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.
23 #include <linux/kvm_host.h>
24 #include <linux/module.h>
25 #include <linux/kernel.h>
27 #include <linux/highmem.h>
28 #include <linux/sched.h>
29 #include <linux/moduleparam.h>
30 #include <linux/mod_devicetable.h>
31 #include <linux/ftrace_event.h>
32 #include <linux/slab.h>
33 #include <linux/tboot.h>
34 #include "kvm_cache_regs.h"
40 #include <asm/virtext.h>
44 #include <asm/perf_event.h>
48 #define __ex(x) __kvm_handle_fault_on_reboot(x)
49 #define __ex_clear(x, reg) \
50 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
52 MODULE_AUTHOR("Qumranet");
53 MODULE_LICENSE("GPL");
55 static const struct x86_cpu_id vmx_cpu_id[] = {
56 X86_FEATURE_MATCH(X86_FEATURE_VMX),
59 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
61 static bool __read_mostly enable_vpid = 1;
62 module_param_named(vpid, enable_vpid, bool, 0444);
64 static bool __read_mostly flexpriority_enabled = 1;
65 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
67 static bool __read_mostly enable_ept = 1;
68 module_param_named(ept, enable_ept, bool, S_IRUGO);
70 static bool __read_mostly enable_unrestricted_guest = 1;
71 module_param_named(unrestricted_guest,
72 enable_unrestricted_guest, bool, S_IRUGO);
74 static bool __read_mostly enable_ept_ad_bits = 1;
75 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
77 static bool __read_mostly emulate_invalid_guest_state = true;
78 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
80 static bool __read_mostly vmm_exclusive = 1;
81 module_param(vmm_exclusive, bool, S_IRUGO);
83 static bool __read_mostly fasteoi = 1;
84 module_param(fasteoi, bool, S_IRUGO);
87 * If nested=1, nested virtualization is supported, i.e., guests may use
88 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
89 * use VMX instructions.
91 static bool __read_mostly nested = 0;
92 module_param(nested, bool, S_IRUGO);
94 #define KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST \
95 (X86_CR0_WP | X86_CR0_NE | X86_CR0_NW | X86_CR0_CD)
96 #define KVM_GUEST_CR0_MASK \
97 (KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
98 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST \
99 (X86_CR0_WP | X86_CR0_NE)
100 #define KVM_VM_CR0_ALWAYS_ON \
101 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
102 #define KVM_CR4_GUEST_OWNED_BITS \
103 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
104 | X86_CR4_OSXMMEXCPT)
106 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
107 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
109 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
112 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
113 * ple_gap: upper bound on the amount of time between two successive
114 * executions of PAUSE in a loop. Also indicate if ple enabled.
115 * According to test, this time is usually smaller than 128 cycles.
116 * ple_window: upper bound on the amount of time a guest is allowed to execute
117 * in a PAUSE loop. Tests indicate that most spinlocks are held for
118 * less than 2^12 cycles
119 * Time is measured based on a counter that runs at the same rate as the TSC,
120 * refer SDM volume 3b section 21.6.13 & 22.1.3.
122 #define KVM_VMX_DEFAULT_PLE_GAP 128
123 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
124 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
125 module_param(ple_gap, int, S_IRUGO);
127 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
128 module_param(ple_window, int, S_IRUGO);
130 extern const ulong vmx_return;
132 #define NR_AUTOLOAD_MSRS 8
133 #define VMCS02_POOL_SIZE 1
142 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
143 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
144 * loaded on this CPU (so we can clear them if the CPU goes down).
150 struct list_head loaded_vmcss_on_cpu_link;
153 struct shared_msr_entry {
160 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
161 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
162 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
163 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
164 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
165 * More than one of these structures may exist, if L1 runs multiple L2 guests.
166 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
167 * underlying hardware which will be used to run L2.
168 * This structure is packed to ensure that its layout is identical across
169 * machines (necessary for live migration).
170 * If there are changes in this struct, VMCS12_REVISION must be changed.
172 typedef u64 natural_width;
173 struct __packed vmcs12 {
174 /* According to the Intel spec, a VMCS region must start with the
175 * following two fields. Then follow implementation-specific data.
180 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
181 u32 padding[7]; /* room for future expansion */
186 u64 vm_exit_msr_store_addr;
187 u64 vm_exit_msr_load_addr;
188 u64 vm_entry_msr_load_addr;
190 u64 virtual_apic_page_addr;
191 u64 apic_access_addr;
193 u64 guest_physical_address;
194 u64 vmcs_link_pointer;
195 u64 guest_ia32_debugctl;
198 u64 guest_ia32_perf_global_ctrl;
205 u64 host_ia32_perf_global_ctrl;
206 u64 padding64[8]; /* room for future expansion */
208 * To allow migration of L1 (complete with its L2 guests) between
209 * machines of different natural widths (32 or 64 bit), we cannot have
210 * unsigned long fields with no explict size. We use u64 (aliased
211 * natural_width) instead. Luckily, x86 is little-endian.
213 natural_width cr0_guest_host_mask;
214 natural_width cr4_guest_host_mask;
215 natural_width cr0_read_shadow;
216 natural_width cr4_read_shadow;
217 natural_width cr3_target_value0;
218 natural_width cr3_target_value1;
219 natural_width cr3_target_value2;
220 natural_width cr3_target_value3;
221 natural_width exit_qualification;
222 natural_width guest_linear_address;
223 natural_width guest_cr0;
224 natural_width guest_cr3;
225 natural_width guest_cr4;
226 natural_width guest_es_base;
227 natural_width guest_cs_base;
228 natural_width guest_ss_base;
229 natural_width guest_ds_base;
230 natural_width guest_fs_base;
231 natural_width guest_gs_base;
232 natural_width guest_ldtr_base;
233 natural_width guest_tr_base;
234 natural_width guest_gdtr_base;
235 natural_width guest_idtr_base;
236 natural_width guest_dr7;
237 natural_width guest_rsp;
238 natural_width guest_rip;
239 natural_width guest_rflags;
240 natural_width guest_pending_dbg_exceptions;
241 natural_width guest_sysenter_esp;
242 natural_width guest_sysenter_eip;
243 natural_width host_cr0;
244 natural_width host_cr3;
245 natural_width host_cr4;
246 natural_width host_fs_base;
247 natural_width host_gs_base;
248 natural_width host_tr_base;
249 natural_width host_gdtr_base;
250 natural_width host_idtr_base;
251 natural_width host_ia32_sysenter_esp;
252 natural_width host_ia32_sysenter_eip;
253 natural_width host_rsp;
254 natural_width host_rip;
255 natural_width paddingl[8]; /* room for future expansion */
256 u32 pin_based_vm_exec_control;
257 u32 cpu_based_vm_exec_control;
258 u32 exception_bitmap;
259 u32 page_fault_error_code_mask;
260 u32 page_fault_error_code_match;
261 u32 cr3_target_count;
262 u32 vm_exit_controls;
263 u32 vm_exit_msr_store_count;
264 u32 vm_exit_msr_load_count;
265 u32 vm_entry_controls;
266 u32 vm_entry_msr_load_count;
267 u32 vm_entry_intr_info_field;
268 u32 vm_entry_exception_error_code;
269 u32 vm_entry_instruction_len;
271 u32 secondary_vm_exec_control;
272 u32 vm_instruction_error;
274 u32 vm_exit_intr_info;
275 u32 vm_exit_intr_error_code;
276 u32 idt_vectoring_info_field;
277 u32 idt_vectoring_error_code;
278 u32 vm_exit_instruction_len;
279 u32 vmx_instruction_info;
286 u32 guest_ldtr_limit;
288 u32 guest_gdtr_limit;
289 u32 guest_idtr_limit;
290 u32 guest_es_ar_bytes;
291 u32 guest_cs_ar_bytes;
292 u32 guest_ss_ar_bytes;
293 u32 guest_ds_ar_bytes;
294 u32 guest_fs_ar_bytes;
295 u32 guest_gs_ar_bytes;
296 u32 guest_ldtr_ar_bytes;
297 u32 guest_tr_ar_bytes;
298 u32 guest_interruptibility_info;
299 u32 guest_activity_state;
300 u32 guest_sysenter_cs;
301 u32 host_ia32_sysenter_cs;
302 u32 padding32[8]; /* room for future expansion */
303 u16 virtual_processor_id;
304 u16 guest_es_selector;
305 u16 guest_cs_selector;
306 u16 guest_ss_selector;
307 u16 guest_ds_selector;
308 u16 guest_fs_selector;
309 u16 guest_gs_selector;
310 u16 guest_ldtr_selector;
311 u16 guest_tr_selector;
312 u16 host_es_selector;
313 u16 host_cs_selector;
314 u16 host_ss_selector;
315 u16 host_ds_selector;
316 u16 host_fs_selector;
317 u16 host_gs_selector;
318 u16 host_tr_selector;
322 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
323 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
324 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
326 #define VMCS12_REVISION 0x11e57ed0
329 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
330 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
331 * current implementation, 4K are reserved to avoid future complications.
333 #define VMCS12_SIZE 0x1000
335 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
337 struct list_head list;
339 struct loaded_vmcs vmcs02;
343 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
344 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
347 /* Has the level1 guest done vmxon? */
350 /* The guest-physical address of the current VMCS L1 keeps for L2 */
352 /* The host-usable pointer to the above */
353 struct page *current_vmcs12_page;
354 struct vmcs12 *current_vmcs12;
356 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
357 struct list_head vmcs02_pool;
359 u64 vmcs01_tsc_offset;
360 /* L2 must run next, and mustn't decide to exit to L1. */
361 bool nested_run_pending;
363 * Guest pages referred to in vmcs02 with host-physical pointers, so
364 * we must keep them pinned while L2 runs.
366 struct page *apic_access_page;
370 struct kvm_vcpu vcpu;
371 unsigned long host_rsp;
374 bool nmi_known_unmasked;
376 u32 idt_vectoring_info;
378 struct shared_msr_entry *guest_msrs;
382 u64 msr_host_kernel_gs_base;
383 u64 msr_guest_kernel_gs_base;
386 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
387 * non-nested (L1) guest, it always points to vmcs01. For a nested
388 * guest (L2), it points to a different VMCS.
390 struct loaded_vmcs vmcs01;
391 struct loaded_vmcs *loaded_vmcs;
392 bool __launched; /* temporary, used in vmx_vcpu_run */
393 struct msr_autoload {
395 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
396 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
400 u16 fs_sel, gs_sel, ldt_sel;
404 int gs_ldt_reload_needed;
405 int fs_reload_needed;
410 struct kvm_segment segs[8];
413 u32 bitmask; /* 4 bits per segment (1 bit per field) */
414 struct kvm_save_segment {
422 bool emulation_required;
424 /* Support for vnmi-less CPUs */
425 int soft_vnmi_blocked;
427 s64 vnmi_blocked_time;
432 /* Support for a guest hypervisor (nested VMX) */
433 struct nested_vmx nested;
436 enum segment_cache_field {
445 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
447 return container_of(vcpu, struct vcpu_vmx, vcpu);
450 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
451 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
452 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
453 [number##_HIGH] = VMCS12_OFFSET(name)+4
455 static const unsigned short vmcs_field_to_offset_table[] = {
456 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
457 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
458 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
459 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
460 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
461 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
462 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
463 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
464 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
465 FIELD(HOST_ES_SELECTOR, host_es_selector),
466 FIELD(HOST_CS_SELECTOR, host_cs_selector),
467 FIELD(HOST_SS_SELECTOR, host_ss_selector),
468 FIELD(HOST_DS_SELECTOR, host_ds_selector),
469 FIELD(HOST_FS_SELECTOR, host_fs_selector),
470 FIELD(HOST_GS_SELECTOR, host_gs_selector),
471 FIELD(HOST_TR_SELECTOR, host_tr_selector),
472 FIELD64(IO_BITMAP_A, io_bitmap_a),
473 FIELD64(IO_BITMAP_B, io_bitmap_b),
474 FIELD64(MSR_BITMAP, msr_bitmap),
475 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
476 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
477 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
478 FIELD64(TSC_OFFSET, tsc_offset),
479 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
480 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
481 FIELD64(EPT_POINTER, ept_pointer),
482 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
483 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
484 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
485 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
486 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
487 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
488 FIELD64(GUEST_PDPTR0, guest_pdptr0),
489 FIELD64(GUEST_PDPTR1, guest_pdptr1),
490 FIELD64(GUEST_PDPTR2, guest_pdptr2),
491 FIELD64(GUEST_PDPTR3, guest_pdptr3),
492 FIELD64(HOST_IA32_PAT, host_ia32_pat),
493 FIELD64(HOST_IA32_EFER, host_ia32_efer),
494 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
495 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
496 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
497 FIELD(EXCEPTION_BITMAP, exception_bitmap),
498 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
499 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
500 FIELD(CR3_TARGET_COUNT, cr3_target_count),
501 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
502 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
503 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
504 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
505 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
506 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
507 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
508 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
509 FIELD(TPR_THRESHOLD, tpr_threshold),
510 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
511 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
512 FIELD(VM_EXIT_REASON, vm_exit_reason),
513 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
514 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
515 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
516 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
517 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
518 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
519 FIELD(GUEST_ES_LIMIT, guest_es_limit),
520 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
521 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
522 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
523 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
524 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
525 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
526 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
527 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
528 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
529 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
530 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
531 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
532 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
533 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
534 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
535 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
536 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
537 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
538 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
539 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
540 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
541 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
542 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
543 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
544 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
545 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
546 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
547 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
548 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
549 FIELD(EXIT_QUALIFICATION, exit_qualification),
550 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
551 FIELD(GUEST_CR0, guest_cr0),
552 FIELD(GUEST_CR3, guest_cr3),
553 FIELD(GUEST_CR4, guest_cr4),
554 FIELD(GUEST_ES_BASE, guest_es_base),
555 FIELD(GUEST_CS_BASE, guest_cs_base),
556 FIELD(GUEST_SS_BASE, guest_ss_base),
557 FIELD(GUEST_DS_BASE, guest_ds_base),
558 FIELD(GUEST_FS_BASE, guest_fs_base),
559 FIELD(GUEST_GS_BASE, guest_gs_base),
560 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
561 FIELD(GUEST_TR_BASE, guest_tr_base),
562 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
563 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
564 FIELD(GUEST_DR7, guest_dr7),
565 FIELD(GUEST_RSP, guest_rsp),
566 FIELD(GUEST_RIP, guest_rip),
567 FIELD(GUEST_RFLAGS, guest_rflags),
568 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
569 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
570 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
571 FIELD(HOST_CR0, host_cr0),
572 FIELD(HOST_CR3, host_cr3),
573 FIELD(HOST_CR4, host_cr4),
574 FIELD(HOST_FS_BASE, host_fs_base),
575 FIELD(HOST_GS_BASE, host_gs_base),
576 FIELD(HOST_TR_BASE, host_tr_base),
577 FIELD(HOST_GDTR_BASE, host_gdtr_base),
578 FIELD(HOST_IDTR_BASE, host_idtr_base),
579 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
580 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
581 FIELD(HOST_RSP, host_rsp),
582 FIELD(HOST_RIP, host_rip),
584 static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
586 static inline short vmcs_field_to_offset(unsigned long field)
588 if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
590 return vmcs_field_to_offset_table[field];
593 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
595 return to_vmx(vcpu)->nested.current_vmcs12;
598 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
600 struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
601 if (is_error_page(page))
607 static void nested_release_page(struct page *page)
609 kvm_release_page_dirty(page);
612 static void nested_release_page_clean(struct page *page)
614 kvm_release_page_clean(page);
617 static u64 construct_eptp(unsigned long root_hpa);
618 static void kvm_cpu_vmxon(u64 addr);
619 static void kvm_cpu_vmxoff(void);
620 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
621 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
622 static void vmx_set_segment(struct kvm_vcpu *vcpu,
623 struct kvm_segment *var, int seg);
624 static void vmx_get_segment(struct kvm_vcpu *vcpu,
625 struct kvm_segment *var, int seg);
627 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
628 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
630 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
631 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
633 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
634 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
636 static unsigned long *vmx_io_bitmap_a;
637 static unsigned long *vmx_io_bitmap_b;
638 static unsigned long *vmx_msr_bitmap_legacy;
639 static unsigned long *vmx_msr_bitmap_longmode;
641 static bool cpu_has_load_ia32_efer;
642 static bool cpu_has_load_perf_global_ctrl;
644 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
645 static DEFINE_SPINLOCK(vmx_vpid_lock);
647 static struct vmcs_config {
651 u32 pin_based_exec_ctrl;
652 u32 cpu_based_exec_ctrl;
653 u32 cpu_based_2nd_exec_ctrl;
658 static struct vmx_capability {
663 #define VMX_SEGMENT_FIELD(seg) \
664 [VCPU_SREG_##seg] = { \
665 .selector = GUEST_##seg##_SELECTOR, \
666 .base = GUEST_##seg##_BASE, \
667 .limit = GUEST_##seg##_LIMIT, \
668 .ar_bytes = GUEST_##seg##_AR_BYTES, \
671 static const struct kvm_vmx_segment_field {
676 } kvm_vmx_segment_fields[] = {
677 VMX_SEGMENT_FIELD(CS),
678 VMX_SEGMENT_FIELD(DS),
679 VMX_SEGMENT_FIELD(ES),
680 VMX_SEGMENT_FIELD(FS),
681 VMX_SEGMENT_FIELD(GS),
682 VMX_SEGMENT_FIELD(SS),
683 VMX_SEGMENT_FIELD(TR),
684 VMX_SEGMENT_FIELD(LDTR),
687 static u64 host_efer;
689 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
692 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
693 * away by decrementing the array size.
695 static const u32 vmx_msr_index[] = {
697 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
699 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
701 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
703 static inline bool is_page_fault(u32 intr_info)
705 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
706 INTR_INFO_VALID_MASK)) ==
707 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
710 static inline bool is_no_device(u32 intr_info)
712 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
713 INTR_INFO_VALID_MASK)) ==
714 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
717 static inline bool is_invalid_opcode(u32 intr_info)
719 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
720 INTR_INFO_VALID_MASK)) ==
721 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
724 static inline bool is_external_interrupt(u32 intr_info)
726 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
727 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
730 static inline bool is_machine_check(u32 intr_info)
732 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
733 INTR_INFO_VALID_MASK)) ==
734 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
737 static inline bool cpu_has_vmx_msr_bitmap(void)
739 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
742 static inline bool cpu_has_vmx_tpr_shadow(void)
744 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
747 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
749 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
752 static inline bool cpu_has_secondary_exec_ctrls(void)
754 return vmcs_config.cpu_based_exec_ctrl &
755 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
758 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
760 return vmcs_config.cpu_based_2nd_exec_ctrl &
761 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
764 static inline bool cpu_has_vmx_flexpriority(void)
766 return cpu_has_vmx_tpr_shadow() &&
767 cpu_has_vmx_virtualize_apic_accesses();
770 static inline bool cpu_has_vmx_ept_execute_only(void)
772 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
775 static inline bool cpu_has_vmx_eptp_uncacheable(void)
777 return vmx_capability.ept & VMX_EPTP_UC_BIT;
780 static inline bool cpu_has_vmx_eptp_writeback(void)
782 return vmx_capability.ept & VMX_EPTP_WB_BIT;
785 static inline bool cpu_has_vmx_ept_2m_page(void)
787 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
790 static inline bool cpu_has_vmx_ept_1g_page(void)
792 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
795 static inline bool cpu_has_vmx_ept_4levels(void)
797 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
800 static inline bool cpu_has_vmx_ept_ad_bits(void)
802 return vmx_capability.ept & VMX_EPT_AD_BIT;
805 static inline bool cpu_has_vmx_invept_individual_addr(void)
807 return vmx_capability.ept & VMX_EPT_EXTENT_INDIVIDUAL_BIT;
810 static inline bool cpu_has_vmx_invept_context(void)
812 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
815 static inline bool cpu_has_vmx_invept_global(void)
817 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
820 static inline bool cpu_has_vmx_invvpid_single(void)
822 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
825 static inline bool cpu_has_vmx_invvpid_global(void)
827 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
830 static inline bool cpu_has_vmx_ept(void)
832 return vmcs_config.cpu_based_2nd_exec_ctrl &
833 SECONDARY_EXEC_ENABLE_EPT;
836 static inline bool cpu_has_vmx_unrestricted_guest(void)
838 return vmcs_config.cpu_based_2nd_exec_ctrl &
839 SECONDARY_EXEC_UNRESTRICTED_GUEST;
842 static inline bool cpu_has_vmx_ple(void)
844 return vmcs_config.cpu_based_2nd_exec_ctrl &
845 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
848 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
850 return flexpriority_enabled && irqchip_in_kernel(kvm);
853 static inline bool cpu_has_vmx_vpid(void)
855 return vmcs_config.cpu_based_2nd_exec_ctrl &
856 SECONDARY_EXEC_ENABLE_VPID;
859 static inline bool cpu_has_vmx_rdtscp(void)
861 return vmcs_config.cpu_based_2nd_exec_ctrl &
862 SECONDARY_EXEC_RDTSCP;
865 static inline bool cpu_has_vmx_invpcid(void)
867 return vmcs_config.cpu_based_2nd_exec_ctrl &
868 SECONDARY_EXEC_ENABLE_INVPCID;
871 static inline bool cpu_has_virtual_nmis(void)
873 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
876 static inline bool cpu_has_vmx_wbinvd_exit(void)
878 return vmcs_config.cpu_based_2nd_exec_ctrl &
879 SECONDARY_EXEC_WBINVD_EXITING;
882 static inline bool report_flexpriority(void)
884 return flexpriority_enabled;
887 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
889 return vmcs12->cpu_based_vm_exec_control & bit;
892 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
894 return (vmcs12->cpu_based_vm_exec_control &
895 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
896 (vmcs12->secondary_vm_exec_control & bit);
899 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12,
900 struct kvm_vcpu *vcpu)
902 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
905 static inline bool is_exception(u32 intr_info)
907 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
908 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
911 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu);
912 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
913 struct vmcs12 *vmcs12,
914 u32 reason, unsigned long qualification);
916 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
920 for (i = 0; i < vmx->nmsrs; ++i)
921 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
926 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
932 } operand = { vpid, 0, gva };
934 asm volatile (__ex(ASM_VMX_INVVPID)
935 /* CF==1 or ZF==1 --> rc = -1 */
937 : : "a"(&operand), "c"(ext) : "cc", "memory");
940 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
944 } operand = {eptp, gpa};
946 asm volatile (__ex(ASM_VMX_INVEPT)
947 /* CF==1 or ZF==1 --> rc = -1 */
948 "; ja 1f ; ud2 ; 1:\n"
949 : : "a" (&operand), "c" (ext) : "cc", "memory");
952 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
956 i = __find_msr_index(vmx, msr);
958 return &vmx->guest_msrs[i];
962 static void vmcs_clear(struct vmcs *vmcs)
964 u64 phys_addr = __pa(vmcs);
967 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
968 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
971 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
975 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
977 vmcs_clear(loaded_vmcs->vmcs);
978 loaded_vmcs->cpu = -1;
979 loaded_vmcs->launched = 0;
982 static void vmcs_load(struct vmcs *vmcs)
984 u64 phys_addr = __pa(vmcs);
987 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
988 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
991 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
995 static void __loaded_vmcs_clear(void *arg)
997 struct loaded_vmcs *loaded_vmcs = arg;
998 int cpu = raw_smp_processor_id();
1000 if (loaded_vmcs->cpu != cpu)
1001 return; /* vcpu migration can race with cpu offline */
1002 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1003 per_cpu(current_vmcs, cpu) = NULL;
1004 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1005 loaded_vmcs_init(loaded_vmcs);
1008 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1010 if (loaded_vmcs->cpu != -1)
1011 smp_call_function_single(
1012 loaded_vmcs->cpu, __loaded_vmcs_clear, loaded_vmcs, 1);
1015 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
1020 if (cpu_has_vmx_invvpid_single())
1021 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
1024 static inline void vpid_sync_vcpu_global(void)
1026 if (cpu_has_vmx_invvpid_global())
1027 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1030 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1032 if (cpu_has_vmx_invvpid_single())
1033 vpid_sync_vcpu_single(vmx);
1035 vpid_sync_vcpu_global();
1038 static inline void ept_sync_global(void)
1040 if (cpu_has_vmx_invept_global())
1041 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1044 static inline void ept_sync_context(u64 eptp)
1047 if (cpu_has_vmx_invept_context())
1048 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1054 static inline void ept_sync_individual_addr(u64 eptp, gpa_t gpa)
1057 if (cpu_has_vmx_invept_individual_addr())
1058 __invept(VMX_EPT_EXTENT_INDIVIDUAL_ADDR,
1061 ept_sync_context(eptp);
1065 static __always_inline unsigned long vmcs_readl(unsigned long field)
1067 unsigned long value;
1069 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1070 : "=a"(value) : "d"(field) : "cc");
1074 static __always_inline u16 vmcs_read16(unsigned long field)
1076 return vmcs_readl(field);
1079 static __always_inline u32 vmcs_read32(unsigned long field)
1081 return vmcs_readl(field);
1084 static __always_inline u64 vmcs_read64(unsigned long field)
1086 #ifdef CONFIG_X86_64
1087 return vmcs_readl(field);
1089 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1093 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1095 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1096 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1100 static void vmcs_writel(unsigned long field, unsigned long value)
1104 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1105 : "=q"(error) : "a"(value), "d"(field) : "cc");
1106 if (unlikely(error))
1107 vmwrite_error(field, value);
1110 static void vmcs_write16(unsigned long field, u16 value)
1112 vmcs_writel(field, value);
1115 static void vmcs_write32(unsigned long field, u32 value)
1117 vmcs_writel(field, value);
1120 static void vmcs_write64(unsigned long field, u64 value)
1122 vmcs_writel(field, value);
1123 #ifndef CONFIG_X86_64
1125 vmcs_writel(field+1, value >> 32);
1129 static void vmcs_clear_bits(unsigned long field, u32 mask)
1131 vmcs_writel(field, vmcs_readl(field) & ~mask);
1134 static void vmcs_set_bits(unsigned long field, u32 mask)
1136 vmcs_writel(field, vmcs_readl(field) | mask);
1139 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1141 vmx->segment_cache.bitmask = 0;
1144 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1148 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1150 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1151 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1152 vmx->segment_cache.bitmask = 0;
1154 ret = vmx->segment_cache.bitmask & mask;
1155 vmx->segment_cache.bitmask |= mask;
1159 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1161 u16 *p = &vmx->segment_cache.seg[seg].selector;
1163 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1164 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1168 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1170 ulong *p = &vmx->segment_cache.seg[seg].base;
1172 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1173 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1177 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1179 u32 *p = &vmx->segment_cache.seg[seg].limit;
1181 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1182 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1186 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1188 u32 *p = &vmx->segment_cache.seg[seg].ar;
1190 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1191 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1195 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1199 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1200 (1u << NM_VECTOR) | (1u << DB_VECTOR);
1201 if ((vcpu->guest_debug &
1202 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1203 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1204 eb |= 1u << BP_VECTOR;
1205 if (to_vmx(vcpu)->rmode.vm86_active)
1208 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1209 if (vcpu->fpu_active)
1210 eb &= ~(1u << NM_VECTOR);
1212 /* When we are running a nested L2 guest and L1 specified for it a
1213 * certain exception bitmap, we must trap the same exceptions and pass
1214 * them to L1. When running L2, we will only handle the exceptions
1215 * specified above if L1 did not want them.
1217 if (is_guest_mode(vcpu))
1218 eb |= get_vmcs12(vcpu)->exception_bitmap;
1220 vmcs_write32(EXCEPTION_BITMAP, eb);
1223 static void clear_atomic_switch_msr_special(unsigned long entry,
1226 vmcs_clear_bits(VM_ENTRY_CONTROLS, entry);
1227 vmcs_clear_bits(VM_EXIT_CONTROLS, exit);
1230 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1233 struct msr_autoload *m = &vmx->msr_autoload;
1237 if (cpu_has_load_ia32_efer) {
1238 clear_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1239 VM_EXIT_LOAD_IA32_EFER);
1243 case MSR_CORE_PERF_GLOBAL_CTRL:
1244 if (cpu_has_load_perf_global_ctrl) {
1245 clear_atomic_switch_msr_special(
1246 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1247 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1253 for (i = 0; i < m->nr; ++i)
1254 if (m->guest[i].index == msr)
1260 m->guest[i] = m->guest[m->nr];
1261 m->host[i] = m->host[m->nr];
1262 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1263 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1266 static void add_atomic_switch_msr_special(unsigned long entry,
1267 unsigned long exit, unsigned long guest_val_vmcs,
1268 unsigned long host_val_vmcs, u64 guest_val, u64 host_val)
1270 vmcs_write64(guest_val_vmcs, guest_val);
1271 vmcs_write64(host_val_vmcs, host_val);
1272 vmcs_set_bits(VM_ENTRY_CONTROLS, entry);
1273 vmcs_set_bits(VM_EXIT_CONTROLS, exit);
1276 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1277 u64 guest_val, u64 host_val)
1280 struct msr_autoload *m = &vmx->msr_autoload;
1284 if (cpu_has_load_ia32_efer) {
1285 add_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1286 VM_EXIT_LOAD_IA32_EFER,
1289 guest_val, host_val);
1293 case MSR_CORE_PERF_GLOBAL_CTRL:
1294 if (cpu_has_load_perf_global_ctrl) {
1295 add_atomic_switch_msr_special(
1296 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1297 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1298 GUEST_IA32_PERF_GLOBAL_CTRL,
1299 HOST_IA32_PERF_GLOBAL_CTRL,
1300 guest_val, host_val);
1306 for (i = 0; i < m->nr; ++i)
1307 if (m->guest[i].index == msr)
1310 if (i == NR_AUTOLOAD_MSRS) {
1311 printk_once(KERN_WARNING"Not enough mst switch entries. "
1312 "Can't add msr %x\n", msr);
1314 } else if (i == m->nr) {
1316 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1317 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1320 m->guest[i].index = msr;
1321 m->guest[i].value = guest_val;
1322 m->host[i].index = msr;
1323 m->host[i].value = host_val;
1326 static void reload_tss(void)
1329 * VT restores TR but not its size. Useless.
1331 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1332 struct desc_struct *descs;
1334 descs = (void *)gdt->address;
1335 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1339 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1344 guest_efer = vmx->vcpu.arch.efer;
1347 * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1350 ignore_bits = EFER_NX | EFER_SCE;
1351 #ifdef CONFIG_X86_64
1352 ignore_bits |= EFER_LMA | EFER_LME;
1353 /* SCE is meaningful only in long mode on Intel */
1354 if (guest_efer & EFER_LMA)
1355 ignore_bits &= ~(u64)EFER_SCE;
1357 guest_efer &= ~ignore_bits;
1358 guest_efer |= host_efer & ignore_bits;
1359 vmx->guest_msrs[efer_offset].data = guest_efer;
1360 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1362 clear_atomic_switch_msr(vmx, MSR_EFER);
1363 /* On ept, can't emulate nx, and must switch nx atomically */
1364 if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1365 guest_efer = vmx->vcpu.arch.efer;
1366 if (!(guest_efer & EFER_LMA))
1367 guest_efer &= ~EFER_LME;
1368 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1375 static unsigned long segment_base(u16 selector)
1377 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1378 struct desc_struct *d;
1379 unsigned long table_base;
1382 if (!(selector & ~3))
1385 table_base = gdt->address;
1387 if (selector & 4) { /* from ldt */
1388 u16 ldt_selector = kvm_read_ldt();
1390 if (!(ldt_selector & ~3))
1393 table_base = segment_base(ldt_selector);
1395 d = (struct desc_struct *)(table_base + (selector & ~7));
1396 v = get_desc_base(d);
1397 #ifdef CONFIG_X86_64
1398 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1399 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1404 static inline unsigned long kvm_read_tr_base(void)
1407 asm("str %0" : "=g"(tr));
1408 return segment_base(tr);
1411 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1413 struct vcpu_vmx *vmx = to_vmx(vcpu);
1416 if (vmx->host_state.loaded)
1419 vmx->host_state.loaded = 1;
1421 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1422 * allow segment selectors with cpl > 0 or ti == 1.
1424 vmx->host_state.ldt_sel = kvm_read_ldt();
1425 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1426 savesegment(fs, vmx->host_state.fs_sel);
1427 if (!(vmx->host_state.fs_sel & 7)) {
1428 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1429 vmx->host_state.fs_reload_needed = 0;
1431 vmcs_write16(HOST_FS_SELECTOR, 0);
1432 vmx->host_state.fs_reload_needed = 1;
1434 savesegment(gs, vmx->host_state.gs_sel);
1435 if (!(vmx->host_state.gs_sel & 7))
1436 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1438 vmcs_write16(HOST_GS_SELECTOR, 0);
1439 vmx->host_state.gs_ldt_reload_needed = 1;
1442 #ifdef CONFIG_X86_64
1443 savesegment(ds, vmx->host_state.ds_sel);
1444 savesegment(es, vmx->host_state.es_sel);
1447 #ifdef CONFIG_X86_64
1448 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1449 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1451 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1452 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1455 #ifdef CONFIG_X86_64
1456 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1457 if (is_long_mode(&vmx->vcpu))
1458 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1460 for (i = 0; i < vmx->save_nmsrs; ++i)
1461 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1462 vmx->guest_msrs[i].data,
1463 vmx->guest_msrs[i].mask);
1466 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1468 if (!vmx->host_state.loaded)
1471 ++vmx->vcpu.stat.host_state_reload;
1472 vmx->host_state.loaded = 0;
1473 #ifdef CONFIG_X86_64
1474 if (is_long_mode(&vmx->vcpu))
1475 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1477 if (vmx->host_state.gs_ldt_reload_needed) {
1478 kvm_load_ldt(vmx->host_state.ldt_sel);
1479 #ifdef CONFIG_X86_64
1480 load_gs_index(vmx->host_state.gs_sel);
1482 loadsegment(gs, vmx->host_state.gs_sel);
1485 if (vmx->host_state.fs_reload_needed)
1486 loadsegment(fs, vmx->host_state.fs_sel);
1487 #ifdef CONFIG_X86_64
1488 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1489 loadsegment(ds, vmx->host_state.ds_sel);
1490 loadsegment(es, vmx->host_state.es_sel);
1494 #ifdef CONFIG_X86_64
1495 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1499 load_gdt(&__get_cpu_var(host_gdt));
1502 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1505 __vmx_load_host_state(vmx);
1510 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1511 * vcpu mutex is already taken.
1513 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1515 struct vcpu_vmx *vmx = to_vmx(vcpu);
1516 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1519 kvm_cpu_vmxon(phys_addr);
1520 else if (vmx->loaded_vmcs->cpu != cpu)
1521 loaded_vmcs_clear(vmx->loaded_vmcs);
1523 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1524 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1525 vmcs_load(vmx->loaded_vmcs->vmcs);
1528 if (vmx->loaded_vmcs->cpu != cpu) {
1529 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1530 unsigned long sysenter_esp;
1532 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1533 local_irq_disable();
1534 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1535 &per_cpu(loaded_vmcss_on_cpu, cpu));
1539 * Linux uses per-cpu TSS and GDT, so set these when switching
1542 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1543 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
1545 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1546 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1547 vmx->loaded_vmcs->cpu = cpu;
1551 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1553 __vmx_load_host_state(to_vmx(vcpu));
1554 if (!vmm_exclusive) {
1555 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1561 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1565 if (vcpu->fpu_active)
1567 vcpu->fpu_active = 1;
1568 cr0 = vmcs_readl(GUEST_CR0);
1569 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1570 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1571 vmcs_writel(GUEST_CR0, cr0);
1572 update_exception_bitmap(vcpu);
1573 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1574 if (is_guest_mode(vcpu))
1575 vcpu->arch.cr0_guest_owned_bits &=
1576 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
1577 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1580 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1583 * Return the cr0 value that a nested guest would read. This is a combination
1584 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1585 * its hypervisor (cr0_read_shadow).
1587 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1589 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1590 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1592 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1594 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1595 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1598 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1600 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
1601 * set this *before* calling this function.
1603 vmx_decache_cr0_guest_bits(vcpu);
1604 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1605 update_exception_bitmap(vcpu);
1606 vcpu->arch.cr0_guest_owned_bits = 0;
1607 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1608 if (is_guest_mode(vcpu)) {
1610 * L1's specified read shadow might not contain the TS bit,
1611 * so now that we turned on shadowing of this bit, we need to
1612 * set this bit of the shadow. Like in nested_vmx_run we need
1613 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
1614 * up-to-date here because we just decached cr0.TS (and we'll
1615 * only update vmcs12->guest_cr0 on nested exit).
1617 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1618 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
1619 (vcpu->arch.cr0 & X86_CR0_TS);
1620 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
1622 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1625 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1627 unsigned long rflags, save_rflags;
1629 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1630 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1631 rflags = vmcs_readl(GUEST_RFLAGS);
1632 if (to_vmx(vcpu)->rmode.vm86_active) {
1633 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1634 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1635 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1637 to_vmx(vcpu)->rflags = rflags;
1639 return to_vmx(vcpu)->rflags;
1642 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1644 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1645 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
1646 to_vmx(vcpu)->rflags = rflags;
1647 if (to_vmx(vcpu)->rmode.vm86_active) {
1648 to_vmx(vcpu)->rmode.save_rflags = rflags;
1649 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1651 vmcs_writel(GUEST_RFLAGS, rflags);
1654 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1656 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1659 if (interruptibility & GUEST_INTR_STATE_STI)
1660 ret |= KVM_X86_SHADOW_INT_STI;
1661 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1662 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1667 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1669 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1670 u32 interruptibility = interruptibility_old;
1672 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1674 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1675 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1676 else if (mask & KVM_X86_SHADOW_INT_STI)
1677 interruptibility |= GUEST_INTR_STATE_STI;
1679 if ((interruptibility != interruptibility_old))
1680 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1683 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1687 rip = kvm_rip_read(vcpu);
1688 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1689 kvm_rip_write(vcpu, rip);
1691 /* skipping an emulated instruction also counts */
1692 vmx_set_interrupt_shadow(vcpu, 0);
1696 * KVM wants to inject page-faults which it got to the guest. This function
1697 * checks whether in a nested guest, we need to inject them to L1 or L2.
1698 * This function assumes it is called with the exit reason in vmcs02 being
1699 * a #PF exception (this is the only case in which KVM injects a #PF when L2
1702 static int nested_pf_handled(struct kvm_vcpu *vcpu)
1704 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1706 /* TODO: also check PFEC_MATCH/MASK, not just EB.PF. */
1707 if (!(vmcs12->exception_bitmap & (1u << PF_VECTOR)))
1710 nested_vmx_vmexit(vcpu);
1714 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1715 bool has_error_code, u32 error_code,
1718 struct vcpu_vmx *vmx = to_vmx(vcpu);
1719 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1721 if (nr == PF_VECTOR && is_guest_mode(vcpu) &&
1722 nested_pf_handled(vcpu))
1725 if (has_error_code) {
1726 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1727 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1730 if (vmx->rmode.vm86_active) {
1732 if (kvm_exception_is_soft(nr))
1733 inc_eip = vcpu->arch.event_exit_inst_len;
1734 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1735 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1739 if (kvm_exception_is_soft(nr)) {
1740 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1741 vmx->vcpu.arch.event_exit_inst_len);
1742 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1744 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1746 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1749 static bool vmx_rdtscp_supported(void)
1751 return cpu_has_vmx_rdtscp();
1754 static bool vmx_invpcid_supported(void)
1756 return cpu_has_vmx_invpcid() && enable_ept;
1760 * Swap MSR entry in host/guest MSR entry array.
1762 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1764 struct shared_msr_entry tmp;
1766 tmp = vmx->guest_msrs[to];
1767 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1768 vmx->guest_msrs[from] = tmp;
1772 * Set up the vmcs to automatically save and restore system
1773 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1774 * mode, as fiddling with msrs is very expensive.
1776 static void setup_msrs(struct vcpu_vmx *vmx)
1778 int save_nmsrs, index;
1779 unsigned long *msr_bitmap;
1782 #ifdef CONFIG_X86_64
1783 if (is_long_mode(&vmx->vcpu)) {
1784 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
1786 move_msr_up(vmx, index, save_nmsrs++);
1787 index = __find_msr_index(vmx, MSR_LSTAR);
1789 move_msr_up(vmx, index, save_nmsrs++);
1790 index = __find_msr_index(vmx, MSR_CSTAR);
1792 move_msr_up(vmx, index, save_nmsrs++);
1793 index = __find_msr_index(vmx, MSR_TSC_AUX);
1794 if (index >= 0 && vmx->rdtscp_enabled)
1795 move_msr_up(vmx, index, save_nmsrs++);
1797 * MSR_STAR is only needed on long mode guests, and only
1798 * if efer.sce is enabled.
1800 index = __find_msr_index(vmx, MSR_STAR);
1801 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
1802 move_msr_up(vmx, index, save_nmsrs++);
1805 index = __find_msr_index(vmx, MSR_EFER);
1806 if (index >= 0 && update_transition_efer(vmx, index))
1807 move_msr_up(vmx, index, save_nmsrs++);
1809 vmx->save_nmsrs = save_nmsrs;
1811 if (cpu_has_vmx_msr_bitmap()) {
1812 if (is_long_mode(&vmx->vcpu))
1813 msr_bitmap = vmx_msr_bitmap_longmode;
1815 msr_bitmap = vmx_msr_bitmap_legacy;
1817 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1822 * reads and returns guest's timestamp counter "register"
1823 * guest_tsc = host_tsc + tsc_offset -- 21.3
1825 static u64 guest_read_tsc(void)
1827 u64 host_tsc, tsc_offset;
1830 tsc_offset = vmcs_read64(TSC_OFFSET);
1831 return host_tsc + tsc_offset;
1835 * Like guest_read_tsc, but always returns L1's notion of the timestamp
1836 * counter, even if a nested guest (L2) is currently running.
1838 u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu)
1840 u64 host_tsc, tsc_offset;
1843 tsc_offset = is_guest_mode(vcpu) ?
1844 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
1845 vmcs_read64(TSC_OFFSET);
1846 return host_tsc + tsc_offset;
1850 * Engage any workarounds for mis-matched TSC rates. Currently limited to
1851 * software catchup for faster rates on slower CPUs.
1853 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
1858 if (user_tsc_khz > tsc_khz) {
1859 vcpu->arch.tsc_catchup = 1;
1860 vcpu->arch.tsc_always_catchup = 1;
1862 WARN(1, "user requested TSC rate below hardware speed\n");
1866 * writes 'offset' into guest's timestamp counter offset register
1868 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
1870 if (is_guest_mode(vcpu)) {
1872 * We're here if L1 chose not to trap WRMSR to TSC. According
1873 * to the spec, this should set L1's TSC; The offset that L1
1874 * set for L2 remains unchanged, and still needs to be added
1875 * to the newly set TSC to get L2's TSC.
1877 struct vmcs12 *vmcs12;
1878 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
1879 /* recalculate vmcs02.TSC_OFFSET: */
1880 vmcs12 = get_vmcs12(vcpu);
1881 vmcs_write64(TSC_OFFSET, offset +
1882 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
1883 vmcs12->tsc_offset : 0));
1885 vmcs_write64(TSC_OFFSET, offset);
1889 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
1891 u64 offset = vmcs_read64(TSC_OFFSET);
1892 vmcs_write64(TSC_OFFSET, offset + adjustment);
1893 if (is_guest_mode(vcpu)) {
1894 /* Even when running L2, the adjustment needs to apply to L1 */
1895 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
1899 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
1901 return target_tsc - native_read_tsc();
1904 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
1906 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
1907 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
1911 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
1912 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
1913 * all guests if the "nested" module option is off, and can also be disabled
1914 * for a single guest by disabling its VMX cpuid bit.
1916 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
1918 return nested && guest_cpuid_has_vmx(vcpu);
1922 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
1923 * returned for the various VMX controls MSRs when nested VMX is enabled.
1924 * The same values should also be used to verify that vmcs12 control fields are
1925 * valid during nested entry from L1 to L2.
1926 * Each of these control msrs has a low and high 32-bit half: A low bit is on
1927 * if the corresponding bit in the (32-bit) control field *must* be on, and a
1928 * bit in the high half is on if the corresponding bit in the control field
1929 * may be on. See also vmx_control_verify().
1930 * TODO: allow these variables to be modified (downgraded) by module options
1933 static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
1934 static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
1935 static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
1936 static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
1937 static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
1938 static __init void nested_vmx_setup_ctls_msrs(void)
1941 * Note that as a general rule, the high half of the MSRs (bits in
1942 * the control fields which may be 1) should be initialized by the
1943 * intersection of the underlying hardware's MSR (i.e., features which
1944 * can be supported) and the list of features we want to expose -
1945 * because they are known to be properly supported in our code.
1946 * Also, usually, the low half of the MSRs (bits which must be 1) can
1947 * be set to 0, meaning that L1 may turn off any of these bits. The
1948 * reason is that if one of these bits is necessary, it will appear
1949 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
1950 * fields of vmcs01 and vmcs02, will turn these bits off - and
1951 * nested_vmx_exit_handled() will not pass related exits to L1.
1952 * These rules have exceptions below.
1955 /* pin-based controls */
1957 * According to the Intel spec, if bit 55 of VMX_BASIC is off (as it is
1958 * in our case), bits 1, 2 and 4 (i.e., 0x16) must be 1 in this MSR.
1960 nested_vmx_pinbased_ctls_low = 0x16 ;
1961 nested_vmx_pinbased_ctls_high = 0x16 |
1962 PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
1963 PIN_BASED_VIRTUAL_NMIS;
1966 nested_vmx_exit_ctls_low = 0;
1967 /* Note that guest use of VM_EXIT_ACK_INTR_ON_EXIT is not supported. */
1968 #ifdef CONFIG_X86_64
1969 nested_vmx_exit_ctls_high = VM_EXIT_HOST_ADDR_SPACE_SIZE;
1971 nested_vmx_exit_ctls_high = 0;
1974 /* entry controls */
1975 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
1976 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
1977 nested_vmx_entry_ctls_low = 0;
1978 nested_vmx_entry_ctls_high &=
1979 VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_IA32E_MODE;
1981 /* cpu-based controls */
1982 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
1983 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
1984 nested_vmx_procbased_ctls_low = 0;
1985 nested_vmx_procbased_ctls_high &=
1986 CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_USE_TSC_OFFSETING |
1987 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
1988 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
1989 CPU_BASED_CR3_STORE_EXITING |
1990 #ifdef CONFIG_X86_64
1991 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
1993 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
1994 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
1995 CPU_BASED_RDPMC_EXITING | CPU_BASED_RDTSC_EXITING |
1996 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1998 * We can allow some features even when not supported by the
1999 * hardware. For example, L1 can specify an MSR bitmap - and we
2000 * can use it to avoid exits to L1 - even when L0 runs L2
2001 * without MSR bitmaps.
2003 nested_vmx_procbased_ctls_high |= CPU_BASED_USE_MSR_BITMAPS;
2005 /* secondary cpu-based controls */
2006 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2007 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
2008 nested_vmx_secondary_ctls_low = 0;
2009 nested_vmx_secondary_ctls_high &=
2010 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
2013 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2016 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2018 return ((control & high) | low) == control;
2021 static inline u64 vmx_control_msr(u32 low, u32 high)
2023 return low | ((u64)high << 32);
2027 * If we allow our guest to use VMX instructions (i.e., nested VMX), we should
2028 * also let it use VMX-specific MSRs.
2029 * vmx_get_vmx_msr() and vmx_set_vmx_msr() return 1 when we handled a
2030 * VMX-specific MSR, or 0 when we haven't (and the caller should handle it
2031 * like all other MSRs).
2033 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2035 if (!nested_vmx_allowed(vcpu) && msr_index >= MSR_IA32_VMX_BASIC &&
2036 msr_index <= MSR_IA32_VMX_TRUE_ENTRY_CTLS) {
2038 * According to the spec, processors which do not support VMX
2039 * should throw a #GP(0) when VMX capability MSRs are read.
2041 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
2045 switch (msr_index) {
2046 case MSR_IA32_FEATURE_CONTROL:
2049 case MSR_IA32_VMX_BASIC:
2051 * This MSR reports some information about VMX support. We
2052 * should return information about the VMX we emulate for the
2053 * guest, and the VMCS structure we give it - not about the
2054 * VMX support of the underlying hardware.
2056 *pdata = VMCS12_REVISION |
2057 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2058 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2060 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2061 case MSR_IA32_VMX_PINBASED_CTLS:
2062 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
2063 nested_vmx_pinbased_ctls_high);
2065 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2066 case MSR_IA32_VMX_PROCBASED_CTLS:
2067 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
2068 nested_vmx_procbased_ctls_high);
2070 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2071 case MSR_IA32_VMX_EXIT_CTLS:
2072 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
2073 nested_vmx_exit_ctls_high);
2075 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2076 case MSR_IA32_VMX_ENTRY_CTLS:
2077 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
2078 nested_vmx_entry_ctls_high);
2080 case MSR_IA32_VMX_MISC:
2084 * These MSRs specify bits which the guest must keep fixed (on or off)
2085 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2086 * We picked the standard core2 setting.
2088 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2089 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2090 case MSR_IA32_VMX_CR0_FIXED0:
2091 *pdata = VMXON_CR0_ALWAYSON;
2093 case MSR_IA32_VMX_CR0_FIXED1:
2096 case MSR_IA32_VMX_CR4_FIXED0:
2097 *pdata = VMXON_CR4_ALWAYSON;
2099 case MSR_IA32_VMX_CR4_FIXED1:
2102 case MSR_IA32_VMX_VMCS_ENUM:
2105 case MSR_IA32_VMX_PROCBASED_CTLS2:
2106 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
2107 nested_vmx_secondary_ctls_high);
2109 case MSR_IA32_VMX_EPT_VPID_CAP:
2110 /* Currently, no nested ept or nested vpid */
2120 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2122 if (!nested_vmx_allowed(vcpu))
2125 if (msr_index == MSR_IA32_FEATURE_CONTROL)
2126 /* TODO: the right thing. */
2129 * No need to treat VMX capability MSRs specially: If we don't handle
2130 * them, handle_wrmsr will #GP(0), which is correct (they are readonly)
2136 * Reads an msr value (of 'msr_index') into 'pdata'.
2137 * Returns 0 on success, non-0 otherwise.
2138 * Assumes vcpu_load() was already called.
2140 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2143 struct shared_msr_entry *msr;
2146 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2150 switch (msr_index) {
2151 #ifdef CONFIG_X86_64
2153 data = vmcs_readl(GUEST_FS_BASE);
2156 data = vmcs_readl(GUEST_GS_BASE);
2158 case MSR_KERNEL_GS_BASE:
2159 vmx_load_host_state(to_vmx(vcpu));
2160 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2164 return kvm_get_msr_common(vcpu, msr_index, pdata);
2166 data = guest_read_tsc();
2168 case MSR_IA32_SYSENTER_CS:
2169 data = vmcs_read32(GUEST_SYSENTER_CS);
2171 case MSR_IA32_SYSENTER_EIP:
2172 data = vmcs_readl(GUEST_SYSENTER_EIP);
2174 case MSR_IA32_SYSENTER_ESP:
2175 data = vmcs_readl(GUEST_SYSENTER_ESP);
2178 if (!to_vmx(vcpu)->rdtscp_enabled)
2180 /* Otherwise falls through */
2182 if (vmx_get_vmx_msr(vcpu, msr_index, pdata))
2184 msr = find_msr_entry(to_vmx(vcpu), msr_index);
2189 return kvm_get_msr_common(vcpu, msr_index, pdata);
2197 * Writes msr value into into the appropriate "register".
2198 * Returns 0 on success, non-0 otherwise.
2199 * Assumes vcpu_load() was already called.
2201 static int vmx_set_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
2203 struct vcpu_vmx *vmx = to_vmx(vcpu);
2204 struct shared_msr_entry *msr;
2207 switch (msr_index) {
2209 ret = kvm_set_msr_common(vcpu, msr_index, data);
2211 #ifdef CONFIG_X86_64
2213 vmx_segment_cache_clear(vmx);
2214 vmcs_writel(GUEST_FS_BASE, data);
2217 vmx_segment_cache_clear(vmx);
2218 vmcs_writel(GUEST_GS_BASE, data);
2220 case MSR_KERNEL_GS_BASE:
2221 vmx_load_host_state(vmx);
2222 vmx->msr_guest_kernel_gs_base = data;
2225 case MSR_IA32_SYSENTER_CS:
2226 vmcs_write32(GUEST_SYSENTER_CS, data);
2228 case MSR_IA32_SYSENTER_EIP:
2229 vmcs_writel(GUEST_SYSENTER_EIP, data);
2231 case MSR_IA32_SYSENTER_ESP:
2232 vmcs_writel(GUEST_SYSENTER_ESP, data);
2235 kvm_write_tsc(vcpu, data);
2237 case MSR_IA32_CR_PAT:
2238 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2239 vmcs_write64(GUEST_IA32_PAT, data);
2240 vcpu->arch.pat = data;
2243 ret = kvm_set_msr_common(vcpu, msr_index, data);
2246 if (!vmx->rdtscp_enabled)
2248 /* Check reserved bit, higher 32 bits should be zero */
2249 if ((data >> 32) != 0)
2251 /* Otherwise falls through */
2253 if (vmx_set_vmx_msr(vcpu, msr_index, data))
2255 msr = find_msr_entry(vmx, msr_index);
2258 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2260 kvm_set_shared_msr(msr->index, msr->data,
2266 ret = kvm_set_msr_common(vcpu, msr_index, data);
2272 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2274 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2277 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2280 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2282 case VCPU_EXREG_PDPTR:
2284 ept_save_pdptrs(vcpu);
2291 static void set_guest_debug(struct kvm_vcpu *vcpu, struct kvm_guest_debug *dbg)
2293 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP)
2294 vmcs_writel(GUEST_DR7, dbg->arch.debugreg[7]);
2296 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
2298 update_exception_bitmap(vcpu);
2301 static __init int cpu_has_kvm_support(void)
2303 return cpu_has_vmx();
2306 static __init int vmx_disabled_by_bios(void)
2310 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2311 if (msr & FEATURE_CONTROL_LOCKED) {
2312 /* launched w/ TXT and VMX disabled */
2313 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2316 /* launched w/o TXT and VMX only enabled w/ TXT */
2317 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2318 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2319 && !tboot_enabled()) {
2320 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2321 "activate TXT before enabling KVM\n");
2324 /* launched w/o TXT and VMX disabled */
2325 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2326 && !tboot_enabled())
2333 static void kvm_cpu_vmxon(u64 addr)
2335 asm volatile (ASM_VMX_VMXON_RAX
2336 : : "a"(&addr), "m"(addr)
2340 static int hardware_enable(void *garbage)
2342 int cpu = raw_smp_processor_id();
2343 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2346 if (read_cr4() & X86_CR4_VMXE)
2349 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2350 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2352 test_bits = FEATURE_CONTROL_LOCKED;
2353 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2354 if (tboot_enabled())
2355 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2357 if ((old & test_bits) != test_bits) {
2358 /* enable and lock */
2359 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2361 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2363 if (vmm_exclusive) {
2364 kvm_cpu_vmxon(phys_addr);
2368 store_gdt(&__get_cpu_var(host_gdt));
2373 static void vmclear_local_loaded_vmcss(void)
2375 int cpu = raw_smp_processor_id();
2376 struct loaded_vmcs *v, *n;
2378 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2379 loaded_vmcss_on_cpu_link)
2380 __loaded_vmcs_clear(v);
2384 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2387 static void kvm_cpu_vmxoff(void)
2389 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2392 static void hardware_disable(void *garbage)
2394 if (vmm_exclusive) {
2395 vmclear_local_loaded_vmcss();
2398 write_cr4(read_cr4() & ~X86_CR4_VMXE);
2401 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2402 u32 msr, u32 *result)
2404 u32 vmx_msr_low, vmx_msr_high;
2405 u32 ctl = ctl_min | ctl_opt;
2407 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2409 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2410 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2412 /* Ensure minimum (required) set of control bits are supported. */
2420 static __init bool allow_1_setting(u32 msr, u32 ctl)
2422 u32 vmx_msr_low, vmx_msr_high;
2424 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2425 return vmx_msr_high & ctl;
2428 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2430 u32 vmx_msr_low, vmx_msr_high;
2431 u32 min, opt, min2, opt2;
2432 u32 _pin_based_exec_control = 0;
2433 u32 _cpu_based_exec_control = 0;
2434 u32 _cpu_based_2nd_exec_control = 0;
2435 u32 _vmexit_control = 0;
2436 u32 _vmentry_control = 0;
2438 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2439 opt = PIN_BASED_VIRTUAL_NMIS;
2440 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2441 &_pin_based_exec_control) < 0)
2444 min = CPU_BASED_HLT_EXITING |
2445 #ifdef CONFIG_X86_64
2446 CPU_BASED_CR8_LOAD_EXITING |
2447 CPU_BASED_CR8_STORE_EXITING |
2449 CPU_BASED_CR3_LOAD_EXITING |
2450 CPU_BASED_CR3_STORE_EXITING |
2451 CPU_BASED_USE_IO_BITMAPS |
2452 CPU_BASED_MOV_DR_EXITING |
2453 CPU_BASED_USE_TSC_OFFSETING |
2454 CPU_BASED_MWAIT_EXITING |
2455 CPU_BASED_MONITOR_EXITING |
2456 CPU_BASED_INVLPG_EXITING |
2457 CPU_BASED_RDPMC_EXITING;
2459 opt = CPU_BASED_TPR_SHADOW |
2460 CPU_BASED_USE_MSR_BITMAPS |
2461 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2462 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2463 &_cpu_based_exec_control) < 0)
2465 #ifdef CONFIG_X86_64
2466 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2467 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2468 ~CPU_BASED_CR8_STORE_EXITING;
2470 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2472 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2473 SECONDARY_EXEC_WBINVD_EXITING |
2474 SECONDARY_EXEC_ENABLE_VPID |
2475 SECONDARY_EXEC_ENABLE_EPT |
2476 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2477 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2478 SECONDARY_EXEC_RDTSCP |
2479 SECONDARY_EXEC_ENABLE_INVPCID;
2480 if (adjust_vmx_controls(min2, opt2,
2481 MSR_IA32_VMX_PROCBASED_CTLS2,
2482 &_cpu_based_2nd_exec_control) < 0)
2485 #ifndef CONFIG_X86_64
2486 if (!(_cpu_based_2nd_exec_control &
2487 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2488 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2490 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2491 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2493 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2494 CPU_BASED_CR3_STORE_EXITING |
2495 CPU_BASED_INVLPG_EXITING);
2496 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2497 vmx_capability.ept, vmx_capability.vpid);
2501 #ifdef CONFIG_X86_64
2502 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2504 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT;
2505 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2506 &_vmexit_control) < 0)
2510 opt = VM_ENTRY_LOAD_IA32_PAT;
2511 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2512 &_vmentry_control) < 0)
2515 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2517 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2518 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2521 #ifdef CONFIG_X86_64
2522 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2523 if (vmx_msr_high & (1u<<16))
2527 /* Require Write-Back (WB) memory type for VMCS accesses. */
2528 if (((vmx_msr_high >> 18) & 15) != 6)
2531 vmcs_conf->size = vmx_msr_high & 0x1fff;
2532 vmcs_conf->order = get_order(vmcs_config.size);
2533 vmcs_conf->revision_id = vmx_msr_low;
2535 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2536 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2537 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2538 vmcs_conf->vmexit_ctrl = _vmexit_control;
2539 vmcs_conf->vmentry_ctrl = _vmentry_control;
2541 cpu_has_load_ia32_efer =
2542 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2543 VM_ENTRY_LOAD_IA32_EFER)
2544 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2545 VM_EXIT_LOAD_IA32_EFER);
2547 cpu_has_load_perf_global_ctrl =
2548 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2549 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
2550 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2551 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
2554 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
2555 * but due to arrata below it can't be used. Workaround is to use
2556 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2558 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
2563 * BC86,AAY89,BD102 (model 44)
2567 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
2568 switch (boot_cpu_data.x86_model) {
2574 cpu_has_load_perf_global_ctrl = false;
2575 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2576 "does not work properly. Using workaround\n");
2586 static struct vmcs *alloc_vmcs_cpu(int cpu)
2588 int node = cpu_to_node(cpu);
2592 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
2595 vmcs = page_address(pages);
2596 memset(vmcs, 0, vmcs_config.size);
2597 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
2601 static struct vmcs *alloc_vmcs(void)
2603 return alloc_vmcs_cpu(raw_smp_processor_id());
2606 static void free_vmcs(struct vmcs *vmcs)
2608 free_pages((unsigned long)vmcs, vmcs_config.order);
2612 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2614 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2616 if (!loaded_vmcs->vmcs)
2618 loaded_vmcs_clear(loaded_vmcs);
2619 free_vmcs(loaded_vmcs->vmcs);
2620 loaded_vmcs->vmcs = NULL;
2623 static void free_kvm_area(void)
2627 for_each_possible_cpu(cpu) {
2628 free_vmcs(per_cpu(vmxarea, cpu));
2629 per_cpu(vmxarea, cpu) = NULL;
2633 static __init int alloc_kvm_area(void)
2637 for_each_possible_cpu(cpu) {
2640 vmcs = alloc_vmcs_cpu(cpu);
2646 per_cpu(vmxarea, cpu) = vmcs;
2651 static __init int hardware_setup(void)
2653 if (setup_vmcs_config(&vmcs_config) < 0)
2656 if (boot_cpu_has(X86_FEATURE_NX))
2657 kvm_enable_efer_bits(EFER_NX);
2659 if (!cpu_has_vmx_vpid())
2662 if (!cpu_has_vmx_ept() ||
2663 !cpu_has_vmx_ept_4levels()) {
2665 enable_unrestricted_guest = 0;
2666 enable_ept_ad_bits = 0;
2669 if (!cpu_has_vmx_ept_ad_bits())
2670 enable_ept_ad_bits = 0;
2672 if (!cpu_has_vmx_unrestricted_guest())
2673 enable_unrestricted_guest = 0;
2675 if (!cpu_has_vmx_flexpriority())
2676 flexpriority_enabled = 0;
2678 if (!cpu_has_vmx_tpr_shadow())
2679 kvm_x86_ops->update_cr8_intercept = NULL;
2681 if (enable_ept && !cpu_has_vmx_ept_2m_page())
2682 kvm_disable_largepages();
2684 if (!cpu_has_vmx_ple())
2688 nested_vmx_setup_ctls_msrs();
2690 return alloc_kvm_area();
2693 static __exit void hardware_unsetup(void)
2698 static void fix_pmode_dataseg(struct kvm_vcpu *vcpu, int seg, struct kvm_segment *save)
2700 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2701 struct kvm_segment tmp = *save;
2703 if (!(vmcs_readl(sf->base) == tmp.base && tmp.s)) {
2704 tmp.base = vmcs_readl(sf->base);
2705 tmp.selector = vmcs_read16(sf->selector);
2708 vmx_set_segment(vcpu, &tmp, seg);
2711 static void enter_pmode(struct kvm_vcpu *vcpu)
2713 unsigned long flags;
2714 struct vcpu_vmx *vmx = to_vmx(vcpu);
2716 vmx->emulation_required = 1;
2717 vmx->rmode.vm86_active = 0;
2719 vmx_segment_cache_clear(vmx);
2721 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2723 flags = vmcs_readl(GUEST_RFLAGS);
2724 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2725 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2726 vmcs_writel(GUEST_RFLAGS, flags);
2728 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
2729 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
2731 update_exception_bitmap(vcpu);
2733 if (emulate_invalid_guest_state)
2736 fix_pmode_dataseg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
2737 fix_pmode_dataseg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
2738 fix_pmode_dataseg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
2739 fix_pmode_dataseg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
2741 vmx_segment_cache_clear(vmx);
2743 vmcs_write16(GUEST_SS_SELECTOR, 0);
2744 vmcs_write32(GUEST_SS_AR_BYTES, 0x93);
2746 vmcs_write16(GUEST_CS_SELECTOR,
2747 vmcs_read16(GUEST_CS_SELECTOR) & ~SELECTOR_RPL_MASK);
2748 vmcs_write32(GUEST_CS_AR_BYTES, 0x9b);
2751 static gva_t rmode_tss_base(struct kvm *kvm)
2753 if (!kvm->arch.tss_addr) {
2754 struct kvm_memslots *slots;
2755 struct kvm_memory_slot *slot;
2758 slots = kvm_memslots(kvm);
2759 slot = id_to_memslot(slots, 0);
2760 base_gfn = slot->base_gfn + slot->npages - 3;
2762 return base_gfn << PAGE_SHIFT;
2764 return kvm->arch.tss_addr;
2767 static void fix_rmode_seg(int seg, struct kvm_segment *save)
2769 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
2771 vmcs_write16(sf->selector, save->base >> 4);
2772 vmcs_write32(sf->base, save->base & 0xffff0);
2773 vmcs_write32(sf->limit, 0xffff);
2774 vmcs_write32(sf->ar_bytes, 0xf3);
2775 if (save->base & 0xf)
2776 printk_once(KERN_WARNING "kvm: segment base is not paragraph"
2777 " aligned when entering protected mode (seg=%d)",
2781 static void enter_rmode(struct kvm_vcpu *vcpu)
2783 unsigned long flags;
2784 struct vcpu_vmx *vmx = to_vmx(vcpu);
2785 struct kvm_segment var;
2787 if (enable_unrestricted_guest)
2790 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
2791 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
2792 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
2793 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
2794 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
2796 vmx->emulation_required = 1;
2797 vmx->rmode.vm86_active = 1;
2801 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
2802 * vcpu. Call it here with phys address pointing 16M below 4G.
2804 if (!vcpu->kvm->arch.tss_addr) {
2805 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
2806 "called before entering vcpu\n");
2807 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
2808 vmx_set_tss_addr(vcpu->kvm, 0xfeffd000);
2809 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
2812 vmx_segment_cache_clear(vmx);
2814 vmcs_writel(GUEST_TR_BASE, rmode_tss_base(vcpu->kvm));
2815 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
2816 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
2818 flags = vmcs_readl(GUEST_RFLAGS);
2819 vmx->rmode.save_rflags = flags;
2821 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2823 vmcs_writel(GUEST_RFLAGS, flags);
2824 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
2825 update_exception_bitmap(vcpu);
2827 if (emulate_invalid_guest_state)
2828 goto continue_rmode;
2830 vmx_get_segment(vcpu, &var, VCPU_SREG_SS);
2831 vmx_set_segment(vcpu, &var, VCPU_SREG_SS);
2833 vmx_get_segment(vcpu, &var, VCPU_SREG_CS);
2834 vmx_set_segment(vcpu, &var, VCPU_SREG_CS);
2836 vmx_get_segment(vcpu, &var, VCPU_SREG_ES);
2837 vmx_set_segment(vcpu, &var, VCPU_SREG_ES);
2839 vmx_get_segment(vcpu, &var, VCPU_SREG_DS);
2840 vmx_set_segment(vcpu, &var, VCPU_SREG_DS);
2842 vmx_get_segment(vcpu, &var, VCPU_SREG_GS);
2843 vmx_set_segment(vcpu, &var, VCPU_SREG_GS);
2845 vmx_get_segment(vcpu, &var, VCPU_SREG_FS);
2846 vmx_set_segment(vcpu, &var, VCPU_SREG_FS);
2849 kvm_mmu_reset_context(vcpu);
2852 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
2854 struct vcpu_vmx *vmx = to_vmx(vcpu);
2855 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
2861 * Force kernel_gs_base reloading before EFER changes, as control
2862 * of this msr depends on is_long_mode().
2864 vmx_load_host_state(to_vmx(vcpu));
2865 vcpu->arch.efer = efer;
2866 if (efer & EFER_LMA) {
2867 vmcs_write32(VM_ENTRY_CONTROLS,
2868 vmcs_read32(VM_ENTRY_CONTROLS) |
2869 VM_ENTRY_IA32E_MODE);
2872 vmcs_write32(VM_ENTRY_CONTROLS,
2873 vmcs_read32(VM_ENTRY_CONTROLS) &
2874 ~VM_ENTRY_IA32E_MODE);
2876 msr->data = efer & ~EFER_LME;
2881 #ifdef CONFIG_X86_64
2883 static void enter_lmode(struct kvm_vcpu *vcpu)
2887 vmx_segment_cache_clear(to_vmx(vcpu));
2889 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
2890 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
2891 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
2893 vmcs_write32(GUEST_TR_AR_BYTES,
2894 (guest_tr_ar & ~AR_TYPE_MASK)
2895 | AR_TYPE_BUSY_64_TSS);
2897 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
2900 static void exit_lmode(struct kvm_vcpu *vcpu)
2902 vmcs_write32(VM_ENTRY_CONTROLS,
2903 vmcs_read32(VM_ENTRY_CONTROLS)
2904 & ~VM_ENTRY_IA32E_MODE);
2905 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
2910 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
2912 vpid_sync_context(to_vmx(vcpu));
2914 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
2916 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
2920 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
2922 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
2924 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
2925 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
2928 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
2930 if (enable_ept && is_paging(vcpu))
2931 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
2932 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
2935 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
2937 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
2939 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
2940 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
2943 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
2945 if (!test_bit(VCPU_EXREG_PDPTR,
2946 (unsigned long *)&vcpu->arch.regs_dirty))
2949 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2950 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
2951 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
2952 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
2953 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
2957 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
2959 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
2960 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
2961 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
2962 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
2963 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
2966 __set_bit(VCPU_EXREG_PDPTR,
2967 (unsigned long *)&vcpu->arch.regs_avail);
2968 __set_bit(VCPU_EXREG_PDPTR,
2969 (unsigned long *)&vcpu->arch.regs_dirty);
2972 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
2974 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
2976 struct kvm_vcpu *vcpu)
2978 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
2979 vmx_decache_cr3(vcpu);
2980 if (!(cr0 & X86_CR0_PG)) {
2981 /* From paging/starting to nonpaging */
2982 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2983 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
2984 (CPU_BASED_CR3_LOAD_EXITING |
2985 CPU_BASED_CR3_STORE_EXITING));
2986 vcpu->arch.cr0 = cr0;
2987 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2988 } else if (!is_paging(vcpu)) {
2989 /* From nonpaging to paging */
2990 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
2991 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
2992 ~(CPU_BASED_CR3_LOAD_EXITING |
2993 CPU_BASED_CR3_STORE_EXITING));
2994 vcpu->arch.cr0 = cr0;
2995 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
2998 if (!(cr0 & X86_CR0_WP))
2999 *hw_cr0 &= ~X86_CR0_WP;
3002 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3004 struct vcpu_vmx *vmx = to_vmx(vcpu);
3005 unsigned long hw_cr0;
3007 if (enable_unrestricted_guest)
3008 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK_UNRESTRICTED_GUEST)
3009 | KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3011 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK) | KVM_VM_CR0_ALWAYS_ON;
3013 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3016 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3019 #ifdef CONFIG_X86_64
3020 if (vcpu->arch.efer & EFER_LME) {
3021 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3023 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3029 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3031 if (!vcpu->fpu_active)
3032 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3034 vmcs_writel(CR0_READ_SHADOW, cr0);
3035 vmcs_writel(GUEST_CR0, hw_cr0);
3036 vcpu->arch.cr0 = cr0;
3037 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3040 static u64 construct_eptp(unsigned long root_hpa)
3044 /* TODO write the value reading from MSR */
3045 eptp = VMX_EPT_DEFAULT_MT |
3046 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3047 if (enable_ept_ad_bits)
3048 eptp |= VMX_EPT_AD_ENABLE_BIT;
3049 eptp |= (root_hpa & PAGE_MASK);
3054 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3056 unsigned long guest_cr3;
3061 eptp = construct_eptp(cr3);
3062 vmcs_write64(EPT_POINTER, eptp);
3063 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
3064 vcpu->kvm->arch.ept_identity_map_addr;
3065 ept_load_pdptrs(vcpu);
3068 vmx_flush_tlb(vcpu);
3069 vmcs_writel(GUEST_CR3, guest_cr3);
3072 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3074 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
3075 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3077 if (cr4 & X86_CR4_VMXE) {
3079 * To use VMXON (and later other VMX instructions), a guest
3080 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3081 * So basically the check on whether to allow nested VMX
3084 if (!nested_vmx_allowed(vcpu))
3086 } else if (to_vmx(vcpu)->nested.vmxon)
3089 vcpu->arch.cr4 = cr4;
3091 if (!is_paging(vcpu)) {
3092 hw_cr4 &= ~X86_CR4_PAE;
3093 hw_cr4 |= X86_CR4_PSE;
3094 } else if (!(cr4 & X86_CR4_PAE)) {
3095 hw_cr4 &= ~X86_CR4_PAE;
3099 vmcs_writel(CR4_READ_SHADOW, cr4);
3100 vmcs_writel(GUEST_CR4, hw_cr4);
3104 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3105 struct kvm_segment *var, int seg)
3107 struct vcpu_vmx *vmx = to_vmx(vcpu);
3110 if (vmx->rmode.vm86_active
3111 && (seg == VCPU_SREG_TR || seg == VCPU_SREG_ES
3112 || seg == VCPU_SREG_DS || seg == VCPU_SREG_FS
3113 || seg == VCPU_SREG_GS)) {
3114 *var = vmx->rmode.segs[seg];
3115 if (seg == VCPU_SREG_TR
3116 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3118 var->base = vmx_read_guest_seg_base(vmx, seg);
3119 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3122 var->base = vmx_read_guest_seg_base(vmx, seg);
3123 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3124 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3125 ar = vmx_read_guest_seg_ar(vmx, seg);
3126 if ((ar & AR_UNUSABLE_MASK) && !emulate_invalid_guest_state)
3128 var->type = ar & 15;
3129 var->s = (ar >> 4) & 1;
3130 var->dpl = (ar >> 5) & 3;
3131 var->present = (ar >> 7) & 1;
3132 var->avl = (ar >> 12) & 1;
3133 var->l = (ar >> 13) & 1;
3134 var->db = (ar >> 14) & 1;
3135 var->g = (ar >> 15) & 1;
3136 var->unusable = (ar >> 16) & 1;
3139 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3141 struct kvm_segment s;
3143 if (to_vmx(vcpu)->rmode.vm86_active) {
3144 vmx_get_segment(vcpu, &s, seg);
3147 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3150 static int __vmx_get_cpl(struct kvm_vcpu *vcpu)
3152 if (!is_protmode(vcpu))
3155 if (!is_long_mode(vcpu)
3156 && (kvm_get_rflags(vcpu) & X86_EFLAGS_VM)) /* if virtual 8086 */
3159 return vmx_read_guest_seg_selector(to_vmx(vcpu), VCPU_SREG_CS) & 3;
3162 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3164 struct vcpu_vmx *vmx = to_vmx(vcpu);
3167 * If we enter real mode with cs.sel & 3 != 0, the normal CPL calculations
3168 * fail; use the cache instead.
3170 if (unlikely(vmx->emulation_required && emulate_invalid_guest_state)) {
3174 if (!test_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail)) {
3175 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3176 vmx->cpl = __vmx_get_cpl(vcpu);
3183 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3187 if (var->unusable || !var->present)
3190 ar = var->type & 15;
3191 ar |= (var->s & 1) << 4;
3192 ar |= (var->dpl & 3) << 5;
3193 ar |= (var->present & 1) << 7;
3194 ar |= (var->avl & 1) << 12;
3195 ar |= (var->l & 1) << 13;
3196 ar |= (var->db & 1) << 14;
3197 ar |= (var->g & 1) << 15;
3203 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3204 struct kvm_segment *var, int seg)
3206 struct vcpu_vmx *vmx = to_vmx(vcpu);
3207 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3210 vmx_segment_cache_clear(vmx);
3212 if (vmx->rmode.vm86_active && seg == VCPU_SREG_TR) {
3213 vmcs_write16(sf->selector, var->selector);
3214 vmx->rmode.segs[VCPU_SREG_TR] = *var;
3217 vmcs_writel(sf->base, var->base);
3218 vmcs_write32(sf->limit, var->limit);
3219 vmcs_write16(sf->selector, var->selector);
3220 if (vmx->rmode.vm86_active && var->s) {
3221 vmx->rmode.segs[seg] = *var;
3223 * Hack real-mode segments into vm86 compatibility.
3225 if (var->base == 0xffff0000 && var->selector == 0xf000)
3226 vmcs_writel(sf->base, 0xf0000);
3229 ar = vmx_segment_access_rights(var);
3232 * Fix the "Accessed" bit in AR field of segment registers for older
3234 * IA32 arch specifies that at the time of processor reset the
3235 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3236 * is setting it to 0 in the userland code. This causes invalid guest
3237 * state vmexit when "unrestricted guest" mode is turned on.
3238 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3239 * tree. Newer qemu binaries with that qemu fix would not need this
3242 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3243 ar |= 0x1; /* Accessed */
3245 vmcs_write32(sf->ar_bytes, ar);
3246 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3249 * Fix segments for real mode guest in hosts that don't have
3250 * "unrestricted_mode" or it was disabled.
3251 * This is done to allow migration of the guests from hosts with
3252 * unrestricted guest like Westmere to older host that don't have
3253 * unrestricted guest like Nehelem.
3255 if (!enable_unrestricted_guest && vmx->rmode.vm86_active) {
3258 vmcs_write32(GUEST_CS_AR_BYTES, 0xf3);
3259 vmcs_write32(GUEST_CS_LIMIT, 0xffff);
3260 if (vmcs_readl(GUEST_CS_BASE) == 0xffff0000)
3261 vmcs_writel(GUEST_CS_BASE, 0xf0000);
3262 vmcs_write16(GUEST_CS_SELECTOR,
3263 vmcs_readl(GUEST_CS_BASE) >> 4);
3269 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3272 vmcs_write16(GUEST_SS_SELECTOR,
3273 vmcs_readl(GUEST_SS_BASE) >> 4);
3274 vmcs_write32(GUEST_SS_LIMIT, 0xffff);
3275 vmcs_write32(GUEST_SS_AR_BYTES, 0xf3);
3281 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3283 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3285 *db = (ar >> 14) & 1;
3286 *l = (ar >> 13) & 1;
3289 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3291 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3292 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3295 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3297 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3298 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3301 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3303 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3304 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3307 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3309 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3310 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3313 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3315 struct kvm_segment var;
3318 vmx_get_segment(vcpu, &var, seg);
3319 ar = vmx_segment_access_rights(&var);
3321 if (var.base != (var.selector << 4))
3323 if (var.limit < 0xffff)
3325 if (((ar | (3 << AR_DPL_SHIFT)) & ~(AR_G_MASK | AR_DB_MASK)) != 0xf3)
3331 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3333 struct kvm_segment cs;
3334 unsigned int cs_rpl;
3336 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3337 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3341 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3345 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3346 if (cs.dpl > cs_rpl)
3349 if (cs.dpl != cs_rpl)
3355 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3359 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3361 struct kvm_segment ss;
3362 unsigned int ss_rpl;
3364 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3365 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3369 if (ss.type != 3 && ss.type != 7)
3373 if (ss.dpl != ss_rpl) /* DPL != RPL */
3381 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3383 struct kvm_segment var;
3386 vmx_get_segment(vcpu, &var, seg);
3387 rpl = var.selector & SELECTOR_RPL_MASK;
3395 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3396 if (var.dpl < rpl) /* DPL < RPL */
3400 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3406 static bool tr_valid(struct kvm_vcpu *vcpu)
3408 struct kvm_segment tr;
3410 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3414 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3416 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3424 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3426 struct kvm_segment ldtr;
3428 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3432 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3442 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3444 struct kvm_segment cs, ss;
3446 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3447 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3449 return ((cs.selector & SELECTOR_RPL_MASK) ==
3450 (ss.selector & SELECTOR_RPL_MASK));
3454 * Check if guest state is valid. Returns true if valid, false if
3456 * We assume that registers are always usable
3458 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3460 /* real mode guest state checks */
3461 if (!is_protmode(vcpu)) {
3462 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3464 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3466 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3468 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3470 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3472 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3475 /* protected mode guest state checks */
3476 if (!cs_ss_rpl_check(vcpu))
3478 if (!code_segment_valid(vcpu))
3480 if (!stack_segment_valid(vcpu))
3482 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3484 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3486 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3488 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3490 if (!tr_valid(vcpu))
3492 if (!ldtr_valid(vcpu))
3496 * - Add checks on RIP
3497 * - Add checks on RFLAGS
3503 static int init_rmode_tss(struct kvm *kvm)
3507 int r, idx, ret = 0;
3509 idx = srcu_read_lock(&kvm->srcu);
3510 fn = rmode_tss_base(kvm) >> PAGE_SHIFT;
3511 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3514 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3515 r = kvm_write_guest_page(kvm, fn++, &data,
3516 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3519 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3522 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3526 r = kvm_write_guest_page(kvm, fn, &data,
3527 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3534 srcu_read_unlock(&kvm->srcu, idx);
3538 static int init_rmode_identity_map(struct kvm *kvm)
3541 pfn_t identity_map_pfn;
3546 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
3547 printk(KERN_ERR "EPT: identity-mapping pagetable "
3548 "haven't been allocated!\n");
3551 if (likely(kvm->arch.ept_identity_pagetable_done))
3554 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3555 idx = srcu_read_lock(&kvm->srcu);
3556 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3559 /* Set up identity-mapping pagetable for EPT in real mode */
3560 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3561 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3562 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3563 r = kvm_write_guest_page(kvm, identity_map_pfn,
3564 &tmp, i * sizeof(tmp), sizeof(tmp));
3568 kvm->arch.ept_identity_pagetable_done = true;
3571 srcu_read_unlock(&kvm->srcu, idx);
3575 static void seg_setup(int seg)
3577 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3580 vmcs_write16(sf->selector, 0);
3581 vmcs_writel(sf->base, 0);
3582 vmcs_write32(sf->limit, 0xffff);
3583 if (enable_unrestricted_guest) {
3585 if (seg == VCPU_SREG_CS)
3586 ar |= 0x08; /* code segment */
3590 vmcs_write32(sf->ar_bytes, ar);
3593 static int alloc_apic_access_page(struct kvm *kvm)
3595 struct kvm_userspace_memory_region kvm_userspace_mem;
3598 mutex_lock(&kvm->slots_lock);
3599 if (kvm->arch.apic_access_page)
3601 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3602 kvm_userspace_mem.flags = 0;
3603 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
3604 kvm_userspace_mem.memory_size = PAGE_SIZE;
3605 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3609 kvm->arch.apic_access_page = gfn_to_page(kvm, 0xfee00);
3611 mutex_unlock(&kvm->slots_lock);
3615 static int alloc_identity_pagetable(struct kvm *kvm)
3617 struct kvm_userspace_memory_region kvm_userspace_mem;
3620 mutex_lock(&kvm->slots_lock);
3621 if (kvm->arch.ept_identity_pagetable)
3623 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
3624 kvm_userspace_mem.flags = 0;
3625 kvm_userspace_mem.guest_phys_addr =
3626 kvm->arch.ept_identity_map_addr;
3627 kvm_userspace_mem.memory_size = PAGE_SIZE;
3628 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem, 0);
3632 kvm->arch.ept_identity_pagetable = gfn_to_page(kvm,
3633 kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
3635 mutex_unlock(&kvm->slots_lock);
3639 static void allocate_vpid(struct vcpu_vmx *vmx)
3646 spin_lock(&vmx_vpid_lock);
3647 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3648 if (vpid < VMX_NR_VPIDS) {
3650 __set_bit(vpid, vmx_vpid_bitmap);
3652 spin_unlock(&vmx_vpid_lock);
3655 static void free_vpid(struct vcpu_vmx *vmx)
3659 spin_lock(&vmx_vpid_lock);
3661 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3662 spin_unlock(&vmx_vpid_lock);
3665 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap, u32 msr)
3667 int f = sizeof(unsigned long);
3669 if (!cpu_has_vmx_msr_bitmap())
3673 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3674 * have the write-low and read-high bitmap offsets the wrong way round.
3675 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3677 if (msr <= 0x1fff) {
3678 __clear_bit(msr, msr_bitmap + 0x000 / f); /* read-low */
3679 __clear_bit(msr, msr_bitmap + 0x800 / f); /* write-low */
3680 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3682 __clear_bit(msr, msr_bitmap + 0x400 / f); /* read-high */
3683 __clear_bit(msr, msr_bitmap + 0xc00 / f); /* write-high */
3687 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
3690 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy, msr);
3691 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode, msr);
3695 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
3696 * will not change in the lifetime of the guest.
3697 * Note that host-state that does change is set elsewhere. E.g., host-state
3698 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
3700 static void vmx_set_constant_host_state(void)
3706 vmcs_writel(HOST_CR0, read_cr0() | X86_CR0_TS); /* 22.2.3 */
3707 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
3708 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
3710 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
3711 #ifdef CONFIG_X86_64
3713 * Load null selectors, so we can avoid reloading them in
3714 * __vmx_load_host_state(), in case userspace uses the null selectors
3715 * too (the expected case).
3717 vmcs_write16(HOST_DS_SELECTOR, 0);
3718 vmcs_write16(HOST_ES_SELECTOR, 0);
3720 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3721 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3723 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
3724 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
3726 native_store_idt(&dt);
3727 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
3729 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
3731 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
3732 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
3733 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
3734 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
3736 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
3737 rdmsr(MSR_IA32_CR_PAT, low32, high32);
3738 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
3742 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
3744 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
3746 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
3747 if (is_guest_mode(&vmx->vcpu))
3748 vmx->vcpu.arch.cr4_guest_owned_bits &=
3749 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
3750 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
3753 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
3755 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
3756 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
3757 exec_control &= ~CPU_BASED_TPR_SHADOW;
3758 #ifdef CONFIG_X86_64
3759 exec_control |= CPU_BASED_CR8_STORE_EXITING |
3760 CPU_BASED_CR8_LOAD_EXITING;
3764 exec_control |= CPU_BASED_CR3_STORE_EXITING |
3765 CPU_BASED_CR3_LOAD_EXITING |
3766 CPU_BASED_INVLPG_EXITING;
3767 return exec_control;
3770 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
3772 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
3773 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3774 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
3776 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
3778 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
3779 enable_unrestricted_guest = 0;
3780 /* Enable INVPCID for non-ept guests may cause performance regression. */
3781 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
3783 if (!enable_unrestricted_guest)
3784 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
3786 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
3787 return exec_control;
3790 static void ept_set_mmio_spte_mask(void)
3793 * EPT Misconfigurations can be generated if the value of bits 2:0
3794 * of an EPT paging-structure entry is 110b (write/execute).
3795 * Also, magic bits (0xffull << 49) is set to quickly identify mmio
3798 kvm_mmu_set_mmio_spte_mask(0xffull << 49 | 0x6ull);
3802 * Sets up the vmcs for emulated real mode.
3804 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
3806 #ifdef CONFIG_X86_64
3812 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
3813 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
3815 if (cpu_has_vmx_msr_bitmap())
3816 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
3818 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
3821 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
3822 vmcs_config.pin_based_exec_ctrl);
3824 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
3826 if (cpu_has_secondary_exec_ctrls()) {
3827 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
3828 vmx_secondary_exec_control(vmx));
3832 vmcs_write32(PLE_GAP, ple_gap);
3833 vmcs_write32(PLE_WINDOW, ple_window);
3836 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
3837 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
3838 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
3840 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
3841 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
3842 vmx_set_constant_host_state();
3843 #ifdef CONFIG_X86_64
3844 rdmsrl(MSR_FS_BASE, a);
3845 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
3846 rdmsrl(MSR_GS_BASE, a);
3847 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
3849 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
3850 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
3853 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
3854 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
3855 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
3856 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
3857 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
3859 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3860 u32 msr_low, msr_high;
3862 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
3863 host_pat = msr_low | ((u64) msr_high << 32);
3864 /* Write the default value follow host pat */
3865 vmcs_write64(GUEST_IA32_PAT, host_pat);
3866 /* Keep arch.pat sync with GUEST_IA32_PAT */
3867 vmx->vcpu.arch.pat = host_pat;
3870 for (i = 0; i < NR_VMX_MSR; ++i) {
3871 u32 index = vmx_msr_index[i];
3872 u32 data_low, data_high;
3875 if (rdmsr_safe(index, &data_low, &data_high) < 0)
3877 if (wrmsr_safe(index, data_low, data_high) < 0)
3879 vmx->guest_msrs[j].index = i;
3880 vmx->guest_msrs[j].data = 0;
3881 vmx->guest_msrs[j].mask = -1ull;
3885 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
3887 /* 22.2.1, 20.8.1 */
3888 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
3890 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
3891 set_cr4_guest_host_mask(vmx);
3893 kvm_write_tsc(&vmx->vcpu, 0);
3898 static int vmx_vcpu_reset(struct kvm_vcpu *vcpu)
3900 struct vcpu_vmx *vmx = to_vmx(vcpu);
3904 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP));
3906 vmx->rmode.vm86_active = 0;
3908 vmx->soft_vnmi_blocked = 0;
3910 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
3911 kvm_set_cr8(&vmx->vcpu, 0);
3912 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
3913 if (kvm_vcpu_is_bsp(&vmx->vcpu))
3914 msr |= MSR_IA32_APICBASE_BSP;
3915 kvm_set_apic_base(&vmx->vcpu, msr);
3917 ret = fx_init(&vmx->vcpu);
3921 vmx_segment_cache_clear(vmx);
3923 seg_setup(VCPU_SREG_CS);
3925 * GUEST_CS_BASE should really be 0xffff0000, but VT vm86 mode
3926 * insists on having GUEST_CS_BASE == GUEST_CS_SELECTOR << 4. Sigh.
3928 if (kvm_vcpu_is_bsp(&vmx->vcpu)) {
3929 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
3930 vmcs_writel(GUEST_CS_BASE, 0x000f0000);
3932 vmcs_write16(GUEST_CS_SELECTOR, vmx->vcpu.arch.sipi_vector << 8);
3933 vmcs_writel(GUEST_CS_BASE, vmx->vcpu.arch.sipi_vector << 12);
3936 seg_setup(VCPU_SREG_DS);
3937 seg_setup(VCPU_SREG_ES);
3938 seg_setup(VCPU_SREG_FS);
3939 seg_setup(VCPU_SREG_GS);
3940 seg_setup(VCPU_SREG_SS);
3942 vmcs_write16(GUEST_TR_SELECTOR, 0);
3943 vmcs_writel(GUEST_TR_BASE, 0);
3944 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
3945 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3947 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
3948 vmcs_writel(GUEST_LDTR_BASE, 0);
3949 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
3950 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
3952 vmcs_write32(GUEST_SYSENTER_CS, 0);
3953 vmcs_writel(GUEST_SYSENTER_ESP, 0);
3954 vmcs_writel(GUEST_SYSENTER_EIP, 0);
3956 vmcs_writel(GUEST_RFLAGS, 0x02);
3957 if (kvm_vcpu_is_bsp(&vmx->vcpu))
3958 kvm_rip_write(vcpu, 0xfff0);
3960 kvm_rip_write(vcpu, 0);
3961 kvm_register_write(vcpu, VCPU_REGS_RSP, 0);
3963 vmcs_writel(GUEST_DR7, 0x400);
3965 vmcs_writel(GUEST_GDTR_BASE, 0);
3966 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
3968 vmcs_writel(GUEST_IDTR_BASE, 0);
3969 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
3971 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
3972 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
3973 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
3975 /* Special registers */
3976 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
3980 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
3982 if (cpu_has_vmx_tpr_shadow()) {
3983 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
3984 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
3985 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
3986 __pa(vmx->vcpu.arch.apic->regs));
3987 vmcs_write32(TPR_THRESHOLD, 0);
3990 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
3991 vmcs_write64(APIC_ACCESS_ADDR,
3992 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
3995 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
3997 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
3998 vcpu->srcu_idx = srcu_read_lock(&vcpu->kvm->srcu);
3999 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
4000 srcu_read_unlock(&vcpu->kvm->srcu, vcpu->srcu_idx);
4001 vmx_set_cr4(&vmx->vcpu, 0);
4002 vmx_set_efer(&vmx->vcpu, 0);
4003 vmx_fpu_activate(&vmx->vcpu);
4004 update_exception_bitmap(&vmx->vcpu);
4006 vpid_sync_context(vmx);
4010 /* HACK: Don't enable emulation on guest boot/reset */
4011 vmx->emulation_required = 0;
4018 * In nested virtualization, check if L1 asked to exit on external interrupts.
4019 * For most existing hypervisors, this will always return true.
4021 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4023 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4024 PIN_BASED_EXT_INTR_MASK;
4027 static void enable_irq_window(struct kvm_vcpu *vcpu)
4029 u32 cpu_based_vm_exec_control;
4030 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
4032 * We get here if vmx_interrupt_allowed() said we can't
4033 * inject to L1 now because L2 must run. Ask L2 to exit
4034 * right after entry, so we can inject to L1 more promptly.
4036 kvm_make_request(KVM_REQ_IMMEDIATE_EXIT, vcpu);
4040 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4041 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4042 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4045 static void enable_nmi_window(struct kvm_vcpu *vcpu)
4047 u32 cpu_based_vm_exec_control;
4049 if (!cpu_has_virtual_nmis()) {
4050 enable_irq_window(vcpu);
4054 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
4055 enable_irq_window(vcpu);
4058 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4059 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4060 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4063 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4065 struct vcpu_vmx *vmx = to_vmx(vcpu);
4067 int irq = vcpu->arch.interrupt.nr;
4069 trace_kvm_inj_virq(irq);
4071 ++vcpu->stat.irq_injections;
4072 if (vmx->rmode.vm86_active) {
4074 if (vcpu->arch.interrupt.soft)
4075 inc_eip = vcpu->arch.event_exit_inst_len;
4076 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4077 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4080 intr = irq | INTR_INFO_VALID_MASK;
4081 if (vcpu->arch.interrupt.soft) {
4082 intr |= INTR_TYPE_SOFT_INTR;
4083 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4084 vmx->vcpu.arch.event_exit_inst_len);
4086 intr |= INTR_TYPE_EXT_INTR;
4087 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4090 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4092 struct vcpu_vmx *vmx = to_vmx(vcpu);
4094 if (is_guest_mode(vcpu))
4097 if (!cpu_has_virtual_nmis()) {
4099 * Tracking the NMI-blocked state in software is built upon
4100 * finding the next open IRQ window. This, in turn, depends on
4101 * well-behaving guests: They have to keep IRQs disabled at
4102 * least as long as the NMI handler runs. Otherwise we may
4103 * cause NMI nesting, maybe breaking the guest. But as this is
4104 * highly unlikely, we can live with the residual risk.
4106 vmx->soft_vnmi_blocked = 1;
4107 vmx->vnmi_blocked_time = 0;
4110 ++vcpu->stat.nmi_injections;
4111 vmx->nmi_known_unmasked = false;
4112 if (vmx->rmode.vm86_active) {
4113 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4114 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4117 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4118 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4121 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4123 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
4126 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4127 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4128 | GUEST_INTR_STATE_NMI));
4131 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4133 if (!cpu_has_virtual_nmis())
4134 return to_vmx(vcpu)->soft_vnmi_blocked;
4135 if (to_vmx(vcpu)->nmi_known_unmasked)
4137 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4140 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4142 struct vcpu_vmx *vmx = to_vmx(vcpu);
4144 if (!cpu_has_virtual_nmis()) {
4145 if (vmx->soft_vnmi_blocked != masked) {
4146 vmx->soft_vnmi_blocked = masked;
4147 vmx->vnmi_blocked_time = 0;
4150 vmx->nmi_known_unmasked = !masked;
4152 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4153 GUEST_INTR_STATE_NMI);
4155 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4156 GUEST_INTR_STATE_NMI);
4160 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4162 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu)) {
4163 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4164 if (to_vmx(vcpu)->nested.nested_run_pending ||
4165 (vmcs12->idt_vectoring_info_field &
4166 VECTORING_INFO_VALID_MASK))
4168 nested_vmx_vmexit(vcpu);
4169 vmcs12->vm_exit_reason = EXIT_REASON_EXTERNAL_INTERRUPT;
4170 vmcs12->vm_exit_intr_info = 0;
4171 /* fall through to normal code, but now in L1, not L2 */
4174 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4175 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4176 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4179 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4182 struct kvm_userspace_memory_region tss_mem = {
4183 .slot = TSS_PRIVATE_MEMSLOT,
4184 .guest_phys_addr = addr,
4185 .memory_size = PAGE_SIZE * 3,
4189 ret = kvm_set_memory_region(kvm, &tss_mem, 0);
4192 kvm->arch.tss_addr = addr;
4193 if (!init_rmode_tss(kvm))
4199 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4200 int vec, u32 err_code)
4203 * Instruction with address size override prefix opcode 0x67
4204 * Cause the #SS fault with 0 error code in VM86 mode.
4206 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0)
4207 if (emulate_instruction(vcpu, 0) == EMULATE_DONE)
4210 * Forward all other exceptions that are valid in real mode.
4211 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4212 * the required debugging infrastructure rework.
4216 if (vcpu->guest_debug &
4217 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4219 kvm_queue_exception(vcpu, vec);
4223 * Update instruction length as we may reinject the exception
4224 * from user space while in guest debugging mode.
4226 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4227 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4228 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4239 kvm_queue_exception(vcpu, vec);
4246 * Trigger machine check on the host. We assume all the MSRs are already set up
4247 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4248 * We pass a fake environment to the machine check handler because we want
4249 * the guest to be always treated like user space, no matter what context
4250 * it used internally.
4252 static void kvm_machine_check(void)
4254 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4255 struct pt_regs regs = {
4256 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4257 .flags = X86_EFLAGS_IF,
4260 do_machine_check(®s, 0);
4264 static int handle_machine_check(struct kvm_vcpu *vcpu)
4266 /* already handled by vcpu_run */
4270 static int handle_exception(struct kvm_vcpu *vcpu)
4272 struct vcpu_vmx *vmx = to_vmx(vcpu);
4273 struct kvm_run *kvm_run = vcpu->run;
4274 u32 intr_info, ex_no, error_code;
4275 unsigned long cr2, rip, dr6;
4277 enum emulation_result er;
4279 vect_info = vmx->idt_vectoring_info;
4280 intr_info = vmx->exit_intr_info;
4282 if (is_machine_check(intr_info))
4283 return handle_machine_check(vcpu);
4285 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4286 !is_page_fault(intr_info)) {
4287 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4288 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4289 vcpu->run->internal.ndata = 2;
4290 vcpu->run->internal.data[0] = vect_info;
4291 vcpu->run->internal.data[1] = intr_info;
4295 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
4296 return 1; /* already handled by vmx_vcpu_run() */
4298 if (is_no_device(intr_info)) {
4299 vmx_fpu_activate(vcpu);
4303 if (is_invalid_opcode(intr_info)) {
4304 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
4305 if (er != EMULATE_DONE)
4306 kvm_queue_exception(vcpu, UD_VECTOR);
4311 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4312 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4313 if (is_page_fault(intr_info)) {
4314 /* EPT won't cause page fault directly */
4316 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4317 trace_kvm_page_fault(cr2, error_code);
4319 if (kvm_event_needs_reinjection(vcpu))
4320 kvm_mmu_unprotect_page_virt(vcpu, cr2);
4321 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
4324 if (vmx->rmode.vm86_active &&
4325 handle_rmode_exception(vcpu, intr_info & INTR_INFO_VECTOR_MASK,
4327 if (vcpu->arch.halt_request) {
4328 vcpu->arch.halt_request = 0;
4329 return kvm_emulate_halt(vcpu);
4334 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4337 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4338 if (!(vcpu->guest_debug &
4339 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4340 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
4341 kvm_queue_exception(vcpu, DB_VECTOR);
4344 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4345 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4349 * Update instruction length as we may reinject #BP from
4350 * user space while in guest debugging mode. Reading it for
4351 * #DB as well causes no harm, it is not used in that case.
4353 vmx->vcpu.arch.event_exit_inst_len =
4354 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4355 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4356 rip = kvm_rip_read(vcpu);
4357 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4358 kvm_run->debug.arch.exception = ex_no;
4361 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4362 kvm_run->ex.exception = ex_no;
4363 kvm_run->ex.error_code = error_code;
4369 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4371 ++vcpu->stat.irq_exits;
4375 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4377 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4381 static int handle_io(struct kvm_vcpu *vcpu)
4383 unsigned long exit_qualification;
4384 int size, in, string;
4387 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4388 string = (exit_qualification & 16) != 0;
4389 in = (exit_qualification & 8) != 0;
4391 ++vcpu->stat.io_exits;
4394 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4396 port = exit_qualification >> 16;
4397 size = (exit_qualification & 7) + 1;
4398 skip_emulated_instruction(vcpu);
4400 return kvm_fast_pio_out(vcpu, size, port);
4404 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4407 * Patch in the VMCALL instruction:
4409 hypercall[0] = 0x0f;
4410 hypercall[1] = 0x01;
4411 hypercall[2] = 0xc1;
4414 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4415 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4417 if (to_vmx(vcpu)->nested.vmxon &&
4418 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
4421 if (is_guest_mode(vcpu)) {
4423 * We get here when L2 changed cr0 in a way that did not change
4424 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4425 * but did change L0 shadowed bits. This can currently happen
4426 * with the TS bit: L0 may want to leave TS on (for lazy fpu
4427 * loading) while pretending to allow the guest to change it.
4429 if (kvm_set_cr0(vcpu, (val & vcpu->arch.cr0_guest_owned_bits) |
4430 (vcpu->arch.cr0 & ~vcpu->arch.cr0_guest_owned_bits)))
4432 vmcs_writel(CR0_READ_SHADOW, val);
4435 return kvm_set_cr0(vcpu, val);
4438 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4440 if (is_guest_mode(vcpu)) {
4441 if (kvm_set_cr4(vcpu, (val & vcpu->arch.cr4_guest_owned_bits) |
4442 (vcpu->arch.cr4 & ~vcpu->arch.cr4_guest_owned_bits)))
4444 vmcs_writel(CR4_READ_SHADOW, val);
4447 return kvm_set_cr4(vcpu, val);
4450 /* called to set cr0 as approriate for clts instruction exit. */
4451 static void handle_clts(struct kvm_vcpu *vcpu)
4453 if (is_guest_mode(vcpu)) {
4455 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
4456 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
4457 * just pretend it's off (also in arch.cr0 for fpu_activate).
4459 vmcs_writel(CR0_READ_SHADOW,
4460 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
4461 vcpu->arch.cr0 &= ~X86_CR0_TS;
4463 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4466 static int handle_cr(struct kvm_vcpu *vcpu)
4468 unsigned long exit_qualification, val;
4473 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4474 cr = exit_qualification & 15;
4475 reg = (exit_qualification >> 8) & 15;
4476 switch ((exit_qualification >> 4) & 3) {
4477 case 0: /* mov to cr */
4478 val = kvm_register_read(vcpu, reg);
4479 trace_kvm_cr_write(cr, val);
4482 err = handle_set_cr0(vcpu, val);
4483 kvm_complete_insn_gp(vcpu, err);
4486 err = kvm_set_cr3(vcpu, val);
4487 kvm_complete_insn_gp(vcpu, err);
4490 err = handle_set_cr4(vcpu, val);
4491 kvm_complete_insn_gp(vcpu, err);
4494 u8 cr8_prev = kvm_get_cr8(vcpu);
4495 u8 cr8 = kvm_register_read(vcpu, reg);
4496 err = kvm_set_cr8(vcpu, cr8);
4497 kvm_complete_insn_gp(vcpu, err);
4498 if (irqchip_in_kernel(vcpu->kvm))
4500 if (cr8_prev <= cr8)
4502 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4509 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4510 skip_emulated_instruction(vcpu);
4511 vmx_fpu_activate(vcpu);
4513 case 1: /*mov from cr*/
4516 val = kvm_read_cr3(vcpu);
4517 kvm_register_write(vcpu, reg, val);
4518 trace_kvm_cr_read(cr, val);
4519 skip_emulated_instruction(vcpu);
4522 val = kvm_get_cr8(vcpu);
4523 kvm_register_write(vcpu, reg, val);
4524 trace_kvm_cr_read(cr, val);
4525 skip_emulated_instruction(vcpu);
4530 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4531 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4532 kvm_lmsw(vcpu, val);
4534 skip_emulated_instruction(vcpu);
4539 vcpu->run->exit_reason = 0;
4540 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4541 (int)(exit_qualification >> 4) & 3, cr);
4545 static int handle_dr(struct kvm_vcpu *vcpu)
4547 unsigned long exit_qualification;
4550 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
4551 if (!kvm_require_cpl(vcpu, 0))
4553 dr = vmcs_readl(GUEST_DR7);
4556 * As the vm-exit takes precedence over the debug trap, we
4557 * need to emulate the latter, either for the host or the
4558 * guest debugging itself.
4560 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
4561 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
4562 vcpu->run->debug.arch.dr7 = dr;
4563 vcpu->run->debug.arch.pc =
4564 vmcs_readl(GUEST_CS_BASE) +
4565 vmcs_readl(GUEST_RIP);
4566 vcpu->run->debug.arch.exception = DB_VECTOR;
4567 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
4570 vcpu->arch.dr7 &= ~DR7_GD;
4571 vcpu->arch.dr6 |= DR6_BD;
4572 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
4573 kvm_queue_exception(vcpu, DB_VECTOR);
4578 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4579 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
4580 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
4581 if (exit_qualification & TYPE_MOV_FROM_DR) {
4583 if (!kvm_get_dr(vcpu, dr, &val))
4584 kvm_register_write(vcpu, reg, val);
4586 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
4587 skip_emulated_instruction(vcpu);
4591 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
4593 vmcs_writel(GUEST_DR7, val);
4596 static int handle_cpuid(struct kvm_vcpu *vcpu)
4598 kvm_emulate_cpuid(vcpu);
4602 static int handle_rdmsr(struct kvm_vcpu *vcpu)
4604 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4607 if (vmx_get_msr(vcpu, ecx, &data)) {
4608 trace_kvm_msr_read_ex(ecx);
4609 kvm_inject_gp(vcpu, 0);
4613 trace_kvm_msr_read(ecx, data);
4615 /* FIXME: handling of bits 32:63 of rax, rdx */
4616 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
4617 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
4618 skip_emulated_instruction(vcpu);
4622 static int handle_wrmsr(struct kvm_vcpu *vcpu)
4624 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
4625 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
4626 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
4628 if (vmx_set_msr(vcpu, ecx, data) != 0) {
4629 trace_kvm_msr_write_ex(ecx, data);
4630 kvm_inject_gp(vcpu, 0);
4634 trace_kvm_msr_write(ecx, data);
4635 skip_emulated_instruction(vcpu);
4639 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
4641 kvm_make_request(KVM_REQ_EVENT, vcpu);
4645 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
4647 u32 cpu_based_vm_exec_control;
4649 /* clear pending irq */
4650 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4651 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
4652 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4654 kvm_make_request(KVM_REQ_EVENT, vcpu);
4656 ++vcpu->stat.irq_window_exits;
4659 * If the user space waits to inject interrupts, exit as soon as
4662 if (!irqchip_in_kernel(vcpu->kvm) &&
4663 vcpu->run->request_interrupt_window &&
4664 !kvm_cpu_has_interrupt(vcpu)) {
4665 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
4671 static int handle_halt(struct kvm_vcpu *vcpu)
4673 skip_emulated_instruction(vcpu);
4674 return kvm_emulate_halt(vcpu);
4677 static int handle_vmcall(struct kvm_vcpu *vcpu)
4679 skip_emulated_instruction(vcpu);
4680 kvm_emulate_hypercall(vcpu);
4684 static int handle_invd(struct kvm_vcpu *vcpu)
4686 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4689 static int handle_invlpg(struct kvm_vcpu *vcpu)
4691 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4693 kvm_mmu_invlpg(vcpu, exit_qualification);
4694 skip_emulated_instruction(vcpu);
4698 static int handle_rdpmc(struct kvm_vcpu *vcpu)
4702 err = kvm_rdpmc(vcpu);
4703 kvm_complete_insn_gp(vcpu, err);
4708 static int handle_wbinvd(struct kvm_vcpu *vcpu)
4710 skip_emulated_instruction(vcpu);
4711 kvm_emulate_wbinvd(vcpu);
4715 static int handle_xsetbv(struct kvm_vcpu *vcpu)
4717 u64 new_bv = kvm_read_edx_eax(vcpu);
4718 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
4720 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
4721 skip_emulated_instruction(vcpu);
4725 static int handle_apic_access(struct kvm_vcpu *vcpu)
4727 if (likely(fasteoi)) {
4728 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4729 int access_type, offset;
4731 access_type = exit_qualification & APIC_ACCESS_TYPE;
4732 offset = exit_qualification & APIC_ACCESS_OFFSET;
4734 * Sane guest uses MOV to write EOI, with written value
4735 * not cared. So make a short-circuit here by avoiding
4736 * heavy instruction emulation.
4738 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
4739 (offset == APIC_EOI)) {
4740 kvm_lapic_set_eoi(vcpu);
4741 skip_emulated_instruction(vcpu);
4745 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4748 static int handle_task_switch(struct kvm_vcpu *vcpu)
4750 struct vcpu_vmx *vmx = to_vmx(vcpu);
4751 unsigned long exit_qualification;
4752 bool has_error_code = false;
4755 int reason, type, idt_v, idt_index;
4757 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
4758 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
4759 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
4761 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4763 reason = (u32)exit_qualification >> 30;
4764 if (reason == TASK_SWITCH_GATE && idt_v) {
4766 case INTR_TYPE_NMI_INTR:
4767 vcpu->arch.nmi_injected = false;
4768 vmx_set_nmi_mask(vcpu, true);
4770 case INTR_TYPE_EXT_INTR:
4771 case INTR_TYPE_SOFT_INTR:
4772 kvm_clear_interrupt_queue(vcpu);
4774 case INTR_TYPE_HARD_EXCEPTION:
4775 if (vmx->idt_vectoring_info &
4776 VECTORING_INFO_DELIVER_CODE_MASK) {
4777 has_error_code = true;
4779 vmcs_read32(IDT_VECTORING_ERROR_CODE);
4782 case INTR_TYPE_SOFT_EXCEPTION:
4783 kvm_clear_exception_queue(vcpu);
4789 tss_selector = exit_qualification;
4791 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
4792 type != INTR_TYPE_EXT_INTR &&
4793 type != INTR_TYPE_NMI_INTR))
4794 skip_emulated_instruction(vcpu);
4796 if (kvm_task_switch(vcpu, tss_selector,
4797 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
4798 has_error_code, error_code) == EMULATE_FAIL) {
4799 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4800 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4801 vcpu->run->internal.ndata = 0;
4805 /* clear all local breakpoint enable flags */
4806 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
4809 * TODO: What about debug traps on tss switch?
4810 * Are we supposed to inject them and update dr6?
4816 static int handle_ept_violation(struct kvm_vcpu *vcpu)
4818 unsigned long exit_qualification;
4823 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4825 if (exit_qualification & (1 << 6)) {
4826 printk(KERN_ERR "EPT: GPA exceeds GAW!\n");
4830 gla_validity = (exit_qualification >> 7) & 0x3;
4831 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
4832 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
4833 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
4834 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
4835 vmcs_readl(GUEST_LINEAR_ADDRESS));
4836 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
4837 (long unsigned int)exit_qualification);
4838 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4839 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
4843 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4844 trace_kvm_page_fault(gpa, exit_qualification);
4846 /* It is a write fault? */
4847 error_code = exit_qualification & (1U << 1);
4848 /* ept page table is present? */
4849 error_code |= (exit_qualification >> 3) & 0x1;
4851 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
4854 static u64 ept_rsvd_mask(u64 spte, int level)
4859 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
4860 mask |= (1ULL << i);
4863 /* bits 7:3 reserved */
4865 else if (level == 2) {
4866 if (spte & (1ULL << 7))
4867 /* 2MB ref, bits 20:12 reserved */
4870 /* bits 6:3 reserved */
4877 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
4880 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
4882 /* 010b (write-only) */
4883 WARN_ON((spte & 0x7) == 0x2);
4885 /* 110b (write/execute) */
4886 WARN_ON((spte & 0x7) == 0x6);
4888 /* 100b (execute-only) and value not supported by logical processor */
4889 if (!cpu_has_vmx_ept_execute_only())
4890 WARN_ON((spte & 0x7) == 0x4);
4894 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
4896 if (rsvd_bits != 0) {
4897 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
4898 __func__, rsvd_bits);
4902 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
4903 u64 ept_mem_type = (spte & 0x38) >> 3;
4905 if (ept_mem_type == 2 || ept_mem_type == 3 ||
4906 ept_mem_type == 7) {
4907 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
4908 __func__, ept_mem_type);
4915 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
4918 int nr_sptes, i, ret;
4921 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
4923 ret = handle_mmio_page_fault_common(vcpu, gpa, true);
4924 if (likely(ret == 1))
4925 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
4930 /* It is the real ept misconfig */
4931 printk(KERN_ERR "EPT: Misconfiguration.\n");
4932 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
4934 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
4936 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
4937 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
4939 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
4940 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
4945 static int handle_nmi_window(struct kvm_vcpu *vcpu)
4947 u32 cpu_based_vm_exec_control;
4949 /* clear pending NMI */
4950 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4951 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
4952 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4953 ++vcpu->stat.nmi_window_exits;
4954 kvm_make_request(KVM_REQ_EVENT, vcpu);
4959 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
4961 struct vcpu_vmx *vmx = to_vmx(vcpu);
4962 enum emulation_result err = EMULATE_DONE;
4965 bool intr_window_requested;
4966 unsigned count = 130;
4968 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4969 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
4971 while (!guest_state_valid(vcpu) && count-- != 0) {
4972 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
4973 return handle_interrupt_window(&vmx->vcpu);
4975 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
4978 err = emulate_instruction(vcpu, 0);
4980 if (err == EMULATE_DO_MMIO) {
4985 if (err != EMULATE_DONE) {
4986 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4987 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
4988 vcpu->run->internal.ndata = 0;
4992 if (signal_pending(current))
4998 vmx->emulation_required = !guest_state_valid(vcpu);
5004 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5005 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5007 static int handle_pause(struct kvm_vcpu *vcpu)
5009 skip_emulated_instruction(vcpu);
5010 kvm_vcpu_on_spin(vcpu);
5015 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5017 kvm_queue_exception(vcpu, UD_VECTOR);
5022 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
5023 * We could reuse a single VMCS for all the L2 guests, but we also want the
5024 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
5025 * allows keeping them loaded on the processor, and in the future will allow
5026 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
5027 * every entry if they never change.
5028 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
5029 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
5031 * The following functions allocate and free a vmcs02 in this pool.
5034 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
5035 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
5037 struct vmcs02_list *item;
5038 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5039 if (item->vmptr == vmx->nested.current_vmptr) {
5040 list_move(&item->list, &vmx->nested.vmcs02_pool);
5041 return &item->vmcs02;
5044 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
5045 /* Recycle the least recently used VMCS. */
5046 item = list_entry(vmx->nested.vmcs02_pool.prev,
5047 struct vmcs02_list, list);
5048 item->vmptr = vmx->nested.current_vmptr;
5049 list_move(&item->list, &vmx->nested.vmcs02_pool);
5050 return &item->vmcs02;
5053 /* Create a new VMCS */
5054 item = (struct vmcs02_list *)
5055 kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
5058 item->vmcs02.vmcs = alloc_vmcs();
5059 if (!item->vmcs02.vmcs) {
5063 loaded_vmcs_init(&item->vmcs02);
5064 item->vmptr = vmx->nested.current_vmptr;
5065 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
5066 vmx->nested.vmcs02_num++;
5067 return &item->vmcs02;
5070 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
5071 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
5073 struct vmcs02_list *item;
5074 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5075 if (item->vmptr == vmptr) {
5076 free_loaded_vmcs(&item->vmcs02);
5077 list_del(&item->list);
5079 vmx->nested.vmcs02_num--;
5085 * Free all VMCSs saved for this vcpu, except the one pointed by
5086 * vmx->loaded_vmcs. These include the VMCSs in vmcs02_pool (except the one
5087 * currently used, if running L2), and vmcs01 when running L2.
5089 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
5091 struct vmcs02_list *item, *n;
5092 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
5093 if (vmx->loaded_vmcs != &item->vmcs02)
5094 free_loaded_vmcs(&item->vmcs02);
5095 list_del(&item->list);
5098 vmx->nested.vmcs02_num = 0;
5100 if (vmx->loaded_vmcs != &vmx->vmcs01)
5101 free_loaded_vmcs(&vmx->vmcs01);
5105 * Emulate the VMXON instruction.
5106 * Currently, we just remember that VMX is active, and do not save or even
5107 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
5108 * do not currently need to store anything in that guest-allocated memory
5109 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
5110 * argument is different from the VMXON pointer (which the spec says they do).
5112 static int handle_vmon(struct kvm_vcpu *vcpu)
5114 struct kvm_segment cs;
5115 struct vcpu_vmx *vmx = to_vmx(vcpu);
5117 /* The Intel VMX Instruction Reference lists a bunch of bits that
5118 * are prerequisite to running VMXON, most notably cr4.VMXE must be
5119 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
5120 * Otherwise, we should fail with #UD. We test these now:
5122 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
5123 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
5124 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
5125 kvm_queue_exception(vcpu, UD_VECTOR);
5129 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5130 if (is_long_mode(vcpu) && !cs.l) {
5131 kvm_queue_exception(vcpu, UD_VECTOR);
5135 if (vmx_get_cpl(vcpu)) {
5136 kvm_inject_gp(vcpu, 0);
5140 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
5141 vmx->nested.vmcs02_num = 0;
5143 vmx->nested.vmxon = true;
5145 skip_emulated_instruction(vcpu);
5150 * Intel's VMX Instruction Reference specifies a common set of prerequisites
5151 * for running VMX instructions (except VMXON, whose prerequisites are
5152 * slightly different). It also specifies what exception to inject otherwise.
5154 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
5156 struct kvm_segment cs;
5157 struct vcpu_vmx *vmx = to_vmx(vcpu);
5159 if (!vmx->nested.vmxon) {
5160 kvm_queue_exception(vcpu, UD_VECTOR);
5164 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5165 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
5166 (is_long_mode(vcpu) && !cs.l)) {
5167 kvm_queue_exception(vcpu, UD_VECTOR);
5171 if (vmx_get_cpl(vcpu)) {
5172 kvm_inject_gp(vcpu, 0);
5180 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
5181 * just stops using VMX.
5183 static void free_nested(struct vcpu_vmx *vmx)
5185 if (!vmx->nested.vmxon)
5187 vmx->nested.vmxon = false;
5188 if (vmx->nested.current_vmptr != -1ull) {
5189 kunmap(vmx->nested.current_vmcs12_page);
5190 nested_release_page(vmx->nested.current_vmcs12_page);
5191 vmx->nested.current_vmptr = -1ull;
5192 vmx->nested.current_vmcs12 = NULL;
5194 /* Unpin physical memory we referred to in current vmcs02 */
5195 if (vmx->nested.apic_access_page) {
5196 nested_release_page(vmx->nested.apic_access_page);
5197 vmx->nested.apic_access_page = 0;
5200 nested_free_all_saved_vmcss(vmx);
5203 /* Emulate the VMXOFF instruction */
5204 static int handle_vmoff(struct kvm_vcpu *vcpu)
5206 if (!nested_vmx_check_permission(vcpu))
5208 free_nested(to_vmx(vcpu));
5209 skip_emulated_instruction(vcpu);
5214 * Decode the memory-address operand of a vmx instruction, as recorded on an
5215 * exit caused by such an instruction (run by a guest hypervisor).
5216 * On success, returns 0. When the operand is invalid, returns 1 and throws
5219 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
5220 unsigned long exit_qualification,
5221 u32 vmx_instruction_info, gva_t *ret)
5224 * According to Vol. 3B, "Information for VM Exits Due to Instruction
5225 * Execution", on an exit, vmx_instruction_info holds most of the
5226 * addressing components of the operand. Only the displacement part
5227 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5228 * For how an actual address is calculated from all these components,
5229 * refer to Vol. 1, "Operand Addressing".
5231 int scaling = vmx_instruction_info & 3;
5232 int addr_size = (vmx_instruction_info >> 7) & 7;
5233 bool is_reg = vmx_instruction_info & (1u << 10);
5234 int seg_reg = (vmx_instruction_info >> 15) & 7;
5235 int index_reg = (vmx_instruction_info >> 18) & 0xf;
5236 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5237 int base_reg = (vmx_instruction_info >> 23) & 0xf;
5238 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
5241 kvm_queue_exception(vcpu, UD_VECTOR);
5245 /* Addr = segment_base + offset */
5246 /* offset = base + [index * scale] + displacement */
5247 *ret = vmx_get_segment_base(vcpu, seg_reg);
5249 *ret += kvm_register_read(vcpu, base_reg);
5251 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
5252 *ret += exit_qualification; /* holds the displacement */
5254 if (addr_size == 1) /* 32 bit */
5258 * TODO: throw #GP (and return 1) in various cases that the VM*
5259 * instructions require it - e.g., offset beyond segment limit,
5260 * unusable or unreadable/unwritable segment, non-canonical 64-bit
5261 * address, and so on. Currently these are not checked.
5267 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
5268 * set the success or error code of an emulated VMX instruction, as specified
5269 * by Vol 2B, VMX Instruction Reference, "Conventions".
5271 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
5273 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
5274 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5275 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
5278 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
5280 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5281 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
5282 X86_EFLAGS_SF | X86_EFLAGS_OF))
5286 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
5287 u32 vm_instruction_error)
5289 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
5291 * failValid writes the error number to the current VMCS, which
5292 * can't be done there isn't a current VMCS.
5294 nested_vmx_failInvalid(vcpu);
5297 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5298 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5299 X86_EFLAGS_SF | X86_EFLAGS_OF))
5301 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
5304 /* Emulate the VMCLEAR instruction */
5305 static int handle_vmclear(struct kvm_vcpu *vcpu)
5307 struct vcpu_vmx *vmx = to_vmx(vcpu);
5310 struct vmcs12 *vmcs12;
5312 struct x86_exception e;
5314 if (!nested_vmx_check_permission(vcpu))
5317 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5318 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5321 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5322 sizeof(vmptr), &e)) {
5323 kvm_inject_page_fault(vcpu, &e);
5327 if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5328 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5329 skip_emulated_instruction(vcpu);
5333 if (vmptr == vmx->nested.current_vmptr) {
5334 kunmap(vmx->nested.current_vmcs12_page);
5335 nested_release_page(vmx->nested.current_vmcs12_page);
5336 vmx->nested.current_vmptr = -1ull;
5337 vmx->nested.current_vmcs12 = NULL;
5340 page = nested_get_page(vcpu, vmptr);
5343 * For accurate processor emulation, VMCLEAR beyond available
5344 * physical memory should do nothing at all. However, it is
5345 * possible that a nested vmx bug, not a guest hypervisor bug,
5346 * resulted in this case, so let's shut down before doing any
5349 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5352 vmcs12 = kmap(page);
5353 vmcs12->launch_state = 0;
5355 nested_release_page(page);
5357 nested_free_vmcs02(vmx, vmptr);
5359 skip_emulated_instruction(vcpu);
5360 nested_vmx_succeed(vcpu);
5364 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
5366 /* Emulate the VMLAUNCH instruction */
5367 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5369 return nested_vmx_run(vcpu, true);
5372 /* Emulate the VMRESUME instruction */
5373 static int handle_vmresume(struct kvm_vcpu *vcpu)
5376 return nested_vmx_run(vcpu, false);
5379 enum vmcs_field_type {
5380 VMCS_FIELD_TYPE_U16 = 0,
5381 VMCS_FIELD_TYPE_U64 = 1,
5382 VMCS_FIELD_TYPE_U32 = 2,
5383 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
5386 static inline int vmcs_field_type(unsigned long field)
5388 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
5389 return VMCS_FIELD_TYPE_U32;
5390 return (field >> 13) & 0x3 ;
5393 static inline int vmcs_field_readonly(unsigned long field)
5395 return (((field >> 10) & 0x3) == 1);
5399 * Read a vmcs12 field. Since these can have varying lengths and we return
5400 * one type, we chose the biggest type (u64) and zero-extend the return value
5401 * to that size. Note that the caller, handle_vmread, might need to use only
5402 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
5403 * 64-bit fields are to be returned).
5405 static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
5406 unsigned long field, u64 *ret)
5408 short offset = vmcs_field_to_offset(field);
5414 p = ((char *)(get_vmcs12(vcpu))) + offset;
5416 switch (vmcs_field_type(field)) {
5417 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5418 *ret = *((natural_width *)p);
5420 case VMCS_FIELD_TYPE_U16:
5423 case VMCS_FIELD_TYPE_U32:
5426 case VMCS_FIELD_TYPE_U64:
5430 return 0; /* can never happen. */
5435 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
5436 * used before) all generate the same failure when it is missing.
5438 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
5440 struct vcpu_vmx *vmx = to_vmx(vcpu);
5441 if (vmx->nested.current_vmptr == -1ull) {
5442 nested_vmx_failInvalid(vcpu);
5443 skip_emulated_instruction(vcpu);
5449 static int handle_vmread(struct kvm_vcpu *vcpu)
5451 unsigned long field;
5453 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5454 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5457 if (!nested_vmx_check_permission(vcpu) ||
5458 !nested_vmx_check_vmcs12(vcpu))
5461 /* Decode instruction info and find the field to read */
5462 field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5463 /* Read the field, zero-extended to a u64 field_value */
5464 if (!vmcs12_read_any(vcpu, field, &field_value)) {
5465 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5466 skip_emulated_instruction(vcpu);
5470 * Now copy part of this value to register or memory, as requested.
5471 * Note that the number of bits actually copied is 32 or 64 depending
5472 * on the guest's mode (32 or 64 bit), not on the given field's length.
5474 if (vmx_instruction_info & (1u << 10)) {
5475 kvm_register_write(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
5478 if (get_vmx_mem_address(vcpu, exit_qualification,
5479 vmx_instruction_info, &gva))
5481 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
5482 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
5483 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
5486 nested_vmx_succeed(vcpu);
5487 skip_emulated_instruction(vcpu);
5492 static int handle_vmwrite(struct kvm_vcpu *vcpu)
5494 unsigned long field;
5496 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5497 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5500 /* The value to write might be 32 or 64 bits, depending on L1's long
5501 * mode, and eventually we need to write that into a field of several
5502 * possible lengths. The code below first zero-extends the value to 64
5503 * bit (field_value), and then copies only the approriate number of
5504 * bits into the vmcs12 field.
5506 u64 field_value = 0;
5507 struct x86_exception e;
5509 if (!nested_vmx_check_permission(vcpu) ||
5510 !nested_vmx_check_vmcs12(vcpu))
5513 if (vmx_instruction_info & (1u << 10))
5514 field_value = kvm_register_read(vcpu,
5515 (((vmx_instruction_info) >> 3) & 0xf));
5517 if (get_vmx_mem_address(vcpu, exit_qualification,
5518 vmx_instruction_info, &gva))
5520 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
5521 &field_value, (is_long_mode(vcpu) ? 8 : 4), &e)) {
5522 kvm_inject_page_fault(vcpu, &e);
5528 field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
5529 if (vmcs_field_readonly(field)) {
5530 nested_vmx_failValid(vcpu,
5531 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
5532 skip_emulated_instruction(vcpu);
5536 offset = vmcs_field_to_offset(field);
5538 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5539 skip_emulated_instruction(vcpu);
5542 p = ((char *) get_vmcs12(vcpu)) + offset;
5544 switch (vmcs_field_type(field)) {
5545 case VMCS_FIELD_TYPE_U16:
5546 *(u16 *)p = field_value;
5548 case VMCS_FIELD_TYPE_U32:
5549 *(u32 *)p = field_value;
5551 case VMCS_FIELD_TYPE_U64:
5552 *(u64 *)p = field_value;
5554 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5555 *(natural_width *)p = field_value;
5558 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
5559 skip_emulated_instruction(vcpu);
5563 nested_vmx_succeed(vcpu);
5564 skip_emulated_instruction(vcpu);
5568 /* Emulate the VMPTRLD instruction */
5569 static int handle_vmptrld(struct kvm_vcpu *vcpu)
5571 struct vcpu_vmx *vmx = to_vmx(vcpu);
5574 struct x86_exception e;
5576 if (!nested_vmx_check_permission(vcpu))
5579 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5580 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5583 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5584 sizeof(vmptr), &e)) {
5585 kvm_inject_page_fault(vcpu, &e);
5589 if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5590 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
5591 skip_emulated_instruction(vcpu);
5595 if (vmx->nested.current_vmptr != vmptr) {
5596 struct vmcs12 *new_vmcs12;
5598 page = nested_get_page(vcpu, vmptr);
5600 nested_vmx_failInvalid(vcpu);
5601 skip_emulated_instruction(vcpu);
5604 new_vmcs12 = kmap(page);
5605 if (new_vmcs12->revision_id != VMCS12_REVISION) {
5607 nested_release_page_clean(page);
5608 nested_vmx_failValid(vcpu,
5609 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
5610 skip_emulated_instruction(vcpu);
5613 if (vmx->nested.current_vmptr != -1ull) {
5614 kunmap(vmx->nested.current_vmcs12_page);
5615 nested_release_page(vmx->nested.current_vmcs12_page);
5618 vmx->nested.current_vmptr = vmptr;
5619 vmx->nested.current_vmcs12 = new_vmcs12;
5620 vmx->nested.current_vmcs12_page = page;
5623 nested_vmx_succeed(vcpu);
5624 skip_emulated_instruction(vcpu);
5628 /* Emulate the VMPTRST instruction */
5629 static int handle_vmptrst(struct kvm_vcpu *vcpu)
5631 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5632 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
5634 struct x86_exception e;
5636 if (!nested_vmx_check_permission(vcpu))
5639 if (get_vmx_mem_address(vcpu, exit_qualification,
5640 vmx_instruction_info, &vmcs_gva))
5642 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
5643 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
5644 (void *)&to_vmx(vcpu)->nested.current_vmptr,
5646 kvm_inject_page_fault(vcpu, &e);
5649 nested_vmx_succeed(vcpu);
5650 skip_emulated_instruction(vcpu);
5655 * The exit handlers return 1 if the exit was handled fully and guest execution
5656 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
5657 * to be done to userspace and return 0.
5659 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
5660 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
5661 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
5662 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
5663 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
5664 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
5665 [EXIT_REASON_CR_ACCESS] = handle_cr,
5666 [EXIT_REASON_DR_ACCESS] = handle_dr,
5667 [EXIT_REASON_CPUID] = handle_cpuid,
5668 [EXIT_REASON_MSR_READ] = handle_rdmsr,
5669 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
5670 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
5671 [EXIT_REASON_HLT] = handle_halt,
5672 [EXIT_REASON_INVD] = handle_invd,
5673 [EXIT_REASON_INVLPG] = handle_invlpg,
5674 [EXIT_REASON_RDPMC] = handle_rdpmc,
5675 [EXIT_REASON_VMCALL] = handle_vmcall,
5676 [EXIT_REASON_VMCLEAR] = handle_vmclear,
5677 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
5678 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
5679 [EXIT_REASON_VMPTRST] = handle_vmptrst,
5680 [EXIT_REASON_VMREAD] = handle_vmread,
5681 [EXIT_REASON_VMRESUME] = handle_vmresume,
5682 [EXIT_REASON_VMWRITE] = handle_vmwrite,
5683 [EXIT_REASON_VMOFF] = handle_vmoff,
5684 [EXIT_REASON_VMON] = handle_vmon,
5685 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
5686 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
5687 [EXIT_REASON_WBINVD] = handle_wbinvd,
5688 [EXIT_REASON_XSETBV] = handle_xsetbv,
5689 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
5690 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
5691 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
5692 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
5693 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
5694 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_invalid_op,
5695 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_invalid_op,
5698 static const int kvm_vmx_max_exit_handlers =
5699 ARRAY_SIZE(kvm_vmx_exit_handlers);
5702 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
5703 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
5704 * disinterest in the current event (read or write a specific MSR) by using an
5705 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
5707 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
5708 struct vmcs12 *vmcs12, u32 exit_reason)
5710 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
5713 if (!nested_cpu_has(get_vmcs12(vcpu), CPU_BASED_USE_MSR_BITMAPS))
5717 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
5718 * for the four combinations of read/write and low/high MSR numbers.
5719 * First we need to figure out which of the four to use:
5721 bitmap = vmcs12->msr_bitmap;
5722 if (exit_reason == EXIT_REASON_MSR_WRITE)
5724 if (msr_index >= 0xc0000000) {
5725 msr_index -= 0xc0000000;
5729 /* Then read the msr_index'th bit from this bitmap: */
5730 if (msr_index < 1024*8) {
5732 kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1);
5733 return 1 & (b >> (msr_index & 7));
5735 return 1; /* let L1 handle the wrong parameter */
5739 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
5740 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
5741 * intercept (via guest_host_mask etc.) the current event.
5743 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
5744 struct vmcs12 *vmcs12)
5746 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5747 int cr = exit_qualification & 15;
5748 int reg = (exit_qualification >> 8) & 15;
5749 unsigned long val = kvm_register_read(vcpu, reg);
5751 switch ((exit_qualification >> 4) & 3) {
5752 case 0: /* mov to cr */
5755 if (vmcs12->cr0_guest_host_mask &
5756 (val ^ vmcs12->cr0_read_shadow))
5760 if ((vmcs12->cr3_target_count >= 1 &&
5761 vmcs12->cr3_target_value0 == val) ||
5762 (vmcs12->cr3_target_count >= 2 &&
5763 vmcs12->cr3_target_value1 == val) ||
5764 (vmcs12->cr3_target_count >= 3 &&
5765 vmcs12->cr3_target_value2 == val) ||
5766 (vmcs12->cr3_target_count >= 4 &&
5767 vmcs12->cr3_target_value3 == val))
5769 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
5773 if (vmcs12->cr4_guest_host_mask &
5774 (vmcs12->cr4_read_shadow ^ val))
5778 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
5784 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
5785 (vmcs12->cr0_read_shadow & X86_CR0_TS))
5788 case 1: /* mov from cr */
5791 if (vmcs12->cpu_based_vm_exec_control &
5792 CPU_BASED_CR3_STORE_EXITING)
5796 if (vmcs12->cpu_based_vm_exec_control &
5797 CPU_BASED_CR8_STORE_EXITING)
5804 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
5805 * cr0. Other attempted changes are ignored, with no exit.
5807 if (vmcs12->cr0_guest_host_mask & 0xe &
5808 (val ^ vmcs12->cr0_read_shadow))
5810 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
5811 !(vmcs12->cr0_read_shadow & 0x1) &&
5820 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
5821 * should handle it ourselves in L0 (and then continue L2). Only call this
5822 * when in is_guest_mode (L2).
5824 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
5826 u32 exit_reason = vmcs_read32(VM_EXIT_REASON);
5827 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
5828 struct vcpu_vmx *vmx = to_vmx(vcpu);
5829 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5831 if (vmx->nested.nested_run_pending)
5834 if (unlikely(vmx->fail)) {
5835 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
5836 vmcs_read32(VM_INSTRUCTION_ERROR));
5840 switch (exit_reason) {
5841 case EXIT_REASON_EXCEPTION_NMI:
5842 if (!is_exception(intr_info))
5844 else if (is_page_fault(intr_info))
5846 return vmcs12->exception_bitmap &
5847 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
5848 case EXIT_REASON_EXTERNAL_INTERRUPT:
5850 case EXIT_REASON_TRIPLE_FAULT:
5852 case EXIT_REASON_PENDING_INTERRUPT:
5853 case EXIT_REASON_NMI_WINDOW:
5855 * prepare_vmcs02() set the CPU_BASED_VIRTUAL_INTR_PENDING bit
5856 * (aka Interrupt Window Exiting) only when L1 turned it on,
5857 * so if we got a PENDING_INTERRUPT exit, this must be for L1.
5858 * Same for NMI Window Exiting.
5861 case EXIT_REASON_TASK_SWITCH:
5863 case EXIT_REASON_CPUID:
5865 case EXIT_REASON_HLT:
5866 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
5867 case EXIT_REASON_INVD:
5869 case EXIT_REASON_INVLPG:
5870 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
5871 case EXIT_REASON_RDPMC:
5872 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
5873 case EXIT_REASON_RDTSC:
5874 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
5875 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
5876 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
5877 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
5878 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
5879 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
5881 * VMX instructions trap unconditionally. This allows L1 to
5882 * emulate them for its L2 guest, i.e., allows 3-level nesting!
5885 case EXIT_REASON_CR_ACCESS:
5886 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
5887 case EXIT_REASON_DR_ACCESS:
5888 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
5889 case EXIT_REASON_IO_INSTRUCTION:
5890 /* TODO: support IO bitmaps */
5892 case EXIT_REASON_MSR_READ:
5893 case EXIT_REASON_MSR_WRITE:
5894 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
5895 case EXIT_REASON_INVALID_STATE:
5897 case EXIT_REASON_MWAIT_INSTRUCTION:
5898 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
5899 case EXIT_REASON_MONITOR_INSTRUCTION:
5900 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
5901 case EXIT_REASON_PAUSE_INSTRUCTION:
5902 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
5903 nested_cpu_has2(vmcs12,
5904 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
5905 case EXIT_REASON_MCE_DURING_VMENTRY:
5907 case EXIT_REASON_TPR_BELOW_THRESHOLD:
5909 case EXIT_REASON_APIC_ACCESS:
5910 return nested_cpu_has2(vmcs12,
5911 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
5912 case EXIT_REASON_EPT_VIOLATION:
5913 case EXIT_REASON_EPT_MISCONFIG:
5915 case EXIT_REASON_WBINVD:
5916 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
5917 case EXIT_REASON_XSETBV:
5924 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
5926 *info1 = vmcs_readl(EXIT_QUALIFICATION);
5927 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
5931 * The guest has exited. See if we can fix it or if we need userspace
5934 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
5936 struct vcpu_vmx *vmx = to_vmx(vcpu);
5937 u32 exit_reason = vmx->exit_reason;
5938 u32 vectoring_info = vmx->idt_vectoring_info;
5940 /* If guest state is invalid, start emulating */
5941 if (vmx->emulation_required && emulate_invalid_guest_state)
5942 return handle_invalid_guest_state(vcpu);
5945 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
5946 * we did not inject a still-pending event to L1 now because of
5947 * nested_run_pending, we need to re-enable this bit.
5949 if (vmx->nested.nested_run_pending)
5950 kvm_make_request(KVM_REQ_EVENT, vcpu);
5952 if (!is_guest_mode(vcpu) && (exit_reason == EXIT_REASON_VMLAUNCH ||
5953 exit_reason == EXIT_REASON_VMRESUME))
5954 vmx->nested.nested_run_pending = 1;
5956 vmx->nested.nested_run_pending = 0;
5958 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
5959 nested_vmx_vmexit(vcpu);
5963 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
5964 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5965 vcpu->run->fail_entry.hardware_entry_failure_reason
5970 if (unlikely(vmx->fail)) {
5971 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
5972 vcpu->run->fail_entry.hardware_entry_failure_reason
5973 = vmcs_read32(VM_INSTRUCTION_ERROR);
5977 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
5978 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
5979 exit_reason != EXIT_REASON_EPT_VIOLATION &&
5980 exit_reason != EXIT_REASON_TASK_SWITCH))
5981 printk(KERN_WARNING "%s: unexpected, valid vectoring info "
5982 "(0x%x) and exit reason is 0x%x\n",
5983 __func__, vectoring_info, exit_reason);
5985 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
5986 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
5987 get_vmcs12(vcpu), vcpu)))) {
5988 if (vmx_interrupt_allowed(vcpu)) {
5989 vmx->soft_vnmi_blocked = 0;
5990 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
5991 vcpu->arch.nmi_pending) {
5993 * This CPU don't support us in finding the end of an
5994 * NMI-blocked window if the guest runs with IRQs
5995 * disabled. So we pull the trigger after 1 s of
5996 * futile waiting, but inform the user about this.
5998 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
5999 "state on VCPU %d after 1 s timeout\n",
6000 __func__, vcpu->vcpu_id);
6001 vmx->soft_vnmi_blocked = 0;
6005 if (exit_reason < kvm_vmx_max_exit_handlers
6006 && kvm_vmx_exit_handlers[exit_reason])
6007 return kvm_vmx_exit_handlers[exit_reason](vcpu);
6009 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6010 vcpu->run->hw.hardware_exit_reason = exit_reason;
6015 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6017 if (irr == -1 || tpr < irr) {
6018 vmcs_write32(TPR_THRESHOLD, 0);
6022 vmcs_write32(TPR_THRESHOLD, irr);
6025 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
6029 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
6030 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
6033 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6034 exit_intr_info = vmx->exit_intr_info;
6036 /* Handle machine checks before interrupts are enabled */
6037 if (is_machine_check(exit_intr_info))
6038 kvm_machine_check();
6040 /* We need to handle NMIs before interrupts are enabled */
6041 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
6042 (exit_intr_info & INTR_INFO_VALID_MASK)) {
6043 kvm_before_handle_nmi(&vmx->vcpu);
6045 kvm_after_handle_nmi(&vmx->vcpu);
6049 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6054 bool idtv_info_valid;
6056 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6058 if (cpu_has_virtual_nmis()) {
6059 if (vmx->nmi_known_unmasked)
6062 * Can't use vmx->exit_intr_info since we're not sure what
6063 * the exit reason is.
6065 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6066 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6067 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6069 * SDM 3: 27.7.1.2 (September 2008)
6070 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6071 * a guest IRET fault.
6072 * SDM 3: 23.2.2 (September 2008)
6073 * Bit 12 is undefined in any of the following cases:
6074 * If the VM exit sets the valid bit in the IDT-vectoring
6075 * information field.
6076 * If the VM exit is due to a double fault.
6078 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6079 vector != DF_VECTOR && !idtv_info_valid)
6080 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6081 GUEST_INTR_STATE_NMI);
6083 vmx->nmi_known_unmasked =
6084 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6085 & GUEST_INTR_STATE_NMI);
6086 } else if (unlikely(vmx->soft_vnmi_blocked))
6087 vmx->vnmi_blocked_time +=
6088 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
6091 static void __vmx_complete_interrupts(struct vcpu_vmx *vmx,
6092 u32 idt_vectoring_info,
6093 int instr_len_field,
6094 int error_code_field)
6098 bool idtv_info_valid;
6100 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6102 vmx->vcpu.arch.nmi_injected = false;
6103 kvm_clear_exception_queue(&vmx->vcpu);
6104 kvm_clear_interrupt_queue(&vmx->vcpu);
6106 if (!idtv_info_valid)
6109 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6111 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6112 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6115 case INTR_TYPE_NMI_INTR:
6116 vmx->vcpu.arch.nmi_injected = true;
6118 * SDM 3: 27.7.1.2 (September 2008)
6119 * Clear bit "block by NMI" before VM entry if a NMI
6122 vmx_set_nmi_mask(&vmx->vcpu, false);
6124 case INTR_TYPE_SOFT_EXCEPTION:
6125 vmx->vcpu.arch.event_exit_inst_len =
6126 vmcs_read32(instr_len_field);
6128 case INTR_TYPE_HARD_EXCEPTION:
6129 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6130 u32 err = vmcs_read32(error_code_field);
6131 kvm_queue_exception_e(&vmx->vcpu, vector, err);
6133 kvm_queue_exception(&vmx->vcpu, vector);
6135 case INTR_TYPE_SOFT_INTR:
6136 vmx->vcpu.arch.event_exit_inst_len =
6137 vmcs_read32(instr_len_field);
6139 case INTR_TYPE_EXT_INTR:
6140 kvm_queue_interrupt(&vmx->vcpu, vector,
6141 type == INTR_TYPE_SOFT_INTR);
6148 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6150 if (is_guest_mode(&vmx->vcpu))
6152 __vmx_complete_interrupts(vmx, vmx->idt_vectoring_info,
6153 VM_EXIT_INSTRUCTION_LEN,
6154 IDT_VECTORING_ERROR_CODE);
6157 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6159 if (is_guest_mode(vcpu))
6161 __vmx_complete_interrupts(to_vmx(vcpu),
6162 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6163 VM_ENTRY_INSTRUCTION_LEN,
6164 VM_ENTRY_EXCEPTION_ERROR_CODE);
6166 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6169 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6172 struct perf_guest_switch_msr *msrs;
6174 msrs = perf_guest_get_msrs(&nr_msrs);
6179 for (i = 0; i < nr_msrs; i++)
6180 if (msrs[i].host == msrs[i].guest)
6181 clear_atomic_switch_msr(vmx, msrs[i].msr);
6183 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6187 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
6189 struct vcpu_vmx *vmx = to_vmx(vcpu);
6190 unsigned long debugctlmsr;
6192 if (is_guest_mode(vcpu) && !vmx->nested.nested_run_pending) {
6193 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6194 if (vmcs12->idt_vectoring_info_field &
6195 VECTORING_INFO_VALID_MASK) {
6196 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
6197 vmcs12->idt_vectoring_info_field);
6198 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
6199 vmcs12->vm_exit_instruction_len);
6200 if (vmcs12->idt_vectoring_info_field &
6201 VECTORING_INFO_DELIVER_CODE_MASK)
6202 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
6203 vmcs12->idt_vectoring_error_code);
6207 /* Record the guest's net vcpu time for enforced NMI injections. */
6208 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
6209 vmx->entry_time = ktime_get();
6211 /* Don't enter VMX if guest state is invalid, let the exit handler
6212 start emulation until we arrive back to a valid state */
6213 if (vmx->emulation_required && emulate_invalid_guest_state)
6216 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
6217 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
6218 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
6219 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
6221 /* When single-stepping over STI and MOV SS, we must clear the
6222 * corresponding interruptibility bits in the guest state. Otherwise
6223 * vmentry fails as it then expects bit 14 (BS) in pending debug
6224 * exceptions being set, but that's not correct for the guest debugging
6226 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
6227 vmx_set_interrupt_shadow(vcpu, 0);
6229 atomic_switch_perf_msrs(vmx);
6230 debugctlmsr = get_debugctlmsr();
6232 vmx->__launched = vmx->loaded_vmcs->launched;
6234 /* Store host registers */
6235 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
6236 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
6237 "push %%" _ASM_CX " \n\t"
6238 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
6240 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
6241 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
6243 /* Reload cr2 if changed */
6244 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
6245 "mov %%cr2, %%" _ASM_DX " \n\t"
6246 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
6248 "mov %%" _ASM_AX", %%cr2 \n\t"
6250 /* Check if vmlaunch of vmresume is needed */
6251 "cmpl $0, %c[launched](%0) \n\t"
6252 /* Load guest registers. Don't clobber flags. */
6253 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
6254 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
6255 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
6256 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
6257 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
6258 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
6259 #ifdef CONFIG_X86_64
6260 "mov %c[r8](%0), %%r8 \n\t"
6261 "mov %c[r9](%0), %%r9 \n\t"
6262 "mov %c[r10](%0), %%r10 \n\t"
6263 "mov %c[r11](%0), %%r11 \n\t"
6264 "mov %c[r12](%0), %%r12 \n\t"
6265 "mov %c[r13](%0), %%r13 \n\t"
6266 "mov %c[r14](%0), %%r14 \n\t"
6267 "mov %c[r15](%0), %%r15 \n\t"
6269 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
6271 /* Enter guest mode */
6273 __ex(ASM_VMX_VMLAUNCH) "\n\t"
6275 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
6277 /* Save guest registers, load host registers, keep flags */
6278 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
6280 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
6281 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
6282 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
6283 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
6284 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
6285 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
6286 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
6287 #ifdef CONFIG_X86_64
6288 "mov %%r8, %c[r8](%0) \n\t"
6289 "mov %%r9, %c[r9](%0) \n\t"
6290 "mov %%r10, %c[r10](%0) \n\t"
6291 "mov %%r11, %c[r11](%0) \n\t"
6292 "mov %%r12, %c[r12](%0) \n\t"
6293 "mov %%r13, %c[r13](%0) \n\t"
6294 "mov %%r14, %c[r14](%0) \n\t"
6295 "mov %%r15, %c[r15](%0) \n\t"
6297 "mov %%cr2, %%" _ASM_AX " \n\t"
6298 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
6300 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
6301 "setbe %c[fail](%0) \n\t"
6302 ".pushsection .rodata \n\t"
6303 ".global vmx_return \n\t"
6304 "vmx_return: " _ASM_PTR " 2b \n\t"
6306 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
6307 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
6308 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
6309 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
6310 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
6311 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
6312 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
6313 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
6314 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
6315 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
6316 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
6317 #ifdef CONFIG_X86_64
6318 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
6319 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
6320 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
6321 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
6322 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
6323 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
6324 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
6325 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
6327 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
6328 [wordsize]"i"(sizeof(ulong))
6330 #ifdef CONFIG_X86_64
6331 , "rax", "rbx", "rdi", "rsi"
6332 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
6334 , "eax", "ebx", "edi", "esi"
6338 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
6340 update_debugctlmsr(debugctlmsr);
6342 #ifndef CONFIG_X86_64
6344 * The sysexit path does not restore ds/es, so we must set them to
6345 * a reasonable value ourselves.
6347 * We can't defer this to vmx_load_host_state() since that function
6348 * may be executed in interrupt context, which saves and restore segments
6349 * around it, nullifying its effect.
6351 loadsegment(ds, __USER_DS);
6352 loadsegment(es, __USER_DS);
6355 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
6356 | (1 << VCPU_EXREG_RFLAGS)
6357 | (1 << VCPU_EXREG_CPL)
6358 | (1 << VCPU_EXREG_PDPTR)
6359 | (1 << VCPU_EXREG_SEGMENTS)
6360 | (1 << VCPU_EXREG_CR3));
6361 vcpu->arch.regs_dirty = 0;
6363 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
6365 if (is_guest_mode(vcpu)) {
6366 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6367 vmcs12->idt_vectoring_info_field = vmx->idt_vectoring_info;
6368 if (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK) {
6369 vmcs12->idt_vectoring_error_code =
6370 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6371 vmcs12->vm_exit_instruction_len =
6372 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
6376 vmx->loaded_vmcs->launched = 1;
6378 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
6379 trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
6381 vmx_complete_atomic_exit(vmx);
6382 vmx_recover_nmi_blocking(vmx);
6383 vmx_complete_interrupts(vmx);
6386 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
6388 struct vcpu_vmx *vmx = to_vmx(vcpu);
6392 free_loaded_vmcs(vmx->loaded_vmcs);
6393 kfree(vmx->guest_msrs);
6394 kvm_vcpu_uninit(vcpu);
6395 kmem_cache_free(kvm_vcpu_cache, vmx);
6398 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
6401 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
6405 return ERR_PTR(-ENOMEM);
6409 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
6413 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
6415 if (!vmx->guest_msrs) {
6419 vmx->loaded_vmcs = &vmx->vmcs01;
6420 vmx->loaded_vmcs->vmcs = alloc_vmcs();
6421 if (!vmx->loaded_vmcs->vmcs)
6424 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
6425 loaded_vmcs_init(vmx->loaded_vmcs);
6430 vmx_vcpu_load(&vmx->vcpu, cpu);
6431 vmx->vcpu.cpu = cpu;
6432 err = vmx_vcpu_setup(vmx);
6433 vmx_vcpu_put(&vmx->vcpu);
6437 if (vm_need_virtualize_apic_accesses(kvm))
6438 err = alloc_apic_access_page(kvm);
6443 if (!kvm->arch.ept_identity_map_addr)
6444 kvm->arch.ept_identity_map_addr =
6445 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
6447 if (alloc_identity_pagetable(kvm) != 0)
6449 if (!init_rmode_identity_map(kvm))
6453 vmx->nested.current_vmptr = -1ull;
6454 vmx->nested.current_vmcs12 = NULL;
6459 free_loaded_vmcs(vmx->loaded_vmcs);
6461 kfree(vmx->guest_msrs);
6463 kvm_vcpu_uninit(&vmx->vcpu);
6466 kmem_cache_free(kvm_vcpu_cache, vmx);
6467 return ERR_PTR(err);
6470 static void __init vmx_check_processor_compat(void *rtn)
6472 struct vmcs_config vmcs_conf;
6475 if (setup_vmcs_config(&vmcs_conf) < 0)
6477 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
6478 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
6479 smp_processor_id());
6484 static int get_ept_level(void)
6486 return VMX_EPT_DEFAULT_GAW + 1;
6489 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
6493 /* For VT-d and EPT combination
6494 * 1. MMIO: always map as UC
6496 * a. VT-d without snooping control feature: can't guarantee the
6497 * result, try to trust guest.
6498 * b. VT-d with snooping control feature: snooping control feature of
6499 * VT-d engine can guarantee the cache correctness. Just set it
6500 * to WB to keep consistent with host. So the same as item 3.
6501 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
6502 * consistent with host MTRR
6505 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
6506 else if (vcpu->kvm->arch.iommu_domain &&
6507 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
6508 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
6509 VMX_EPT_MT_EPTE_SHIFT;
6511 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
6517 static int vmx_get_lpage_level(void)
6519 if (enable_ept && !cpu_has_vmx_ept_1g_page())
6520 return PT_DIRECTORY_LEVEL;
6522 /* For shadow and EPT supported 1GB page */
6523 return PT_PDPE_LEVEL;
6526 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
6528 struct kvm_cpuid_entry2 *best;
6529 struct vcpu_vmx *vmx = to_vmx(vcpu);
6532 vmx->rdtscp_enabled = false;
6533 if (vmx_rdtscp_supported()) {
6534 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6535 if (exec_control & SECONDARY_EXEC_RDTSCP) {
6536 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
6537 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
6538 vmx->rdtscp_enabled = true;
6540 exec_control &= ~SECONDARY_EXEC_RDTSCP;
6541 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
6547 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6548 /* Exposing INVPCID only when PCID is exposed */
6549 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
6550 if (vmx_invpcid_supported() &&
6551 best && (best->ecx & bit(X86_FEATURE_INVPCID)) &&
6552 guest_cpuid_has_pcid(vcpu)) {
6553 exec_control |= SECONDARY_EXEC_ENABLE_INVPCID;
6554 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
6557 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
6558 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
6561 best->ecx &= ~bit(X86_FEATURE_INVPCID);
6565 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
6567 if (func == 1 && nested)
6568 entry->ecx |= bit(X86_FEATURE_VMX);
6572 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
6573 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
6574 * with L0's requirements for its guest (a.k.a. vmsc01), so we can run the L2
6575 * guest in a way that will both be appropriate to L1's requests, and our
6576 * needs. In addition to modifying the active vmcs (which is vmcs02), this
6577 * function also has additional necessary side-effects, like setting various
6578 * vcpu->arch fields.
6580 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6582 struct vcpu_vmx *vmx = to_vmx(vcpu);
6585 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
6586 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
6587 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
6588 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
6589 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
6590 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
6591 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
6592 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
6593 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
6594 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
6595 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
6596 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
6597 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
6598 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
6599 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
6600 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
6601 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
6602 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
6603 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
6604 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
6605 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
6606 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
6607 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
6608 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
6609 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
6610 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
6611 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
6612 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
6613 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
6614 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
6615 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
6616 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
6617 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
6618 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
6619 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
6620 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
6622 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
6623 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
6624 vmcs12->vm_entry_intr_info_field);
6625 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
6626 vmcs12->vm_entry_exception_error_code);
6627 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
6628 vmcs12->vm_entry_instruction_len);
6629 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
6630 vmcs12->guest_interruptibility_info);
6631 vmcs_write32(GUEST_ACTIVITY_STATE, vmcs12->guest_activity_state);
6632 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
6633 vmcs_writel(GUEST_DR7, vmcs12->guest_dr7);
6634 vmcs_writel(GUEST_RFLAGS, vmcs12->guest_rflags);
6635 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
6636 vmcs12->guest_pending_dbg_exceptions);
6637 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
6638 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
6640 vmcs_write64(VMCS_LINK_POINTER, -1ull);
6642 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
6643 (vmcs_config.pin_based_exec_ctrl |
6644 vmcs12->pin_based_vm_exec_control));
6647 * Whether page-faults are trapped is determined by a combination of
6648 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
6649 * If enable_ept, L0 doesn't care about page faults and we should
6650 * set all of these to L1's desires. However, if !enable_ept, L0 does
6651 * care about (at least some) page faults, and because it is not easy
6652 * (if at all possible?) to merge L0 and L1's desires, we simply ask
6653 * to exit on each and every L2 page fault. This is done by setting
6654 * MASK=MATCH=0 and (see below) EB.PF=1.
6655 * Note that below we don't need special code to set EB.PF beyond the
6656 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
6657 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
6658 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
6660 * A problem with this approach (when !enable_ept) is that L1 may be
6661 * injected with more page faults than it asked for. This could have
6662 * caused problems, but in practice existing hypervisors don't care.
6663 * To fix this, we will need to emulate the PFEC checking (on the L1
6664 * page tables), using walk_addr(), when injecting PFs to L1.
6666 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
6667 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
6668 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
6669 enable_ept ? vmcs12->page_fault_error_code_match : 0);
6671 if (cpu_has_secondary_exec_ctrls()) {
6672 u32 exec_control = vmx_secondary_exec_control(vmx);
6673 if (!vmx->rdtscp_enabled)
6674 exec_control &= ~SECONDARY_EXEC_RDTSCP;
6675 /* Take the following fields only from vmcs12 */
6676 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6677 if (nested_cpu_has(vmcs12,
6678 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
6679 exec_control |= vmcs12->secondary_vm_exec_control;
6681 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
6683 * Translate L1 physical address to host physical
6684 * address for vmcs02. Keep the page pinned, so this
6685 * physical address remains valid. We keep a reference
6686 * to it so we can release it later.
6688 if (vmx->nested.apic_access_page) /* shouldn't happen */
6689 nested_release_page(vmx->nested.apic_access_page);
6690 vmx->nested.apic_access_page =
6691 nested_get_page(vcpu, vmcs12->apic_access_addr);
6693 * If translation failed, no matter: This feature asks
6694 * to exit when accessing the given address, and if it
6695 * can never be accessed, this feature won't do
6698 if (!vmx->nested.apic_access_page)
6700 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6702 vmcs_write64(APIC_ACCESS_ADDR,
6703 page_to_phys(vmx->nested.apic_access_page));
6706 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6711 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
6712 * Some constant fields are set here by vmx_set_constant_host_state().
6713 * Other fields are different per CPU, and will be set later when
6714 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
6716 vmx_set_constant_host_state();
6719 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
6720 * entry, but only if the current (host) sp changed from the value
6721 * we wrote last (vmx->host_rsp). This cache is no longer relevant
6722 * if we switch vmcs, and rather than hold a separate cache per vmcs,
6723 * here we just force the write to happen on entry.
6727 exec_control = vmx_exec_control(vmx); /* L0's desires */
6728 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
6729 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
6730 exec_control &= ~CPU_BASED_TPR_SHADOW;
6731 exec_control |= vmcs12->cpu_based_vm_exec_control;
6733 * Merging of IO and MSR bitmaps not currently supported.
6734 * Rather, exit every time.
6736 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
6737 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
6738 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
6740 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
6742 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
6743 * bitwise-or of what L1 wants to trap for L2, and what we want to
6744 * trap. Note that CR0.TS also needs updating - we do this later.
6746 update_exception_bitmap(vcpu);
6747 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
6748 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
6750 /* Note: IA32_MODE, LOAD_IA32_EFER are modified by vmx_set_efer below */
6751 vmcs_write32(VM_EXIT_CONTROLS,
6752 vmcs12->vm_exit_controls | vmcs_config.vmexit_ctrl);
6753 vmcs_write32(VM_ENTRY_CONTROLS, vmcs12->vm_entry_controls |
6754 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
6756 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)
6757 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
6758 else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
6759 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
6762 set_cr4_guest_host_mask(vmx);
6764 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
6765 vmcs_write64(TSC_OFFSET,
6766 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
6768 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
6772 * Trivially support vpid by letting L2s share their parent
6773 * L1's vpid. TODO: move to a more elaborate solution, giving
6774 * each L2 its own vpid and exposing the vpid feature to L1.
6776 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
6777 vmx_flush_tlb(vcpu);
6780 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
6781 vcpu->arch.efer = vmcs12->guest_ia32_efer;
6782 if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
6783 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
6785 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
6786 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
6787 vmx_set_efer(vcpu, vcpu->arch.efer);
6790 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
6791 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
6792 * The CR0_READ_SHADOW is what L2 should have expected to read given
6793 * the specifications by L1; It's not enough to take
6794 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
6795 * have more bits than L1 expected.
6797 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
6798 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
6800 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
6801 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
6803 /* shadow page tables on either EPT or shadow page tables */
6804 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
6805 kvm_mmu_reset_context(vcpu);
6807 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
6808 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
6812 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
6813 * for running an L2 nested guest.
6815 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
6817 struct vmcs12 *vmcs12;
6818 struct vcpu_vmx *vmx = to_vmx(vcpu);
6820 struct loaded_vmcs *vmcs02;
6822 if (!nested_vmx_check_permission(vcpu) ||
6823 !nested_vmx_check_vmcs12(vcpu))
6826 skip_emulated_instruction(vcpu);
6827 vmcs12 = get_vmcs12(vcpu);
6830 * The nested entry process starts with enforcing various prerequisites
6831 * on vmcs12 as required by the Intel SDM, and act appropriately when
6832 * they fail: As the SDM explains, some conditions should cause the
6833 * instruction to fail, while others will cause the instruction to seem
6834 * to succeed, but return an EXIT_REASON_INVALID_STATE.
6835 * To speed up the normal (success) code path, we should avoid checking
6836 * for misconfigurations which will anyway be caught by the processor
6837 * when using the merged vmcs02.
6839 if (vmcs12->launch_state == launch) {
6840 nested_vmx_failValid(vcpu,
6841 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
6842 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
6846 if ((vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_MSR_BITMAPS) &&
6847 !IS_ALIGNED(vmcs12->msr_bitmap, PAGE_SIZE)) {
6848 /*TODO: Also verify bits beyond physical address width are 0*/
6849 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6853 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
6854 !IS_ALIGNED(vmcs12->apic_access_addr, PAGE_SIZE)) {
6855 /*TODO: Also verify bits beyond physical address width are 0*/
6856 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6860 if (vmcs12->vm_entry_msr_load_count > 0 ||
6861 vmcs12->vm_exit_msr_load_count > 0 ||
6862 vmcs12->vm_exit_msr_store_count > 0) {
6863 pr_warn_ratelimited("%s: VMCS MSR_{LOAD,STORE} unsupported\n",
6865 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6869 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
6870 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high) ||
6871 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
6872 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high) ||
6873 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
6874 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high) ||
6875 !vmx_control_verify(vmcs12->vm_exit_controls,
6876 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high) ||
6877 !vmx_control_verify(vmcs12->vm_entry_controls,
6878 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high))
6880 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
6884 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
6885 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
6886 nested_vmx_failValid(vcpu,
6887 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
6891 if (((vmcs12->guest_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
6892 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
6893 nested_vmx_entry_failure(vcpu, vmcs12,
6894 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
6897 if (vmcs12->vmcs_link_pointer != -1ull) {
6898 nested_vmx_entry_failure(vcpu, vmcs12,
6899 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
6904 * We're finally done with prerequisite checking, and can start with
6908 vmcs02 = nested_get_current_vmcs02(vmx);
6912 enter_guest_mode(vcpu);
6914 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
6917 vmx->loaded_vmcs = vmcs02;
6919 vmx_vcpu_load(vcpu, cpu);
6923 vmcs12->launch_state = 1;
6925 prepare_vmcs02(vcpu, vmcs12);
6928 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
6929 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
6930 * returned as far as L1 is concerned. It will only return (and set
6931 * the success flag) when L2 exits (see nested_vmx_vmexit()).
6937 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
6938 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
6939 * This function returns the new value we should put in vmcs12.guest_cr0.
6940 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
6941 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
6942 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
6943 * didn't trap the bit, because if L1 did, so would L0).
6944 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
6945 * been modified by L2, and L1 knows it. So just leave the old value of
6946 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
6947 * isn't relevant, because if L0 traps this bit it can set it to anything.
6948 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
6949 * changed these bits, and therefore they need to be updated, but L0
6950 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
6951 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
6953 static inline unsigned long
6954 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6957 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
6958 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
6959 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
6960 vcpu->arch.cr0_guest_owned_bits));
6963 static inline unsigned long
6964 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6967 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
6968 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
6969 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
6970 vcpu->arch.cr4_guest_owned_bits));
6974 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
6975 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
6976 * and this function updates it to reflect the changes to the guest state while
6977 * L2 was running (and perhaps made some exits which were handled directly by L0
6978 * without going back to L1), and to reflect the exit reason.
6979 * Note that we do not have to copy here all VMCS fields, just those that
6980 * could have changed by the L2 guest or the exit - i.e., the guest-state and
6981 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
6982 * which already writes to vmcs12 directly.
6984 void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
6986 /* update guest state fields: */
6987 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
6988 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
6990 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
6991 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
6992 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
6993 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
6995 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
6996 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
6997 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
6998 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
6999 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
7000 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
7001 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
7002 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
7003 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
7004 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
7005 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
7006 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
7007 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
7008 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
7009 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
7010 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
7011 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
7012 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
7013 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
7014 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
7015 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
7016 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
7017 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
7018 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
7019 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
7020 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
7021 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
7022 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
7023 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
7024 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
7025 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
7026 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
7027 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
7028 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
7029 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
7030 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
7032 vmcs12->guest_activity_state = vmcs_read32(GUEST_ACTIVITY_STATE);
7033 vmcs12->guest_interruptibility_info =
7034 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
7035 vmcs12->guest_pending_dbg_exceptions =
7036 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
7038 /* TODO: These cannot have changed unless we have MSR bitmaps and
7039 * the relevant bit asks not to trap the change */
7040 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
7041 if (vmcs12->vm_entry_controls & VM_EXIT_SAVE_IA32_PAT)
7042 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
7043 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
7044 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
7045 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
7047 /* update exit information fields: */
7049 vmcs12->vm_exit_reason = vmcs_read32(VM_EXIT_REASON);
7050 vmcs12->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7052 vmcs12->vm_exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7053 vmcs12->vm_exit_intr_error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
7054 vmcs12->idt_vectoring_info_field =
7055 vmcs_read32(IDT_VECTORING_INFO_FIELD);
7056 vmcs12->idt_vectoring_error_code =
7057 vmcs_read32(IDT_VECTORING_ERROR_CODE);
7058 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
7059 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7061 /* clear vm-entry fields which are to be cleared on exit */
7062 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
7063 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
7067 * A part of what we need to when the nested L2 guest exits and we want to
7068 * run its L1 parent, is to reset L1's guest state to the host state specified
7070 * This function is to be called not only on normal nested exit, but also on
7071 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
7072 * Failures During or After Loading Guest State").
7073 * This function should be called when the active VMCS is L1's (vmcs01).
7075 void load_vmcs12_host_state(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7077 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
7078 vcpu->arch.efer = vmcs12->host_ia32_efer;
7079 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
7080 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
7082 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
7083 vmx_set_efer(vcpu, vcpu->arch.efer);
7085 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
7086 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
7088 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
7089 * actually changed, because it depends on the current state of
7090 * fpu_active (which may have changed).
7091 * Note that vmx_set_cr0 refers to efer set above.
7093 kvm_set_cr0(vcpu, vmcs12->host_cr0);
7095 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
7096 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
7097 * but we also need to update cr0_guest_host_mask and exception_bitmap.
7099 update_exception_bitmap(vcpu);
7100 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
7101 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
7104 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
7105 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
7107 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
7108 kvm_set_cr4(vcpu, vmcs12->host_cr4);
7110 /* shadow page tables on either EPT or shadow page tables */
7111 kvm_set_cr3(vcpu, vmcs12->host_cr3);
7112 kvm_mmu_reset_context(vcpu);
7116 * Trivially support vpid by letting L2s share their parent
7117 * L1's vpid. TODO: move to a more elaborate solution, giving
7118 * each L2 its own vpid and exposing the vpid feature to L1.
7120 vmx_flush_tlb(vcpu);
7124 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
7125 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
7126 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
7127 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
7128 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
7129 vmcs_writel(GUEST_TR_BASE, vmcs12->host_tr_base);
7130 vmcs_writel(GUEST_GS_BASE, vmcs12->host_gs_base);
7131 vmcs_writel(GUEST_FS_BASE, vmcs12->host_fs_base);
7132 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->host_es_selector);
7133 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->host_cs_selector);
7134 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->host_ss_selector);
7135 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->host_ds_selector);
7136 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->host_fs_selector);
7137 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->host_gs_selector);
7138 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->host_tr_selector);
7140 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT)
7141 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
7142 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
7143 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
7144 vmcs12->host_ia32_perf_global_ctrl);
7148 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
7149 * and modify vmcs12 to make it see what it would expect to see there if
7150 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
7152 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu)
7154 struct vcpu_vmx *vmx = to_vmx(vcpu);
7156 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7158 leave_guest_mode(vcpu);
7159 prepare_vmcs12(vcpu, vmcs12);
7162 vmx->loaded_vmcs = &vmx->vmcs01;
7164 vmx_vcpu_load(vcpu, cpu);
7168 /* if no vmcs02 cache requested, remove the one we used */
7169 if (VMCS02_POOL_SIZE == 0)
7170 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
7172 load_vmcs12_host_state(vcpu, vmcs12);
7174 /* Update TSC_OFFSET if TSC was changed while L2 ran */
7175 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
7177 /* This is needed for same reason as it was needed in prepare_vmcs02 */
7180 /* Unpin physical memory we referred to in vmcs02 */
7181 if (vmx->nested.apic_access_page) {
7182 nested_release_page(vmx->nested.apic_access_page);
7183 vmx->nested.apic_access_page = 0;
7187 * Exiting from L2 to L1, we're now back to L1 which thinks it just
7188 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
7189 * success or failure flag accordingly.
7191 if (unlikely(vmx->fail)) {
7193 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
7195 nested_vmx_succeed(vcpu);
7199 * L1's failure to enter L2 is a subset of a normal exit, as explained in
7200 * 23.7 "VM-entry failures during or after loading guest state" (this also
7201 * lists the acceptable exit-reason and exit-qualification parameters).
7202 * It should only be called before L2 actually succeeded to run, and when
7203 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
7205 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
7206 struct vmcs12 *vmcs12,
7207 u32 reason, unsigned long qualification)
7209 load_vmcs12_host_state(vcpu, vmcs12);
7210 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
7211 vmcs12->exit_qualification = qualification;
7212 nested_vmx_succeed(vcpu);
7215 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
7216 struct x86_instruction_info *info,
7217 enum x86_intercept_stage stage)
7219 return X86EMUL_CONTINUE;
7222 static struct kvm_x86_ops vmx_x86_ops = {
7223 .cpu_has_kvm_support = cpu_has_kvm_support,
7224 .disabled_by_bios = vmx_disabled_by_bios,
7225 .hardware_setup = hardware_setup,
7226 .hardware_unsetup = hardware_unsetup,
7227 .check_processor_compatibility = vmx_check_processor_compat,
7228 .hardware_enable = hardware_enable,
7229 .hardware_disable = hardware_disable,
7230 .cpu_has_accelerated_tpr = report_flexpriority,
7232 .vcpu_create = vmx_create_vcpu,
7233 .vcpu_free = vmx_free_vcpu,
7234 .vcpu_reset = vmx_vcpu_reset,
7236 .prepare_guest_switch = vmx_save_host_state,
7237 .vcpu_load = vmx_vcpu_load,
7238 .vcpu_put = vmx_vcpu_put,
7240 .set_guest_debug = set_guest_debug,
7241 .get_msr = vmx_get_msr,
7242 .set_msr = vmx_set_msr,
7243 .get_segment_base = vmx_get_segment_base,
7244 .get_segment = vmx_get_segment,
7245 .set_segment = vmx_set_segment,
7246 .get_cpl = vmx_get_cpl,
7247 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
7248 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
7249 .decache_cr3 = vmx_decache_cr3,
7250 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
7251 .set_cr0 = vmx_set_cr0,
7252 .set_cr3 = vmx_set_cr3,
7253 .set_cr4 = vmx_set_cr4,
7254 .set_efer = vmx_set_efer,
7255 .get_idt = vmx_get_idt,
7256 .set_idt = vmx_set_idt,
7257 .get_gdt = vmx_get_gdt,
7258 .set_gdt = vmx_set_gdt,
7259 .set_dr7 = vmx_set_dr7,
7260 .cache_reg = vmx_cache_reg,
7261 .get_rflags = vmx_get_rflags,
7262 .set_rflags = vmx_set_rflags,
7263 .fpu_activate = vmx_fpu_activate,
7264 .fpu_deactivate = vmx_fpu_deactivate,
7266 .tlb_flush = vmx_flush_tlb,
7268 .run = vmx_vcpu_run,
7269 .handle_exit = vmx_handle_exit,
7270 .skip_emulated_instruction = skip_emulated_instruction,
7271 .set_interrupt_shadow = vmx_set_interrupt_shadow,
7272 .get_interrupt_shadow = vmx_get_interrupt_shadow,
7273 .patch_hypercall = vmx_patch_hypercall,
7274 .set_irq = vmx_inject_irq,
7275 .set_nmi = vmx_inject_nmi,
7276 .queue_exception = vmx_queue_exception,
7277 .cancel_injection = vmx_cancel_injection,
7278 .interrupt_allowed = vmx_interrupt_allowed,
7279 .nmi_allowed = vmx_nmi_allowed,
7280 .get_nmi_mask = vmx_get_nmi_mask,
7281 .set_nmi_mask = vmx_set_nmi_mask,
7282 .enable_nmi_window = enable_nmi_window,
7283 .enable_irq_window = enable_irq_window,
7284 .update_cr8_intercept = update_cr8_intercept,
7286 .set_tss_addr = vmx_set_tss_addr,
7287 .get_tdp_level = get_ept_level,
7288 .get_mt_mask = vmx_get_mt_mask,
7290 .get_exit_info = vmx_get_exit_info,
7292 .get_lpage_level = vmx_get_lpage_level,
7294 .cpuid_update = vmx_cpuid_update,
7296 .rdtscp_supported = vmx_rdtscp_supported,
7297 .invpcid_supported = vmx_invpcid_supported,
7299 .set_supported_cpuid = vmx_set_supported_cpuid,
7301 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
7303 .set_tsc_khz = vmx_set_tsc_khz,
7304 .write_tsc_offset = vmx_write_tsc_offset,
7305 .adjust_tsc_offset = vmx_adjust_tsc_offset,
7306 .compute_tsc_offset = vmx_compute_tsc_offset,
7307 .read_l1_tsc = vmx_read_l1_tsc,
7309 .set_tdp_cr3 = vmx_set_cr3,
7311 .check_intercept = vmx_check_intercept,
7314 static int __init vmx_init(void)
7318 rdmsrl_safe(MSR_EFER, &host_efer);
7320 for (i = 0; i < NR_VMX_MSR; ++i)
7321 kvm_define_shared_msr(i, vmx_msr_index[i]);
7323 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
7324 if (!vmx_io_bitmap_a)
7329 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
7330 if (!vmx_io_bitmap_b)
7333 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
7334 if (!vmx_msr_bitmap_legacy)
7338 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
7339 if (!vmx_msr_bitmap_longmode)
7344 * Allow direct access to the PC debug port (it is often used for I/O
7345 * delays, but the vmexits simply slow things down).
7347 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
7348 clear_bit(0x80, vmx_io_bitmap_a);
7350 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
7352 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
7353 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
7355 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
7357 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
7358 __alignof__(struct vcpu_vmx), THIS_MODULE);
7362 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
7363 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
7364 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
7365 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
7366 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
7367 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
7370 kvm_mmu_set_mask_ptes(0ull,
7371 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
7372 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
7373 0ull, VMX_EPT_EXECUTABLE_MASK);
7374 ept_set_mmio_spte_mask();
7382 free_page((unsigned long)vmx_msr_bitmap_longmode);
7384 free_page((unsigned long)vmx_msr_bitmap_legacy);
7386 free_page((unsigned long)vmx_io_bitmap_b);
7388 free_page((unsigned long)vmx_io_bitmap_a);
7392 static void __exit vmx_exit(void)
7394 free_page((unsigned long)vmx_msr_bitmap_legacy);
7395 free_page((unsigned long)vmx_msr_bitmap_longmode);
7396 free_page((unsigned long)vmx_io_bitmap_b);
7397 free_page((unsigned long)vmx_io_bitmap_a);
7402 module_init(vmx_init)
7403 module_exit(vmx_exit)