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.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
22 #include <linux/kvm_host.h>
23 #include <linux/module.h>
24 #include <linux/kernel.h>
26 #include <linux/highmem.h>
27 #include <linux/sched.h>
28 #include <linux/moduleparam.h>
29 #include <linux/ftrace_event.h>
30 #include <linux/slab.h>
31 #include <linux/tboot.h>
32 #include "kvm_cache_regs.h"
38 #include <asm/virtext.h>
45 #define __ex(x) __kvm_handle_fault_on_reboot(x)
46 #define __ex_clear(x, reg) \
47 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
49 MODULE_AUTHOR("Qumranet");
50 MODULE_LICENSE("GPL");
52 static int __read_mostly bypass_guest_pf = 1;
53 module_param(bypass_guest_pf, bool, S_IRUGO);
55 static int __read_mostly enable_vpid = 1;
56 module_param_named(vpid, enable_vpid, bool, 0444);
58 static int __read_mostly flexpriority_enabled = 1;
59 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
61 static int __read_mostly enable_ept = 1;
62 module_param_named(ept, enable_ept, bool, S_IRUGO);
64 static int __read_mostly enable_unrestricted_guest = 1;
65 module_param_named(unrestricted_guest,
66 enable_unrestricted_guest, bool, S_IRUGO);
68 static int __read_mostly emulate_invalid_guest_state = 0;
69 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
71 static int __read_mostly vmm_exclusive = 1;
72 module_param(vmm_exclusive, bool, S_IRUGO);
74 static int __read_mostly yield_on_hlt = 1;
75 module_param(yield_on_hlt, bool, S_IRUGO);
78 * If nested=1, nested virtualization is supported, i.e., guests may use
79 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
80 * use VMX instructions.
82 static int __read_mostly nested = 0;
83 module_param(nested, bool, S_IRUGO);
85 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST \
86 (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
87 #define KVM_GUEST_CR0_MASK \
88 (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
89 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST \
90 (X86_CR0_WP | X86_CR0_NE)
91 #define KVM_VM_CR0_ALWAYS_ON \
92 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
93 #define KVM_CR4_GUEST_OWNED_BITS \
94 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
97 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
98 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
100 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
103 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
104 * ple_gap: upper bound on the amount of time between two successive
105 * executions of PAUSE in a loop. Also indicate if ple enabled.
106 * According to test, this time is usually smaller than 128 cycles.
107 * ple_window: upper bound on the amount of time a guest is allowed to execute
108 * in a PAUSE loop. Tests indicate that most spinlocks are held for
109 * less than 2^12 cycles
110 * Time is measured based on a counter that runs at the same rate as the TSC,
111 * refer SDM volume 3b section 21.6.13 & 22.1.3.
113 #define KVM_VMX_DEFAULT_PLE_GAP 128
114 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
115 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
116 module_param(ple_gap, int, S_IRUGO);
118 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
119 module_param(ple_window, int, S_IRUGO);
121 #define NR_AUTOLOAD_MSRS 1
122 #define VMCS02_POOL_SIZE 1
131 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
132 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
133 * loaded on this CPU (so we can clear them if the CPU goes down).
139 struct list_head loaded_vmcss_on_cpu_link;
142 struct shared_msr_entry {
149 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
150 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
151 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
152 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
153 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
154 * More than one of these structures may exist, if L1 runs multiple L2 guests.
155 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
156 * underlying hardware which will be used to run L2.
157 * This structure is packed to ensure that its layout is identical across
158 * machines (necessary for live migration).
159 * If there are changes in this struct, VMCS12_REVISION must be changed.
161 typedef u64 natural_width;
162 struct __packed vmcs12 {
163 /* According to the Intel spec, a VMCS region must start with the
164 * following two fields. Then follow implementation-specific data.
169 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
170 u32 padding[7]; /* room for future expansion */
175 u64 vm_exit_msr_store_addr;
176 u64 vm_exit_msr_load_addr;
177 u64 vm_entry_msr_load_addr;
179 u64 virtual_apic_page_addr;
180 u64 apic_access_addr;
182 u64 guest_physical_address;
183 u64 vmcs_link_pointer;
184 u64 guest_ia32_debugctl;
187 u64 guest_ia32_perf_global_ctrl;
194 u64 host_ia32_perf_global_ctrl;
195 u64 padding64[8]; /* room for future expansion */
197 * To allow migration of L1 (complete with its L2 guests) between
198 * machines of different natural widths (32 or 64 bit), we cannot have
199 * unsigned long fields with no explict size. We use u64 (aliased
200 * natural_width) instead. Luckily, x86 is little-endian.
202 natural_width cr0_guest_host_mask;
203 natural_width cr4_guest_host_mask;
204 natural_width cr0_read_shadow;
205 natural_width cr4_read_shadow;
206 natural_width cr3_target_value0;
207 natural_width cr3_target_value1;
208 natural_width cr3_target_value2;
209 natural_width cr3_target_value3;
210 natural_width exit_qualification;
211 natural_width guest_linear_address;
212 natural_width guest_cr0;
213 natural_width guest_cr3;
214 natural_width guest_cr4;
215 natural_width guest_es_base;
216 natural_width guest_cs_base;
217 natural_width guest_ss_base;
218 natural_width guest_ds_base;
219 natural_width guest_fs_base;
220 natural_width guest_gs_base;
221 natural_width guest_ldtr_base;
222 natural_width guest_tr_base;
223 natural_width guest_gdtr_base;
224 natural_width guest_idtr_base;
225 natural_width guest_dr7;
226 natural_width guest_rsp;
227 natural_width guest_rip;
228 natural_width guest_rflags;
229 natural_width guest_pending_dbg_exceptions;
230 natural_width guest_sysenter_esp;
231 natural_width guest_sysenter_eip;
232 natural_width host_cr0;
233 natural_width host_cr3;
234 natural_width host_cr4;
235 natural_width host_fs_base;
236 natural_width host_gs_base;
237 natural_width host_tr_base;
238 natural_width host_gdtr_base;
239 natural_width host_idtr_base;
240 natural_width host_ia32_sysenter_esp;
241 natural_width host_ia32_sysenter_eip;
242 natural_width host_rsp;
243 natural_width host_rip;
244 natural_width paddingl[8]; /* room for future expansion */
245 u32 pin_based_vm_exec_control;
246 u32 cpu_based_vm_exec_control;
247 u32 exception_bitmap;
248 u32 page_fault_error_code_mask;
249 u32 page_fault_error_code_match;
250 u32 cr3_target_count;
251 u32 vm_exit_controls;
252 u32 vm_exit_msr_store_count;
253 u32 vm_exit_msr_load_count;
254 u32 vm_entry_controls;
255 u32 vm_entry_msr_load_count;
256 u32 vm_entry_intr_info_field;
257 u32 vm_entry_exception_error_code;
258 u32 vm_entry_instruction_len;
260 u32 secondary_vm_exec_control;
261 u32 vm_instruction_error;
263 u32 vm_exit_intr_info;
264 u32 vm_exit_intr_error_code;
265 u32 idt_vectoring_info_field;
266 u32 idt_vectoring_error_code;
267 u32 vm_exit_instruction_len;
268 u32 vmx_instruction_info;
275 u32 guest_ldtr_limit;
277 u32 guest_gdtr_limit;
278 u32 guest_idtr_limit;
279 u32 guest_es_ar_bytes;
280 u32 guest_cs_ar_bytes;
281 u32 guest_ss_ar_bytes;
282 u32 guest_ds_ar_bytes;
283 u32 guest_fs_ar_bytes;
284 u32 guest_gs_ar_bytes;
285 u32 guest_ldtr_ar_bytes;
286 u32 guest_tr_ar_bytes;
287 u32 guest_interruptibility_info;
288 u32 guest_activity_state;
289 u32 guest_sysenter_cs;
290 u32 host_ia32_sysenter_cs;
291 u32 padding32[8]; /* room for future expansion */
292 u16 virtual_processor_id;
293 u16 guest_es_selector;
294 u16 guest_cs_selector;
295 u16 guest_ss_selector;
296 u16 guest_ds_selector;
297 u16 guest_fs_selector;
298 u16 guest_gs_selector;
299 u16 guest_ldtr_selector;
300 u16 guest_tr_selector;
301 u16 host_es_selector;
302 u16 host_cs_selector;
303 u16 host_ss_selector;
304 u16 host_ds_selector;
305 u16 host_fs_selector;
306 u16 host_gs_selector;
307 u16 host_tr_selector;
311 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
312 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
313 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
315 #define VMCS12_REVISION 0x11e57ed0
318 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
319 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
320 * current implementation, 4K are reserved to avoid future complications.
322 #define VMCS12_SIZE 0x1000
324 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
326 struct list_head list;
328 struct loaded_vmcs vmcs02;
332 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
333 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
336 /* Has the level1 guest done vmxon? */
339 /* The guest-physical address of the current VMCS L1 keeps for L2 */
341 /* The host-usable pointer to the above */
342 struct page *current_vmcs12_page;
343 struct vmcs12 *current_vmcs12;
345 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
346 struct list_head vmcs02_pool;
351 struct kvm_vcpu vcpu;
352 unsigned long host_rsp;
355 bool nmi_known_unmasked;
357 u32 idt_vectoring_info;
359 struct shared_msr_entry *guest_msrs;
363 u64 msr_host_kernel_gs_base;
364 u64 msr_guest_kernel_gs_base;
367 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
368 * non-nested (L1) guest, it always points to vmcs01. For a nested
369 * guest (L2), it points to a different VMCS.
371 struct loaded_vmcs vmcs01;
372 struct loaded_vmcs *loaded_vmcs;
373 bool __launched; /* temporary, used in vmx_vcpu_run */
374 struct msr_autoload {
376 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
377 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
381 u16 fs_sel, gs_sel, ldt_sel;
382 int gs_ldt_reload_needed;
383 int fs_reload_needed;
388 struct kvm_save_segment {
393 } tr, es, ds, fs, gs;
396 u32 bitmask; /* 4 bits per segment (1 bit per field) */
397 struct kvm_save_segment seg[8];
400 bool emulation_required;
402 /* Support for vnmi-less CPUs */
403 int soft_vnmi_blocked;
405 s64 vnmi_blocked_time;
410 /* Support for a guest hypervisor (nested VMX) */
411 struct nested_vmx nested;
414 enum segment_cache_field {
423 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
425 return container_of(vcpu, struct vcpu_vmx, vcpu);
428 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
429 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
430 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
431 [number##_HIGH] = VMCS12_OFFSET(name)+4
433 static unsigned short vmcs_field_to_offset_table[] = {
434 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
435 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
436 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
437 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
438 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
439 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
440 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
441 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
442 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
443 FIELD(HOST_ES_SELECTOR, host_es_selector),
444 FIELD(HOST_CS_SELECTOR, host_cs_selector),
445 FIELD(HOST_SS_SELECTOR, host_ss_selector),
446 FIELD(HOST_DS_SELECTOR, host_ds_selector),
447 FIELD(HOST_FS_SELECTOR, host_fs_selector),
448 FIELD(HOST_GS_SELECTOR, host_gs_selector),
449 FIELD(HOST_TR_SELECTOR, host_tr_selector),
450 FIELD64(IO_BITMAP_A, io_bitmap_a),
451 FIELD64(IO_BITMAP_B, io_bitmap_b),
452 FIELD64(MSR_BITMAP, msr_bitmap),
453 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
454 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
455 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
456 FIELD64(TSC_OFFSET, tsc_offset),
457 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
458 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
459 FIELD64(EPT_POINTER, ept_pointer),
460 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
461 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
462 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
463 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
464 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
465 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
466 FIELD64(GUEST_PDPTR0, guest_pdptr0),
467 FIELD64(GUEST_PDPTR1, guest_pdptr1),
468 FIELD64(GUEST_PDPTR2, guest_pdptr2),
469 FIELD64(GUEST_PDPTR3, guest_pdptr3),
470 FIELD64(HOST_IA32_PAT, host_ia32_pat),
471 FIELD64(HOST_IA32_EFER, host_ia32_efer),
472 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
473 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
474 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
475 FIELD(EXCEPTION_BITMAP, exception_bitmap),
476 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
477 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
478 FIELD(CR3_TARGET_COUNT, cr3_target_count),
479 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
480 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
481 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
482 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
483 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
484 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
485 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
486 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
487 FIELD(TPR_THRESHOLD, tpr_threshold),
488 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
489 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
490 FIELD(VM_EXIT_REASON, vm_exit_reason),
491 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
492 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
493 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
494 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
495 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
496 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
497 FIELD(GUEST_ES_LIMIT, guest_es_limit),
498 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
499 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
500 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
501 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
502 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
503 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
504 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
505 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
506 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
507 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
508 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
509 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
510 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
511 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
512 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
513 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
514 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
515 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
516 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
517 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
518 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
519 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
520 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
521 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
522 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
523 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
524 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
525 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
526 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
527 FIELD(EXIT_QUALIFICATION, exit_qualification),
528 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
529 FIELD(GUEST_CR0, guest_cr0),
530 FIELD(GUEST_CR3, guest_cr3),
531 FIELD(GUEST_CR4, guest_cr4),
532 FIELD(GUEST_ES_BASE, guest_es_base),
533 FIELD(GUEST_CS_BASE, guest_cs_base),
534 FIELD(GUEST_SS_BASE, guest_ss_base),
535 FIELD(GUEST_DS_BASE, guest_ds_base),
536 FIELD(GUEST_FS_BASE, guest_fs_base),
537 FIELD(GUEST_GS_BASE, guest_gs_base),
538 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
539 FIELD(GUEST_TR_BASE, guest_tr_base),
540 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
541 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
542 FIELD(GUEST_DR7, guest_dr7),
543 FIELD(GUEST_RSP, guest_rsp),
544 FIELD(GUEST_RIP, guest_rip),
545 FIELD(GUEST_RFLAGS, guest_rflags),
546 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
547 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
548 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
549 FIELD(HOST_CR0, host_cr0),
550 FIELD(HOST_CR3, host_cr3),
551 FIELD(HOST_CR4, host_cr4),
552 FIELD(HOST_FS_BASE, host_fs_base),
553 FIELD(HOST_GS_BASE, host_gs_base),
554 FIELD(HOST_TR_BASE, host_tr_base),
555 FIELD(HOST_GDTR_BASE, host_gdtr_base),
556 FIELD(HOST_IDTR_BASE, host_idtr_base),
557 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
558 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
559 FIELD(HOST_RSP, host_rsp),
560 FIELD(HOST_RIP, host_rip),
562 static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
564 static inline short vmcs_field_to_offset(unsigned long field)
566 if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
568 return vmcs_field_to_offset_table[field];
571 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
573 return to_vmx(vcpu)->nested.current_vmcs12;
576 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
578 struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
579 if (is_error_page(page)) {
580 kvm_release_page_clean(page);
586 static void nested_release_page(struct page *page)
588 kvm_release_page_dirty(page);
591 static void nested_release_page_clean(struct page *page)
593 kvm_release_page_clean(page);
596 static u64 construct_eptp(unsigned long root_hpa);
597 static void kvm_cpu_vmxon(u64 addr);
598 static void kvm_cpu_vmxoff(void);
599 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
600 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
602 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
603 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
605 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
606 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
608 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
609 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
611 static unsigned long *vmx_io_bitmap_a;
612 static unsigned long *vmx_io_bitmap_b;
613 static unsigned long *vmx_msr_bitmap_legacy;
614 static unsigned long *vmx_msr_bitmap_longmode;
616 static bool cpu_has_load_ia32_efer;
618 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
619 static DEFINE_SPINLOCK(vmx_vpid_lock);
621 static struct vmcs_config {
625 u32 pin_based_exec_ctrl;
626 u32 cpu_based_exec_ctrl;
627 u32 cpu_based_2nd_exec_ctrl;
632 static struct vmx_capability {
637 #define VMX_SEGMENT_FIELD(seg) \
638 [VCPU_SREG_##seg] = { \
639 .selector = GUEST_##seg##_SELECTOR, \
640 .base = GUEST_##seg##_BASE, \
641 .limit = GUEST_##seg##_LIMIT, \
642 .ar_bytes = GUEST_##seg##_AR_BYTES, \
645 static struct kvm_vmx_segment_field {
650 } kvm_vmx_segment_fields[] = {
651 VMX_SEGMENT_FIELD(CS),
652 VMX_SEGMENT_FIELD(DS),
653 VMX_SEGMENT_FIELD(ES),
654 VMX_SEGMENT_FIELD(FS),
655 VMX_SEGMENT_FIELD(GS),
656 VMX_SEGMENT_FIELD(SS),
657 VMX_SEGMENT_FIELD(TR),
658 VMX_SEGMENT_FIELD(LDTR),
661 static u64 host_efer;
663 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
666 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
667 * away by decrementing the array size.
669 static const u32 vmx_msr_index[] = {
671 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
673 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
675 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
677 static inline bool is_page_fault(u32 intr_info)
679 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
680 INTR_INFO_VALID_MASK)) ==
681 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
684 static inline bool is_no_device(u32 intr_info)
686 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
687 INTR_INFO_VALID_MASK)) ==
688 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
691 static inline bool is_invalid_opcode(u32 intr_info)
693 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
694 INTR_INFO_VALID_MASK)) ==
695 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
698 static inline bool is_external_interrupt(u32 intr_info)
700 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
701 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
704 static inline bool is_machine_check(u32 intr_info)
706 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
707 INTR_INFO_VALID_MASK)) ==
708 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
711 static inline bool cpu_has_vmx_msr_bitmap(void)
713 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
716 static inline bool cpu_has_vmx_tpr_shadow(void)
718 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
721 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
723 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
726 static inline bool cpu_has_secondary_exec_ctrls(void)
728 return vmcs_config.cpu_based_exec_ctrl &
729 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
732 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
734 return vmcs_config.cpu_based_2nd_exec_ctrl &
735 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
738 static inline bool cpu_has_vmx_flexpriority(void)
740 return cpu_has_vmx_tpr_shadow() &&
741 cpu_has_vmx_virtualize_apic_accesses();
744 static inline bool cpu_has_vmx_ept_execute_only(void)
746 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
749 static inline bool cpu_has_vmx_eptp_uncacheable(void)
751 return vmx_capability.ept & VMX_EPTP_UC_BIT;
754 static inline bool cpu_has_vmx_eptp_writeback(void)
756 return vmx_capability.ept & VMX_EPTP_WB_BIT;
759 static inline bool cpu_has_vmx_ept_2m_page(void)
761 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
764 static inline bool cpu_has_vmx_ept_1g_page(void)
766 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
769 static inline bool cpu_has_vmx_ept_4levels(void)
771 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
774 static inline bool cpu_has_vmx_invept_individual_addr(void)
776 return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
779 static inline bool cpu_has_vmx_invept_context(void)
781 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
784 static inline bool cpu_has_vmx_invept_global(void)
786 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
789 static inline bool cpu_has_vmx_invvpid_single(void)
791 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
794 static inline bool cpu_has_vmx_invvpid_global(void)
796 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
799 static inline bool cpu_has_vmx_ept(void)
801 return vmcs_config.cpu_based_2nd_exec_ctrl &
802 SECONDARY_EXEC_ENABLE_EPT;
805 static inline bool cpu_has_vmx_unrestricted_guest(void)
807 return vmcs_config.cpu_based_2nd_exec_ctrl &
808 SECONDARY_EXEC_UNRESTRICTED_GUEST;
811 static inline bool cpu_has_vmx_ple(void)
813 return vmcs_config.cpu_based_2nd_exec_ctrl &
814 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
817 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
819 return flexpriority_enabled && irqchip_in_kernel(kvm);
822 static inline bool cpu_has_vmx_vpid(void)
824 return vmcs_config.cpu_based_2nd_exec_ctrl &
825 SECONDARY_EXEC_ENABLE_VPID;
828 static inline bool cpu_has_vmx_rdtscp(void)
830 return vmcs_config.cpu_based_2nd_exec_ctrl &
831 SECONDARY_EXEC_RDTSCP;
834 static inline bool cpu_has_virtual_nmis(void)
836 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
839 static inline bool cpu_has_vmx_wbinvd_exit(void)
841 return vmcs_config.cpu_based_2nd_exec_ctrl &
842 SECONDARY_EXEC_WBINVD_EXITING;
845 static inline bool report_flexpriority(void)
847 return flexpriority_enabled;
850 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
854 for (i = 0; i < vmx->nmsrs; ++i)
855 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
860 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
866 } operand = { vpid, 0, gva };
868 asm volatile (__ex(ASM_VMX_INVVPID)
869 /* CF==1 or ZF==1 --> rc = -1 */
871 : : "a"(&operand), "c"(ext) : "cc", "memory");
874 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
878 } operand = {eptp, gpa};
880 asm volatile (__ex(ASM_VMX_INVEPT)
881 /* CF==1 or ZF==1 --> rc = -1 */
882 "; ja 1f ; ud2 ; 1:\n"
883 : : "a" (&operand), "c" (ext) : "cc", "memory");
886 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
890 i = __find_msr_index(vmx, msr);
892 return &vmx->guest_msrs[i];
896 static void vmcs_clear(struct vmcs *vmcs)
898 u64 phys_addr = __pa(vmcs);
901 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
902 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
905 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
909 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
911 vmcs_clear(loaded_vmcs->vmcs);
912 loaded_vmcs->cpu = -1;
913 loaded_vmcs->launched = 0;
916 static void vmcs_load(struct vmcs *vmcs)
918 u64 phys_addr = __pa(vmcs);
921 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
922 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
925 printk(KERN_ERR "kvm: vmptrld %p/%llx fail\n",
929 static void __loaded_vmcs_clear(void *arg)
931 struct loaded_vmcs *loaded_vmcs = arg;
932 int cpu = raw_smp_processor_id();
934 if (loaded_vmcs->cpu != cpu)
935 return; /* vcpu migration can race with cpu offline */
936 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
937 per_cpu(current_vmcs, cpu) = NULL;
938 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
939 loaded_vmcs_init(loaded_vmcs);
942 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
944 if (loaded_vmcs->cpu != -1)
945 smp_call_function_single(
946 loaded_vmcs->cpu, __loaded_vmcs_clear, loaded_vmcs, 1);
949 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
954 if (cpu_has_vmx_invvpid_single())
955 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
958 static inline void vpid_sync_vcpu_global(void)
960 if (cpu_has_vmx_invvpid_global())
961 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
964 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
966 if (cpu_has_vmx_invvpid_single())
967 vpid_sync_vcpu_single(vmx);
969 vpid_sync_vcpu_global();
972 static inline void ept_sync_global(void)
974 if (cpu_has_vmx_invept_global())
975 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
978 static inline void ept_sync_context(u64 eptp)
981 if (cpu_has_vmx_invept_context())
982 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
988 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
991 if (cpu_has_vmx_invept_individual_addr())
992 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
995 ept_sync_context(eptp);
999 static __always_inline unsigned long vmcs_readl(unsigned long field)
1001 unsigned long value;
1003 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1004 : "=a"(value) : "d"(field) : "cc");
1008 static __always_inline u16 vmcs_read16(unsigned long field)
1010 return vmcs_readl(field);
1013 static __always_inline u32 vmcs_read32(unsigned long field)
1015 return vmcs_readl(field);
1018 static __always_inline u64 vmcs_read64(unsigned long field)
1020 #ifdef CONFIG_X86_64
1021 return vmcs_readl(field);
1023 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1027 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1029 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1030 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1034 static void vmcs_writel(unsigned long field, unsigned long value)
1038 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1039 : "=q"(error) : "a"(value), "d"(field) : "cc");
1040 if (unlikely(error))
1041 vmwrite_error(field, value);
1044 static void vmcs_write16(unsigned long field, u16 value)
1046 vmcs_writel(field, value);
1049 static void vmcs_write32(unsigned long field, u32 value)
1051 vmcs_writel(field, value);
1054 static void vmcs_write64(unsigned long field, u64 value)
1056 vmcs_writel(field, value);
1057 #ifndef CONFIG_X86_64
1059 vmcs_writel(field+1, value >> 32);
1063 static void vmcs_clear_bits(unsigned long field, u32 mask)
1065 vmcs_writel(field, vmcs_readl(field) & ~mask);
1068 static void vmcs_set_bits(unsigned long field, u32 mask)
1070 vmcs_writel(field, vmcs_readl(field) | mask);
1073 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1075 vmx->segment_cache.bitmask = 0;
1078 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1082 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1084 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1085 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1086 vmx->segment_cache.bitmask = 0;
1088 ret = vmx->segment_cache.bitmask & mask;
1089 vmx->segment_cache.bitmask |= mask;
1093 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1095 u16 *p = &vmx->segment_cache.seg[seg].selector;
1097 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1098 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1102 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1104 ulong *p = &vmx->segment_cache.seg[seg].base;
1106 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1107 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1111 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1113 u32 *p = &vmx->segment_cache.seg[seg].limit;
1115 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1116 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1120 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1122 u32 *p = &vmx->segment_cache.seg[seg].ar;
1124 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1125 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1129 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1133 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1134 (1u << NM_VECTOR) | (1u << DB_VECTOR);
1135 if ((vcpu->guest_debug &
1136 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1137 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1138 eb |= 1u << BP_VECTOR;
1139 if (to_vmx(vcpu)->rmode.vm86_active)
1142 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1143 if (vcpu->fpu_active)
1144 eb &= ~(1u << NM_VECTOR);
1145 vmcs_write32(EXCEPTION_BITMAP, eb);
1148 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1151 struct msr_autoload *m = &vmx->msr_autoload;
1153 if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
1154 vmcs_clear_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
1155 vmcs_clear_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
1159 for (i = 0; i < m->nr; ++i)
1160 if (m->guest[i].index == msr)
1166 m->guest[i] = m->guest[m->nr];
1167 m->host[i] = m->host[m->nr];
1168 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1169 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1172 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1173 u64 guest_val, u64 host_val)
1176 struct msr_autoload *m = &vmx->msr_autoload;
1178 if (msr == MSR_EFER && cpu_has_load_ia32_efer) {
1179 vmcs_write64(GUEST_IA32_EFER, guest_val);
1180 vmcs_write64(HOST_IA32_EFER, host_val);
1181 vmcs_set_bits(VM_ENTRY_CONTROLS, VM_ENTRY_LOAD_IA32_EFER);
1182 vmcs_set_bits(VM_EXIT_CONTROLS, VM_EXIT_LOAD_IA32_EFER);
1186 for (i = 0; i < m->nr; ++i)
1187 if (m->guest[i].index == msr)
1192 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1193 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1196 m->guest[i].index = msr;
1197 m->guest[i].value = guest_val;
1198 m->host[i].index = msr;
1199 m->host[i].value = host_val;
1202 static void reload_tss(void)
1205 * VT restores TR but not its size. Useless.
1207 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1208 struct desc_struct *descs;
1210 descs = (void *)gdt->address;
1211 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1215 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1220 guest_efer = vmx->vcpu.arch.efer;
1223 * NX is emulated; LMA and LME handled by hardware; SCE meaninless
1226 ignore_bits = EFER_NX | EFER_SCE;
1227 #ifdef CONFIG_X86_64
1228 ignore_bits |= EFER_LMA | EFER_LME;
1229 /* SCE is meaningful only in long mode on Intel */
1230 if (guest_efer & EFER_LMA)
1231 ignore_bits &= ~(u64)EFER_SCE;
1233 guest_efer &= ~ignore_bits;
1234 guest_efer |= host_efer & ignore_bits;
1235 vmx->guest_msrs[efer_offset].data = guest_efer;
1236 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1238 clear_atomic_switch_msr(vmx, MSR_EFER);
1239 /* On ept, can't emulate nx, and must switch nx atomically */
1240 if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1241 guest_efer = vmx->vcpu.arch.efer;
1242 if (!(guest_efer & EFER_LMA))
1243 guest_efer &= ~EFER_LME;
1244 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1251 static unsigned long segment_base(u16 selector)
1253 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1254 struct desc_struct *d;
1255 unsigned long table_base;
1258 if (!(selector & ~3))
1261 table_base = gdt->address;
1263 if (selector & 4) { /* from ldt */
1264 u16 ldt_selector = kvm_read_ldt();
1266 if (!(ldt_selector & ~3))
1269 table_base = segment_base(ldt_selector);
1271 d = (struct desc_struct *)(table_base + (selector & ~7));
1272 v = get_desc_base(d);
1273 #ifdef CONFIG_X86_64
1274 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1275 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1280 static inline unsigned long kvm_read_tr_base(void)
1283 asm("str %0" : "=g"(tr));
1284 return segment_base(tr);
1287 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1289 struct vcpu_vmx *vmx = to_vmx(vcpu);
1292 if (vmx->host_state.loaded)
1295 vmx->host_state.loaded = 1;
1297 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1298 * allow segment selectors with cpl > 0 or ti == 1.
1300 vmx->host_state.ldt_sel = kvm_read_ldt();
1301 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1302 savesegment(fs, vmx->host_state.fs_sel);
1303 if (!(vmx->host_state.fs_sel & 7)) {
1304 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1305 vmx->host_state.fs_reload_needed = 0;
1307 vmcs_write16(HOST_FS_SELECTOR, 0);
1308 vmx->host_state.fs_reload_needed = 1;
1310 savesegment(gs, vmx->host_state.gs_sel);
1311 if (!(vmx->host_state.gs_sel & 7))
1312 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1314 vmcs_write16(HOST_GS_SELECTOR, 0);
1315 vmx->host_state.gs_ldt_reload_needed = 1;
1318 #ifdef CONFIG_X86_64
1319 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1320 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1322 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1323 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1326 #ifdef CONFIG_X86_64
1327 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1328 if (is_long_mode(&vmx->vcpu))
1329 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1331 for (i = 0; i < vmx->save_nmsrs; ++i)
1332 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1333 vmx->guest_msrs[i].data,
1334 vmx->guest_msrs[i].mask);
1337 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1339 if (!vmx->host_state.loaded)
1342 ++vmx->vcpu.stat.host_state_reload;
1343 vmx->host_state.loaded = 0;
1344 #ifdef CONFIG_X86_64
1345 if (is_long_mode(&vmx->vcpu))
1346 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1348 if (vmx->host_state.gs_ldt_reload_needed) {
1349 kvm_load_ldt(vmx->host_state.ldt_sel);
1350 #ifdef CONFIG_X86_64
1351 load_gs_index(vmx->host_state.gs_sel);
1353 loadsegment(gs, vmx->host_state.gs_sel);
1356 if (vmx->host_state.fs_reload_needed)
1357 loadsegment(fs, vmx->host_state.fs_sel);
1359 #ifdef CONFIG_X86_64
1360 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1362 if (current_thread_info()->status & TS_USEDFPU)
1364 load_gdt(&__get_cpu_var(host_gdt));
1367 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1370 __vmx_load_host_state(vmx);
1375 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1376 * vcpu mutex is already taken.
1378 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1380 struct vcpu_vmx *vmx = to_vmx(vcpu);
1381 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1384 kvm_cpu_vmxon(phys_addr);
1385 else if (vmx->loaded_vmcs->cpu != cpu)
1386 loaded_vmcs_clear(vmx->loaded_vmcs);
1388 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1389 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1390 vmcs_load(vmx->loaded_vmcs->vmcs);
1393 if (vmx->loaded_vmcs->cpu != cpu) {
1394 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1395 unsigned long sysenter_esp;
1397 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1398 local_irq_disable();
1399 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1400 &per_cpu(loaded_vmcss_on_cpu, cpu));
1404 * Linux uses per-cpu TSS and GDT, so set these when switching
1407 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1408 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
1410 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1411 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1412 vmx->loaded_vmcs->cpu = cpu;
1416 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1418 __vmx_load_host_state(to_vmx(vcpu));
1419 if (!vmm_exclusive) {
1420 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1426 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1430 if (vcpu->fpu_active)
1432 vcpu->fpu_active = 1;
1433 cr0 = vmcs_readl(GUEST_CR0);
1434 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1435 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1436 vmcs_writel(GUEST_CR0, cr0);
1437 update_exception_bitmap(vcpu);
1438 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1439 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1442 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1444 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1446 vmx_decache_cr0_guest_bits(vcpu);
1447 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1448 update_exception_bitmap(vcpu);
1449 vcpu->arch.cr0_guest_owned_bits = 0;
1450 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1451 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1454 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1456 unsigned long rflags, save_rflags;
1458 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1459 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1460 rflags = vmcs_readl(GUEST_RFLAGS);
1461 if (to_vmx(vcpu)->rmode.vm86_active) {
1462 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1463 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1464 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1466 to_vmx(vcpu)->rflags = rflags;
1468 return to_vmx(vcpu)->rflags;
1471 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1473 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1474 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
1475 to_vmx(vcpu)->rflags = rflags;
1476 if (to_vmx(vcpu)->rmode.vm86_active) {
1477 to_vmx(vcpu)->rmode.save_rflags = rflags;
1478 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1480 vmcs_writel(GUEST_RFLAGS, rflags);
1483 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1485 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1488 if (interruptibility & GUEST_INTR_STATE_STI)
1489 ret |= KVM_X86_SHADOW_INT_STI;
1490 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1491 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1496 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1498 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1499 u32 interruptibility = interruptibility_old;
1501 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1503 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1504 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1505 else if (mask & KVM_X86_SHADOW_INT_STI)
1506 interruptibility |= GUEST_INTR_STATE_STI;
1508 if ((interruptibility != interruptibility_old))
1509 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1512 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1516 rip = kvm_rip_read(vcpu);
1517 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1518 kvm_rip_write(vcpu, rip);
1520 /* skipping an emulated instruction also counts */
1521 vmx_set_interrupt_shadow(vcpu, 0);
1524 static void vmx_clear_hlt(struct kvm_vcpu *vcpu)
1526 /* Ensure that we clear the HLT state in the VMCS. We don't need to
1527 * explicitly skip the instruction because if the HLT state is set, then
1528 * the instruction is already executing and RIP has already been
1530 if (!yield_on_hlt &&
1531 vmcs_read32(GUEST_ACTIVITY_STATE) == GUEST_ACTIVITY_HLT)
1532 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
1535 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1536 bool has_error_code, u32 error_code,
1539 struct vcpu_vmx *vmx = to_vmx(vcpu);
1540 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1542 if (has_error_code) {
1543 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1544 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1547 if (vmx->rmode.vm86_active) {
1549 if (kvm_exception_is_soft(nr))
1550 inc_eip = vcpu->arch.event_exit_inst_len;
1551 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1552 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1556 if (kvm_exception_is_soft(nr)) {
1557 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1558 vmx->vcpu.arch.event_exit_inst_len);
1559 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1561 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1563 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1564 vmx_clear_hlt(vcpu);
1567 static bool vmx_rdtscp_supported(void)
1569 return cpu_has_vmx_rdtscp();
1573 * Swap MSR entry in host/guest MSR entry array.
1575 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1577 struct shared_msr_entry tmp;
1579 tmp = vmx->guest_msrs[to];
1580 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1581 vmx->guest_msrs[from] = tmp;
1585 * Set up the vmcs to automatically save and restore system
1586 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1587 * mode, as fiddling with msrs is very expensive.
1589 static void setup_msrs(struct vcpu_vmx *vmx)
1591 int save_nmsrs, index;
1592 unsigned long *msr_bitmap;
1594 vmx_load_host_state(vmx);
1596 #ifdef CONFIG_X86_64
1597 if (is_long_mode(&vmx->vcpu)) {
1598 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1600 move_msr_up(vmx, index, save_nmsrs++);
1601 index = __find_msr_index(vmx, MSR_LSTAR);
1603 move_msr_up(vmx, index, save_nmsrs++);
1604 index = __find_msr_index(vmx, MSR_CSTAR);
1606 move_msr_up(vmx, index, save_nmsrs++);
1607 index = __find_msr_index(vmx, MSR_TSC_AUX);
1608 if (index >= 0 && vmx->rdtscp_enabled)
1609 move_msr_up(vmx, index, save_nmsrs++);
1611 * MSR_STAR is only needed on long mode guests, and only
1612 * if efer.sce is enabled.
1614 index = __find_msr_index(vmx, MSR_STAR);
1615 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1616 move_msr_up(vmx, index, save_nmsrs++);
1619 index = __find_msr_index(vmx, MSR_EFER);
1620 if (index >= 0 && update_transition_efer(vmx, index))
1621 move_msr_up(vmx, index, save_nmsrs++);
1623 vmx->save_nmsrs = save_nmsrs;
1625 if (cpu_has_vmx_msr_bitmap()) {
1626 if (is_long_mode(&vmx->vcpu))
1627 msr_bitmap = vmx_msr_bitmap_longmode;
1629 msr_bitmap = vmx_msr_bitmap_legacy;
1631 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1636 * reads and returns guest's timestamp counter "register"
1637 * guest_tsc = host_tsc + tsc_offset -- 21.3
1639 static u64 guest_read_tsc(void)
1641 u64 host_tsc, tsc_offset;
1644 tsc_offset = vmcs_read64(TSC_OFFSET);
1645 return host_tsc + tsc_offset;
1649 * Empty call-back. Needs to be implemented when VMX enables the SET_TSC_KHZ
1650 * ioctl. In this case the call-back should update internal vmx state to make
1651 * the changes effective.
1653 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz)
1655 /* Nothing to do here */
1659 * writes 'offset' into guest's timestamp counter offset register
1661 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1663 vmcs_write64(TSC_OFFSET, offset);
1666 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment)
1668 u64 offset = vmcs_read64(TSC_OFFSET);
1669 vmcs_write64(TSC_OFFSET, offset + adjustment);
1672 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1674 return target_tsc - native_read_tsc();
1677 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
1679 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
1680 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
1684 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1685 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1686 * all guests if the "nested" module option is off, and can also be disabled
1687 * for a single guest by disabling its VMX cpuid bit.
1689 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1691 return nested && guest_cpuid_has_vmx(vcpu);
1695 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
1696 * returned for the various VMX controls MSRs when nested VMX is enabled.
1697 * The same values should also be used to verify that vmcs12 control fields are
1698 * valid during nested entry from L1 to L2.
1699 * Each of these control msrs has a low and high 32-bit half: A low bit is on
1700 * if the corresponding bit in the (32-bit) control field *must* be on, and a
1701 * bit in the high half is on if the corresponding bit in the control field
1702 * may be on. See also vmx_control_verify().
1703 * TODO: allow these variables to be modified (downgraded) by module options
1706 static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
1707 static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
1708 static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
1709 static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
1710 static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
1711 static __init void nested_vmx_setup_ctls_msrs(void)
1714 * Note that as a general rule, the high half of the MSRs (bits in
1715 * the control fields which may be 1) should be initialized by the
1716 * intersection of the underlying hardware's MSR (i.e., features which
1717 * can be supported) and the list of features we want to expose -
1718 * because they are known to be properly supported in our code.
1719 * Also, usually, the low half of the MSRs (bits which must be 1) can
1720 * be set to 0, meaning that L1 may turn off any of these bits. The
1721 * reason is that if one of these bits is necessary, it will appear
1722 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
1723 * fields of vmcs01 and vmcs02, will turn these bits off - and
1724 * nested_vmx_exit_handled() will not pass related exits to L1.
1725 * These rules have exceptions below.
1728 /* pin-based controls */
1730 * According to the Intel spec, if bit 55 of VMX_BASIC is off (as it is
1731 * in our case), bits 1, 2 and 4 (i.e., 0x16) must be 1 in this MSR.
1733 nested_vmx_pinbased_ctls_low = 0x16 ;
1734 nested_vmx_pinbased_ctls_high = 0x16 |
1735 PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
1736 PIN_BASED_VIRTUAL_NMIS;
1739 nested_vmx_exit_ctls_low = 0;
1740 #ifdef CONFIG_X86_64
1741 nested_vmx_exit_ctls_high = VM_EXIT_HOST_ADDR_SPACE_SIZE;
1743 nested_vmx_exit_ctls_high = 0;
1746 /* entry controls */
1747 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
1748 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
1749 nested_vmx_entry_ctls_low = 0;
1750 nested_vmx_entry_ctls_high &=
1751 VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_IA32E_MODE;
1753 /* cpu-based controls */
1754 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
1755 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
1756 nested_vmx_procbased_ctls_low = 0;
1757 nested_vmx_procbased_ctls_high &=
1758 CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_USE_TSC_OFFSETING |
1759 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
1760 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
1761 CPU_BASED_CR3_STORE_EXITING |
1762 #ifdef CONFIG_X86_64
1763 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
1765 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
1766 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
1767 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1769 * We can allow some features even when not supported by the
1770 * hardware. For example, L1 can specify an MSR bitmap - and we
1771 * can use it to avoid exits to L1 - even when L0 runs L2
1772 * without MSR bitmaps.
1774 nested_vmx_procbased_ctls_high |= CPU_BASED_USE_MSR_BITMAPS;
1776 /* secondary cpu-based controls */
1777 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
1778 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
1779 nested_vmx_secondary_ctls_low = 0;
1780 nested_vmx_secondary_ctls_high &=
1781 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1784 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
1787 * Bits 0 in high must be 0, and bits 1 in low must be 1.
1789 return ((control & high) | low) == control;
1792 static inline u64 vmx_control_msr(u32 low, u32 high)
1794 return low | ((u64)high << 32);
1798 * If we allow our guest to use VMX instructions (i.e., nested VMX), we should
1799 * also let it use VMX-specific MSRs.
1800 * vmx_get_vmx_msr() and vmx_set_vmx_msr() return 1 when we handled a
1801 * VMX-specific MSR, or 0 when we haven't (and the caller should handle it
1802 * like all other MSRs).
1804 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1806 if (!nested_vmx_allowed(vcpu) && msr_index >= MSR_IA32_VMX_BASIC &&
1807 msr_index <= MSR_IA32_VMX_TRUE_ENTRY_CTLS) {
1809 * According to the spec, processors which do not support VMX
1810 * should throw a #GP(0) when VMX capability MSRs are read.
1812 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
1816 switch (msr_index) {
1817 case MSR_IA32_FEATURE_CONTROL:
1820 case MSR_IA32_VMX_BASIC:
1822 * This MSR reports some information about VMX support. We
1823 * should return information about the VMX we emulate for the
1824 * guest, and the VMCS structure we give it - not about the
1825 * VMX support of the underlying hardware.
1827 *pdata = VMCS12_REVISION |
1828 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
1829 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
1831 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
1832 case MSR_IA32_VMX_PINBASED_CTLS:
1833 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
1834 nested_vmx_pinbased_ctls_high);
1836 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
1837 case MSR_IA32_VMX_PROCBASED_CTLS:
1838 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
1839 nested_vmx_procbased_ctls_high);
1841 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
1842 case MSR_IA32_VMX_EXIT_CTLS:
1843 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
1844 nested_vmx_exit_ctls_high);
1846 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
1847 case MSR_IA32_VMX_ENTRY_CTLS:
1848 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
1849 nested_vmx_entry_ctls_high);
1851 case MSR_IA32_VMX_MISC:
1855 * These MSRs specify bits which the guest must keep fixed (on or off)
1856 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
1857 * We picked the standard core2 setting.
1859 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
1860 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
1861 case MSR_IA32_VMX_CR0_FIXED0:
1862 *pdata = VMXON_CR0_ALWAYSON;
1864 case MSR_IA32_VMX_CR0_FIXED1:
1867 case MSR_IA32_VMX_CR4_FIXED0:
1868 *pdata = VMXON_CR4_ALWAYSON;
1870 case MSR_IA32_VMX_CR4_FIXED1:
1873 case MSR_IA32_VMX_VMCS_ENUM:
1876 case MSR_IA32_VMX_PROCBASED_CTLS2:
1877 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
1878 nested_vmx_secondary_ctls_high);
1880 case MSR_IA32_VMX_EPT_VPID_CAP:
1881 /* Currently, no nested ept or nested vpid */
1891 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1893 if (!nested_vmx_allowed(vcpu))
1896 if (msr_index == MSR_IA32_FEATURE_CONTROL)
1897 /* TODO: the right thing. */
1900 * No need to treat VMX capability MSRs specially: If we don't handle
1901 * them, handle_wrmsr will #GP(0), which is correct (they are readonly)
1907 * Reads an msr value (of 'msr_index') into 'pdata'.
1908 * Returns 0 on success, non-0 otherwise.
1909 * Assumes vcpu_load() was already called.
1911 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
1914 struct shared_msr_entry *msr;
1917 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
1921 switch (msr_index) {
1922 #ifdef CONFIG_X86_64
1924 data = vmcs_readl(GUEST_FS_BASE);
1927 data = vmcs_readl(GUEST_GS_BASE);
1929 case MSR_KERNEL_GS_BASE:
1930 vmx_load_host_state(to_vmx(vcpu));
1931 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
1935 return kvm_get_msr_common(vcpu, msr_index, pdata);
1937 data = guest_read_tsc();
1939 case MSR_IA32_SYSENTER_CS:
1940 data = vmcs_read32(GUEST_SYSENTER_CS);
1942 case MSR_IA32_SYSENTER_EIP:
1943 data = vmcs_readl(GUEST_SYSENTER_EIP);
1945 case MSR_IA32_SYSENTER_ESP:
1946 data = vmcs_readl(GUEST_SYSENTER_ESP);
1949 if (!to_vmx(vcpu)->rdtscp_enabled)
1951 /* Otherwise falls through */
1953 vmx_load_host_state(to_vmx(vcpu));
1954 if (vmx_get_vmx_msr(vcpu, msr_index, pdata))
1956 msr = find_msr_entry(to_vmx(vcpu), msr_index);
1958 vmx_load_host_state(to_vmx(vcpu));
1962 return kvm_get_msr_common(vcpu, msr_index, pdata);
1970 * Writes msr value into into the appropriate "register".
1971 * Returns 0 on success, non-0 otherwise.
1972 * Assumes vcpu_load() was already called.
1974 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
1976 struct vcpu_vmx *vmx = to_vmx(vcpu);
1977 struct shared_msr_entry *msr;
1980 switch (msr_index) {
1982 vmx_load_host_state(vmx);
1983 ret = kvm_set_msr_common(vcpu, msr_index, data);
1985 #ifdef CONFIG_X86_64
1987 vmx_segment_cache_clear(vmx);
1988 vmcs_writel(GUEST_FS_BASE, data);
1991 vmx_segment_cache_clear(vmx);
1992 vmcs_writel(GUEST_GS_BASE, data);
1994 case MSR_KERNEL_GS_BASE:
1995 vmx_load_host_state(vmx);
1996 vmx->msr_guest_kernel_gs_base = data;
1999 case MSR_IA32_SYSENTER_CS:
2000 vmcs_write32(GUEST_SYSENTER_CS, data);
2002 case MSR_IA32_SYSENTER_EIP:
2003 vmcs_writel(GUEST_SYSENTER_EIP, data);
2005 case MSR_IA32_SYSENTER_ESP:
2006 vmcs_writel(GUEST_SYSENTER_ESP, data);
2009 kvm_write_tsc(vcpu, data);
2011 case MSR_IA32_CR_PAT:
2012 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2013 vmcs_write64(GUEST_IA32_PAT, data);
2014 vcpu->arch.pat = data;
2017 ret = kvm_set_msr_common(vcpu, msr_index, data);
2020 if (!vmx->rdtscp_enabled)
2022 /* Check reserved bit, higher 32 bits should be zero */
2023 if ((data >> 32) != 0)
2025 /* Otherwise falls through */
2027 if (vmx_set_vmx_msr(vcpu, msr_index, data))
2029 msr = find_msr_entry(vmx, msr_index);
2031 vmx_load_host_state(vmx);
2035 ret = kvm_set_msr_common(vcpu, msr_index, data);
2041 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2043 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2046 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2049 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2051 case VCPU_EXREG_PDPTR:
2053 ept_save_pdptrs(vcpu);
2060 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
2062 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
2063 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
2065 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2067 update_exception_bitmap(vcpu);
2070 static __init int cpu_has_kvm_support(void)
2072 return cpu_has_vmx();
2075 static __init int vmx_disabled_by_bios(void)
2079 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2080 if (msr & FEATURE_CONTROL_LOCKED) {
2081 /* launched w/ TXT and VMX disabled */
2082 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2085 /* launched w/o TXT and VMX only enabled w/ TXT */
2086 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2087 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2088 && !tboot_enabled()) {
2089 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2090 "activate TXT before enabling KVM\n");
2093 /* launched w/o TXT and VMX disabled */
2094 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2095 && !tboot_enabled())
2102 static void kvm_cpu_vmxon(u64 addr)
2104 asm volatile (ASM_VMX_VMXON_RAX
2105 : : "a"(&addr), "m"(addr)
2109 static int hardware_enable(void *garbage)
2111 int cpu = raw_smp_processor_id();
2112 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2115 if (read_cr4() & X86_CR4_VMXE)
2118 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2119 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2121 test_bits = FEATURE_CONTROL_LOCKED;
2122 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2123 if (tboot_enabled())
2124 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2126 if ((old & test_bits) != test_bits) {
2127 /* enable and lock */
2128 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2130 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2132 if (vmm_exclusive) {
2133 kvm_cpu_vmxon(phys_addr);
2137 store_gdt(&__get_cpu_var(host_gdt));
2142 static void vmclear_local_loaded_vmcss(void)
2144 int cpu = raw_smp_processor_id();
2145 struct loaded_vmcs *v, *n;
2147 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2148 loaded_vmcss_on_cpu_link)
2149 __loaded_vmcs_clear(v);
2153 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2156 static void kvm_cpu_vmxoff(void)
2158 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2161 static void hardware_disable(void *garbage)
2163 if (vmm_exclusive) {
2164 vmclear_local_loaded_vmcss();
2167 write_cr4(read_cr4() & ~X86_CR4_VMXE);
2170 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2171 u32 msr, u32 *result)
2173 u32 vmx_msr_low, vmx_msr_high;
2174 u32 ctl = ctl_min | ctl_opt;
2176 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2178 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2179 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2181 /* Ensure minimum (required) set of control bits are supported. */
2189 static __init bool allow_1_setting(u32 msr, u32 ctl)
2191 u32 vmx_msr_low, vmx_msr_high;
2193 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2194 return vmx_msr_high & ctl;
2197 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2199 u32 vmx_msr_low, vmx_msr_high;
2200 u32 min, opt, min2, opt2;
2201 u32 _pin_based_exec_control = 0;
2202 u32 _cpu_based_exec_control = 0;
2203 u32 _cpu_based_2nd_exec_control = 0;
2204 u32 _vmexit_control = 0;
2205 u32 _vmentry_control = 0;
2207 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2208 opt = PIN_BASED_VIRTUAL_NMIS;
2209 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2210 &_pin_based_exec_control) < 0)
2214 #ifdef CONFIG_X86_64
2215 CPU_BASED_CR8_LOAD_EXITING |
2216 CPU_BASED_CR8_STORE_EXITING |
2218 CPU_BASED_CR3_LOAD_EXITING |
2219 CPU_BASED_CR3_STORE_EXITING |
2220 CPU_BASED_USE_IO_BITMAPS |
2221 CPU_BASED_MOV_DR_EXITING |
2222 CPU_BASED_USE_TSC_OFFSETING |
2223 CPU_BASED_MWAIT_EXITING |
2224 CPU_BASED_MONITOR_EXITING |
2225 CPU_BASED_INVLPG_EXITING;
2228 min |= CPU_BASED_HLT_EXITING;
2230 opt = CPU_BASED_TPR_SHADOW |
2231 CPU_BASED_USE_MSR_BITMAPS |
2232 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2233 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2234 &_cpu_based_exec_control) < 0)
2236 #ifdef CONFIG_X86_64
2237 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2238 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2239 ~CPU_BASED_CR8_STORE_EXITING;
2241 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2243 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2244 SECONDARY_EXEC_WBINVD_EXITING |
2245 SECONDARY_EXEC_ENABLE_VPID |
2246 SECONDARY_EXEC_ENABLE_EPT |
2247 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2248 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2249 SECONDARY_EXEC_RDTSCP;
2250 if (adjust_vmx_controls(min2, opt2,
2251 MSR_IA32_VMX_PROCBASED_CTLS2,
2252 &_cpu_based_2nd_exec_control) < 0)
2255 #ifndef CONFIG_X86_64
2256 if (!(_cpu_based_2nd_exec_control &
2257 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2258 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2260 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2261 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2263 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2264 CPU_BASED_CR3_STORE_EXITING |
2265 CPU_BASED_INVLPG_EXITING);
2266 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2267 vmx_capability.ept, vmx_capability.vpid);
2271 #ifdef CONFIG_X86_64
2272 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2274 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
2275 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2276 &_vmexit_control) < 0)
2280 opt = VM_ENTRY_LOAD_IA32_PAT;
2281 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2282 &_vmentry_control) < 0)
2285 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2287 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2288 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2291 #ifdef CONFIG_X86_64
2292 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2293 if (vmx_msr_high & (1u<<16))
2297 /* Require Write-Back (WB) memory type for VMCS accesses. */
2298 if (((vmx_msr_high >> 18) & 15) != 6)
2301 vmcs_conf->size = vmx_msr_high & 0x1fff;
2302 vmcs_conf->order = get_order(vmcs_config.size);
2303 vmcs_conf->revision_id = vmx_msr_low;
2305 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2306 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2307 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2308 vmcs_conf->vmexit_ctrl = _vmexit_control;
2309 vmcs_conf->vmentry_ctrl = _vmentry_control;
2311 cpu_has_load_ia32_efer =
2312 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2313 VM_ENTRY_LOAD_IA32_EFER)
2314 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2315 VM_EXIT_LOAD_IA32_EFER);
2320 static struct vmcs *alloc_vmcs_cpu(int cpu)
2322 int node = cpu_to_node(cpu);
2326 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
2329 vmcs = page_address(pages);
2330 memset(vmcs, 0, vmcs_config.size);
2331 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
2335 static struct vmcs *alloc_vmcs(void)
2337 return alloc_vmcs_cpu(raw_smp_processor_id());
2340 static void free_vmcs(struct vmcs *vmcs)
2342 free_pages((unsigned long)vmcs, vmcs_config.order);
2346 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2348 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2350 if (!loaded_vmcs->vmcs)
2352 loaded_vmcs_clear(loaded_vmcs);
2353 free_vmcs(loaded_vmcs->vmcs);
2354 loaded_vmcs->vmcs = NULL;
2357 static void free_kvm_area(void)
2361 for_each_possible_cpu(cpu) {
2362 free_vmcs(per_cpu(vmxarea, cpu));
2363 per_cpu(vmxarea, cpu) = NULL;
2367 static __init int alloc_kvm_area(void)
2371 for_each_possible_cpu(cpu) {
2374 vmcs = alloc_vmcs_cpu(cpu);
2380 per_cpu(vmxarea, cpu) = vmcs;
2385 static __init int hardware_setup(void)
2387 if (setup_vmcs_config(&vmcs_config) < 0)
2390 if (boot_cpu_has(X86_FEATURE_NX))
2391 kvm_enable_efer_bits(EFER_NX);
2393 if (!cpu_has_vmx_vpid())
2396 if (!cpu_has_vmx_ept() ||
2397 !cpu_has_vmx_ept_4levels()) {
2399 enable_unrestricted_guest = 0;
2402 if (!cpu_has_vmx_unrestricted_guest())
2403 enable_unrestricted_guest = 0;
2405 if (!cpu_has_vmx_flexpriority())
2406 flexpriority_enabled = 0;
2408 if (!cpu_has_vmx_tpr_shadow())
2409 kvm_x86_ops->update_cr8_intercept = NULL;
2411 if (enable_ept && !cpu_has_vmx_ept_2m_page())
2412 kvm_disable_largepages();
2414 if (!cpu_has_vmx_ple())
2418 nested_vmx_setup_ctls_msrs();
2420 return alloc_kvm_area();
2423 static __exit void hardware_unsetup(void)
2428 static void fix_pmode_dataseg(int seg, struct kvm_save_segment *save)
2430 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2432 if (vmcs_readl(sf->base) == save->base && (save->base & AR_S_MASK)) {
2433 vmcs_write16(sf->selector, save->selector);
2434 vmcs_writel(sf->base, save->base);
2435 vmcs_write32(sf->limit, save->limit);
2436 vmcs_write32(sf->ar_bytes, save->ar);
2438 u32 dpl = (vmcs_read16(sf->selector) & SELECTOR_RPL_MASK)
2440 vmcs_write32(sf->ar_bytes, 0x93 | dpl);
2444 static void enter_pmode(struct kvm_vcpu *vcpu)
2446 unsigned long flags;
2447 struct vcpu_vmx *vmx = to_vmx(vcpu);
2449 vmx->emulation_required = 1;
2450 vmx->rmode.vm86_active = 0;
2452 vmx_segment_cache_clear(vmx);
2454 vmcs_write16(GUEST_TR_SELECTOR, vmx->rmode.tr.selector);
2455 vmcs_writel(GUEST_TR_BASE, vmx->rmode.tr.base);
2456 vmcs_write32(GUEST_TR_LIMIT, vmx->rmode.tr.limit);
2457 vmcs_write32(GUEST_TR_AR_BYTES, vmx->rmode.tr.ar);
2459 flags = vmcs_readl(GUEST_RFLAGS);
2460 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2461 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2462 vmcs_writel(GUEST_RFLAGS, flags);
2464 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2465 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2467 update_exception_bitmap(vcpu);
2469 if (emulate_invalid_guest_state)
2472 fix_pmode_dataseg(VCPU_SREG_ES, &vmx->rmode.es);
2473 fix_pmode_dataseg(VCPU_SREG_DS, &vmx->rmode.ds);
2474 fix_pmode_dataseg(VCPU_SREG_GS, &vmx->rmode.gs);
2475 fix_pmode_dataseg(VCPU_SREG_FS, &vmx->rmode.fs);
2477 vmx_segment_cache_clear(vmx);
2479 vmcs_write16(GUEST_SS_SELECTOR, 0);
2480 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
2482 vmcs_write16(GUEST_CS_SELECTOR,
2483 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
2484 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
2487 static gva_t rmode_tss_base(struct kvm *kvm)
2489 if (!kvm->arch.tss_addr) {
2490 struct kvm_memslots *slots;
2493 slots = kvm_memslots(kvm);
2494 base_gfn = slots->memslots[0].base_gfn +
2495 kvm->memslots->memslots[0].npages - 3;
2496 return base_gfn << PAGE_SHIFT;
2498 return kvm->arch.tss_addr;
2501 static void fix_rmode_seg(int seg, struct kvm_save_segment *save)
2503 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2505 save->selector = vmcs_read16(sf->selector);
2506 save->base = vmcs_readl(sf->base);
2507 save->limit = vmcs_read32(sf->limit);
2508 save->ar = vmcs_read32(sf->ar_bytes);
2509 vmcs_write16(sf->selector, save->base >> 4);
2510 vmcs_write32(sf->base, save->base & 0xffff0);
2511 vmcs_write32(sf->limit, 0xffff);
2512 vmcs_write32(sf->ar_bytes, 0xf3);
2513 if (save->base & 0xf)
2514 printk_once(KERN_WARNING "kvm: segment base is not paragraph"
2515 " aligned when entering protected mode (seg=%d)",
2519 static void enter_rmode(struct kvm_vcpu *vcpu)
2521 unsigned long flags;
2522 struct vcpu_vmx *vmx = to_vmx(vcpu);
2524 if (enable_unrestricted_guest)
2527 vmx->emulation_required = 1;
2528 vmx->rmode.vm86_active = 1;
2531 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2532 * vcpu. Call it here with phys address pointing 16M below 4G.
2534 if (!vcpu->kvm->arch.tss_addr) {
2535 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2536 "called before entering vcpu\n");
2537 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2538 vmx_set_tss_addr(vcpu->kvm, 0xfeffd000);
2539 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2542 vmx_segment_cache_clear(vmx);
2544 vmx->rmode.tr.selector = vmcs_read16(GUEST_TR_SELECTOR);
2545 vmx->rmode.tr.base = vmcs_readl(GUEST_TR_BASE);
2546 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
2548 vmx->rmode.tr.limit = vmcs_read32(GUEST_TR_LIMIT);
2549 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2551 vmx->rmode.tr.ar = vmcs_read32(GUEST_TR_AR_BYTES);
2552 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2554 flags = vmcs_readl(GUEST_RFLAGS);
2555 vmx->rmode.save_rflags = flags;
2557 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2559 vmcs_writel(GUEST_RFLAGS, flags);
2560 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2561 update_exception_bitmap(vcpu);
2563 if (emulate_invalid_guest_state)
2564 goto continue_rmode;
2566 vmcs_write16(GUEST_SS_SELECTOR, vmcs_readl(GUEST_SS_BASE) >> 4);
2567 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
2568 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
2570 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
2571 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
2572 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
2573 vmcs_writel(GUEST_CS_BASE, 0xf0000);
2574 vmcs_write16(GUEST_CS_SELECTOR, vmcs_readl(GUEST_CS_BASE) >> 4);
2576 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.es);
2577 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.ds);
2578 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.gs);
2579 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.fs);
2582 kvm_mmu_reset_context(vcpu);
2585 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2587 struct vcpu_vmx *vmx = to_vmx(vcpu);
2588 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2594 * Force kernel_gs_base reloading before EFER changes, as control
2595 * of this msr depends on is_long_mode().
2597 vmx_load_host_state(to_vmx(vcpu));
2598 vcpu->arch.efer = efer;
2599 if (efer & EFER_LMA) {
2600 vmcs_write32(VM_ENTRY_CONTROLS,
2601 vmcs_read32(VM_ENTRY_CONTROLS) |
2602 VM_ENTRY_IA32E_MODE);
2605 vmcs_write32(VM_ENTRY_CONTROLS,
2606 vmcs_read32(VM_ENTRY_CONTROLS) &
2607 ~VM_ENTRY_IA32E_MODE);
2609 msr->data = efer & ~EFER_LME;
2614 #ifdef CONFIG_X86_64
2616 static void enter_lmode(struct kvm_vcpu *vcpu)
2620 vmx_segment_cache_clear(to_vmx(vcpu));
2622 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2623 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
2624 printk(KERN_DEBUG "%s: tss fixup for long mode. \n",
2626 vmcs_write32(GUEST_TR_AR_BYTES,
2627 (guest_tr_ar & ~AR_TYPE_MASK)
2628 | AR_TYPE_BUSY_64_TSS);
2630 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2633 static void exit_lmode(struct kvm_vcpu *vcpu)
2635 vmcs_write32(VM_ENTRY_CONTROLS,
2636 vmcs_read32(VM_ENTRY_CONTROLS)
2637 & ~VM_ENTRY_IA32E_MODE);
2638 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2643 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
2645 vpid_sync_context(to_vmx(vcpu));
2647 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
2649 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
2653 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2655 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2657 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2658 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2661 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2663 if (enable_ept && is_paging(vcpu))
2664 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2665 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2668 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2670 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2672 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2673 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2676 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2678 if (!test_bit(VCPU_EXREG_PDPTR,
2679 (unsigned long *)&vcpu->arch.regs_dirty))
2682 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2683 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
2684 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
2685 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
2686 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
2690 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2692 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2693 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2694 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2695 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2696 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2699 __set_bit(VCPU_EXREG_PDPTR,
2700 (unsigned long *)&vcpu->arch.regs_avail);
2701 __set_bit(VCPU_EXREG_PDPTR,
2702 (unsigned long *)&vcpu->arch.regs_dirty);
2705 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
2707 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2709 struct kvm_vcpu *vcpu)
2711 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2712 vmx_decache_cr3(vcpu);
2713 if (!(cr0 & X86_CR0_PG)) {
2714 /* From paging/starting to nonpaging */
2715 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2716 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
2717 (CPU_BASED_CR3_LOAD_EXITING |
2718 CPU_BASED_CR3_STORE_EXITING));
2719 vcpu->arch.cr0 = cr0;
2720 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2721 } else if (!is_paging(vcpu)) {
2722 /* From nonpaging to paging */
2723 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2724 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2725 ~(CPU_BASED_CR3_LOAD_EXITING |
2726 CPU_BASED_CR3_STORE_EXITING));
2727 vcpu->arch.cr0 = cr0;
2728 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2731 if (!(cr0 & X86_CR0_WP))
2732 *hw_cr0 &= ~X86_CR0_WP;
2735 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
2737 struct vcpu_vmx *vmx = to_vmx(vcpu);
2738 unsigned long hw_cr0;
2740 if (enable_unrestricted_guest)
2741 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
2742 | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
2744 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
2746 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
2749 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
2752 #ifdef CONFIG_X86_64
2753 if (vcpu->arch.efer & EFER_LME) {
2754 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
2756 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
2762 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
2764 if (!vcpu->fpu_active)
2765 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
2767 vmcs_writel(CR0_READ_SHADOW, cr0);
2768 vmcs_writel(GUEST_CR0, hw_cr0);
2769 vcpu->arch.cr0 = cr0;
2770 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
2773 static u64 construct_eptp(unsigned long root_hpa)
2777 /* TODO write the value reading from MSR */
2778 eptp = VMX_EPT_DEFAULT_MT |
2779 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
2780 eptp |= (root_hpa & PAGE_MASK);
2785 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
2787 unsigned long guest_cr3;
2792 eptp = construct_eptp(cr3);
2793 vmcs_write64(EPT_POINTER, eptp);
2794 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
2795 vcpu->kvm->arch.ept_identity_map_addr;
2796 ept_load_pdptrs(vcpu);
2799 vmx_flush_tlb(vcpu);
2800 vmcs_writel(GUEST_CR3, guest_cr3);
2803 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
2805 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
2806 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
2808 if (cr4 & X86_CR4_VMXE) {
2810 * To use VMXON (and later other VMX instructions), a guest
2811 * must first be able to turn on cr4.VMXE (see handle_vmon()).
2812 * So basically the check on whether to allow nested VMX
2815 if (!nested_vmx_allowed(vcpu))
2817 } else if (to_vmx(vcpu)->nested.vmxon)
2820 vcpu->arch.cr4 = cr4;
2822 if (!is_paging(vcpu)) {
2823 hw_cr4 &= ~X86_CR4_PAE;
2824 hw_cr4 |= X86_CR4_PSE;
2825 } else if (!(cr4 & X86_CR4_PAE)) {
2826 hw_cr4 &= ~X86_CR4_PAE;
2830 vmcs_writel(CR4_READ_SHADOW, cr4);
2831 vmcs_writel(GUEST_CR4, hw_cr4);
2835 static void vmx_get_segment(struct kvm_vcpu *vcpu,
2836 struct kvm_segment *var, int seg)
2838 struct vcpu_vmx *vmx = to_vmx(vcpu);
2839 struct kvm_save_segment *save;
2842 if (vmx->rmode.vm86_active
2843 && (seg == VCPU_SREG_TR || seg == VCPU_SREG_ES
2844 || seg == VCPU_SREG_DS || seg == VCPU_SREG_FS
2845 || seg == VCPU_SREG_GS)
2846 && !emulate_invalid_guest_state) {
2848 case VCPU_SREG_TR: save = &vmx->rmode.tr; break;
2849 case VCPU_SREG_ES: save = &vmx->rmode.es; break;
2850 case VCPU_SREG_DS: save = &vmx->rmode.ds; break;
2851 case VCPU_SREG_FS: save = &vmx->rmode.fs; break;
2852 case VCPU_SREG_GS: save = &vmx->rmode.gs; break;
2855 var->selector = save->selector;
2856 var->base = save->base;
2857 var->limit = save->limit;
2859 if (seg == VCPU_SREG_TR
2860 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
2861 goto use_saved_rmode_seg;
2863 var->base = vmx_read_guest_seg_base(vmx, seg);
2864 var->limit = vmx_read_guest_seg_limit(vmx, seg);
2865 var->selector = vmx_read_guest_seg_selector(vmx, seg);
2866 ar = vmx_read_guest_seg_ar(vmx, seg);
2867 use_saved_rmode_seg:
2868 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
2870 var->type = ar & 15;
2871 var->s = (ar >> 4) & 1;
2872 var->dpl = (ar >> 5) & 3;
2873 var->present = (ar >> 7) & 1;
2874 var->avl = (ar >> 12) & 1;
2875 var->l = (ar >> 13) & 1;
2876 var->db = (ar >> 14) & 1;
2877 var->g = (ar >> 15) & 1;
2878 var->unusable = (ar >> 16) & 1;
2881 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
2883 struct kvm_segment s;
2885 if (to_vmx(vcpu)->rmode.vm86_active) {
2886 vmx_get_segment(vcpu, &s, seg);
2889 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
2892 static int __vmx_get_cpl(struct kvm_vcpu *vcpu)
2894 if (!is_protmode(vcpu))
2897 if (!is_long_mode(vcpu)
2898 && (kvm_get_rflags(vcpu) & X86_EFLAGS_VM)) /* if virtual 8086 */
2901 return vmx_read_guest_seg_selector(to_vmx(vcpu), VCPU_SREG_CS) & 3;
2904 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
2906 if (!test_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail)) {
2907 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
2908 to_vmx(vcpu)->cpl = __vmx_get_cpl(vcpu);
2910 return to_vmx(vcpu)->cpl;
2914 static u32 vmx_segment_access_rights(struct kvm_segment *var)
2921 ar = var->type & 15;
2922 ar |= (var->s & 1) << 4;
2923 ar |= (var->dpl & 3) << 5;
2924 ar |= (var->present & 1) << 7;
2925 ar |= (var->avl & 1) << 12;
2926 ar |= (var->l & 1) << 13;
2927 ar |= (var->db & 1) << 14;
2928 ar |= (var->g & 1) << 15;
2930 if (ar == 0) /* a 0 value means unusable */
2931 ar = AR_UNUSABLE_MASK;
2936 static void vmx_set_segment(struct kvm_vcpu *vcpu,
2937 struct kvm_segment *var, int seg)
2939 struct vcpu_vmx *vmx = to_vmx(vcpu);
2940 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2943 vmx_segment_cache_clear(vmx);
2945 if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
2946 vmcs_write16(sf->selector, var->selector);
2947 vmx->rmode.tr.selector = var->selector;
2948 vmx->rmode.tr.base = var->base;
2949 vmx->rmode.tr.limit = var->limit;
2950 vmx->rmode.tr.ar = vmx_segment_access_rights(var);
2953 vmcs_writel(sf->base, var->base);
2954 vmcs_write32(sf->limit, var->limit);
2955 vmcs_write16(sf->selector, var->selector);
2956 if (vmx->rmode.vm86_active && var->s) {
2958 * Hack real-mode segments into vm86 compatibility.
2960 if (var->base == 0xffff0000 && var->selector == 0xf000)
2961 vmcs_writel(sf->base, 0xf0000);
2964 ar = vmx_segment_access_rights(var);
2967 * Fix the "Accessed" bit in AR field of segment registers for older
2969 * IA32 arch specifies that at the time of processor reset the
2970 * "Accessed" bit in the AR field of segment registers is 1. And qemu
2971 * is setting it to 0 in the usedland code. This causes invalid guest
2972 * state vmexit when "unrestricted guest" mode is turned on.
2973 * Fix for this setup issue in cpu_reset is being pushed in the qemu
2974 * tree. Newer qemu binaries with that qemu fix would not need this
2977 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
2978 ar |= 0x1; /* Accessed */
2980 vmcs_write32(sf->ar_bytes, ar);
2981 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
2984 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
2986 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
2988 *db = (ar >> 14) & 1;
2989 *l = (ar >> 13) & 1;
2992 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
2994 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
2995 dt->address = vmcs_readl(GUEST_IDTR_BASE);
2998 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3000 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3001 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3004 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3006 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3007 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3010 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3012 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3013 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3016 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3018 struct kvm_segment var;
3021 vmx_get_segment(vcpu, &var, seg);
3022 ar = vmx_segment_access_rights(&var);
3024 if (var.base != (var.selector << 4))
3026 if (var.limit != 0xffff)
3034 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3036 struct kvm_segment cs;
3037 unsigned int cs_rpl;
3039 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3040 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3044 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3048 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3049 if (cs.dpl > cs_rpl)
3052 if (cs.dpl != cs_rpl)
3058 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3062 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3064 struct kvm_segment ss;
3065 unsigned int ss_rpl;
3067 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3068 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3072 if (ss.type != 3 && ss.type != 7)
3076 if (ss.dpl != ss_rpl) /* DPL != RPL */
3084 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3086 struct kvm_segment var;
3089 vmx_get_segment(vcpu, &var, seg);
3090 rpl = var.selector & SELECTOR_RPL_MASK;
3098 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3099 if (var.dpl < rpl) /* DPL < RPL */
3103 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3109 static bool tr_valid(struct kvm_vcpu *vcpu)
3111 struct kvm_segment tr;
3113 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3117 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3119 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3127 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3129 struct kvm_segment ldtr;
3131 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3135 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3145 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3147 struct kvm_segment cs, ss;
3149 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3150 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3152 return ((cs.selector & SELECTOR_RPL_MASK) ==
3153 (ss.selector & SELECTOR_RPL_MASK));
3157 * Check if guest state is valid. Returns true if valid, false if
3159 * We assume that registers are always usable
3161 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3163 /* real mode guest state checks */
3164 if (!is_protmode(vcpu)) {
3165 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3167 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3169 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3171 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3173 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3175 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3178 /* protected mode guest state checks */
3179 if (!cs_ss_rpl_check(vcpu))
3181 if (!code_segment_valid(vcpu))
3183 if (!stack_segment_valid(vcpu))
3185 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3187 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3189 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3191 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3193 if (!tr_valid(vcpu))
3195 if (!ldtr_valid(vcpu))
3199 * - Add checks on RIP
3200 * - Add checks on RFLAGS
3206 static int init_rmode_tss(struct kvm *kvm)
3210 int r, idx, ret = 0;
3212 idx = srcu_read_lock(&kvm->srcu);
3213 fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
3214 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3217 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3218 r = kvm_write_guest_page(kvm, fn++, &data,
3219 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3222 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3225 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3229 r = kvm_write_guest_page(kvm, fn, &data,
3230 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3237 srcu_read_unlock(&kvm->srcu, idx);
3241 static int init_rmode_identity_map(struct kvm *kvm)
3244 pfn_t identity_map_pfn;
3249 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
3250 printk(KERN_ERR "EPT: identity-mapping pagetable "
3251 "haven't been allocated!\n");
3254 if (likely(kvm->arch.ept_identity_pagetable_done))
3257 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3258 idx = srcu_read_lock(&kvm->srcu);
3259 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3262 /* Set up identity-mapping pagetable for EPT in real mode */
3263 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3264 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3265 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3266 r = kvm_write_guest_page(kvm, identity_map_pfn,
3267 &tmp, i * sizeof(tmp), sizeof(tmp));
3271 kvm->arch.ept_identity_pagetable_done = true;
3274 srcu_read_unlock(&kvm->srcu, idx);
3278 static void seg_setup(int seg)
3280 struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3283 vmcs_write16(sf->selector, 0);
3284 vmcs_writel(sf->base, 0);
3285 vmcs_write32(sf->limit, 0xffff);
3286 if (enable_unrestricted_guest) {
3288 if (seg == VCPU_SREG_CS)
3289 ar |= 0x08; /* code segment */
3293 vmcs_write32(sf->ar_bytes, ar);
3296 static int alloc_apic_access_page(struct kvm *kvm)
3298 struct kvm_userspace_memory_region kvm_userspace_mem;
3301 mutex_lock(&kvm->slots_lock);
3302 if (kvm->arch.apic_access_page)
3304 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3305 kvm_userspace_mem.flags = 0;
3306 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
3307 kvm_userspace_mem.memory_size = PAGE_SIZE;
3308 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3312 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
3314 mutex_unlock(&kvm->slots_lock);
3318 static int alloc_identity_pagetable(struct kvm *kvm)
3320 struct kvm_userspace_memory_region kvm_userspace_mem;
3323 mutex_lock(&kvm->slots_lock);
3324 if (kvm->arch.ept_identity_pagetable)
3326 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
3327 kvm_userspace_mem.flags = 0;
3328 kvm_userspace_mem.guest_phys_addr =
3329 kvm->arch.ept_identity_map_addr;
3330 kvm_userspace_mem.memory_size = PAGE_SIZE;
3331 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3335 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
3336 kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
3338 mutex_unlock(&kvm->slots_lock);
3342 static void allocate_vpid(struct vcpu_vmx *vmx)
3349 spin_lock(&vmx_vpid_lock);
3350 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3351 if (vpid < VMX_NR_VPIDS) {
3353 __set_bit(vpid, vmx_vpid_bitmap);
3355 spin_unlock(&vmx_vpid_lock);
3358 static void free_vpid(struct vcpu_vmx *vmx)
3362 spin_lock(&vmx_vpid_lock);
3364 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3365 spin_unlock(&vmx_vpid_lock);
3368 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
3370 int f = sizeof(unsigned long);
3372 if (!cpu_has_vmx_msr_bitmap())
3376 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3377 * have the write-low and read-high bitmap offsets the wrong way round.
3378 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3380 if (msr <= 0x1fff) {
3381 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
3382 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
3383 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3385 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
3386 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
3390 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
3393 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
3394 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
3398 * Sets up the vmcs for emulated real mode.
3400 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
3402 u32 host_sysenter_cs, msr_low, msr_high;
3408 unsigned long kvm_vmx_return;
3412 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
3413 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
3415 if (cpu_has_vmx_msr_bitmap())
3416 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
3418 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
3421 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
3422 vmcs_config.pin_based_exec_ctrl);
3424 exec_control = vmcs_config.cpu_based_exec_ctrl;
3425 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
3426 exec_control &= ~CPU_BASED_TPR_SHADOW;
3427 #ifdef CONFIG_X86_64
3428 exec_control |= CPU_BASED_CR8_STORE_EXITING |
3429 CPU_BASED_CR8_LOAD_EXITING;
3433 exec_control |= CPU_BASED_CR3_STORE_EXITING |
3434 CPU_BASED_CR3_LOAD_EXITING |
3435 CPU_BASED_INVLPG_EXITING;
3436 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
3438 if (cpu_has_secondary_exec_ctrls()) {
3439 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3440 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3442 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3444 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3446 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3447 enable_unrestricted_guest = 0;
3449 if (!enable_unrestricted_guest)
3450 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3452 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3453 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
3457 vmcs_write32(PLE_GAP, ple_gap);
3458 vmcs_write32(PLE_WINDOW, ple_window);
3461 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, !!bypass_guest_pf);
3462 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, !!bypass_guest_pf);
3463 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
3465 vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS); /* 22.2.3 */
3466 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
3467 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
3469 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
3470 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3471 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3472 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
3473 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
3474 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3475 #ifdef CONFIG_X86_64
3476 rdmsrl(MSR_FS_BASE, a);
3477 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
3478 rdmsrl(MSR_GS_BASE, a);
3479 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
3481 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
3482 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
3485 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
3487 native_store_idt(&dt);
3488 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
3490 asm("mov $.Lkvm_vmx_return, %0" : "=r"(kvm_vmx_return));
3491 vmcs_writel(HOST_RIP, kvm_vmx_return); /* 22.2.5 */
3492 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
3493 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3494 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
3495 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3496 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
3498 rdmsr(MSR_IA32_SYSENTER_CS, host_sysenter_cs, junk);
3499 vmcs_write32(HOST_IA32_SYSENTER_CS, host_sysenter_cs);
3500 rdmsrl(MSR_IA32_SYSENTER_ESP, a);
3501 vmcs_writel(HOST_IA32_SYSENTER_ESP, a); /* 22.2.3 */
3502 rdmsrl(MSR_IA32_SYSENTER_EIP, a);
3503 vmcs_writel(HOST_IA32_SYSENTER_EIP, a); /* 22.2.3 */
3505 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3506 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
3507 host_pat = msr_low | ((u64) msr_high << 32);
3508 vmcs_write64(HOST_IA32_PAT, host_pat);
3510 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3511 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
3512 host_pat = msr_low | ((u64) msr_high << 32);
3513 /* Write the default value follow host pat */
3514 vmcs_write64(GUEST_IA32_PAT, host_pat);
3515 /* Keep arch.pat sync with GUEST_IA32_PAT */
3516 vmx->vcpu.arch.pat = host_pat;
3519 for (i = 0; i < NR_VMX_MSR; ++i) {
3520 u32 index = vmx_msr_index[i];
3521 u32 data_low, data_high;
3524 if (rdmsr_safe(index, &data_low, &data_high) < 0)
3526 if (wrmsr_safe(index, data_low, data_high) < 0)
3528 vmx->guest_msrs[j].index = i;
3529 vmx->guest_msrs[j].data = 0;
3530 vmx->guest_msrs[j].mask = -1ull;
3534 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
3536 /* 22.2.1, 20.8.1 */
3537 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
3539 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
3540 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3542 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3543 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3545 kvm_write_tsc(&vmx->vcpu, 0);
3550 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
3552 struct vcpu_vmx *vmx = to_vmx(vcpu);
3556 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3558 vmx->rmode.vm86_active = 0;
3560 vmx->soft_vnmi_blocked = 0;
3562 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
3563 kvm_set_cr8(&vmx->vcpu, 0);
3564 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
3565 if (kvm_vcpu_is_bsp(&vmx->vcpu))
3566 msr |= MSR_IA32_APICBASE_BSP;
3567 kvm_set_apic_base(&vmx->vcpu, msr);
3569 ret = fx_init(&vmx->vcpu);
3573 vmx_segment_cache_clear(vmx);
3575 seg_setup(VCPU_SREG_CS);
3577 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
3578 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
3580 if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
3581 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
3582 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
3584 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
3585 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
3588 seg_setup(VCPU_SREG_DS);
3589 seg_setup(VCPU_SREG_ES);
3590 seg_setup(VCPU_SREG_FS);
3591 seg_setup(VCPU_SREG_GS);
3592 seg_setup(VCPU_SREG_SS);
3594 vmcs_write16(GUEST_TR_SELECTOR, 0);
3595 vmcs_writel(GUEST_TR_BASE, 0);
3596 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
3597 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3599 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
3600 vmcs_writel(GUEST_LDTR_BASE, 0);
3601 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
3602 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
3604 vmcs_write32(GUEST_SYSENTER_CS, 0);
3605 vmcs_writel(GUEST_SYSENTER_ESP, 0);
3606 vmcs_writel(GUEST_SYSENTER_EIP, 0);
3608 vmcs_writel(GUEST_RFLAGS, 0x02);
3609 if (kvm_vcpu_is_bsp(&vmx->vcpu))
3610 kvm_rip_write(vcpu, 0xfff0);
3612 kvm_rip_write(vcpu, 0);
3613 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
3615 vmcs_writel(GUEST_DR7, 0x400);
3617 vmcs_writel(GUEST_GDTR_BASE, 0);
3618 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
3620 vmcs_writel(GUEST_IDTR_BASE, 0);
3621 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
3623 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
3624 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
3625 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
3627 /* Special registers */
3628 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
3632 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
3634 if (cpu_has_vmx_tpr_shadow()) {
3635 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
3636 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
3637 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
3638 __pa(vmx->vcpu.arch.apic->regs));
3639 vmcs_write32(TPR_THRESHOLD, 0);
3642 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3643 vmcs_write64(APIC_ACCESS_ADDR,
3644 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
3647 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
3649 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
3650 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
3651 vmx_set_cr4(&vmx->vcpu, 0);
3652 vmx_set_efer(&vmx->vcpu, 0);
3653 vmx_fpu_activate(&vmx->vcpu);
3654 update_exception_bitmap(&vmx->vcpu);
3656 vpid_sync_context(vmx);
3660 /* HACK: Don't enable emulation on guest boot/reset */
3661 vmx->emulation_required = 0;
3667 static void enable_irq_window(struct kvm_vcpu *vcpu)
3669 u32 cpu_based_vm_exec_control;
3671 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3672 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
3673 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3676 static void enable_nmi_window(struct kvm_vcpu *vcpu)
3678 u32 cpu_based_vm_exec_control;
3680 if (!cpu_has_virtual_nmis()) {
3681 enable_irq_window(vcpu);
3685 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
3686 enable_irq_window(vcpu);
3689 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
3690 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
3691 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
3694 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
3696 struct vcpu_vmx *vmx = to_vmx(vcpu);
3698 int irq = vcpu->arch.interrupt.nr;
3700 trace_kvm_inj_virq(irq);
3702 ++vcpu->stat.irq_injections;
3703 if (vmx->rmode.vm86_active) {
3705 if (vcpu->arch.interrupt.soft)
3706 inc_eip = vcpu->arch.event_exit_inst_len;
3707 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
3708 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3711 intr = irq | INTR_INFO_VALID_MASK;
3712 if (vcpu->arch.interrupt.soft) {
3713 intr |= INTR_TYPE_SOFT_INTR;
3714 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
3715 vmx->vcpu.arch.event_exit_inst_len);
3717 intr |= INTR_TYPE_EXT_INTR;
3718 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
3719 vmx_clear_hlt(vcpu);
3722 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
3724 struct vcpu_vmx *vmx = to_vmx(vcpu);
3726 if (!cpu_has_virtual_nmis()) {
3728 * Tracking the NMI-blocked state in software is built upon
3729 * finding the next open IRQ window. This, in turn, depends on
3730 * well-behaving guests: They have to keep IRQs disabled at
3731 * least as long as the NMI handler runs. Otherwise we may
3732 * cause NMI nesting, maybe breaking the guest. But as this is
3733 * highly unlikely, we can live with the residual risk.
3735 vmx->soft_vnmi_blocked = 1;
3736 vmx->vnmi_blocked_time = 0;
3739 ++vcpu->stat.nmi_injections;
3740 vmx->nmi_known_unmasked = false;
3741 if (vmx->rmode.vm86_active) {
3742 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
3743 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
3746 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
3747 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
3748 vmx_clear_hlt(vcpu);
3751 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
3753 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
3756 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
3757 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
3758 | GUEST_INTR_STATE_NMI));
3761 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
3763 if (!cpu_has_virtual_nmis())
3764 return to_vmx(vcpu)->soft_vnmi_blocked;
3765 if (to_vmx(vcpu)->nmi_known_unmasked)
3767 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
3770 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
3772 struct vcpu_vmx *vmx = to_vmx(vcpu);
3774 if (!cpu_has_virtual_nmis()) {
3775 if (vmx->soft_vnmi_blocked != masked) {
3776 vmx->soft_vnmi_blocked = masked;
3777 vmx->vnmi_blocked_time = 0;
3780 vmx->nmi_known_unmasked = !masked;
3782 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
3783 GUEST_INTR_STATE_NMI);
3785 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
3786 GUEST_INTR_STATE_NMI);
3790 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
3792 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
3793 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
3794 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
3797 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
3800 struct kvm_userspace_memory_region tss_mem = {
3801 .slot = TSS_PRIVATE_MEMSLOT,
3802 .guest_phys_addr = addr,
3803 .memory_size = PAGE_SIZE * 3,
3807 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
3810 kvm->arch.tss_addr = addr;
3811 if (!init_rmode_tss(kvm))
3817 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
3818 int vec, u32 err_code)
3821 * Instruction with address size override prefix opcode 0x67
3822 * Cause the #SS fault with 0 error code in VM86 mode.
3824 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
3825 if (emulate_instruction(vcpu, 0) == EMULATE_DONE)
3828 * Forward all other exceptions that are valid in real mode.
3829 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
3830 * the required debugging infrastructure rework.
3834 if (vcpu->guest_debug &
3835 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
3837 kvm_queue_exception(vcpu, vec);
3841 * Update instruction length as we may reinject the exception
3842 * from user space while in guest debugging mode.
3844 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
3845 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3846 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
3857 kvm_queue_exception(vcpu, vec);
3864 * Trigger machine check on the host. We assume all the MSRs are already set up
3865 * by the CPU and that we still run on the same CPU as the MCE occurred on.
3866 * We pass a fake environment to the machine check handler because we want
3867 * the guest to be always treated like user space, no matter what context
3868 * it used internally.
3870 static void kvm_machine_check(void)
3872 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
3873 struct pt_regs regs = {
3874 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
3875 .flags = X86_EFLAGS_IF,
3878 do_machine_check(®s, 0);
3882 static int handle_machine_check(struct kvm_vcpu *vcpu)
3884 /* already handled by vcpu_run */
3888 static int handle_exception(struct kvm_vcpu *vcpu)
3890 struct vcpu_vmx *vmx = to_vmx(vcpu);
3891 struct kvm_run *kvm_run = vcpu->run;
3892 u32 intr_info, ex_no, error_code;
3893 unsigned long cr2, rip, dr6;
3895 enum emulation_result er;
3897 vect_info = vmx->idt_vectoring_info;
3898 intr_info = vmx->exit_intr_info;
3900 if (is_machine_check(intr_info))
3901 return handle_machine_check(vcpu);
3903 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
3904 !is_page_fault(intr_info)) {
3905 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
3906 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
3907 vcpu->run->internal.ndata = 2;
3908 vcpu->run->internal.data[0] = vect_info;
3909 vcpu->run->internal.data[1] = intr_info;
3913 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
3914 return 1; /* already handled by vmx_vcpu_run() */
3916 if (is_no_device(intr_info)) {
3917 vmx_fpu_activate(vcpu);
3921 if (is_invalid_opcode(intr_info)) {
3922 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
3923 if (er != EMULATE_DONE)
3924 kvm_queue_exception(vcpu, UD_VECTOR);
3929 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
3930 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
3931 if (is_page_fault(intr_info)) {
3932 /* EPT won't cause page fault directly */
3935 cr2 = vmcs_readl(EXIT_QUALIFICATION);
3936 trace_kvm_page_fault(cr2, error_code);
3938 if (kvm_event_needs_reinjection(vcpu))
3939 kvm_mmu_unprotect_page_virt(vcpu, cr2);
3940 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
3943 if (vmx->rmode.vm86_active &&
3944 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
3946 if (vcpu->arch.halt_request) {
3947 vcpu->arch.halt_request = 0;
3948 return kvm_emulate_halt(vcpu);
3953 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
3956 dr6 = vmcs_readl(EXIT_QUALIFICATION);
3957 if (!(vcpu->guest_debug &
3958 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
3959 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
3960 kvm_queue_exception(vcpu, DB_VECTOR);
3963 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
3964 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
3968 * Update instruction length as we may reinject #BP from
3969 * user space while in guest debugging mode. Reading it for
3970 * #DB as well causes no harm, it is not used in that case.
3972 vmx->vcpu.arch.event_exit_inst_len =
3973 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
3974 kvm_run->exit_reason = KVM_EXIT_DEBUG;
3975 rip = kvm_rip_read(vcpu);
3976 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
3977 kvm_run->debug.arch.exception = ex_no;
3980 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
3981 kvm_run->ex.exception = ex_no;
3982 kvm_run->ex.error_code = error_code;
3988 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
3990 ++vcpu->stat.irq_exits;
3994 static int handle_triple_fault(struct kvm_vcpu *vcpu)
3996 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4000 static int handle_io(struct kvm_vcpu *vcpu)
4002 unsigned long exit_qualification;
4003 int size, in, string;
4006 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4007 string = (exit_qualification & 16) != 0;
4008 in = (exit_qualification & 8) != 0;
4010 ++vcpu->stat.io_exits;
4013 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4015 port = exit_qualification >> 16;
4016 size = (exit_qualification & 7) + 1;
4017 skip_emulated_instruction(vcpu);
4019 return kvm_fast_pio_out(vcpu, size, port);
4023 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4026 * Patch in the VMCALL instruction:
4028 hypercall[0] = 0x0f;
4029 hypercall[1] = 0x01;
4030 hypercall[2] = 0xc1;
4033 static int handle_cr(struct kvm_vcpu *vcpu)
4035 unsigned long exit_qualification, val;
4040 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4041 cr = exit_qualification & 15;
4042 reg = (exit_qualification >> 8) & 15;
4043 switch ((exit_qualification >> 4) & 3) {
4044 case 0: /* mov to cr */
4045 val = kvm_register_read(vcpu, reg);
4046 trace_kvm_cr_write(cr, val);
4049 err = kvm_set_cr0(vcpu, val);
4050 kvm_complete_insn_gp(vcpu, err);
4053 err = kvm_set_cr3(vcpu, val);
4054 kvm_complete_insn_gp(vcpu, err);
4057 err = kvm_set_cr4(vcpu, val);
4058 kvm_complete_insn_gp(vcpu, err);
4061 u8 cr8_prev = kvm_get_cr8(vcpu);
4062 u8 cr8 = kvm_register_read(vcpu, reg);
4063 err = kvm_set_cr8(vcpu, cr8);
4064 kvm_complete_insn_gp(vcpu, err);
4065 if (irqchip_in_kernel(vcpu->kvm))
4067 if (cr8_prev <= cr8)
4069 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4075 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4076 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4077 skip_emulated_instruction(vcpu);
4078 vmx_fpu_activate(vcpu);
4080 case 1: /*mov from cr*/
4083 val = kvm_read_cr3(vcpu);
4084 kvm_register_write(vcpu, reg, val);
4085 trace_kvm_cr_read(cr, val);
4086 skip_emulated_instruction(vcpu);
4089 val = kvm_get_cr8(vcpu);
4090 kvm_register_write(vcpu, reg, val);
4091 trace_kvm_cr_read(cr, val);
4092 skip_emulated_instruction(vcpu);
4097 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4098 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4099 kvm_lmsw(vcpu, val);
4101 skip_emulated_instruction(vcpu);
4106 vcpu->run->exit_reason = 0;
4107 pr_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4108 (int)(exit_qualification >> 4) & 3, cr);
4112 static int handle_dr(struct kvm_vcpu *vcpu)
4114 unsigned long exit_qualification;
4117 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
4118 if (!kvm_require_cpl(vcpu, 0))
4120 dr = vmcs_readl(GUEST_DR7);
4123 * As the vm-exit takes precedence over the debug trap, we
4124 * need to emulate the latter, either for the host or the
4125 * guest debugging itself.
4127 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4128 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4129 vcpu->run->debug.arch.dr7 = dr;
4130 vcpu->run->debug.arch.pc =
4131 vmcs_readl(GUEST_CS_BASE) +
4132 vmcs_readl(GUEST_RIP);
4133 vcpu->run->debug.arch.exception = DB_VECTOR;
4134 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4137 vcpu->arch.dr7 &= ~DR7_GD;
4138 vcpu->arch.dr6 |= DR6_BD;
4139 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
4140 kvm_queue_exception(vcpu, DB_VECTOR);
4145 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4146 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4147 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4148 if (exit_qualification & TYPE_MOV_FROM_DR) {
4150 if (!kvm_get_dr(vcpu, dr, &val))
4151 kvm_register_write(vcpu, reg, val);
4153 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
4154 skip_emulated_instruction(vcpu);
4158 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4160 vmcs_writel(GUEST_DR7, val);
4163 static int handle_cpuid(struct kvm_vcpu *vcpu)
4165 kvm_emulate_cpuid(vcpu);
4169 static int handle_rdmsr(struct kvm_vcpu *vcpu)
4171 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4174 if (vmx_get_msr(vcpu, ecx, &data)) {
4175 trace_kvm_msr_read_ex(ecx);
4176 kvm_inject_gp(vcpu, 0);
4180 trace_kvm_msr_read(ecx, data);
4182 /* FIXME: handling of bits 32:63 of rax, rdx */
4183 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
4184 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
4185 skip_emulated_instruction(vcpu);
4189 static int handle_wrmsr(struct kvm_vcpu *vcpu)
4191 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4192 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
4193 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
4195 if (vmx_set_msr(vcpu, ecx, data) != 0) {
4196 trace_kvm_msr_write_ex(ecx, data);
4197 kvm_inject_gp(vcpu, 0);
4201 trace_kvm_msr_write(ecx, data);
4202 skip_emulated_instruction(vcpu);
4206 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4208 kvm_make_request(KVM_REQ_EVENT, vcpu);
4212 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4214 u32 cpu_based_vm_exec_control;
4216 /* clear pending irq */
4217 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4218 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
4219 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4221 kvm_make_request(KVM_REQ_EVENT, vcpu);
4223 ++vcpu->stat.irq_window_exits;
4226 * If the user space waits to inject interrupts, exit as soon as
4229 if (!irqchip_in_kernel(vcpu->kvm) &&
4230 vcpu->run->request_interrupt_window &&
4231 !kvm_cpu_has_interrupt(vcpu)) {
4232 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
4238 static int handle_halt(struct kvm_vcpu *vcpu)
4240 skip_emulated_instruction(vcpu);
4241 return kvm_emulate_halt(vcpu);
4244 static int handle_vmcall(struct kvm_vcpu *vcpu)
4246 skip_emulated_instruction(vcpu);
4247 kvm_emulate_hypercall(vcpu);
4251 static int handle_vmx_insn(struct kvm_vcpu *vcpu)
4253 kvm_queue_exception(vcpu, UD_VECTOR);
4257 static int handle_invd(struct kvm_vcpu *vcpu)
4259 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4262 static int handle_invlpg(struct kvm_vcpu *vcpu)
4264 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4266 kvm_mmu_invlpg(vcpu, exit_qualification);
4267 skip_emulated_instruction(vcpu);
4271 static int handle_wbinvd(struct kvm_vcpu *vcpu)
4273 skip_emulated_instruction(vcpu);
4274 kvm_emulate_wbinvd(vcpu);
4278 static int handle_xsetbv(struct kvm_vcpu *vcpu)
4280 u64 new_bv = kvm_read_edx_eax(vcpu);
4281 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4283 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4284 skip_emulated_instruction(vcpu);
4288 static int handle_apic_access(struct kvm_vcpu *vcpu)
4290 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4293 static int handle_task_switch(struct kvm_vcpu *vcpu)
4295 struct vcpu_vmx *vmx = to_vmx(vcpu);
4296 unsigned long exit_qualification;
4297 bool has_error_code = false;
4300 int reason, type, idt_v;
4302 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
4303 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
4305 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4307 reason = (u32)exit_qualification >> 30;
4308 if (reason == TASK_SWITCH_GATE && idt_v) {
4310 case INTR_TYPE_NMI_INTR:
4311 vcpu->arch.nmi_injected = false;
4312 vmx_set_nmi_mask(vcpu, true);
4314 case INTR_TYPE_EXT_INTR:
4315 case INTR_TYPE_SOFT_INTR:
4316 kvm_clear_interrupt_queue(vcpu);
4318 case INTR_TYPE_HARD_EXCEPTION:
4319 if (vmx->idt_vectoring_info &
4320 VECTORING_INFO_DELIVER_CODE_MASK) {
4321 has_error_code = true;
4323 vmcs_read32(IDT_VECTORING_ERROR_CODE);
4326 case INTR_TYPE_SOFT_EXCEPTION:
4327 kvm_clear_exception_queue(vcpu);
4333 tss_selector = exit_qualification;
4335 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
4336 type != INTR_TYPE_EXT_INTR &&
4337 type != INTR_TYPE_NMI_INTR))
4338 skip_emulated_instruction(vcpu);
4340 if (kvm_task_switch(vcpu, tss_selector, reason,
4341 has_error_code, error_code) == EMULATE_FAIL) {
4342 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4343 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4344 vcpu->run->internal.ndata = 0;
4348 /* clear all local breakpoint enable flags */
4349 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
4352 * TODO: What about debug traps on tss switch?
4353 * Are we supposed to inject them and update dr6?
4359 static int handle_ept_violation(struct kvm_vcpu *vcpu)
4361 unsigned long exit_qualification;
4365 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4367 if (exit_qualification & (1 << 6)) {
4368 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
4372 gla_validity = (exit_qualification >> 7) & 0x3;
4373 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
4374 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
4375 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
4376 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
4377 vmcs_readl(GUEST_LINEAR_ADDRESS));
4378 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
4379 (long unsigned int)exit_qualification);
4380 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4381 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
4385 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4386 trace_kvm_page_fault(gpa, exit_qualification);
4387 return kvm_mmu_page_fault(vcpu, gpa, exit_qualification & 0x3, NULL, 0);
4390 static u64 ept_rsvd_mask(u64 spte, int level)
4395 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
4396 mask |= (1ULL << i);
4399 /* bits 7:3 reserved */
4401 else if (level == 2) {
4402 if (spte & (1ULL << 7))
4403 /* 2MB ref, bits 20:12 reserved */
4406 /* bits 6:3 reserved */
4413 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
4416 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
4418 /* 010b (write-only) */
4419 WARN_ON((spte & 0x7) == 0x2);
4421 /* 110b (write/execute) */
4422 WARN_ON((spte & 0x7) == 0x6);
4424 /* 100b (execute-only) and value not supported by logical processor */
4425 if (!cpu_has_vmx_ept_execute_only())
4426 WARN_ON((spte & 0x7) == 0x4);
4430 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
4432 if (rsvd_bits != 0) {
4433 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
4434 __func__, rsvd_bits);
4438 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
4439 u64 ept_mem_type = (spte & 0x38) >> 3;
4441 if (ept_mem_type == 2 || ept_mem_type == 3 ||
4442 ept_mem_type == 7) {
4443 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
4444 __func__, ept_mem_type);
4451 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
4457 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4459 printk(KERN_ERR "EPT: Misconfiguration.\n");
4460 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
4462 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
4464 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
4465 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
4467 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4468 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
4473 static int handle_nmi_window(struct kvm_vcpu *vcpu)
4475 u32 cpu_based_vm_exec_control;
4477 /* clear pending NMI */
4478 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4479 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
4480 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4481 ++vcpu->stat.nmi_window_exits;
4482 kvm_make_request(KVM_REQ_EVENT, vcpu);
4487 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
4489 struct vcpu_vmx *vmx = to_vmx(vcpu);
4490 enum emulation_result err = EMULATE_DONE;
4493 bool intr_window_requested;
4495 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4496 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
4498 while (!guest_state_valid(vcpu)) {
4499 if (intr_window_requested
4500 && (kvm_get_rflags(&vmx->vcpu) & X86_EFLAGS_IF))
4501 return handle_interrupt_window(&vmx->vcpu);
4503 err = emulate_instruction(vcpu, 0);
4505 if (err == EMULATE_DO_MMIO) {
4510 if (err != EMULATE_DONE)
4513 if (signal_pending(current))
4519 vmx->emulation_required = 0;
4525 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
4526 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
4528 static int handle_pause(struct kvm_vcpu *vcpu)
4530 skip_emulated_instruction(vcpu);
4531 kvm_vcpu_on_spin(vcpu);
4536 static int handle_invalid_op(struct kvm_vcpu *vcpu)
4538 kvm_queue_exception(vcpu, UD_VECTOR);
4543 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
4544 * We could reuse a single VMCS for all the L2 guests, but we also want the
4545 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
4546 * allows keeping them loaded on the processor, and in the future will allow
4547 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
4548 * every entry if they never change.
4549 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
4550 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
4552 * The following functions allocate and free a vmcs02 in this pool.
4555 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
4556 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
4558 struct vmcs02_list *item;
4559 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
4560 if (item->vmptr == vmx->nested.current_vmptr) {
4561 list_move(&item->list, &vmx->nested.vmcs02_pool);
4562 return &item->vmcs02;
4565 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
4566 /* Recycle the least recently used VMCS. */
4567 item = list_entry(vmx->nested.vmcs02_pool.prev,
4568 struct vmcs02_list, list);
4569 item->vmptr = vmx->nested.current_vmptr;
4570 list_move(&item->list, &vmx->nested.vmcs02_pool);
4571 return &item->vmcs02;
4574 /* Create a new VMCS */
4575 item = (struct vmcs02_list *)
4576 kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
4579 item->vmcs02.vmcs = alloc_vmcs();
4580 if (!item->vmcs02.vmcs) {
4584 loaded_vmcs_init(&item->vmcs02);
4585 item->vmptr = vmx->nested.current_vmptr;
4586 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
4587 vmx->nested.vmcs02_num++;
4588 return &item->vmcs02;
4591 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
4592 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
4594 struct vmcs02_list *item;
4595 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
4596 if (item->vmptr == vmptr) {
4597 free_loaded_vmcs(&item->vmcs02);
4598 list_del(&item->list);
4600 vmx->nested.vmcs02_num--;
4606 * Free all VMCSs saved for this vcpu, except the one pointed by
4607 * vmx->loaded_vmcs. These include the VMCSs in vmcs02_pool (except the one
4608 * currently used, if running L2), and vmcs01 when running L2.
4610 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
4612 struct vmcs02_list *item, *n;
4613 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
4614 if (vmx->loaded_vmcs != &item->vmcs02)
4615 free_loaded_vmcs(&item->vmcs02);
4616 list_del(&item->list);
4619 vmx->nested.vmcs02_num = 0;
4621 if (vmx->loaded_vmcs != &vmx->vmcs01)
4622 free_loaded_vmcs(&vmx->vmcs01);
4626 * Emulate the VMXON instruction.
4627 * Currently, we just remember that VMX is active, and do not save or even
4628 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
4629 * do not currently need to store anything in that guest-allocated memory
4630 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
4631 * argument is different from the VMXON pointer (which the spec says they do).
4633 static int handle_vmon(struct kvm_vcpu *vcpu)
4635 struct kvm_segment cs;
4636 struct vcpu_vmx *vmx = to_vmx(vcpu);
4638 /* The Intel VMX Instruction Reference lists a bunch of bits that
4639 * are prerequisite to running VMXON, most notably cr4.VMXE must be
4640 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
4641 * Otherwise, we should fail with #UD. We test these now:
4643 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
4644 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
4645 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4646 kvm_queue_exception(vcpu, UD_VECTOR);
4650 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4651 if (is_long_mode(vcpu) && !cs.l) {
4652 kvm_queue_exception(vcpu, UD_VECTOR);
4656 if (vmx_get_cpl(vcpu)) {
4657 kvm_inject_gp(vcpu, 0);
4661 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
4662 vmx->nested.vmcs02_num = 0;
4664 vmx->nested.vmxon = true;
4666 skip_emulated_instruction(vcpu);
4671 * Intel's VMX Instruction Reference specifies a common set of prerequisites
4672 * for running VMX instructions (except VMXON, whose prerequisites are
4673 * slightly different). It also specifies what exception to inject otherwise.
4675 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
4677 struct kvm_segment cs;
4678 struct vcpu_vmx *vmx = to_vmx(vcpu);
4680 if (!vmx->nested.vmxon) {
4681 kvm_queue_exception(vcpu, UD_VECTOR);
4685 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4686 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
4687 (is_long_mode(vcpu) && !cs.l)) {
4688 kvm_queue_exception(vcpu, UD_VECTOR);
4692 if (vmx_get_cpl(vcpu)) {
4693 kvm_inject_gp(vcpu, 0);
4701 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
4702 * just stops using VMX.
4704 static void free_nested(struct vcpu_vmx *vmx)
4706 if (!vmx->nested.vmxon)
4708 vmx->nested.vmxon = false;
4709 if (vmx->nested.current_vmptr != -1ull) {
4710 kunmap(vmx->nested.current_vmcs12_page);
4711 nested_release_page(vmx->nested.current_vmcs12_page);
4712 vmx->nested.current_vmptr = -1ull;
4713 vmx->nested.current_vmcs12 = NULL;
4716 nested_free_all_saved_vmcss(vmx);
4719 /* Emulate the VMXOFF instruction */
4720 static int handle_vmoff(struct kvm_vcpu *vcpu)
4722 if (!nested_vmx_check_permission(vcpu))
4724 free_nested(to_vmx(vcpu));
4725 skip_emulated_instruction(vcpu);
4730 * Decode the memory-address operand of a vmx instruction, as recorded on an
4731 * exit caused by such an instruction (run by a guest hypervisor).
4732 * On success, returns 0. When the operand is invalid, returns 1 and throws
4735 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
4736 unsigned long exit_qualification,
4737 u32 vmx_instruction_info, gva_t *ret)
4740 * According to Vol. 3B, "Information for VM Exits Due to Instruction
4741 * Execution", on an exit, vmx_instruction_info holds most of the
4742 * addressing components of the operand. Only the displacement part
4743 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
4744 * For how an actual address is calculated from all these components,
4745 * refer to Vol. 1, "Operand Addressing".
4747 int scaling = vmx_instruction_info & 3;
4748 int addr_size = (vmx_instruction_info >> 7) & 7;
4749 bool is_reg = vmx_instruction_info & (1u << 10);
4750 int seg_reg = (vmx_instruction_info >> 15) & 7;
4751 int index_reg = (vmx_instruction_info >> 18) & 0xf;
4752 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
4753 int base_reg = (vmx_instruction_info >> 23) & 0xf;
4754 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
4757 kvm_queue_exception(vcpu, UD_VECTOR);
4761 /* Addr = segment_base + offset */
4762 /* offset = base + [index * scale] + displacement */
4763 *ret = vmx_get_segment_base(vcpu, seg_reg);
4765 *ret += kvm_register_read(vcpu, base_reg);
4767 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
4768 *ret += exit_qualification; /* holds the displacement */
4770 if (addr_size == 1) /* 32 bit */
4774 * TODO: throw #GP (and return 1) in various cases that the VM*
4775 * instructions require it - e.g., offset beyond segment limit,
4776 * unusable or unreadable/unwritable segment, non-canonical 64-bit
4777 * address, and so on. Currently these are not checked.
4783 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
4784 * set the success or error code of an emulated VMX instruction, as specified
4785 * by Vol 2B, VMX Instruction Reference, "Conventions".
4787 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
4789 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
4790 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
4791 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
4794 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
4796 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
4797 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
4798 X86_EFLAGS_SF | X86_EFLAGS_OF))
4802 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
4803 u32 vm_instruction_error)
4805 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
4807 * failValid writes the error number to the current VMCS, which
4808 * can't be done there isn't a current VMCS.
4810 nested_vmx_failInvalid(vcpu);
4813 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
4814 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
4815 X86_EFLAGS_SF | X86_EFLAGS_OF))
4817 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
4820 /* Emulate the VMCLEAR instruction */
4821 static int handle_vmclear(struct kvm_vcpu *vcpu)
4823 struct vcpu_vmx *vmx = to_vmx(vcpu);
4826 struct vmcs12 *vmcs12;
4828 struct x86_exception e;
4830 if (!nested_vmx_check_permission(vcpu))
4833 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
4834 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
4837 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
4838 sizeof(vmptr), &e)) {
4839 kvm_inject_page_fault(vcpu, &e);
4843 if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
4844 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
4845 skip_emulated_instruction(vcpu);
4849 if (vmptr == vmx->nested.current_vmptr) {
4850 kunmap(vmx->nested.current_vmcs12_page);
4851 nested_release_page(vmx->nested.current_vmcs12_page);
4852 vmx->nested.current_vmptr = -1ull;
4853 vmx->nested.current_vmcs12 = NULL;
4856 page = nested_get_page(vcpu, vmptr);
4859 * For accurate processor emulation, VMCLEAR beyond available
4860 * physical memory should do nothing at all. However, it is
4861 * possible that a nested vmx bug, not a guest hypervisor bug,
4862 * resulted in this case, so let's shut down before doing any
4865 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4868 vmcs12 = kmap(page);
4869 vmcs12->launch_state = 0;
4871 nested_release_page(page);
4873 nested_free_vmcs02(vmx, vmptr);
4875 skip_emulated_instruction(vcpu);
4876 nested_vmx_succeed(vcpu);
4880 enum vmcs_field_type {
4881 VMCS_FIELD_TYPE_U16 = 0,
4882 VMCS_FIELD_TYPE_U64 = 1,
4883 VMCS_FIELD_TYPE_U32 = 2,
4884 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
4887 static inline int vmcs_field_type(unsigned long field)
4889 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
4890 return VMCS_FIELD_TYPE_U32;
4891 return (field >> 13) & 0x3 ;
4894 static inline int vmcs_field_readonly(unsigned long field)
4896 return (((field >> 10) & 0x3) == 1);
4900 * Read a vmcs12 field. Since these can have varying lengths and we return
4901 * one type, we chose the biggest type (u64) and zero-extend the return value
4902 * to that size. Note that the caller, handle_vmread, might need to use only
4903 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
4904 * 64-bit fields are to be returned).
4906 static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
4907 unsigned long field, u64 *ret)
4909 short offset = vmcs_field_to_offset(field);
4915 p = ((char *)(get_vmcs12(vcpu))) + offset;
4917 switch (vmcs_field_type(field)) {
4918 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
4919 *ret = *((natural_width *)p);
4921 case VMCS_FIELD_TYPE_U16:
4924 case VMCS_FIELD_TYPE_U32:
4927 case VMCS_FIELD_TYPE_U64:
4931 return 0; /* can never happen. */
4936 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
4937 * used before) all generate the same failure when it is missing.
4939 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
4941 struct vcpu_vmx *vmx = to_vmx(vcpu);
4942 if (vmx->nested.current_vmptr == -1ull) {
4943 nested_vmx_failInvalid(vcpu);
4944 skip_emulated_instruction(vcpu);
4950 static int handle_vmread(struct kvm_vcpu *vcpu)
4952 unsigned long field;
4954 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4955 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
4958 if (!nested_vmx_check_permission(vcpu) ||
4959 !nested_vmx_check_vmcs12(vcpu))
4962 /* Decode instruction info and find the field to read */
4963 field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
4964 /* Read the field, zero-extended to a u64 field_value */
4965 if (!vmcs12_read_any(vcpu, field, &field_value)) {
4966 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
4967 skip_emulated_instruction(vcpu);
4971 * Now copy part of this value to register or memory, as requested.
4972 * Note that the number of bits actually copied is 32 or 64 depending
4973 * on the guest's mode (32 or 64 bit), not on the given field's length.
4975 if (vmx_instruction_info & (1u << 10)) {
4976 kvm_register_write(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
4979 if (get_vmx_mem_address(vcpu, exit_qualification,
4980 vmx_instruction_info, &gva))
4982 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
4983 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
4984 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
4987 nested_vmx_succeed(vcpu);
4988 skip_emulated_instruction(vcpu);
4993 static int handle_vmwrite(struct kvm_vcpu *vcpu)
4995 unsigned long field;
4997 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4998 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5001 /* The value to write might be 32 or 64 bits, depending on L1's long
5002 * mode, and eventually we need to write that into a field of several
5003 * possible lengths. The code below first zero-extends the value to 64
5004 * bit (field_value), and then copies only the approriate number of
5005 * bits into the vmcs12 field.
5007 u64 field_value = 0;
5008 struct x86_exception e;
5010 if (!nested_vmx_check_permission(vcpu) ||
5011 !nested_vmx_check_vmcs12(vcpu))
5014 if (vmx_instruction_info & (1u << 10))
5015 field_value = kvm_register_read(vcpu,
5016 (((vmx_instruction_info) >> 3) & 0xf));
5018 if (get_vmx_mem_address(vcpu, exit_qualification,
5019 vmx_instruction_info, &gva))
5021 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
5022 &field_value, (is_long_mode(vcpu) ? 8 : 4), &e)) {
5023 kvm_inject_page_fault(vcpu, &e);
5029 field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5030 if (vmcs_field_readonly(field)) {
5031 nested_vmx_failValid(vcpu,
5032 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5033 skip_emulated_instruction(vcpu);
5037 offset = vmcs_field_to_offset(field);
5039 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5040 skip_emulated_instruction(vcpu);
5043 p = ((char *) get_vmcs12(vcpu)) + offset;
5045 switch (vmcs_field_type(field)) {
5046 case VMCS_FIELD_TYPE_U16:
5047 *(u16 *)p = field_value;
5049 case VMCS_FIELD_TYPE_U32:
5050 *(u32 *)p = field_value;
5052 case VMCS_FIELD_TYPE_U64:
5053 *(u64 *)p = field_value;
5055 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5056 *(natural_width *)p = field_value;
5059 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5060 skip_emulated_instruction(vcpu);
5064 nested_vmx_succeed(vcpu);
5065 skip_emulated_instruction(vcpu);
5069 /* Emulate the VMPTRLD instruction */
5070 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5072 struct vcpu_vmx *vmx = to_vmx(vcpu);
5075 struct x86_exception e;
5077 if (!nested_vmx_check_permission(vcpu))
5080 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5081 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5084 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5085 sizeof(vmptr), &e)) {
5086 kvm_inject_page_fault(vcpu, &e);
5090 if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5091 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5092 skip_emulated_instruction(vcpu);
5096 if (vmx->nested.current_vmptr != vmptr) {
5097 struct vmcs12 *new_vmcs12;
5099 page = nested_get_page(vcpu, vmptr);
5101 nested_vmx_failInvalid(vcpu);
5102 skip_emulated_instruction(vcpu);
5105 new_vmcs12 = kmap(page);
5106 if (new_vmcs12->revision_id != VMCS12_REVISION) {
5108 nested_release_page_clean(page);
5109 nested_vmx_failValid(vcpu,
5110 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5111 skip_emulated_instruction(vcpu);
5114 if (vmx->nested.current_vmptr != -1ull) {
5115 kunmap(vmx->nested.current_vmcs12_page);
5116 nested_release_page(vmx->nested.current_vmcs12_page);
5119 vmx->nested.current_vmptr = vmptr;
5120 vmx->nested.current_vmcs12 = new_vmcs12;
5121 vmx->nested.current_vmcs12_page = page;
5124 nested_vmx_succeed(vcpu);
5125 skip_emulated_instruction(vcpu);
5129 /* Emulate the VMPTRST instruction */
5130 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5132 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5133 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5135 struct x86_exception e;
5137 if (!nested_vmx_check_permission(vcpu))
5140 if (get_vmx_mem_address(vcpu, exit_qualification,
5141 vmx_instruction_info, &vmcs_gva))
5143 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
5144 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
5145 (void *)&to_vmx(vcpu)->nested.current_vmptr,
5147 kvm_inject_page_fault(vcpu, &e);
5150 nested_vmx_succeed(vcpu);
5151 skip_emulated_instruction(vcpu);
5156 * The exit handlers return 1 if the exit was handled fully and guest execution
5157 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
5158 * to be done to userspace and return 0.
5160 static int (*kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5161 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
5162 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
5163 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
5164 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
5165 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
5166 [EXIT_REASON_CR_ACCESS] = handle_cr,
5167 [EXIT_REASON_DR_ACCESS] = handle_dr,
5168 [EXIT_REASON_CPUID] = handle_cpuid,
5169 [EXIT_REASON_MSR_READ] = handle_rdmsr,
5170 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
5171 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
5172 [EXIT_REASON_HLT] = handle_halt,
5173 [EXIT_REASON_INVD] = handle_invd,
5174 [EXIT_REASON_INVLPG] = handle_invlpg,
5175 [EXIT_REASON_VMCALL] = handle_vmcall,
5176 [EXIT_REASON_VMCLEAR] = handle_vmclear,
5177 [EXIT_REASON_VMLAUNCH] = handle_vmx_insn,
5178 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
5179 [EXIT_REASON_VMPTRST] = handle_vmptrst,
5180 [EXIT_REASON_VMREAD] = handle_vmread,
5181 [EXIT_REASON_VMRESUME] = handle_vmx_insn,
5182 [EXIT_REASON_VMWRITE] = handle_vmwrite,
5183 [EXIT_REASON_VMOFF] = handle_vmoff,
5184 [EXIT_REASON_VMON] = handle_vmon,
5185 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
5186 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
5187 [EXIT_REASON_WBINVD] = handle_wbinvd,
5188 [EXIT_REASON_XSETBV] = handle_xsetbv,
5189 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
5190 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
5191 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
5192 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
5193 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
5194 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_invalid_op,
5195 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_invalid_op,
5198 static const int kvm_vmx_max_exit_handlers =
5199 ARRAY_SIZE(kvm_vmx_exit_handlers);
5201 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5203 *info1 = vmcs_readl(EXIT_QUALIFICATION);
5204 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5208 * The guest has exited. See if we can fix it or if we need userspace
5211 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5213 struct vcpu_vmx *vmx = to_vmx(vcpu);
5214 u32 exit_reason = vmx->exit_reason;
5215 u32 vectoring_info = vmx->idt_vectoring_info;
5217 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
5219 /* If guest state is invalid, start emulating */
5220 if (vmx->emulation_required && emulate_invalid_guest_state)
5221 return handle_invalid_guest_state(vcpu);
5223 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5224 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5225 vcpu->run->fail_entry.hardware_entry_failure_reason
5230 if (unlikely(vmx->fail)) {
5231 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5232 vcpu->run->fail_entry.hardware_entry_failure_reason
5233 = vmcs_read32(VM_INSTRUCTION_ERROR);
5237 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5238 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5239 exit_reason != EXIT_REASON_EPT_VIOLATION &&
5240 exit_reason != EXIT_REASON_TASK_SWITCH))
5241 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
5242 "(0x%x) and exit reason is 0x%x\n",
5243 __func__, vectoring_info, exit_reason);
5245 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked)) {
5246 if (vmx_interrupt_allowed(vcpu)) {
5247 vmx->soft_vnmi_blocked = 0;
5248 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
5249 vcpu->arch.nmi_pending) {
5251 * This CPU don't support us in finding the end of an
5252 * NMI-blocked window if the guest runs with IRQs
5253 * disabled. So we pull the trigger after 1 s of
5254 * futile waiting, but inform the user about this.
5256 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5257 "state on VCPU %d after 1 s timeout\n",
5258 __func__, vcpu->vcpu_id);
5259 vmx->soft_vnmi_blocked = 0;
5263 if (exit_reason < kvm_vmx_max_exit_handlers
5264 && kvm_vmx_exit_handlers[exit_reason])
5265 return kvm_vmx_exit_handlers[exit_reason](vcpu);
5267 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5268 vcpu->run->hw.hardware_exit_reason = exit_reason;
5273 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
5275 if (irr == -1 || tpr < irr) {
5276 vmcs_write32(TPR_THRESHOLD, 0);
5280 vmcs_write32(TPR_THRESHOLD, irr);
5283 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
5287 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
5288 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
5291 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5292 exit_intr_info = vmx->exit_intr_info;
5294 /* Handle machine checks before interrupts are enabled */
5295 if (is_machine_check(exit_intr_info))
5296 kvm_machine_check();
5298 /* We need to handle NMIs before interrupts are enabled */
5299 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
5300 (exit_intr_info & INTR_INFO_VALID_MASK)) {
5301 kvm_before_handle_nmi(&vmx->vcpu);
5303 kvm_after_handle_nmi(&vmx->vcpu);
5307 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
5312 bool idtv_info_valid;
5314 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
5316 if (cpu_has_virtual_nmis()) {
5317 if (vmx->nmi_known_unmasked)
5320 * Can't use vmx->exit_intr_info since we're not sure what
5321 * the exit reason is.
5323 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5324 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
5325 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
5327 * SDM 3: 27.7.1.2 (September 2008)
5328 * Re-set bit "block by NMI" before VM entry if vmexit caused by
5329 * a guest IRET fault.
5330 * SDM 3: 23.2.2 (September 2008)
5331 * Bit 12 is undefined in any of the following cases:
5332 * If the VM exit sets the valid bit in the IDT-vectoring
5333 * information field.
5334 * If the VM exit is due to a double fault.
5336 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
5337 vector != DF_VECTOR && !idtv_info_valid)
5338 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5339 GUEST_INTR_STATE_NMI);
5341 vmx->nmi_known_unmasked =
5342 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
5343 & GUEST_INTR_STATE_NMI);
5344 } else if (unlikely(vmx->soft_vnmi_blocked))
5345 vmx->vnmi_blocked_time +=
5346 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
5349 static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
5350 u32 idt_vectoring_info,
5351 int instr_len_field,
5352 int error_code_field)
5356 bool idtv_info_valid;
5358 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
5360 vmx->vcpu.arch.nmi_injected = false;
5361 kvm_clear_exception_queue(&vmx->vcpu);
5362 kvm_clear_interrupt_queue(&vmx->vcpu);
5364 if (!idtv_info_valid)
5367 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
5369 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
5370 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
5373 case INTR_TYPE_NMI_INTR:
5374 vmx->vcpu.arch.nmi_injected = true;
5376 * SDM 3: 27.7.1.2 (September 2008)
5377 * Clear bit "block by NMI" before VM entry if a NMI
5380 vmx_set_nmi_mask(&vmx->vcpu, false);
5382 case INTR_TYPE_SOFT_EXCEPTION:
5383 vmx->vcpu.arch.event_exit_inst_len =
5384 vmcs_read32(instr_len_field);
5386 case INTR_TYPE_HARD_EXCEPTION:
5387 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
5388 u32 err = vmcs_read32(error_code_field);
5389 kvm_queue_exception_e(&vmx->vcpu, vector, err);
5391 kvm_queue_exception(&vmx->vcpu, vector);
5393 case INTR_TYPE_SOFT_INTR:
5394 vmx->vcpu.arch.event_exit_inst_len =
5395 vmcs_read32(instr_len_field);
5397 case INTR_TYPE_EXT_INTR:
5398 kvm_queue_interrupt(&vmx->vcpu, vector,
5399 type == INTR_TYPE_SOFT_INTR);
5406 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
5408 __vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
5409 VM_EXIT_INSTRUCTION_LEN,
5410 IDT_VECTORING_ERROR_CODE);
5413 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
5415 __vmx_complete_interrupts(to_vmx(vcpu),
5416 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
5417 VM_ENTRY_INSTRUCTION_LEN,
5418 VM_ENTRY_EXCEPTION_ERROR_CODE);
5420 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
5423 #ifdef CONFIG_X86_64
5431 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
5433 struct vcpu_vmx *vmx = to_vmx(vcpu);
5435 /* Record the guest's net vcpu time for enforced NMI injections. */
5436 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
5437 vmx->entry_time = ktime_get();
5439 /* Don't enter VMX if guest state is invalid, let the exit handler
5440 start emulation until we arrive back to a valid state */
5441 if (vmx->emulation_required && emulate_invalid_guest_state)
5444 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
5445 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
5446 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
5447 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
5449 /* When single-stepping over STI and MOV SS, we must clear the
5450 * corresponding interruptibility bits in the guest state. Otherwise
5451 * vmentry fails as it then expects bit 14 (BS) in pending debug
5452 * exceptions being set, but that's not correct for the guest debugging
5454 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
5455 vmx_set_interrupt_shadow(vcpu, 0);
5457 vmx->__launched = vmx->loaded_vmcs->launched;
5459 /* Store host registers */
5460 "push %%"R"dx; push %%"R"bp;"
5461 "push %%"R"cx \n\t" /* placeholder for guest rcx */
5463 "cmp %%"R"sp, %c[host_rsp](%0) \n\t"
5465 "mov %%"R"sp, %c[host_rsp](%0) \n\t"
5466 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
5468 /* Reload cr2 if changed */
5469 "mov %c[cr2](%0), %%"R"ax \n\t"
5470 "mov %%cr2, %%"R"dx \n\t"
5471 "cmp %%"R"ax, %%"R"dx \n\t"
5473 "mov %%"R"ax, %%cr2 \n\t"
5475 /* Check if vmlaunch of vmresume is needed */
5476 "cmpl $0, %c[launched](%0) \n\t"
5477 /* Load guest registers. Don't clobber flags. */
5478 "mov %c[rax](%0), %%"R"ax \n\t"
5479 "mov %c[rbx](%0), %%"R"bx \n\t"
5480 "mov %c[rdx](%0), %%"R"dx \n\t"
5481 "mov %c[rsi](%0), %%"R"si \n\t"
5482 "mov %c[rdi](%0), %%"R"di \n\t"
5483 "mov %c[rbp](%0), %%"R"bp \n\t"
5484 #ifdef CONFIG_X86_64
5485 "mov %c[r8](%0), %%r8 \n\t"
5486 "mov %c[r9](%0), %%r9 \n\t"
5487 "mov %c[r10](%0), %%r10 \n\t"
5488 "mov %c[r11](%0), %%r11 \n\t"
5489 "mov %c[r12](%0), %%r12 \n\t"
5490 "mov %c[r13](%0), %%r13 \n\t"
5491 "mov %c[r14](%0), %%r14 \n\t"
5492 "mov %c[r15](%0), %%r15 \n\t"
5494 "mov %c[rcx](%0), %%"R"cx \n\t" /* kills %0 (ecx) */
5496 /* Enter guest mode */
5497 "jne .Llaunched \n\t"
5498 __ex(ASM_VMX_VMLAUNCH) "\n\t"
5499 "jmp .Lkvm_vmx_return \n\t"
5500 ".Llaunched: " __ex(ASM_VMX_VMRESUME) "\n\t"
5501 ".Lkvm_vmx_return: "
5502 /* Save guest registers, load host registers, keep flags */
5503 "mov %0, %c[wordsize](%%"R"sp) \n\t"
5505 "mov %%"R"ax, %c[rax](%0) \n\t"
5506 "mov %%"R"bx, %c[rbx](%0) \n\t"
5507 "pop"Q" %c[rcx](%0) \n\t"
5508 "mov %%"R"dx, %c[rdx](%0) \n\t"
5509 "mov %%"R"si, %c[rsi](%0) \n\t"
5510 "mov %%"R"di, %c[rdi](%0) \n\t"
5511 "mov %%"R"bp, %c[rbp](%0) \n\t"
5512 #ifdef CONFIG_X86_64
5513 "mov %%r8, %c[r8](%0) \n\t"
5514 "mov %%r9, %c[r9](%0) \n\t"
5515 "mov %%r10, %c[r10](%0) \n\t"
5516 "mov %%r11, %c[r11](%0) \n\t"
5517 "mov %%r12, %c[r12](%0) \n\t"
5518 "mov %%r13, %c[r13](%0) \n\t"
5519 "mov %%r14, %c[r14](%0) \n\t"
5520 "mov %%r15, %c[r15](%0) \n\t"
5522 "mov %%cr2, %%"R"ax \n\t"
5523 "mov %%"R"ax, %c[cr2](%0) \n\t"
5525 "pop %%"R"bp; pop %%"R"dx \n\t"
5526 "setbe %c[fail](%0) \n\t"
5527 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
5528 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
5529 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
5530 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
5531 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
5532 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
5533 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
5534 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
5535 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
5536 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
5537 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
5538 #ifdef CONFIG_X86_64
5539 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
5540 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
5541 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
5542 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
5543 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
5544 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
5545 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
5546 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
5548 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
5549 [wordsize]"i"(sizeof(ulong))
5551 , R"ax", R"bx", R"di", R"si"
5552 #ifdef CONFIG_X86_64
5553 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
5557 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
5558 | (1 << VCPU_EXREG_RFLAGS)
5559 | (1 << VCPU_EXREG_CPL)
5560 | (1 << VCPU_EXREG_PDPTR)
5561 | (1 << VCPU_EXREG_SEGMENTS)
5562 | (1 << VCPU_EXREG_CR3));
5563 vcpu->arch.regs_dirty = 0;
5565 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
5567 asm("mov %0, %%ds; mov %0, %%es" : : "r"(__USER_DS));
5568 vmx->loaded_vmcs->launched = 1;
5570 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
5572 vmx_complete_atomic_exit(vmx);
5573 vmx_recover_nmi_blocking(vmx);
5574 vmx_complete_interrupts(vmx);
5580 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
5582 struct vcpu_vmx *vmx = to_vmx(vcpu);
5586 free_loaded_vmcs(vmx->loaded_vmcs);
5587 kfree(vmx->guest_msrs);
5588 kvm_vcpu_uninit(vcpu);
5589 kmem_cache_free(kvm_vcpu_cache, vmx);
5592 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
5595 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
5599 return ERR_PTR(-ENOMEM);
5603 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
5607 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
5609 if (!vmx->guest_msrs) {
5613 vmx->loaded_vmcs = &vmx->vmcs01;
5614 vmx->loaded_vmcs->vmcs = alloc_vmcs();
5615 if (!vmx->loaded_vmcs->vmcs)
5618 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
5619 loaded_vmcs_init(vmx->loaded_vmcs);
5624 vmx_vcpu_load(&vmx->vcpu, cpu);
5625 vmx->vcpu.cpu = cpu;
5626 err = vmx_vcpu_setup(vmx);
5627 vmx_vcpu_put(&vmx->vcpu);
5631 if (vm_need_virtualize_apic_accesses(kvm))
5632 err = alloc_apic_access_page(kvm);
5637 if (!kvm->arch.ept_identity_map_addr)
5638 kvm->arch.ept_identity_map_addr =
5639 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
5641 if (alloc_identity_pagetable(kvm) != 0)
5643 if (!init_rmode_identity_map(kvm))
5647 vmx->nested.current_vmptr = -1ull;
5648 vmx->nested.current_vmcs12 = NULL;
5653 free_vmcs(vmx->loaded_vmcs->vmcs);
5655 kfree(vmx->guest_msrs);
5657 kvm_vcpu_uninit(&vmx->vcpu);
5660 kmem_cache_free(kvm_vcpu_cache, vmx);
5661 return ERR_PTR(err);
5664 static void __init vmx_check_processor_compat(void *rtn)
5666 struct vmcs_config vmcs_conf;
5669 if (setup_vmcs_config(&vmcs_conf) < 0)
5671 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
5672 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
5673 smp_processor_id());
5678 static int get_ept_level(void)
5680 return VMX_EPT_DEFAULT_GAW + 1;
5683 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
5687 /* For VT-d and EPT combination
5688 * 1. MMIO: always map as UC
5690 * a. VT-d without snooping control feature: can't guarantee the
5691 * result, try to trust guest.
5692 * b. VT-d with snooping control feature: snooping control feature of
5693 * VT-d engine can guarantee the cache correctness. Just set it
5694 * to WB to keep consistent with host. So the same as item 3.
5695 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
5696 * consistent with host MTRR
5699 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
5700 else if (vcpu->kvm->arch.iommu_domain &&
5701 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
5702 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
5703 VMX_EPT_MT_EPTE_SHIFT;
5705 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
5711 #define _ER(x) { EXIT_REASON_##x, #x }
5713 static const struct trace_print_flags vmx_exit_reasons_str[] = {
5715 _ER(EXTERNAL_INTERRUPT),
5717 _ER(PENDING_INTERRUPT),
5737 _ER(IO_INSTRUCTION),
5740 _ER(MWAIT_INSTRUCTION),
5741 _ER(MONITOR_INSTRUCTION),
5742 _ER(PAUSE_INSTRUCTION),
5743 _ER(MCE_DURING_VMENTRY),
5744 _ER(TPR_BELOW_THRESHOLD),
5754 static int vmx_get_lpage_level(void)
5756 if (enable_ept && !cpu_has_vmx_ept_1g_page())
5757 return PT_DIRECTORY_LEVEL;
5759 /* For shadow and EPT supported 1GB page */
5760 return PT_PDPE_LEVEL;
5763 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
5765 struct kvm_cpuid_entry2 *best;
5766 struct vcpu_vmx *vmx = to_vmx(vcpu);
5769 vmx->rdtscp_enabled = false;
5770 if (vmx_rdtscp_supported()) {
5771 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5772 if (exec_control & SECONDARY_EXEC_RDTSCP) {
5773 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
5774 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
5775 vmx->rdtscp_enabled = true;
5777 exec_control &= ~SECONDARY_EXEC_RDTSCP;
5778 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5785 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
5789 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
5790 struct x86_instruction_info *info,
5791 enum x86_intercept_stage stage)
5793 return X86EMUL_CONTINUE;
5796 static struct kvm_x86_ops vmx_x86_ops = {
5797 .cpu_has_kvm_support = cpu_has_kvm_support,
5798 .disabled_by_bios = vmx_disabled_by_bios,
5799 .hardware_setup = hardware_setup,
5800 .hardware_unsetup = hardware_unsetup,
5801 .check_processor_compatibility = vmx_check_processor_compat,
5802 .hardware_enable = hardware_enable,
5803 .hardware_disable = hardware_disable,
5804 .cpu_has_accelerated_tpr = report_flexpriority,
5806 .vcpu_create = vmx_create_vcpu,
5807 .vcpu_free = vmx_free_vcpu,
5808 .vcpu_reset = vmx_vcpu_reset,
5810 .prepare_guest_switch = vmx_save_host_state,
5811 .vcpu_load = vmx_vcpu_load,
5812 .vcpu_put = vmx_vcpu_put,
5814 .set_guest_debug = set_guest_debug,
5815 .get_msr = vmx_get_msr,
5816 .set_msr = vmx_set_msr,
5817 .get_segment_base = vmx_get_segment_base,
5818 .get_segment = vmx_get_segment,
5819 .set_segment = vmx_set_segment,
5820 .get_cpl = vmx_get_cpl,
5821 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
5822 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
5823 .decache_cr3 = vmx_decache_cr3,
5824 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
5825 .set_cr0 = vmx_set_cr0,
5826 .set_cr3 = vmx_set_cr3,
5827 .set_cr4 = vmx_set_cr4,
5828 .set_efer = vmx_set_efer,
5829 .get_idt = vmx_get_idt,
5830 .set_idt = vmx_set_idt,
5831 .get_gdt = vmx_get_gdt,
5832 .set_gdt = vmx_set_gdt,
5833 .set_dr7 = vmx_set_dr7,
5834 .cache_reg = vmx_cache_reg,
5835 .get_rflags = vmx_get_rflags,
5836 .set_rflags = vmx_set_rflags,
5837 .fpu_activate = vmx_fpu_activate,
5838 .fpu_deactivate = vmx_fpu_deactivate,
5840 .tlb_flush = vmx_flush_tlb,
5842 .run = vmx_vcpu_run,
5843 .handle_exit = vmx_handle_exit,
5844 .skip_emulated_instruction = skip_emulated_instruction,
5845 .set_interrupt_shadow = vmx_set_interrupt_shadow,
5846 .get_interrupt_shadow = vmx_get_interrupt_shadow,
5847 .patch_hypercall = vmx_patch_hypercall,
5848 .set_irq = vmx_inject_irq,
5849 .set_nmi = vmx_inject_nmi,
5850 .queue_exception = vmx_queue_exception,
5851 .cancel_injection = vmx_cancel_injection,
5852 .interrupt_allowed = vmx_interrupt_allowed,
5853 .nmi_allowed = vmx_nmi_allowed,
5854 .get_nmi_mask = vmx_get_nmi_mask,
5855 .set_nmi_mask = vmx_set_nmi_mask,
5856 .enable_nmi_window = enable_nmi_window,
5857 .enable_irq_window = enable_irq_window,
5858 .update_cr8_intercept = update_cr8_intercept,
5860 .set_tss_addr = vmx_set_tss_addr,
5861 .get_tdp_level = get_ept_level,
5862 .get_mt_mask = vmx_get_mt_mask,
5864 .get_exit_info = vmx_get_exit_info,
5865 .exit_reasons_str = vmx_exit_reasons_str,
5867 .get_lpage_level = vmx_get_lpage_level,
5869 .cpuid_update = vmx_cpuid_update,
5871 .rdtscp_supported = vmx_rdtscp_supported,
5873 .set_supported_cpuid = vmx_set_supported_cpuid,
5875 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
5877 .set_tsc_khz = vmx_set_tsc_khz,
5878 .write_tsc_offset = vmx_write_tsc_offset,
5879 .adjust_tsc_offset = vmx_adjust_tsc_offset,
5880 .compute_tsc_offset = vmx_compute_tsc_offset,
5882 .set_tdp_cr3 = vmx_set_cr3,
5884 .check_intercept = vmx_check_intercept,
5887 static int __init vmx_init(void)
5891 rdmsrl_safe(MSR_EFER, &host_efer);
5893 for (i = 0; i < NR_VMX_MSR; ++i)
5894 kvm_define_shared_msr(i, vmx_msr_index[i]);
5896 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
5897 if (!vmx_io_bitmap_a)
5900 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
5901 if (!vmx_io_bitmap_b) {
5906 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
5907 if (!vmx_msr_bitmap_legacy) {
5912 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
5913 if (!vmx_msr_bitmap_longmode) {
5919 * Allow direct access to the PC debug port (it is often used for I/O
5920 * delays, but the vmexits simply slow things down).
5922 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
5923 clear_bit(0x80, vmx_io_bitmap_a);
5925 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
5927 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
5928 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
5930 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
5932 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
5933 __alignof__(struct vcpu_vmx), THIS_MODULE);
5937 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
5938 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
5939 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
5940 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
5941 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
5942 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
5945 bypass_guest_pf = 0;
5946 kvm_mmu_set_mask_ptes(0ull, 0ull, 0ull, 0ull,
5947 VMX_EPT_EXECUTABLE_MASK);
5952 if (bypass_guest_pf)
5953 kvm_mmu_set_nonpresent_ptes(~0xffeull, 0ull);
5958 free_page((unsigned long)vmx_msr_bitmap_longmode);
5960 free_page((unsigned long)vmx_msr_bitmap_legacy);
5962 free_page((unsigned long)vmx_io_bitmap_b);
5964 free_page((unsigned long)vmx_io_bitmap_a);
5968 static void __exit vmx_exit(void)
5970 free_page((unsigned long)vmx_msr_bitmap_legacy);
5971 free_page((unsigned long)vmx_msr_bitmap_longmode);
5972 free_page((unsigned long)vmx_io_bitmap_b);
5973 free_page((unsigned long)vmx_io_bitmap_a);
5978 module_init(vmx_init)
5979 module_exit(vmx_exit)