871 if (idtv_info_field & INTR_INFO_VALID_MASK) {
872 if (is_external_interrupt(idtv_info_field))
873 return idtv_info_field & VECTORING_INFO_VECTOR_MASK;
875 printk(KERN_DEBUG "pending exception: not handled yet\n");
880 static __init int cpu_has_kvm_support(void)
882 unsigned long ecx = cpuid_ecx(1);
883 return test_bit(5, &ecx); /* CPUID.1:ECX.VMX[bit 5] -> VT */
886 static __init int vmx_disabled_by_bios(void)
890 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
891 return (msr & (MSR_IA32_FEATURE_CONTROL_LOCKED |
892 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
893 == MSR_IA32_FEATURE_CONTROL_LOCKED;
894 /* locked but not enabled */
897 static void hardware_enable(void *garbage)
899 int cpu = raw_smp_processor_id();
900 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
903 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
904 if ((old & (MSR_IA32_FEATURE_CONTROL_LOCKED |
905 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
906 != (MSR_IA32_FEATURE_CONTROL_LOCKED |
907 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED))
908 /* enable and lock */
909 wrmsrl(MSR_IA32_FEATURE_CONTROL, old |
910 MSR_IA32_FEATURE_CONTROL_LOCKED |
911 MSR_IA32_FEATURE_CONTROL_VMXON_ENABLED);
912 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
913 asm volatile (ASM_VMX_VMXON_RAX : : "a"(&phys_addr), "m"(phys_addr)
917 static void hardware_disable(void *garbage)
919 asm volatile (ASM_VMX_VMXOFF : : : "cc");
922 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
923 u32 msr, u32 *result)
925 u32 vmx_msr_low, vmx_msr_high;
926 u32 ctl = ctl_min | ctl_opt;
928 rdmsr(msr, vmx_msr_low, vmx_msr_high);
930 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
931 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
933 /* Ensure minimum (required) set of control bits are supported. */
941 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
943 u32 vmx_msr_low, vmx_msr_high;
945 u32 _pin_based_exec_control = 0;
946 u32 _cpu_based_exec_control = 0;
947 u32 _cpu_based_2nd_exec_control = 0;
948 u32 _vmexit_control = 0;
949 u32 _vmentry_control = 0;
951 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
953 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
954 &_pin_based_exec_control) < 0)
957 min = CPU_BASED_HLT_EXITING |
959 CPU_BASED_CR8_LOAD_EXITING |
960 CPU_BASED_CR8_STORE_EXITING |
962 CPU_BASED_USE_IO_BITMAPS |
963 CPU_BASED_MOV_DR_EXITING |
964 CPU_BASED_USE_TSC_OFFSETING;
965 opt = CPU_BASED_TPR_SHADOW |
966 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
967 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
968 &_cpu_based_exec_control) < 0)
971 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
972 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
973 ~CPU_BASED_CR8_STORE_EXITING;
975 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
977 opt = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
978 SECONDARY_EXEC_WBINVD_EXITING;
979 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS2,
980 &_cpu_based_2nd_exec_control) < 0)
983 #ifndef CONFIG_X86_64
984 if (!(_cpu_based_2nd_exec_control &
985 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
986 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
991 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
994 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
995 &_vmexit_control) < 0)
999 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
1000 &_vmentry_control) < 0)
1003 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
1005 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
1006 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
1009 #ifdef CONFIG_X86_64
1010 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
1011 if (vmx_msr_high & (1u<<16))
1015 /* Require Write-Back (WB) memory type for VMCS accesses. */
1016 if (((vmx_msr_high >> 18) & 15) != 6)
1019 vmcs_conf->size = vmx_msr_high & 0x1fff;
1020 vmcs_conf->order = get_order(vmcs_config.size);
1021 vmcs_conf->revision_id = vmx_msr_low;
1023 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
1024 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
1025 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
1026 vmcs_conf->vmexit_ctrl = _vmexit_control;
1027 vmcs_conf->vmentry_ctrl = _vmentry_control;
1032 static struct vmcs *alloc_vmcs_cpu(int cpu)
1034 int node = cpu_to_node(cpu);
1038 pages = alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
1041 vmcs = page_address(pages);
1042 memset(vmcs, 0, vmcs_config.size);
1043 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
1047 static struct vmcs *alloc_vmcs(void)
1049 return alloc_vmcs_cpu(raw_smp_processor_id());
1052 static void free_vmcs(struct vmcs *vmcs)
1054 free_pages((unsigned long)vmcs, vmcs_config.order);
1057 static void free_kvm_area(void)
1061 for_each_online_cpu(cpu)
1062 free_vmcs(per_cpu(vmxarea, cpu));
1065 static __init int alloc_kvm_area(void)
1069 for_each_online_cpu(cpu) {
1072 vmcs = alloc_vmcs_cpu(cpu);
1078 per_cpu(vmxarea, cpu) = vmcs;
1083 static __init int hardware_setup(void)
1085 if (setup_vmcs_config(&vmcs_config) < 0)
1087 return alloc_kvm_area();
1090 static __exit void hardware_unsetup(void)
1095 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
1097 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1099 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
1100 vmcs_write16(sf->selector, save->selector);
1101 vmcs_writel(sf->base, save->base);
1102 vmcs_write32(sf->limit, save->limit);
1103 vmcs_write32(sf->ar_bytes, save->ar);
1105 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
1107 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
1111 static void enter_pmode(struct kvm_vcpu *vcpu)
1113 unsigned long flags;
1115 vcpu->arch.rmode.active = 0;
1117 vmcs_writel(GUEST_TR_BASE, vcpu->arch.rmode.tr.base);
1118 vmcs_write32(GUEST_TR_LIMIT, vcpu->arch.rmode.tr.limit);
1119 vmcs_write32(GUEST_TR_AR_BYTES, vcpu->arch.rmode.tr.ar);
1121 flags = vmcs_readl(GUEST_RFLAGS);
1122 flags &= ~(X86_EFLAGS_IOPL | X86_EFLAGS_VM);
1123 flags |= (vcpu->arch.rmode.save_iopl << IOPL_SHIFT);
1124 vmcs_writel(GUEST_RFLAGS, flags);
1126 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
1127 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
1129 update_exception_bitmap(vcpu);
1131 fix_pmode_dataseg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1132 fix_pmode_dataseg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1133 fix_pmode_dataseg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1134 fix_pmode_dataseg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1136 vmcs_write16(GUEST_SS_SELECTOR, 0);
1137 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
1139 vmcs_write16(GUEST_CS_SELECTOR,
1140 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
1141 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1144 static gva_t rmode_tss_base(struct kvm *kvm)
1146 if (!kvm->arch.tss_addr) {
1147 gfn_t base_gfn = kvm->memslots[0].base_gfn +
1148 kvm->memslots[0].npages - 3;
1149 return base_gfn << PAGE_SHIFT;
1151 return kvm->arch.tss_addr;
1154 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
1156 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1158 save->selector = vmcs_read16(sf->selector);
1159 save->base = vmcs_readl(sf->base);
1160 save->limit = vmcs_read32(sf->limit);
1161 save->ar = vmcs_read32(sf->ar_bytes);
1162 vmcs_write16(sf->selector, save->base >> 4);
1163 vmcs_write32(sf->base, save->base & 0xfffff);
1164 vmcs_write32(sf->limit, 0xffff);
1165 vmcs_write32(sf->ar_bytes, 0xf3);
1168 static void enter_rmode(struct kvm_vcpu *vcpu)
1170 unsigned long flags;
1172 vcpu->arch.rmode.active = 1;
1174 vcpu->arch.rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
1175 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
1177 vcpu->arch.rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
1178 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
1180 vcpu->arch.rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
1181 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1183 flags = vmcs_readl(GUEST_RFLAGS);
1184 vcpu->arch.rmode.save_iopl
1185 = (flags & X86_EFLAGS_IOPL) >> IOPL_SHIFT;
1187 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1189 vmcs_writel(GUEST_RFLAGS, flags);
1190 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
1191 update_exception_bitmap(vcpu);
1193 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
1194 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
1195 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
1197 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
1198 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1199 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
1200 vmcs_writel(GUEST_CS_BASE, 0xf0000);
1201 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
1203 fix_rmode_seg(VCPU_SREG_ES, &vcpu->arch.rmode.es);
1204 fix_rmode_seg(VCPU_SREG_DS, &vcpu->arch.rmode.ds);
1205 fix_rmode_seg(VCPU_SREG_GS, &vcpu->arch.rmode.gs);
1206 fix_rmode_seg(VCPU_SREG_FS, &vcpu->arch.rmode.fs);
1208 kvm_mmu_reset_context(vcpu);
1209 init_rmode_tss(vcpu->kvm);
1212 #ifdef CONFIG_X86_64
1214 static void enter_lmode(struct kvm_vcpu *vcpu)
1218 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
1219 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
1220 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
1222 vmcs_write32(GUEST_TR_AR_BYTES,
1223 (guest_tr_ar & ~AR_TYPE_MASK)
1224 | AR_TYPE_BUSY_64_TSS);
1227 vcpu->arch.shadow_efer |= EFER_LMA;
1229 find_msr_entry(to_vmx(vcpu), MSR_EFER)->data |= EFER_LMA | EFER_LME;
1230 vmcs_write32(VM_ENTRY_CONTROLS,
1231 vmcs_read32(VM_ENTRY_CONTROLS)
1232 | VM_ENTRY_IA32E_MODE);
1235 static void exit_lmode(struct kvm_vcpu *vcpu)
1237 vcpu->arch.shadow_efer &= ~EFER_LMA;
1239 vmcs_write32(VM_ENTRY_CONTROLS,
1240 vmcs_read32(VM_ENTRY_CONTROLS)
1241 & ~VM_ENTRY_IA32E_MODE);
1246 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
1248 vcpu->arch.cr4 &= KVM_GUEST_CR4_MASK;
1249 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & ~KVM_GUEST_CR4_MASK;
1252 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
1254 vmx_fpu_deactivate(vcpu);
1256 if (vcpu->arch.rmode.active && (cr0 & X86_CR0_PE))
1259 if (!vcpu->arch.rmode.active && !(cr0 & X86_CR0_PE))
1262 #ifdef CONFIG_X86_64
1263 if (vcpu->arch.shadow_efer & EFER_LME) {
1264 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
1266 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
1271 vmcs_writel(CR0_READ_SHADOW, cr0);
1272 vmcs_writel(GUEST_CR0,
1273 (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON);
1274 vcpu->arch.cr0 = cr0;
1276 if (!(cr0 & X86_CR0_TS) || !(cr0 & X86_CR0_PE))
1277 vmx_fpu_activate(vcpu);
1280 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
1282 vmcs_writel(GUEST_CR3, cr3);
1283 if (vcpu->arch.cr0 & X86_CR0_PE)
1284 vmx_fpu_deactivate(vcpu);
1287 static void vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
1289 vmcs_writel(CR4_READ_SHADOW, cr4);
1290 vmcs_writel(GUEST_CR4, cr4 | (vcpu->arch.rmode.active ?
1291 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON));
1292 vcpu->arch.cr4 = cr4;
1295 #ifdef CONFIG_X86_64
1297 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
1299 struct vcpu_vmx *vmx = to_vmx(vcpu);
1300 struct kvm_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
1302 vcpu->arch.shadow_efer = efer;
1303 if (efer & EFER_LMA) {
1304 vmcs_write32(VM_ENTRY_CONTROLS,
1305 vmcs_read32(VM_ENTRY_CONTROLS) |
1306 VM_ENTRY_IA32E_MODE);
1310 vmcs_write32(VM_ENTRY_CONTROLS,
1311 vmcs_read32(VM_ENTRY_CONTROLS) &
1312 ~VM_ENTRY_IA32E_MODE);
1314 msr->data = efer & ~EFER_LME;
1321 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
1323 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1325 return vmcs_readl(sf->base);
1328 static void vmx_get_segment(struct kvm_vcpu *vcpu,
1329 struct kvm_segment *var, int seg)
1331 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1334 var->base = vmcs_readl(sf->base);
1335 var->limit = vmcs_read32(sf->limit);
1336 var->selector = vmcs_read16(sf->selector);
1337 ar = vmcs_read32(sf->ar_bytes);
1338 if (ar & AR_UNUSABLE_MASK)
1340 var->type = ar & 15;
1341 var->s = (ar >> 4) & 1;
1342 var->dpl = (ar >> 5) & 3;
1343 var->present = (ar >> 7) & 1;
1344 var->avl = (ar >> 12) & 1;
1345 var->l = (ar >> 13) & 1;
1346 var->db = (ar >> 14) & 1;
1347 var->g = (ar >> 15) & 1;
1348 var->unusable = (ar >> 16) & 1;
1351 static u32 vmx_segment_access_rights(struct kvm_segment *var)
1358 ar = var->type & 15;
1359 ar |= (var->s & 1) << 4;
1360 ar |= (var->dpl & 3) << 5;
1361 ar |= (var->present & 1) << 7;
1362 ar |= (var->avl & 1) << 12;
1363 ar |= (var->l & 1) << 13;
1364 ar |= (var->db & 1) << 14;
1365 ar |= (var->g & 1) << 15;
1367 if (ar == 0) /* a 0 value means unusable */
1368 ar = AR_UNUSABLE_MASK;
1373 static void vmx_set_segment(struct kvm_vcpu *vcpu,
1374 struct kvm_segment *var, int seg)
1376 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1379 if (vcpu->arch.rmode.active && seg == VCPU_SREG_TR) {
1380 vcpu->arch.rmode.tr.selector = var->selector;
1381 vcpu->arch.rmode.tr.base = var->base;
1382 vcpu->arch.rmode.tr.limit = var->limit;
1383 vcpu->arch.rmode.tr.ar = vmx_segment_access_rights(var);
1386 vmcs_writel(sf->base, var->base);
1387 vmcs_write32(sf->limit, var->limit);
1388 vmcs_write16(sf->selector, var->selector);
1389 if (vcpu->arch.rmode.active && var->s) {
1391 * Hack real-mode segments into vm86 compatibility.
1393 if (var->base == 0xffff0000 && var->selector == 0xf000)
1394 vmcs_writel(sf->base, 0xf0000);
1397 ar = vmx_segment_access_rights(var);
1398 vmcs_write32(sf->ar_bytes, ar);
1401 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
1403 u32 ar = vmcs_read32(GUEST_CS_AR_BYTES);
1405 *db = (ar >> 14) & 1;
1406 *l = (ar >> 13) & 1;
1409 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1411 dt->limit = vmcs_read32(GUEST_IDTR_LIMIT);
1412 dt->base = vmcs_readl(GUEST_IDTR_BASE);
1415 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1417 vmcs_write32(GUEST_IDTR_LIMIT, dt->limit);
1418 vmcs_writel(GUEST_IDTR_BASE, dt->base);
1421 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1423 dt->limit = vmcs_read32(GUEST_GDTR_LIMIT);
1424 dt->base = vmcs_readl(GUEST_GDTR_BASE);
1427 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct descriptor_table *dt)
1429 vmcs_write32(GUEST_GDTR_LIMIT, dt->limit);
1430 vmcs_writel(GUEST_GDTR_BASE, dt->base);
1433 static int init_rmode_tss(struct kvm *kvm)
1435 gfn_t fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
1439 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1442 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
1443 r = kvm_write_guest_page(kvm, fn++, &data, 0x66, sizeof(u16));
1446 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
1449 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
1453 r = kvm_write_guest_page(kvm, fn, &data, RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
1460 static void seg_setup(int seg)
1462 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
1464 vmcs_write16(sf->selector, 0);
1465 vmcs_writel(sf->base, 0);
1466 vmcs_write32(sf->limit, 0xffff);
1467 vmcs_write32(sf->ar_bytes, 0x93);
1470 static int alloc_apic_access_page(struct kvm *kvm)
1472 struct kvm_userspace_memory_region kvm_userspace_mem;
1475 mutex_lock(&kvm->lock);
1476 if (kvm->arch.apic_access_page)
1478 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
1479 kvm_userspace_mem.flags = 0;
1480 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
1481 kvm_userspace_mem.memory_size = PAGE_SIZE;
1482 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
1485 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
1487 mutex_unlock(&kvm->lock);
1492 * Sets up the vmcs for emulated real mode.
1494 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
1496 u32 host_sysenter_cs;
1499 struct descriptor_table dt;
1501 unsigned long kvm_vmx_return;
1505 vmcs_write64(IO_BITMAP_A, page_to_phys(vmx_io_bitmap_a));
1506 vmcs_write64(IO_BITMAP_B, page_to_phys(vmx_io_bitmap_b));
1508 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
1511 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
1512 vmcs_config.pin_based_exec_ctrl);
1514 exec_control = vmcs_config.cpu_based_exec_ctrl;
1515 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
1516 exec_control &= ~CPU_BASED_TPR_SHADOW;
1517 #ifdef CONFIG_X86_64
1518 exec_control |= CPU_BASED_CR8_STORE_EXITING |
1519 CPU_BASED_CR8_LOAD_EXITING;
1522 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
1524 if (cpu_has_secondary_exec_ctrls()) {
1525 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
1526 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
1528 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1529 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
1532 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
1533 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
1534 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
1536 vmcs_writel(HOST_CR0, read_cr0()); /* 22.2.3 */
1537 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
1538 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
1540 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
1541 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1542 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1543 vmcs_write16(HOST_FS_SELECTOR, read_fs()); /* 22.2.4 */
1544 vmcs_write16(HOST_GS_SELECTOR, read_gs()); /* 22.2.4 */
1545 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
1546 #ifdef CONFIG_X86_64
1547 rdmsrl(MSR_FS_BASE, a);
1548 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
1549 rdmsrl(MSR_GS_BASE, a);
1550 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
1552 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
1553 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
1556 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
1559 vmcs_writel(HOST_IDTR_BASE, dt.base); /* 22.2.4 */
1561 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
1562 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
1563 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
1564 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
1565 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
1567 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
1568 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
1569 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
1570 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
1571 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
1572 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
1574 for (i = 0; i < NR_VMX_MSR; ++i) {
1575 u32 index = vmx_msr_index[i];
1576 u32 data_low, data_high;
1580 if (rdmsr_safe(index, &data_low, &data_high) < 0)
1582 if (wrmsr_safe(index, data_low, data_high) < 0)
1584 data = data_low | ((u64)data_high << 32);
1585 vmx->host_msrs[j].index = index;
1586 vmx->host_msrs[j].reserved = 0;
1587 vmx->host_msrs[j].data = data;
1588 vmx->guest_msrs[j] = vmx->host_msrs[j];
1592 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
1594 /* 22.2.1, 20.8.1 */
1595 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
1597 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
1598 vmcs_writel(CR4_GUEST_HOST_MASK, KVM_GUEST_CR4_MASK);
1600 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
1601 if (alloc_apic_access_page(vmx->vcpu.kvm) != 0)
1607 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
1609 struct vcpu_vmx *vmx = to_vmx(vcpu);
1613 if (!init_rmode_tss(vmx->vcpu.kvm)) {
1618 vmx->vcpu.arch.rmode.active = 0;
1620 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
1621 set_cr8(&vmx->vcpu, 0);
1622 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
1623 if (vmx->vcpu.vcpu_id == 0)
1624 msr |= MSR_IA32_APICBASE_BSP;
1625 kvm_set_apic_base(&vmx->vcpu, msr);
1627 fx_init(&vmx->vcpu);
1630 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
1631 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
1633 if (vmx->vcpu.vcpu_id == 0) {
1634 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
1635 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
1637 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
1638 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
1640 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
1641 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
1643 seg_setup(VCPU_SREG_DS);
1644 seg_setup(VCPU_SREG_ES);
1645 seg_setup(VCPU_SREG_FS);
1646 seg_setup(VCPU_SREG_GS);
1647 seg_setup(VCPU_SREG_SS);
1649 vmcs_write16(GUEST_TR_SELECTOR, 0);
1650 vmcs_writel(GUEST_TR_BASE, 0);
1651 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
1652 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
1654 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
1655 vmcs_writel(GUEST_LDTR_BASE, 0);
1656 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
1657 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
1659 vmcs_write32(GUEST_SYSENTER_CS, 0);
1660 vmcs_writel(GUEST_SYSENTER_ESP, 0);
1661 vmcs_writel(GUEST_SYSENTER_EIP, 0);
1663 vmcs_writel(GUEST_RFLAGS, 0x02);
1664 if (vmx->vcpu.vcpu_id == 0)
1665 vmcs_writel(GUEST_RIP, 0xfff0);
1667 vmcs_writel(GUEST_RIP, 0);
1668 vmcs_writel(GUEST_RSP, 0);
1670 /* todo: dr0 = dr1 = dr2 = dr3 = 0; dr6 = 0xffff0ff0 */
1671 vmcs_writel(GUEST_DR7, 0x400);
1673 vmcs_writel(GUEST_GDTR_BASE, 0);
1674 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
1676 vmcs_writel(GUEST_IDTR_BASE, 0);
1677 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
1679 vmcs_write32(GUEST_ACTIVITY_STATE, 0);
1680 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
1681 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
1685 /* Special registers */
1686 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
1690 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
1692 if (cpu_has_vmx_tpr_shadow()) {
1693 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
1694 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
1695 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
1696 page_to_phys(vmx->vcpu.arch.apic->regs_page));
1697 vmcs_write32(TPR_THRESHOLD, 0);
1700 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
1701 vmcs_write64(APIC_ACCESS_ADDR,
1702 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
1704 vmx->vcpu.arch.cr0 = 0x60000010;
1705 vmx_set_cr0(&vmx->vcpu, vmx->vcpu.arch.cr0); /* enter rmode */
1706 vmx_set_cr4(&vmx->vcpu, 0);
1707 #ifdef CONFIG_X86_64
1708 vmx_set_efer(&vmx->vcpu, 0);
1710 vmx_fpu_activate(&vmx->vcpu);
1711 update_exception_bitmap(&vmx->vcpu);
1719 static void vmx_inject_irq(struct kvm_vcpu *vcpu, int irq)
1721 struct vcpu_vmx *vmx = to_vmx(vcpu);
1723 if (vcpu->arch.rmode.active) {
1724 vmx->rmode.irq.pending = true;
1725 vmx->rmode.irq.vector = irq;
1726 vmx->rmode.irq.rip = vmcs_readl(GUEST_RIP);
1727 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1728 irq | INTR_TYPE_SOFT_INTR | INTR_INFO_VALID_MASK);
1729 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN, 1);
1730 vmcs_writel(GUEST_RIP, vmx->rmode.irq.rip - 1);
1733 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
1734 irq | INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1737 static void kvm_do_inject_irq(struct kvm_vcpu *vcpu)
1739 int word_index = __ffs(vcpu->arch.irq_summary);
1740 int bit_index = __ffs(vcpu->arch.irq_pending[word_index]);
1741 int irq = word_index * BITS_PER_LONG + bit_index;
1743 clear_bit(bit_index, &vcpu->arch.irq_pending[word_index]);
1744 if (!vcpu->arch.irq_pending[word_index])
1745 clear_bit(word_index, &vcpu->arch.irq_summary);
1746 vmx_inject_irq(vcpu, irq);
1750 static void do_interrupt_requests(struct kvm_vcpu *vcpu,
1751 struct kvm_run *kvm_run)
1753 u32 cpu_based_vm_exec_control;
1755 vcpu->arch.interrupt_window_open =
1756 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
1757 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
1759 if (vcpu->arch.interrupt_window_open &&
1760 vcpu->arch.irq_summary &&
1761 !(vmcs_read32(VM_ENTRY_INTR_INFO_FIELD) & INTR_INFO_VALID_MASK))
1763 * If interrupts enabled, and not blocked by sti or mov ss. Good.
1765 kvm_do_inject_irq(vcpu);
1767 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
1768 if (!vcpu->arch.interrupt_window_open &&
1769 (vcpu->arch.irq_summary || kvm_run->request_interrupt_window))
1771 * Interrupts blocked. Wait for unblock.
1773 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
1775 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
1776 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
1779 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
1782 struct kvm_userspace_memory_region tss_mem = {
1784 .guest_phys_addr = addr,
1785 .memory_size = PAGE_SIZE * 3,
1789 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
1792 kvm->arch.tss_addr = addr;
1796 static void kvm_guest_debug_pre(struct kvm_vcpu *vcpu)
1798 struct kvm_guest_debug *dbg = &vcpu->guest_debug;
1800 set_debugreg(dbg->bp[0], 0);
1801 set_debugreg(dbg->bp[1], 1);
1802 set_debugreg(dbg->bp[2], 2);
1803 set_debugreg(dbg->bp[3], 3);
1805 if (dbg->singlestep) {
1806 unsigned long flags;
1808 flags = vmcs_readl(GUEST_RFLAGS);
1809 flags |= X86_EFLAGS_TF | X86_EFLAGS_RF;
1810 vmcs_writel(GUEST_RFLAGS, flags);
1814 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
1815 int vec, u32 err_code)
1817 if (!vcpu->arch.rmode.active)
1821 * Instruction with address size override prefix opcode 0x67
1822 * Cause the #SS fault with 0 error code in VM86 mode.
1824 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
1825 if (emulate_instruction(vcpu, NULL, 0, 0, 0) == EMULATE_DONE)
1830 static int handle_exception(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1832 struct vcpu_vmx *vmx = to_vmx(vcpu);
1833 u32 intr_info, error_code;
1834 unsigned long cr2, rip;
1836 enum emulation_result er;
1838 vect_info = vmx->idt_vectoring_info;
1839 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
1841 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
1842 !is_page_fault(intr_info))
1843 printk(KERN_ERR "%s: unexpected, vectoring info 0x%x "
1844 "intr info 0x%x\n", __FUNCTION__, vect_info, intr_info);
1846 if (!irqchip_in_kernel(vcpu->kvm) && is_external_interrupt(vect_info)) {
1847 int irq = vect_info & VECTORING_INFO_VECTOR_MASK;
1848 set_bit(irq, vcpu->arch.irq_pending);
1849 set_bit(irq / BITS_PER_LONG, &vcpu->arch.irq_summary);
1852 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) /* nmi */
1853 return 1; /* already handled by vmx_vcpu_run() */
1855 if (is_no_device(intr_info)) {
1856 vmx_fpu_activate(vcpu);
1860 if (is_invalid_opcode(intr_info)) {
1861 er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
1862 if (er != EMULATE_DONE)
1863 kvm_queue_exception(vcpu, UD_VECTOR);
1868 rip = vmcs_readl(GUEST_RIP);
1869 if (intr_info & INTR_INFO_DELIEVER_CODE_MASK)
1870 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
1871 if (is_page_fault(intr_info)) {
1872 cr2 = vmcs_readl(EXIT_QUALIFICATION);
1873 return kvm_mmu_page_fault(vcpu, cr2, error_code);
1876 if (vcpu->arch.rmode.active &&
1877 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
1879 if (vcpu->arch.halt_request) {
1880 vcpu->arch.halt_request = 0;
1881 return kvm_emulate_halt(vcpu);
1886 if ((intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK)) ==
1887 (INTR_TYPE_EXCEPTION | 1)) {
1888 kvm_run->exit_reason = KVM_EXIT_DEBUG;
1891 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
1892 kvm_run->ex.exception = intr_info & INTR_INFO_VECTOR_MASK;
1893 kvm_run->ex.error_code = error_code;
1897 static int handle_external_interrupt(struct kvm_vcpu *vcpu,
1898 struct kvm_run *kvm_run)
1900 ++vcpu->stat.irq_exits;
1904 static int handle_triple_fault(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1906 kvm_run->exit_reason = KVM_EXIT_SHUTDOWN;
1910 static int handle_io(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1912 unsigned long exit_qualification;
1913 int size, down, in, string, rep;
1916 ++vcpu->stat.io_exits;
1917 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
1918 string = (exit_qualification & 16) != 0;
1921 if (emulate_instruction(vcpu,
1922 kvm_run, 0, 0, 0) == EMULATE_DO_MMIO)
1927 size = (exit_qualification & 7) + 1;
1928 in = (exit_qualification & 8) != 0;
1929 down = (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_DF) != 0;
1930 rep = (exit_qualification & 32) != 0;
1931 port = exit_qualification >> 16;
1933 return kvm_emulate_pio(vcpu, kvm_run, in, size, port);
1937 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
1940 * Patch in the VMCALL instruction:
1942 hypercall[0] = 0x0f;
1943 hypercall[1] = 0x01;
1944 hypercall[2] = 0xc1;
1947 static int handle_cr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
1949 unsigned long exit_qualification;
1953 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
1954 cr = exit_qualification & 15;
1955 reg = (exit_qualification >> 8) & 15;
1956 switch ((exit_qualification >> 4) & 3) {
1957 case 0: /* mov to cr */
1960 vcpu_load_rsp_rip(vcpu);
1961 set_cr0(vcpu, vcpu->arch.regs[reg]);
1962 skip_emulated_instruction(vcpu);
1965 vcpu_load_rsp_rip(vcpu);
1966 set_cr3(vcpu, vcpu->arch.regs[reg]);
1967 skip_emulated_instruction(vcpu);
1970 vcpu_load_rsp_rip(vcpu);
1971 set_cr4(vcpu, vcpu->arch.regs[reg]);
1972 skip_emulated_instruction(vcpu);
1975 vcpu_load_rsp_rip(vcpu);
1976 set_cr8(vcpu, vcpu->arch.regs[reg]);
1977 skip_emulated_instruction(vcpu);
1978 if (irqchip_in_kernel(vcpu->kvm))
1980 kvm_run->exit_reason = KVM_EXIT_SET_TPR;
1985 vcpu_load_rsp_rip(vcpu);
1986 vmx_fpu_deactivate(vcpu);
1987 vcpu->arch.cr0 &= ~X86_CR0_TS;
1988 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1989 vmx_fpu_activate(vcpu);
1990 skip_emulated_instruction(vcpu);
1992 case 1: /*mov from cr*/
1995 vcpu_load_rsp_rip(vcpu);
1996 vcpu->arch.regs[reg] = vcpu->arch.cr3;
1997 vcpu_put_rsp_rip(vcpu);
1998 skip_emulated_instruction(vcpu);
2001 vcpu_load_rsp_rip(vcpu);
2002 vcpu->arch.regs[reg] = get_cr8(vcpu);
2003 vcpu_put_rsp_rip(vcpu);
2004 skip_emulated_instruction(vcpu);
2009 lmsw(vcpu, (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f);
2011 skip_emulated_instruction(vcpu);
2016 kvm_run->exit_reason = 0;
2017 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
2018 (int)(exit_qualification >> 4) & 3, cr);
2022 static int handle_dr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2024 unsigned long exit_qualification;
2029 * FIXME: this code assumes the host is debugging the guest.
2030 * need to deal with guest debugging itself too.
2032 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
2033 dr = exit_qualification & 7;
2034 reg = (exit_qualification >> 8) & 15;
2035 vcpu_load_rsp_rip(vcpu);
2036 if (exit_qualification & 16) {
2048 vcpu->arch.regs[reg] = val;
2052 vcpu_put_rsp_rip(vcpu);
2053 skip_emulated_instruction(vcpu);
2057 static int handle_cpuid(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2059 kvm_emulate_cpuid(vcpu);
2063 static int handle_rdmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2065 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2068 if (vmx_get_msr(vcpu, ecx, &data)) {
2069 kvm_inject_gp(vcpu, 0);
2073 /* FIXME: handling of bits 32:63 of rax, rdx */
2074 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
2075 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
2076 skip_emulated_instruction(vcpu);
2080 static int handle_wrmsr(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2082 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
2083 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
2084 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
2086 if (vmx_set_msr(vcpu, ecx, data) != 0) {
2087 kvm_inject_gp(vcpu, 0);
2091 skip_emulated_instruction(vcpu);
2095 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu,
2096 struct kvm_run *kvm_run)
2101 static int handle_interrupt_window(struct kvm_vcpu *vcpu,
2102 struct kvm_run *kvm_run)
2104 u32 cpu_based_vm_exec_control;
2106 /* clear pending irq */
2107 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2108 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
2109 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2111 * If the user space waits to inject interrupts, exit as soon as
2114 if (kvm_run->request_interrupt_window &&
2115 !vcpu->arch.irq_summary) {
2116 kvm_run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
2117 ++vcpu->stat.irq_window_exits;
2123 static int handle_halt(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2125 skip_emulated_instruction(vcpu);
2126 return kvm_emulate_halt(vcpu);
2129 static int handle_vmcall(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2131 skip_emulated_instruction(vcpu);
2132 kvm_emulate_hypercall(vcpu);
2136 static int handle_wbinvd(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2138 skip_emulated_instruction(vcpu);
2139 /* TODO: Add support for VT-d/pass-through device */
2143 static int handle_apic_access(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2145 u64 exit_qualification;
2146 enum emulation_result er;
2147 unsigned long offset;
2149 exit_qualification = vmcs_read64(EXIT_QUALIFICATION);
2150 offset = exit_qualification & 0xffful;
2152 er = emulate_instruction(vcpu, kvm_run, 0, 0, 0);
2154 if (er != EMULATE_DONE) {
2156 "Fail to handle apic access vmexit! Offset is 0x%lx\n",
2164 * The exit handlers return 1 if the exit was handled fully and guest execution
2165 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
2166 * to be done to userspace and return 0.
2168 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu,
2169 struct kvm_run *kvm_run) = {
2170 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
2171 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
2172 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
2173 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
2174 [EXIT_REASON_CR_ACCESS] = handle_cr,
2175 [EXIT_REASON_DR_ACCESS] = handle_dr,
2176 [EXIT_REASON_CPUID] = handle_cpuid,
2177 [EXIT_REASON_MSR_READ] = handle_rdmsr,
2178 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
2179 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
2180 [EXIT_REASON_HLT] = handle_halt,
2181 [EXIT_REASON_VMCALL] = handle_vmcall,
2182 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
2183 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
2184 [EXIT_REASON_WBINVD] = handle_wbinvd,
2187 static const int kvm_vmx_max_exit_handlers =
2188 ARRAY_SIZE(kvm_vmx_exit_handlers);
2191 * The guest has exited. See if we can fix it or if we need userspace
2194 static int kvm_handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
2196 u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
2197 struct vcpu_vmx *vmx = to_vmx(vcpu);
2198 u32 vectoring_info = vmx->idt_vectoring_info;
2200 if (unlikely(vmx->fail)) {
2201 kvm_run->exit_reason = KVM_EXIT_FAIL_ENTRY;
2202 kvm_run->fail_entry.hardware_entry_failure_reason
2203 = vmcs_read32(VM_INSTRUCTION_ERROR);
2207 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
2208 exit_reason != EXIT_REASON_EXCEPTION_NMI)
2209 printk(KERN_WARNING "%s: unexpected, valid vectoring info and "
2210 "exit reason is 0x%x\n", __FUNCTION__, exit_reason);
2211 if (exit_reason < kvm_vmx_max_exit_handlers
2212 && kvm_vmx_exit_handlers[exit_reason])
2213 return kvm_vmx_exit_handlers[exit_reason](vcpu, kvm_run);
2215 kvm_run->exit_reason = KVM_EXIT_UNKNOWN;
2216 kvm_run->hw.hardware_exit_reason = exit_reason;
2221 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
2225 static void update_tpr_threshold(struct kvm_vcpu *vcpu)
2229 if (!vm_need_tpr_shadow(vcpu->kvm))
2232 if (!kvm_lapic_enabled(vcpu) ||
2233 ((max_irr = kvm_lapic_find_highest_irr(vcpu)) == -1)) {
2234 vmcs_write32(TPR_THRESHOLD, 0);
2238 tpr = (kvm_lapic_get_cr8(vcpu) & 0x0f) << 4;
2239 vmcs_write32(TPR_THRESHOLD, (max_irr > tpr) ? tpr >> 4 : max_irr >> 4);
2242 static void enable_irq_window(struct kvm_vcpu *vcpu)
2244 u32 cpu_based_vm_exec_control;
2246 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
2247 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
2248 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
2251 static void vmx_intr_assist(struct kvm_vcpu *vcpu)
2253 struct vcpu_vmx *vmx = to_vmx(vcpu);
2254 u32 idtv_info_field, intr_info_field;
2255 int has_ext_irq, interrupt_window_open;
2258 update_tpr_threshold(vcpu);
2260 has_ext_irq = kvm_cpu_has_interrupt(vcpu);
2261 intr_info_field = vmcs_read32(VM_ENTRY_INTR_INFO_FIELD);
2262 idtv_info_field = vmx->idt_vectoring_info;
2263 if (intr_info_field & INTR_INFO_VALID_MASK) {
2264 if (idtv_info_field & INTR_INFO_VALID_MASK) {
2265 /* TODO: fault when IDT_Vectoring */
2266 if (printk_ratelimit())
2267 printk(KERN_ERR "Fault when IDT_Vectoring\n");
2270 enable_irq_window(vcpu);
2273 if (unlikely(idtv_info_field & INTR_INFO_VALID_MASK)) {
2274 if ((idtv_info_field & VECTORING_INFO_TYPE_MASK)
2275 == INTR_TYPE_EXT_INTR
2276 && vcpu->arch.rmode.active) {
2277 u8 vect = idtv_info_field & VECTORING_INFO_VECTOR_MASK;
2279 vmx_inject_irq(vcpu, vect);
2280 if (unlikely(has_ext_irq))
2281 enable_irq_window(vcpu);
2285 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, idtv_info_field);
2286 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2287 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
2289 if (unlikely(idtv_info_field & INTR_INFO_DELIEVER_CODE_MASK))
2290 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
2291 vmcs_read32(IDT_VECTORING_ERROR_CODE));
2292 if (unlikely(has_ext_irq))
2293 enable_irq_window(vcpu);
2298 interrupt_window_open =
2299 ((vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
2300 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0);
2301 if (interrupt_window_open) {
2302 vector = kvm_cpu_get_interrupt(vcpu);
2303 vmx_inject_irq(vcpu, vector);
2304 kvm_timer_intr_post(vcpu, vector);
2306 enable_irq_window(vcpu);
2310 * Failure to inject an interrupt should give us the information
2311 * in IDT_VECTORING_INFO_FIELD. However, if the failure occurs
2312 * when fetching the interrupt redirection bitmap in the real-mode
2313 * tss, this doesn't happen. So we do it ourselves.
2315 static void fixup_rmode_irq(struct vcpu_vmx *vmx)
2317 vmx->rmode.irq.pending = 0;
2318 if (vmcs_readl(GUEST_RIP) + 1 != vmx->rmode.irq.rip)
2320 vmcs_writel(GUEST_RIP, vmx->rmode.irq.rip);
2321 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
2322 vmx->idt_vectoring_info &= ~VECTORING_INFO_TYPE_MASK;
2323 vmx->idt_vectoring_info |= INTR_TYPE_EXT_INTR;
2326 vmx->idt_vectoring_info =
2327 VECTORING_INFO_VALID_MASK
2328 | INTR_TYPE_EXT_INTR
2329 | vmx->rmode.irq.vector;
2332 static void vmx_vcpu_run(struct kvm_vcpu *vcpu, struct kvm_run *kvm_run)
2334 struct vcpu_vmx *vmx = to_vmx(vcpu);
2338 * Loading guest fpu may have cleared host cr0.ts
2340 vmcs_writel(HOST_CR0, read_cr0());
2343 /* Store host registers */
2344 #ifdef CONFIG_X86_64
2345 "push %%rdx; push %%rbp;"
2348 "push %%edx; push %%ebp;"
2351 ASM_VMX_VMWRITE_RSP_RDX "\n\t"
2352 /* Check if vmlaunch of vmresume is needed */
2353 "cmpl $0, %c[launched](%0) \n\t"
2354 /* Load guest registers. Don't clobber flags. */
2355 #ifdef CONFIG_X86_64
2356 "mov %c[cr2](%0), %%rax \n\t"
2357 "mov %%rax, %%cr2 \n\t"
2358 "mov %c[rax](%0), %%rax \n\t"
2359 "mov %c[rbx](%0), %%rbx \n\t"
2360 "mov %c[rdx](%0), %%rdx \n\t"
2361 "mov %c[rsi](%0), %%rsi \n\t"
2362 "mov %c[rdi](%0), %%rdi \n\t"
2363 "mov %c[rbp](%0), %%rbp \n\t"
2364 "mov %c[r8](%0), %%r8 \n\t"
2365 "mov %c[r9](%0), %%r9 \n\t"
2366 "mov %c[r10](%0), %%r10 \n\t"
2367 "mov %c[r11](%0), %%r11 \n\t"
2368 "mov %c[r12](%0), %%r12 \n\t"
2369 "mov %c[r13](%0), %%r13 \n\t"
2370 "mov %c[r14](%0), %%r14 \n\t"
2371 "mov %c[r15](%0), %%r15 \n\t"
2372 "mov %c[rcx](%0), %%rcx \n\t" /* kills %0 (rcx) */
2374 "mov %c[cr2](%0), %%eax \n\t"
2375 "mov %%eax, %%cr2 \n\t"
2376 "mov %c[rax](%0), %%eax \n\t"
2377 "mov %c[rbx](%0), %%ebx \n\t"
2378 "mov %c[rdx](%0), %%edx \n\t"
2379 "mov %c[rsi](%0), %%esi \n\t"
2380 "mov %c[rdi](%0), %%edi \n\t"
2381 "mov %c[rbp](%0), %%ebp \n\t"
2382 "mov %c[rcx](%0), %%ecx \n\t" /* kills %0 (ecx) */
2384 /* Enter guest mode */
2385 "jne .Llaunched \n\t"
2386 ASM_VMX_VMLAUNCH "\n\t"
2387 "jmp .Lkvm_vmx_return \n\t"
2388 ".Llaunched: " ASM_VMX_VMRESUME "\n\t"
2389 ".Lkvm_vmx_return: "
2390 /* Save guest registers, load host registers, keep flags */
2391 #ifdef CONFIG_X86_64
2392 "xchg %0, (%%rsp) \n\t"
2393 "mov %%rax, %c[rax](%0) \n\t"
2394 "mov %%rbx, %c[rbx](%0) \n\t"
2395 "pushq (%%rsp); popq %c[rcx](%0) \n\t"
2396 "mov %%rdx, %c[rdx](%0) \n\t"
2397 "mov %%rsi, %c[rsi](%0) \n\t"
2398 "mov %%rdi, %c[rdi](%0) \n\t"
2399 "mov %%rbp, %c[rbp](%0) \n\t"
2400 "mov %%r8, %c[r8](%0) \n\t"
2401 "mov %%r9, %c[r9](%0) \n\t"
2402 "mov %%r10, %c[r10](%0) \n\t"
2403 "mov %%r11, %c[r11](%0) \n\t"
2404 "mov %%r12, %c[r12](%0) \n\t"
2405 "mov %%r13, %c[r13](%0) \n\t"
2406 "mov %%r14, %c[r14](%0) \n\t"
2407 "mov %%r15, %c[r15](%0) \n\t"
2408 "mov %%cr2, %%rax \n\t"
2409 "mov %%rax, %c[cr2](%0) \n\t"
2411 "pop %%rbp; pop %%rbp; pop %%rdx \n\t"
2413 "xchg %0, (%%esp) \n\t"
2414 "mov %%eax, %c[rax](%0) \n\t"
2415 "mov %%ebx, %c[rbx](%0) \n\t"
2416 "pushl (%%esp); popl %c[rcx](%0) \n\t"
2417 "mov %%edx, %c[rdx](%0) \n\t"
2418 "mov %%esi, %c[rsi](%0) \n\t"
2419 "mov %%edi, %c[rdi](%0) \n\t"
2420 "mov %%ebp, %c[rbp](%0) \n\t"
2421 "mov %%cr2, %%eax \n\t"
2422 "mov %%eax, %c[cr2](%0) \n\t"
2424 "pop %%ebp; pop %%ebp; pop %%edx \n\t"
2426 "setbe %c[fail](%0) \n\t"
2427 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
2428 [launched]"i"(offsetof(struct vcpu_vmx, launched)),
2429 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
2430 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
2431 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
2432 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
2433 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
2434 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
2435 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
2436 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
2437 #ifdef CONFIG_X86_64
2438 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
2439 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
2440 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
2441 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
2442 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
2443 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
2444 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
2445 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
2447 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2))
2449 #ifdef CONFIG_X86_64
2450 , "rbx", "rdi", "rsi"
2451 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
2453 , "ebx", "edi", "rsi"
2457 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
2458 if (vmx->rmode.irq.pending)
2459 fixup_rmode_irq(vmx);
2461 vcpu->arch.interrupt_window_open =
2462 (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & 3) == 0;
2464 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
2467 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
2469 /* We need to handle NMIs before interrupts are enabled */
2470 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == 0x200) /* nmi */
2474 static void vmx_free_vmcs(struct kvm_vcpu *vcpu)
2476 struct vcpu_vmx *vmx = to_vmx(vcpu);
2479 on_each_cpu(__vcpu_clear, vmx, 0, 1);
2480 free_vmcs(vmx->vmcs);
2485 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
2487 struct vcpu_vmx *vmx = to_vmx(vcpu);
2489 vmx_free_vmcs(vcpu);
2490 kfree(vmx->host_msrs);
2491 kfree(vmx->guest_msrs);
2492 kvm_vcpu_uninit(vcpu);
2493 kmem_cache_free(kvm_vcpu_cache, vmx);
2496 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
2499 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
2503 return ERR_PTR(-ENOMEM);
2505 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
2509 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2510 if (!vmx->guest_msrs) {
2515 vmx->host_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
2516 if (!vmx->host_msrs)
2517 goto free_guest_msrs;
2519 vmx->vmcs = alloc_vmcs();
2523 vmcs_clear(vmx->vmcs);
2526 vmx_vcpu_load(&vmx->vcpu, cpu);
2527 err = vmx_vcpu_setup(vmx);
2528 vmx_vcpu_put(&vmx->vcpu);
2536 free_vmcs(vmx->vmcs);
2538 kfree(vmx->host_msrs);
2540 kfree(vmx->guest_msrs);
2542 kvm_vcpu_uninit(&vmx->vcpu);
2544 kmem_cache_free(kvm_vcpu_cache, vmx);
2545 return ERR_PTR(err);
2548 static void __init vmx_check_processor_compat(void *rtn)
2550 struct vmcs_config vmcs_conf;
2553 if (setup_vmcs_config(&vmcs_conf) < 0)
2555 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
2556 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
2557 smp_processor_id());
2562 static struct kvm_x86_ops vmx_x86_ops = {
2563 .cpu_has_kvm_support = cpu_has_kvm_support,
2564 .disabled_by_bios = vmx_disabled_by_bios,
2565 .hardware_setup = hardware_setup,
2566 .hardware_unsetup = hardware_unsetup,
2567 .check_processor_compatibility = vmx_check_processor_compat,
2568 .hardware_enable = hardware_enable,
2569 .hardware_disable = hardware_disable,
2571 .vcpu_create = vmx_create_vcpu,
2572 .vcpu_free = vmx_free_vcpu,
2573 .vcpu_reset = vmx_vcpu_reset,
2575 .prepare_guest_switch = vmx_save_host_state,
2576 .vcpu_load = vmx_vcpu_load,
2577 .vcpu_put = vmx_vcpu_put,
2578 .vcpu_decache = vmx_vcpu_decache,
2580 .set_guest_debug = set_guest_debug,
2581 .guest_debug_pre = kvm_guest_debug_pre,
2582 .get_msr = vmx_get_msr,
2583 .set_msr = vmx_set_msr,
2584 .get_segment_base = vmx_get_segment_base,
2585 .get_segment = vmx_get_segment,
2586 .set_segment = vmx_set_segment,
2587 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
2588 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
2589 .set_cr0 = vmx_set_cr0,
2590 .set_cr3 = vmx_set_cr3,
2591 .set_cr4 = vmx_set_cr4,
2592 #ifdef CONFIG_X86_64
2593 .set_efer = vmx_set_efer,
2595 .get_idt = vmx_get_idt,
2596 .set_idt = vmx_set_idt,
2597 .get_gdt = vmx_get_gdt,
2598 .set_gdt = vmx_set_gdt,
2599 .cache_regs = vcpu_load_rsp_rip,
2600 .decache_regs = vcpu_put_rsp_rip,
2601 .get_rflags = vmx_get_rflags,
2602 .set_rflags = vmx_set_rflags,
2604 .tlb_flush = vmx_flush_tlb,
2606 .run = vmx_vcpu_run,
2607 .handle_exit = kvm_handle_exit,
2608 .skip_emulated_instruction = skip_emulated_instruction,
2609 .patch_hypercall = vmx_patch_hypercall,
2610 .get_irq = vmx_get_irq,
2611 .set_irq = vmx_inject_irq,
2612 .queue_exception = vmx_queue_exception,
2613 .exception_injected = vmx_exception_injected,
2614 .inject_pending_irq = vmx_intr_assist,
2615 .inject_pending_vectors = do_interrupt_requests,
2617 .set_tss_addr = vmx_set_tss_addr,
2620 static int __init vmx_init(void)
2625 vmx_io_bitmap_a = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2626 if (!vmx_io_bitmap_a)
2629 vmx_io_bitmap_b = alloc_page(GFP_KERNEL | __GFP_HIGHMEM);
2630 if (!vmx_io_bitmap_b) {
2636 * Allow direct access to the PC debug port (it is often used for I/O
2637 * delays, but the vmexits simply slow things down).
2639 iova = kmap(vmx_io_bitmap_a);
2640 memset(iova, 0xff, PAGE_SIZE);
2641 clear_bit(0x80, iova);
2642 kunmap(vmx_io_bitmap_a);
2644 iova = kmap(vmx_io_bitmap_b);
2645 memset(iova, 0xff, PAGE_SIZE);
2646 kunmap(vmx_io_bitmap_b);
2648 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx), THIS_MODULE);
2652 if (bypass_guest_pf)
2653 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
2658 __free_page(vmx_io_bitmap_b);
2660 __free_page(vmx_io_bitmap_a);
2664 static void __exit vmx_exit(void)
2666 __free_page(vmx_io_bitmap_b);
2667 __free_page(vmx_io_bitmap_a);
2672 module_init(vmx_init)
2673 module_exit(vmx_exit)