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>
45 #include <asm/kexec.h>
49 #define __ex(x) __kvm_handle_fault_on_reboot(x)
50 #define __ex_clear(x, reg) \
51 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
53 MODULE_AUTHOR("Qumranet");
54 MODULE_LICENSE("GPL");
56 static const struct x86_cpu_id vmx_cpu_id[] = {
57 X86_FEATURE_MATCH(X86_FEATURE_VMX),
60 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
62 static bool __read_mostly enable_vpid = 1;
63 module_param_named(vpid, enable_vpid, bool, 0444);
65 static bool __read_mostly flexpriority_enabled = 1;
66 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
68 static bool __read_mostly enable_ept = 1;
69 module_param_named(ept, enable_ept, bool, S_IRUGO);
71 static bool __read_mostly enable_unrestricted_guest = 1;
72 module_param_named(unrestricted_guest,
73 enable_unrestricted_guest, bool, S_IRUGO);
75 static bool __read_mostly enable_ept_ad_bits = 1;
76 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
78 static bool __read_mostly emulate_invalid_guest_state = true;
79 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
81 static bool __read_mostly vmm_exclusive = 1;
82 module_param(vmm_exclusive, bool, S_IRUGO);
84 static bool __read_mostly fasteoi = 1;
85 module_param(fasteoi, bool, S_IRUGO);
87 static bool __read_mostly enable_apicv = 1;
88 module_param(enable_apicv, bool, S_IRUGO);
90 static bool __read_mostly enable_shadow_vmcs = 1;
91 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
93 * If nested=1, nested virtualization is supported, i.e., guests may use
94 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
95 * use VMX instructions.
97 static bool __read_mostly nested = 0;
98 module_param(nested, bool, S_IRUGO);
100 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
101 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
102 #define KVM_VM_CR0_ALWAYS_ON \
103 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
104 #define KVM_CR4_GUEST_OWNED_BITS \
105 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
106 | X86_CR4_OSXMMEXCPT)
108 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
109 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
111 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
114 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
115 * ple_gap: upper bound on the amount of time between two successive
116 * executions of PAUSE in a loop. Also indicate if ple enabled.
117 * According to test, this time is usually smaller than 128 cycles.
118 * ple_window: upper bound on the amount of time a guest is allowed to execute
119 * in a PAUSE loop. Tests indicate that most spinlocks are held for
120 * less than 2^12 cycles
121 * Time is measured based on a counter that runs at the same rate as the TSC,
122 * refer SDM volume 3b section 21.6.13 & 22.1.3.
124 #define KVM_VMX_DEFAULT_PLE_GAP 128
125 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
126 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
127 module_param(ple_gap, int, S_IRUGO);
129 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
130 module_param(ple_window, int, S_IRUGO);
132 extern const ulong vmx_return;
134 #define NR_AUTOLOAD_MSRS 8
135 #define VMCS02_POOL_SIZE 1
144 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
145 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
146 * loaded on this CPU (so we can clear them if the CPU goes down).
152 struct list_head loaded_vmcss_on_cpu_link;
155 struct shared_msr_entry {
162 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
163 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
164 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
165 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
166 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
167 * More than one of these structures may exist, if L1 runs multiple L2 guests.
168 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
169 * underlying hardware which will be used to run L2.
170 * This structure is packed to ensure that its layout is identical across
171 * machines (necessary for live migration).
172 * If there are changes in this struct, VMCS12_REVISION must be changed.
174 typedef u64 natural_width;
175 struct __packed vmcs12 {
176 /* According to the Intel spec, a VMCS region must start with the
177 * following two fields. Then follow implementation-specific data.
182 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
183 u32 padding[7]; /* room for future expansion */
188 u64 vm_exit_msr_store_addr;
189 u64 vm_exit_msr_load_addr;
190 u64 vm_entry_msr_load_addr;
192 u64 virtual_apic_page_addr;
193 u64 apic_access_addr;
195 u64 guest_physical_address;
196 u64 vmcs_link_pointer;
197 u64 guest_ia32_debugctl;
200 u64 guest_ia32_perf_global_ctrl;
207 u64 host_ia32_perf_global_ctrl;
208 u64 padding64[8]; /* room for future expansion */
210 * To allow migration of L1 (complete with its L2 guests) between
211 * machines of different natural widths (32 or 64 bit), we cannot have
212 * unsigned long fields with no explict size. We use u64 (aliased
213 * natural_width) instead. Luckily, x86 is little-endian.
215 natural_width cr0_guest_host_mask;
216 natural_width cr4_guest_host_mask;
217 natural_width cr0_read_shadow;
218 natural_width cr4_read_shadow;
219 natural_width cr3_target_value0;
220 natural_width cr3_target_value1;
221 natural_width cr3_target_value2;
222 natural_width cr3_target_value3;
223 natural_width exit_qualification;
224 natural_width guest_linear_address;
225 natural_width guest_cr0;
226 natural_width guest_cr3;
227 natural_width guest_cr4;
228 natural_width guest_es_base;
229 natural_width guest_cs_base;
230 natural_width guest_ss_base;
231 natural_width guest_ds_base;
232 natural_width guest_fs_base;
233 natural_width guest_gs_base;
234 natural_width guest_ldtr_base;
235 natural_width guest_tr_base;
236 natural_width guest_gdtr_base;
237 natural_width guest_idtr_base;
238 natural_width guest_dr7;
239 natural_width guest_rsp;
240 natural_width guest_rip;
241 natural_width guest_rflags;
242 natural_width guest_pending_dbg_exceptions;
243 natural_width guest_sysenter_esp;
244 natural_width guest_sysenter_eip;
245 natural_width host_cr0;
246 natural_width host_cr3;
247 natural_width host_cr4;
248 natural_width host_fs_base;
249 natural_width host_gs_base;
250 natural_width host_tr_base;
251 natural_width host_gdtr_base;
252 natural_width host_idtr_base;
253 natural_width host_ia32_sysenter_esp;
254 natural_width host_ia32_sysenter_eip;
255 natural_width host_rsp;
256 natural_width host_rip;
257 natural_width paddingl[8]; /* room for future expansion */
258 u32 pin_based_vm_exec_control;
259 u32 cpu_based_vm_exec_control;
260 u32 exception_bitmap;
261 u32 page_fault_error_code_mask;
262 u32 page_fault_error_code_match;
263 u32 cr3_target_count;
264 u32 vm_exit_controls;
265 u32 vm_exit_msr_store_count;
266 u32 vm_exit_msr_load_count;
267 u32 vm_entry_controls;
268 u32 vm_entry_msr_load_count;
269 u32 vm_entry_intr_info_field;
270 u32 vm_entry_exception_error_code;
271 u32 vm_entry_instruction_len;
273 u32 secondary_vm_exec_control;
274 u32 vm_instruction_error;
276 u32 vm_exit_intr_info;
277 u32 vm_exit_intr_error_code;
278 u32 idt_vectoring_info_field;
279 u32 idt_vectoring_error_code;
280 u32 vm_exit_instruction_len;
281 u32 vmx_instruction_info;
288 u32 guest_ldtr_limit;
290 u32 guest_gdtr_limit;
291 u32 guest_idtr_limit;
292 u32 guest_es_ar_bytes;
293 u32 guest_cs_ar_bytes;
294 u32 guest_ss_ar_bytes;
295 u32 guest_ds_ar_bytes;
296 u32 guest_fs_ar_bytes;
297 u32 guest_gs_ar_bytes;
298 u32 guest_ldtr_ar_bytes;
299 u32 guest_tr_ar_bytes;
300 u32 guest_interruptibility_info;
301 u32 guest_activity_state;
302 u32 guest_sysenter_cs;
303 u32 host_ia32_sysenter_cs;
304 u32 vmx_preemption_timer_value;
305 u32 padding32[7]; /* room for future expansion */
306 u16 virtual_processor_id;
307 u16 guest_es_selector;
308 u16 guest_cs_selector;
309 u16 guest_ss_selector;
310 u16 guest_ds_selector;
311 u16 guest_fs_selector;
312 u16 guest_gs_selector;
313 u16 guest_ldtr_selector;
314 u16 guest_tr_selector;
315 u16 host_es_selector;
316 u16 host_cs_selector;
317 u16 host_ss_selector;
318 u16 host_ds_selector;
319 u16 host_fs_selector;
320 u16 host_gs_selector;
321 u16 host_tr_selector;
325 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
326 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
327 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
329 #define VMCS12_REVISION 0x11e57ed0
332 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
333 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
334 * current implementation, 4K are reserved to avoid future complications.
336 #define VMCS12_SIZE 0x1000
338 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
340 struct list_head list;
342 struct loaded_vmcs vmcs02;
346 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
347 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
350 /* Has the level1 guest done vmxon? */
353 /* The guest-physical address of the current VMCS L1 keeps for L2 */
355 /* The host-usable pointer to the above */
356 struct page *current_vmcs12_page;
357 struct vmcs12 *current_vmcs12;
358 struct vmcs *current_shadow_vmcs;
360 * Indicates if the shadow vmcs must be updated with the
361 * data hold by vmcs12
363 bool sync_shadow_vmcs;
365 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
366 struct list_head vmcs02_pool;
368 u64 vmcs01_tsc_offset;
369 /* L2 must run next, and mustn't decide to exit to L1. */
370 bool nested_run_pending;
372 * Guest pages referred to in vmcs02 with host-physical pointers, so
373 * we must keep them pinned while L2 runs.
375 struct page *apic_access_page;
376 u64 msr_ia32_feature_control;
379 #define POSTED_INTR_ON 0
380 /* Posted-Interrupt Descriptor */
382 u32 pir[8]; /* Posted interrupt requested */
383 u32 control; /* bit 0 of control is outstanding notification bit */
387 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
389 return test_and_set_bit(POSTED_INTR_ON,
390 (unsigned long *)&pi_desc->control);
393 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
395 return test_and_clear_bit(POSTED_INTR_ON,
396 (unsigned long *)&pi_desc->control);
399 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
401 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
405 struct kvm_vcpu vcpu;
406 unsigned long host_rsp;
409 bool nmi_known_unmasked;
411 u32 idt_vectoring_info;
413 struct shared_msr_entry *guest_msrs;
416 unsigned long host_idt_base;
418 u64 msr_host_kernel_gs_base;
419 u64 msr_guest_kernel_gs_base;
422 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
423 * non-nested (L1) guest, it always points to vmcs01. For a nested
424 * guest (L2), it points to a different VMCS.
426 struct loaded_vmcs vmcs01;
427 struct loaded_vmcs *loaded_vmcs;
428 bool __launched; /* temporary, used in vmx_vcpu_run */
429 struct msr_autoload {
431 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
432 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
436 u16 fs_sel, gs_sel, ldt_sel;
440 int gs_ldt_reload_needed;
441 int fs_reload_needed;
446 struct kvm_segment segs[8];
449 u32 bitmask; /* 4 bits per segment (1 bit per field) */
450 struct kvm_save_segment {
458 bool emulation_required;
460 /* Support for vnmi-less CPUs */
461 int soft_vnmi_blocked;
463 s64 vnmi_blocked_time;
468 /* Posted interrupt descriptor */
469 struct pi_desc pi_desc;
471 /* Support for a guest hypervisor (nested VMX) */
472 struct nested_vmx nested;
475 enum segment_cache_field {
484 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
486 return container_of(vcpu, struct vcpu_vmx, vcpu);
489 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
490 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
491 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
492 [number##_HIGH] = VMCS12_OFFSET(name)+4
495 static const unsigned long shadow_read_only_fields[] = {
497 * We do NOT shadow fields that are modified when L0
498 * traps and emulates any vmx instruction (e.g. VMPTRLD,
499 * VMXON...) executed by L1.
500 * For example, VM_INSTRUCTION_ERROR is read
501 * by L1 if a vmx instruction fails (part of the error path).
502 * Note the code assumes this logic. If for some reason
503 * we start shadowing these fields then we need to
504 * force a shadow sync when L0 emulates vmx instructions
505 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
506 * by nested_vmx_failValid)
510 VM_EXIT_INSTRUCTION_LEN,
511 IDT_VECTORING_INFO_FIELD,
512 IDT_VECTORING_ERROR_CODE,
513 VM_EXIT_INTR_ERROR_CODE,
515 GUEST_LINEAR_ADDRESS,
516 GUEST_PHYSICAL_ADDRESS
518 static const int max_shadow_read_only_fields =
519 ARRAY_SIZE(shadow_read_only_fields);
521 static const unsigned long shadow_read_write_fields[] = {
527 GUEST_INTERRUPTIBILITY_INFO,
539 CPU_BASED_VM_EXEC_CONTROL,
540 VM_ENTRY_EXCEPTION_ERROR_CODE,
541 VM_ENTRY_INTR_INFO_FIELD,
542 VM_ENTRY_INSTRUCTION_LEN,
543 VM_ENTRY_EXCEPTION_ERROR_CODE,
549 static const int max_shadow_read_write_fields =
550 ARRAY_SIZE(shadow_read_write_fields);
552 static const unsigned short vmcs_field_to_offset_table[] = {
553 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
554 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
555 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
556 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
557 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
558 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
559 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
560 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
561 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
562 FIELD(HOST_ES_SELECTOR, host_es_selector),
563 FIELD(HOST_CS_SELECTOR, host_cs_selector),
564 FIELD(HOST_SS_SELECTOR, host_ss_selector),
565 FIELD(HOST_DS_SELECTOR, host_ds_selector),
566 FIELD(HOST_FS_SELECTOR, host_fs_selector),
567 FIELD(HOST_GS_SELECTOR, host_gs_selector),
568 FIELD(HOST_TR_SELECTOR, host_tr_selector),
569 FIELD64(IO_BITMAP_A, io_bitmap_a),
570 FIELD64(IO_BITMAP_B, io_bitmap_b),
571 FIELD64(MSR_BITMAP, msr_bitmap),
572 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
573 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
574 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
575 FIELD64(TSC_OFFSET, tsc_offset),
576 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
577 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
578 FIELD64(EPT_POINTER, ept_pointer),
579 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
580 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
581 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
582 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
583 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
584 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
585 FIELD64(GUEST_PDPTR0, guest_pdptr0),
586 FIELD64(GUEST_PDPTR1, guest_pdptr1),
587 FIELD64(GUEST_PDPTR2, guest_pdptr2),
588 FIELD64(GUEST_PDPTR3, guest_pdptr3),
589 FIELD64(HOST_IA32_PAT, host_ia32_pat),
590 FIELD64(HOST_IA32_EFER, host_ia32_efer),
591 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
592 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
593 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
594 FIELD(EXCEPTION_BITMAP, exception_bitmap),
595 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
596 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
597 FIELD(CR3_TARGET_COUNT, cr3_target_count),
598 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
599 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
600 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
601 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
602 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
603 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
604 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
605 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
606 FIELD(TPR_THRESHOLD, tpr_threshold),
607 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
608 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
609 FIELD(VM_EXIT_REASON, vm_exit_reason),
610 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
611 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
612 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
613 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
614 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
615 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
616 FIELD(GUEST_ES_LIMIT, guest_es_limit),
617 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
618 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
619 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
620 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
621 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
622 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
623 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
624 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
625 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
626 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
627 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
628 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
629 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
630 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
631 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
632 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
633 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
634 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
635 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
636 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
637 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
638 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
639 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
640 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
641 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
642 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
643 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
644 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
645 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
646 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
647 FIELD(EXIT_QUALIFICATION, exit_qualification),
648 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
649 FIELD(GUEST_CR0, guest_cr0),
650 FIELD(GUEST_CR3, guest_cr3),
651 FIELD(GUEST_CR4, guest_cr4),
652 FIELD(GUEST_ES_BASE, guest_es_base),
653 FIELD(GUEST_CS_BASE, guest_cs_base),
654 FIELD(GUEST_SS_BASE, guest_ss_base),
655 FIELD(GUEST_DS_BASE, guest_ds_base),
656 FIELD(GUEST_FS_BASE, guest_fs_base),
657 FIELD(GUEST_GS_BASE, guest_gs_base),
658 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
659 FIELD(GUEST_TR_BASE, guest_tr_base),
660 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
661 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
662 FIELD(GUEST_DR7, guest_dr7),
663 FIELD(GUEST_RSP, guest_rsp),
664 FIELD(GUEST_RIP, guest_rip),
665 FIELD(GUEST_RFLAGS, guest_rflags),
666 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
667 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
668 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
669 FIELD(HOST_CR0, host_cr0),
670 FIELD(HOST_CR3, host_cr3),
671 FIELD(HOST_CR4, host_cr4),
672 FIELD(HOST_FS_BASE, host_fs_base),
673 FIELD(HOST_GS_BASE, host_gs_base),
674 FIELD(HOST_TR_BASE, host_tr_base),
675 FIELD(HOST_GDTR_BASE, host_gdtr_base),
676 FIELD(HOST_IDTR_BASE, host_idtr_base),
677 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
678 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
679 FIELD(HOST_RSP, host_rsp),
680 FIELD(HOST_RIP, host_rip),
682 static const int max_vmcs_field = ARRAY_SIZE(vmcs_field_to_offset_table);
684 static inline short vmcs_field_to_offset(unsigned long field)
686 if (field >= max_vmcs_field || vmcs_field_to_offset_table[field] == 0)
688 return vmcs_field_to_offset_table[field];
691 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
693 return to_vmx(vcpu)->nested.current_vmcs12;
696 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
698 struct page *page = gfn_to_page(vcpu->kvm, addr >> PAGE_SHIFT);
699 if (is_error_page(page))
705 static void nested_release_page(struct page *page)
707 kvm_release_page_dirty(page);
710 static void nested_release_page_clean(struct page *page)
712 kvm_release_page_clean(page);
715 static u64 construct_eptp(unsigned long root_hpa);
716 static void kvm_cpu_vmxon(u64 addr);
717 static void kvm_cpu_vmxoff(void);
718 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3);
719 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
720 static void vmx_set_segment(struct kvm_vcpu *vcpu,
721 struct kvm_segment *var, int seg);
722 static void vmx_get_segment(struct kvm_vcpu *vcpu,
723 struct kvm_segment *var, int seg);
724 static bool guest_state_valid(struct kvm_vcpu *vcpu);
725 static u32 vmx_segment_access_rights(struct kvm_segment *var);
726 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu);
727 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
728 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
730 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
731 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
733 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
734 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
736 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
737 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
739 static unsigned long *vmx_io_bitmap_a;
740 static unsigned long *vmx_io_bitmap_b;
741 static unsigned long *vmx_msr_bitmap_legacy;
742 static unsigned long *vmx_msr_bitmap_longmode;
743 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
744 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
745 static unsigned long *vmx_vmread_bitmap;
746 static unsigned long *vmx_vmwrite_bitmap;
748 static bool cpu_has_load_ia32_efer;
749 static bool cpu_has_load_perf_global_ctrl;
751 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
752 static DEFINE_SPINLOCK(vmx_vpid_lock);
754 static struct vmcs_config {
758 u32 pin_based_exec_ctrl;
759 u32 cpu_based_exec_ctrl;
760 u32 cpu_based_2nd_exec_ctrl;
765 static struct vmx_capability {
770 #define VMX_SEGMENT_FIELD(seg) \
771 [VCPU_SREG_##seg] = { \
772 .selector = GUEST_##seg##_SELECTOR, \
773 .base = GUEST_##seg##_BASE, \
774 .limit = GUEST_##seg##_LIMIT, \
775 .ar_bytes = GUEST_##seg##_AR_BYTES, \
778 static const struct kvm_vmx_segment_field {
783 } kvm_vmx_segment_fields[] = {
784 VMX_SEGMENT_FIELD(CS),
785 VMX_SEGMENT_FIELD(DS),
786 VMX_SEGMENT_FIELD(ES),
787 VMX_SEGMENT_FIELD(FS),
788 VMX_SEGMENT_FIELD(GS),
789 VMX_SEGMENT_FIELD(SS),
790 VMX_SEGMENT_FIELD(TR),
791 VMX_SEGMENT_FIELD(LDTR),
794 static u64 host_efer;
796 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
799 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
800 * away by decrementing the array size.
802 static const u32 vmx_msr_index[] = {
804 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
806 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
808 #define NR_VMX_MSR ARRAY_SIZE(vmx_msr_index)
810 static inline bool is_page_fault(u32 intr_info)
812 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
813 INTR_INFO_VALID_MASK)) ==
814 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
817 static inline bool is_no_device(u32 intr_info)
819 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
820 INTR_INFO_VALID_MASK)) ==
821 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
824 static inline bool is_invalid_opcode(u32 intr_info)
826 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
827 INTR_INFO_VALID_MASK)) ==
828 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
831 static inline bool is_external_interrupt(u32 intr_info)
833 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
834 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
837 static inline bool is_machine_check(u32 intr_info)
839 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
840 INTR_INFO_VALID_MASK)) ==
841 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
844 static inline bool cpu_has_vmx_msr_bitmap(void)
846 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
849 static inline bool cpu_has_vmx_tpr_shadow(void)
851 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
854 static inline bool vm_need_tpr_shadow(struct kvm *kvm)
856 return (cpu_has_vmx_tpr_shadow()) && (irqchip_in_kernel(kvm));
859 static inline bool cpu_has_secondary_exec_ctrls(void)
861 return vmcs_config.cpu_based_exec_ctrl &
862 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
865 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
867 return vmcs_config.cpu_based_2nd_exec_ctrl &
868 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
871 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
873 return vmcs_config.cpu_based_2nd_exec_ctrl &
874 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
877 static inline bool cpu_has_vmx_apic_register_virt(void)
879 return vmcs_config.cpu_based_2nd_exec_ctrl &
880 SECONDARY_EXEC_APIC_REGISTER_VIRT;
883 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
885 return vmcs_config.cpu_based_2nd_exec_ctrl &
886 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
889 static inline bool cpu_has_vmx_posted_intr(void)
891 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
894 static inline bool cpu_has_vmx_apicv(void)
896 return cpu_has_vmx_apic_register_virt() &&
897 cpu_has_vmx_virtual_intr_delivery() &&
898 cpu_has_vmx_posted_intr();
901 static inline bool cpu_has_vmx_flexpriority(void)
903 return cpu_has_vmx_tpr_shadow() &&
904 cpu_has_vmx_virtualize_apic_accesses();
907 static inline bool cpu_has_vmx_ept_execute_only(void)
909 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
912 static inline bool cpu_has_vmx_eptp_uncacheable(void)
914 return vmx_capability.ept & VMX_EPTP_UC_BIT;
917 static inline bool cpu_has_vmx_eptp_writeback(void)
919 return vmx_capability.ept & VMX_EPTP_WB_BIT;
922 static inline bool cpu_has_vmx_ept_2m_page(void)
924 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
927 static inline bool cpu_has_vmx_ept_1g_page(void)
929 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
932 static inline bool cpu_has_vmx_ept_4levels(void)
934 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
937 static inline bool cpu_has_vmx_ept_ad_bits(void)
939 return vmx_capability.ept & VMX_EPT_AD_BIT;
942 static inline bool cpu_has_vmx_invept_context(void)
944 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
947 static inline bool cpu_has_vmx_invept_global(void)
949 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
952 static inline bool cpu_has_vmx_invvpid_single(void)
954 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
957 static inline bool cpu_has_vmx_invvpid_global(void)
959 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
962 static inline bool cpu_has_vmx_ept(void)
964 return vmcs_config.cpu_based_2nd_exec_ctrl &
965 SECONDARY_EXEC_ENABLE_EPT;
968 static inline bool cpu_has_vmx_unrestricted_guest(void)
970 return vmcs_config.cpu_based_2nd_exec_ctrl &
971 SECONDARY_EXEC_UNRESTRICTED_GUEST;
974 static inline bool cpu_has_vmx_ple(void)
976 return vmcs_config.cpu_based_2nd_exec_ctrl &
977 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
980 static inline bool vm_need_virtualize_apic_accesses(struct kvm *kvm)
982 return flexpriority_enabled && irqchip_in_kernel(kvm);
985 static inline bool cpu_has_vmx_vpid(void)
987 return vmcs_config.cpu_based_2nd_exec_ctrl &
988 SECONDARY_EXEC_ENABLE_VPID;
991 static inline bool cpu_has_vmx_rdtscp(void)
993 return vmcs_config.cpu_based_2nd_exec_ctrl &
994 SECONDARY_EXEC_RDTSCP;
997 static inline bool cpu_has_vmx_invpcid(void)
999 return vmcs_config.cpu_based_2nd_exec_ctrl &
1000 SECONDARY_EXEC_ENABLE_INVPCID;
1003 static inline bool cpu_has_virtual_nmis(void)
1005 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1008 static inline bool cpu_has_vmx_wbinvd_exit(void)
1010 return vmcs_config.cpu_based_2nd_exec_ctrl &
1011 SECONDARY_EXEC_WBINVD_EXITING;
1014 static inline bool cpu_has_vmx_shadow_vmcs(void)
1017 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1018 /* check if the cpu supports writing r/o exit information fields */
1019 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1022 return vmcs_config.cpu_based_2nd_exec_ctrl &
1023 SECONDARY_EXEC_SHADOW_VMCS;
1026 static inline bool report_flexpriority(void)
1028 return flexpriority_enabled;
1031 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1033 return vmcs12->cpu_based_vm_exec_control & bit;
1036 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1038 return (vmcs12->cpu_based_vm_exec_control &
1039 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1040 (vmcs12->secondary_vm_exec_control & bit);
1043 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12,
1044 struct kvm_vcpu *vcpu)
1046 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1049 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1051 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1054 static inline bool is_exception(u32 intr_info)
1056 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1057 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1060 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu);
1061 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1062 struct vmcs12 *vmcs12,
1063 u32 reason, unsigned long qualification);
1065 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1069 for (i = 0; i < vmx->nmsrs; ++i)
1070 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1075 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1081 } operand = { vpid, 0, gva };
1083 asm volatile (__ex(ASM_VMX_INVVPID)
1084 /* CF==1 or ZF==1 --> rc = -1 */
1085 "; ja 1f ; ud2 ; 1:"
1086 : : "a"(&operand), "c"(ext) : "cc", "memory");
1089 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1093 } operand = {eptp, gpa};
1095 asm volatile (__ex(ASM_VMX_INVEPT)
1096 /* CF==1 or ZF==1 --> rc = -1 */
1097 "; ja 1f ; ud2 ; 1:\n"
1098 : : "a" (&operand), "c" (ext) : "cc", "memory");
1101 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1105 i = __find_msr_index(vmx, msr);
1107 return &vmx->guest_msrs[i];
1111 static void vmcs_clear(struct vmcs *vmcs)
1113 u64 phys_addr = __pa(vmcs);
1116 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1117 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1120 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1124 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1126 vmcs_clear(loaded_vmcs->vmcs);
1127 loaded_vmcs->cpu = -1;
1128 loaded_vmcs->launched = 0;
1131 static void vmcs_load(struct vmcs *vmcs)
1133 u64 phys_addr = __pa(vmcs);
1136 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1137 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1140 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1146 * This bitmap is used to indicate whether the vmclear
1147 * operation is enabled on all cpus. All disabled by
1150 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1152 static inline void crash_enable_local_vmclear(int cpu)
1154 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1157 static inline void crash_disable_local_vmclear(int cpu)
1159 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1162 static inline int crash_local_vmclear_enabled(int cpu)
1164 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1167 static void crash_vmclear_local_loaded_vmcss(void)
1169 int cpu = raw_smp_processor_id();
1170 struct loaded_vmcs *v;
1172 if (!crash_local_vmclear_enabled(cpu))
1175 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1176 loaded_vmcss_on_cpu_link)
1177 vmcs_clear(v->vmcs);
1180 static inline void crash_enable_local_vmclear(int cpu) { }
1181 static inline void crash_disable_local_vmclear(int cpu) { }
1182 #endif /* CONFIG_KEXEC */
1184 static void __loaded_vmcs_clear(void *arg)
1186 struct loaded_vmcs *loaded_vmcs = arg;
1187 int cpu = raw_smp_processor_id();
1189 if (loaded_vmcs->cpu != cpu)
1190 return; /* vcpu migration can race with cpu offline */
1191 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1192 per_cpu(current_vmcs, cpu) = NULL;
1193 crash_disable_local_vmclear(cpu);
1194 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1197 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1198 * is before setting loaded_vmcs->vcpu to -1 which is done in
1199 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1200 * then adds the vmcs into percpu list before it is deleted.
1204 loaded_vmcs_init(loaded_vmcs);
1205 crash_enable_local_vmclear(cpu);
1208 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1210 int cpu = loaded_vmcs->cpu;
1213 smp_call_function_single(cpu,
1214 __loaded_vmcs_clear, loaded_vmcs, 1);
1217 static inline void vpid_sync_vcpu_single(struct vcpu_vmx *vmx)
1222 if (cpu_has_vmx_invvpid_single())
1223 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vmx->vpid, 0);
1226 static inline void vpid_sync_vcpu_global(void)
1228 if (cpu_has_vmx_invvpid_global())
1229 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1232 static inline void vpid_sync_context(struct vcpu_vmx *vmx)
1234 if (cpu_has_vmx_invvpid_single())
1235 vpid_sync_vcpu_single(vmx);
1237 vpid_sync_vcpu_global();
1240 static inline void ept_sync_global(void)
1242 if (cpu_has_vmx_invept_global())
1243 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1246 static inline void ept_sync_context(u64 eptp)
1249 if (cpu_has_vmx_invept_context())
1250 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1256 static __always_inline unsigned long vmcs_readl(unsigned long field)
1258 unsigned long value;
1260 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1261 : "=a"(value) : "d"(field) : "cc");
1265 static __always_inline u16 vmcs_read16(unsigned long field)
1267 return vmcs_readl(field);
1270 static __always_inline u32 vmcs_read32(unsigned long field)
1272 return vmcs_readl(field);
1275 static __always_inline u64 vmcs_read64(unsigned long field)
1277 #ifdef CONFIG_X86_64
1278 return vmcs_readl(field);
1280 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1284 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1286 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1287 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1291 static void vmcs_writel(unsigned long field, unsigned long value)
1295 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1296 : "=q"(error) : "a"(value), "d"(field) : "cc");
1297 if (unlikely(error))
1298 vmwrite_error(field, value);
1301 static void vmcs_write16(unsigned long field, u16 value)
1303 vmcs_writel(field, value);
1306 static void vmcs_write32(unsigned long field, u32 value)
1308 vmcs_writel(field, value);
1311 static void vmcs_write64(unsigned long field, u64 value)
1313 vmcs_writel(field, value);
1314 #ifndef CONFIG_X86_64
1316 vmcs_writel(field+1, value >> 32);
1320 static void vmcs_clear_bits(unsigned long field, u32 mask)
1322 vmcs_writel(field, vmcs_readl(field) & ~mask);
1325 static void vmcs_set_bits(unsigned long field, u32 mask)
1327 vmcs_writel(field, vmcs_readl(field) | mask);
1330 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1332 vmx->segment_cache.bitmask = 0;
1335 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1339 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1341 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1342 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1343 vmx->segment_cache.bitmask = 0;
1345 ret = vmx->segment_cache.bitmask & mask;
1346 vmx->segment_cache.bitmask |= mask;
1350 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1352 u16 *p = &vmx->segment_cache.seg[seg].selector;
1354 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1355 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1359 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1361 ulong *p = &vmx->segment_cache.seg[seg].base;
1363 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1364 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1368 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1370 u32 *p = &vmx->segment_cache.seg[seg].limit;
1372 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1373 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1377 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1379 u32 *p = &vmx->segment_cache.seg[seg].ar;
1381 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1382 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1386 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1390 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1391 (1u << NM_VECTOR) | (1u << DB_VECTOR);
1392 if ((vcpu->guest_debug &
1393 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1394 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1395 eb |= 1u << BP_VECTOR;
1396 if (to_vmx(vcpu)->rmode.vm86_active)
1399 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1400 if (vcpu->fpu_active)
1401 eb &= ~(1u << NM_VECTOR);
1403 /* When we are running a nested L2 guest and L1 specified for it a
1404 * certain exception bitmap, we must trap the same exceptions and pass
1405 * them to L1. When running L2, we will only handle the exceptions
1406 * specified above if L1 did not want them.
1408 if (is_guest_mode(vcpu))
1409 eb |= get_vmcs12(vcpu)->exception_bitmap;
1411 vmcs_write32(EXCEPTION_BITMAP, eb);
1414 static void clear_atomic_switch_msr_special(unsigned long entry,
1417 vmcs_clear_bits(VM_ENTRY_CONTROLS, entry);
1418 vmcs_clear_bits(VM_EXIT_CONTROLS, exit);
1421 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1424 struct msr_autoload *m = &vmx->msr_autoload;
1428 if (cpu_has_load_ia32_efer) {
1429 clear_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1430 VM_EXIT_LOAD_IA32_EFER);
1434 case MSR_CORE_PERF_GLOBAL_CTRL:
1435 if (cpu_has_load_perf_global_ctrl) {
1436 clear_atomic_switch_msr_special(
1437 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1438 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1444 for (i = 0; i < m->nr; ++i)
1445 if (m->guest[i].index == msr)
1451 m->guest[i] = m->guest[m->nr];
1452 m->host[i] = m->host[m->nr];
1453 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1454 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1457 static void add_atomic_switch_msr_special(unsigned long entry,
1458 unsigned long exit, unsigned long guest_val_vmcs,
1459 unsigned long host_val_vmcs, u64 guest_val, u64 host_val)
1461 vmcs_write64(guest_val_vmcs, guest_val);
1462 vmcs_write64(host_val_vmcs, host_val);
1463 vmcs_set_bits(VM_ENTRY_CONTROLS, entry);
1464 vmcs_set_bits(VM_EXIT_CONTROLS, exit);
1467 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1468 u64 guest_val, u64 host_val)
1471 struct msr_autoload *m = &vmx->msr_autoload;
1475 if (cpu_has_load_ia32_efer) {
1476 add_atomic_switch_msr_special(VM_ENTRY_LOAD_IA32_EFER,
1477 VM_EXIT_LOAD_IA32_EFER,
1480 guest_val, host_val);
1484 case MSR_CORE_PERF_GLOBAL_CTRL:
1485 if (cpu_has_load_perf_global_ctrl) {
1486 add_atomic_switch_msr_special(
1487 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1488 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1489 GUEST_IA32_PERF_GLOBAL_CTRL,
1490 HOST_IA32_PERF_GLOBAL_CTRL,
1491 guest_val, host_val);
1497 for (i = 0; i < m->nr; ++i)
1498 if (m->guest[i].index == msr)
1501 if (i == NR_AUTOLOAD_MSRS) {
1502 printk_once(KERN_WARNING"Not enough mst switch entries. "
1503 "Can't add msr %x\n", msr);
1505 } else if (i == m->nr) {
1507 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1508 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1511 m->guest[i].index = msr;
1512 m->guest[i].value = guest_val;
1513 m->host[i].index = msr;
1514 m->host[i].value = host_val;
1517 static void reload_tss(void)
1520 * VT restores TR but not its size. Useless.
1522 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1523 struct desc_struct *descs;
1525 descs = (void *)gdt->address;
1526 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1530 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1535 guest_efer = vmx->vcpu.arch.efer;
1538 * NX is emulated; LMA and LME handled by hardware; SCE meaningless
1541 ignore_bits = EFER_NX | EFER_SCE;
1542 #ifdef CONFIG_X86_64
1543 ignore_bits |= EFER_LMA | EFER_LME;
1544 /* SCE is meaningful only in long mode on Intel */
1545 if (guest_efer & EFER_LMA)
1546 ignore_bits &= ~(u64)EFER_SCE;
1548 guest_efer &= ~ignore_bits;
1549 guest_efer |= host_efer & ignore_bits;
1550 vmx->guest_msrs[efer_offset].data = guest_efer;
1551 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1553 clear_atomic_switch_msr(vmx, MSR_EFER);
1554 /* On ept, can't emulate nx, and must switch nx atomically */
1555 if (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX)) {
1556 guest_efer = vmx->vcpu.arch.efer;
1557 if (!(guest_efer & EFER_LMA))
1558 guest_efer &= ~EFER_LME;
1559 add_atomic_switch_msr(vmx, MSR_EFER, guest_efer, host_efer);
1566 static unsigned long segment_base(u16 selector)
1568 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1569 struct desc_struct *d;
1570 unsigned long table_base;
1573 if (!(selector & ~3))
1576 table_base = gdt->address;
1578 if (selector & 4) { /* from ldt */
1579 u16 ldt_selector = kvm_read_ldt();
1581 if (!(ldt_selector & ~3))
1584 table_base = segment_base(ldt_selector);
1586 d = (struct desc_struct *)(table_base + (selector & ~7));
1587 v = get_desc_base(d);
1588 #ifdef CONFIG_X86_64
1589 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1590 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1595 static inline unsigned long kvm_read_tr_base(void)
1598 asm("str %0" : "=g"(tr));
1599 return segment_base(tr);
1602 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1604 struct vcpu_vmx *vmx = to_vmx(vcpu);
1607 if (vmx->host_state.loaded)
1610 vmx->host_state.loaded = 1;
1612 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1613 * allow segment selectors with cpl > 0 or ti == 1.
1615 vmx->host_state.ldt_sel = kvm_read_ldt();
1616 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1617 savesegment(fs, vmx->host_state.fs_sel);
1618 if (!(vmx->host_state.fs_sel & 7)) {
1619 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1620 vmx->host_state.fs_reload_needed = 0;
1622 vmcs_write16(HOST_FS_SELECTOR, 0);
1623 vmx->host_state.fs_reload_needed = 1;
1625 savesegment(gs, vmx->host_state.gs_sel);
1626 if (!(vmx->host_state.gs_sel & 7))
1627 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1629 vmcs_write16(HOST_GS_SELECTOR, 0);
1630 vmx->host_state.gs_ldt_reload_needed = 1;
1633 #ifdef CONFIG_X86_64
1634 savesegment(ds, vmx->host_state.ds_sel);
1635 savesegment(es, vmx->host_state.es_sel);
1638 #ifdef CONFIG_X86_64
1639 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1640 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1642 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1643 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1646 #ifdef CONFIG_X86_64
1647 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1648 if (is_long_mode(&vmx->vcpu))
1649 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1651 for (i = 0; i < vmx->save_nmsrs; ++i)
1652 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1653 vmx->guest_msrs[i].data,
1654 vmx->guest_msrs[i].mask);
1657 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1659 if (!vmx->host_state.loaded)
1662 ++vmx->vcpu.stat.host_state_reload;
1663 vmx->host_state.loaded = 0;
1664 #ifdef CONFIG_X86_64
1665 if (is_long_mode(&vmx->vcpu))
1666 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1668 if (vmx->host_state.gs_ldt_reload_needed) {
1669 kvm_load_ldt(vmx->host_state.ldt_sel);
1670 #ifdef CONFIG_X86_64
1671 load_gs_index(vmx->host_state.gs_sel);
1673 loadsegment(gs, vmx->host_state.gs_sel);
1676 if (vmx->host_state.fs_reload_needed)
1677 loadsegment(fs, vmx->host_state.fs_sel);
1678 #ifdef CONFIG_X86_64
1679 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1680 loadsegment(ds, vmx->host_state.ds_sel);
1681 loadsegment(es, vmx->host_state.es_sel);
1685 #ifdef CONFIG_X86_64
1686 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1689 * If the FPU is not active (through the host task or
1690 * the guest vcpu), then restore the cr0.TS bit.
1692 if (!user_has_fpu() && !vmx->vcpu.guest_fpu_loaded)
1694 load_gdt(&__get_cpu_var(host_gdt));
1697 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1700 __vmx_load_host_state(vmx);
1705 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
1706 * vcpu mutex is already taken.
1708 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
1710 struct vcpu_vmx *vmx = to_vmx(vcpu);
1711 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
1714 kvm_cpu_vmxon(phys_addr);
1715 else if (vmx->loaded_vmcs->cpu != cpu)
1716 loaded_vmcs_clear(vmx->loaded_vmcs);
1718 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
1719 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
1720 vmcs_load(vmx->loaded_vmcs->vmcs);
1723 if (vmx->loaded_vmcs->cpu != cpu) {
1724 struct desc_ptr *gdt = &__get_cpu_var(host_gdt);
1725 unsigned long sysenter_esp;
1727 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
1728 local_irq_disable();
1729 crash_disable_local_vmclear(cpu);
1732 * Read loaded_vmcs->cpu should be before fetching
1733 * loaded_vmcs->loaded_vmcss_on_cpu_link.
1734 * See the comments in __loaded_vmcs_clear().
1738 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
1739 &per_cpu(loaded_vmcss_on_cpu, cpu));
1740 crash_enable_local_vmclear(cpu);
1744 * Linux uses per-cpu TSS and GDT, so set these when switching
1747 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
1748 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
1750 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
1751 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
1752 vmx->loaded_vmcs->cpu = cpu;
1756 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
1758 __vmx_load_host_state(to_vmx(vcpu));
1759 if (!vmm_exclusive) {
1760 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
1766 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
1770 if (vcpu->fpu_active)
1772 vcpu->fpu_active = 1;
1773 cr0 = vmcs_readl(GUEST_CR0);
1774 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
1775 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
1776 vmcs_writel(GUEST_CR0, cr0);
1777 update_exception_bitmap(vcpu);
1778 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
1779 if (is_guest_mode(vcpu))
1780 vcpu->arch.cr0_guest_owned_bits &=
1781 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
1782 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1785 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
1788 * Return the cr0 value that a nested guest would read. This is a combination
1789 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
1790 * its hypervisor (cr0_read_shadow).
1792 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
1794 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
1795 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
1797 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
1799 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
1800 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
1803 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
1805 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
1806 * set this *before* calling this function.
1808 vmx_decache_cr0_guest_bits(vcpu);
1809 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
1810 update_exception_bitmap(vcpu);
1811 vcpu->arch.cr0_guest_owned_bits = 0;
1812 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
1813 if (is_guest_mode(vcpu)) {
1815 * L1's specified read shadow might not contain the TS bit,
1816 * so now that we turned on shadowing of this bit, we need to
1817 * set this bit of the shadow. Like in nested_vmx_run we need
1818 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
1819 * up-to-date here because we just decached cr0.TS (and we'll
1820 * only update vmcs12->guest_cr0 on nested exit).
1822 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1823 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
1824 (vcpu->arch.cr0 & X86_CR0_TS);
1825 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
1827 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
1830 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
1832 unsigned long rflags, save_rflags;
1834 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
1835 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1836 rflags = vmcs_readl(GUEST_RFLAGS);
1837 if (to_vmx(vcpu)->rmode.vm86_active) {
1838 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
1839 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
1840 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
1842 to_vmx(vcpu)->rflags = rflags;
1844 return to_vmx(vcpu)->rflags;
1847 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
1849 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
1850 to_vmx(vcpu)->rflags = rflags;
1851 if (to_vmx(vcpu)->rmode.vm86_active) {
1852 to_vmx(vcpu)->rmode.save_rflags = rflags;
1853 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
1855 vmcs_writel(GUEST_RFLAGS, rflags);
1858 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1860 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1863 if (interruptibility & GUEST_INTR_STATE_STI)
1864 ret |= KVM_X86_SHADOW_INT_STI;
1865 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
1866 ret |= KVM_X86_SHADOW_INT_MOV_SS;
1871 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
1873 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
1874 u32 interruptibility = interruptibility_old;
1876 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
1878 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
1879 interruptibility |= GUEST_INTR_STATE_MOV_SS;
1880 else if (mask & KVM_X86_SHADOW_INT_STI)
1881 interruptibility |= GUEST_INTR_STATE_STI;
1883 if ((interruptibility != interruptibility_old))
1884 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
1887 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
1891 rip = kvm_rip_read(vcpu);
1892 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
1893 kvm_rip_write(vcpu, rip);
1895 /* skipping an emulated instruction also counts */
1896 vmx_set_interrupt_shadow(vcpu, 0);
1900 * KVM wants to inject page-faults which it got to the guest. This function
1901 * checks whether in a nested guest, we need to inject them to L1 or L2.
1902 * This function assumes it is called with the exit reason in vmcs02 being
1903 * a #PF exception (this is the only case in which KVM injects a #PF when L2
1906 static int nested_pf_handled(struct kvm_vcpu *vcpu)
1908 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
1910 /* TODO: also check PFEC_MATCH/MASK, not just EB.PF. */
1911 if (!(vmcs12->exception_bitmap & (1u << PF_VECTOR)))
1914 nested_vmx_vmexit(vcpu);
1918 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
1919 bool has_error_code, u32 error_code,
1922 struct vcpu_vmx *vmx = to_vmx(vcpu);
1923 u32 intr_info = nr | INTR_INFO_VALID_MASK;
1925 if (nr == PF_VECTOR && is_guest_mode(vcpu) &&
1926 !vmx->nested.nested_run_pending && nested_pf_handled(vcpu))
1929 if (has_error_code) {
1930 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
1931 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
1934 if (vmx->rmode.vm86_active) {
1936 if (kvm_exception_is_soft(nr))
1937 inc_eip = vcpu->arch.event_exit_inst_len;
1938 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
1939 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
1943 if (kvm_exception_is_soft(nr)) {
1944 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
1945 vmx->vcpu.arch.event_exit_inst_len);
1946 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
1948 intr_info |= INTR_TYPE_HARD_EXCEPTION;
1950 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
1953 static bool vmx_rdtscp_supported(void)
1955 return cpu_has_vmx_rdtscp();
1958 static bool vmx_invpcid_supported(void)
1960 return cpu_has_vmx_invpcid() && enable_ept;
1964 * Swap MSR entry in host/guest MSR entry array.
1966 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
1968 struct shared_msr_entry tmp;
1970 tmp = vmx->guest_msrs[to];
1971 vmx->guest_msrs[to] = vmx->guest_msrs[from];
1972 vmx->guest_msrs[from] = tmp;
1975 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
1977 unsigned long *msr_bitmap;
1979 if (irqchip_in_kernel(vcpu->kvm) && apic_x2apic_mode(vcpu->arch.apic)) {
1980 if (is_long_mode(vcpu))
1981 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
1983 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
1985 if (is_long_mode(vcpu))
1986 msr_bitmap = vmx_msr_bitmap_longmode;
1988 msr_bitmap = vmx_msr_bitmap_legacy;
1991 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
1995 * Set up the vmcs to automatically save and restore system
1996 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
1997 * mode, as fiddling with msrs is very expensive.
1999 static void setup_msrs(struct vcpu_vmx *vmx)
2001 int save_nmsrs, index;
2004 #ifdef CONFIG_X86_64
2005 if (is_long_mode(&vmx->vcpu)) {
2006 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2008 move_msr_up(vmx, index, save_nmsrs++);
2009 index = __find_msr_index(vmx, MSR_LSTAR);
2011 move_msr_up(vmx, index, save_nmsrs++);
2012 index = __find_msr_index(vmx, MSR_CSTAR);
2014 move_msr_up(vmx, index, save_nmsrs++);
2015 index = __find_msr_index(vmx, MSR_TSC_AUX);
2016 if (index >= 0 && vmx->rdtscp_enabled)
2017 move_msr_up(vmx, index, save_nmsrs++);
2019 * MSR_STAR is only needed on long mode guests, and only
2020 * if efer.sce is enabled.
2022 index = __find_msr_index(vmx, MSR_STAR);
2023 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2024 move_msr_up(vmx, index, save_nmsrs++);
2027 index = __find_msr_index(vmx, MSR_EFER);
2028 if (index >= 0 && update_transition_efer(vmx, index))
2029 move_msr_up(vmx, index, save_nmsrs++);
2031 vmx->save_nmsrs = save_nmsrs;
2033 if (cpu_has_vmx_msr_bitmap())
2034 vmx_set_msr_bitmap(&vmx->vcpu);
2038 * reads and returns guest's timestamp counter "register"
2039 * guest_tsc = host_tsc + tsc_offset -- 21.3
2041 static u64 guest_read_tsc(void)
2043 u64 host_tsc, tsc_offset;
2046 tsc_offset = vmcs_read64(TSC_OFFSET);
2047 return host_tsc + tsc_offset;
2051 * Like guest_read_tsc, but always returns L1's notion of the timestamp
2052 * counter, even if a nested guest (L2) is currently running.
2054 u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2058 tsc_offset = is_guest_mode(vcpu) ?
2059 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2060 vmcs_read64(TSC_OFFSET);
2061 return host_tsc + tsc_offset;
2065 * Engage any workarounds for mis-matched TSC rates. Currently limited to
2066 * software catchup for faster rates on slower CPUs.
2068 static void vmx_set_tsc_khz(struct kvm_vcpu *vcpu, u32 user_tsc_khz, bool scale)
2073 if (user_tsc_khz > tsc_khz) {
2074 vcpu->arch.tsc_catchup = 1;
2075 vcpu->arch.tsc_always_catchup = 1;
2077 WARN(1, "user requested TSC rate below hardware speed\n");
2080 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2082 return vmcs_read64(TSC_OFFSET);
2086 * writes 'offset' into guest's timestamp counter offset register
2088 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2090 if (is_guest_mode(vcpu)) {
2092 * We're here if L1 chose not to trap WRMSR to TSC. According
2093 * to the spec, this should set L1's TSC; The offset that L1
2094 * set for L2 remains unchanged, and still needs to be added
2095 * to the newly set TSC to get L2's TSC.
2097 struct vmcs12 *vmcs12;
2098 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2099 /* recalculate vmcs02.TSC_OFFSET: */
2100 vmcs12 = get_vmcs12(vcpu);
2101 vmcs_write64(TSC_OFFSET, offset +
2102 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2103 vmcs12->tsc_offset : 0));
2105 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2106 vmcs_read64(TSC_OFFSET), offset);
2107 vmcs_write64(TSC_OFFSET, offset);
2111 static void vmx_adjust_tsc_offset(struct kvm_vcpu *vcpu, s64 adjustment, bool host)
2113 u64 offset = vmcs_read64(TSC_OFFSET);
2115 vmcs_write64(TSC_OFFSET, offset + adjustment);
2116 if (is_guest_mode(vcpu)) {
2117 /* Even when running L2, the adjustment needs to apply to L1 */
2118 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2120 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2121 offset + adjustment);
2124 static u64 vmx_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
2126 return target_tsc - native_read_tsc();
2129 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2131 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2132 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2136 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2137 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2138 * all guests if the "nested" module option is off, and can also be disabled
2139 * for a single guest by disabling its VMX cpuid bit.
2141 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2143 return nested && guest_cpuid_has_vmx(vcpu);
2147 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2148 * returned for the various VMX controls MSRs when nested VMX is enabled.
2149 * The same values should also be used to verify that vmcs12 control fields are
2150 * valid during nested entry from L1 to L2.
2151 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2152 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2153 * bit in the high half is on if the corresponding bit in the control field
2154 * may be on. See also vmx_control_verify().
2155 * TODO: allow these variables to be modified (downgraded) by module options
2158 static u32 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high;
2159 static u32 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high;
2160 static u32 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high;
2161 static u32 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high;
2162 static u32 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high;
2163 static u32 nested_vmx_misc_low, nested_vmx_misc_high;
2164 static __init void nested_vmx_setup_ctls_msrs(void)
2167 * Note that as a general rule, the high half of the MSRs (bits in
2168 * the control fields which may be 1) should be initialized by the
2169 * intersection of the underlying hardware's MSR (i.e., features which
2170 * can be supported) and the list of features we want to expose -
2171 * because they are known to be properly supported in our code.
2172 * Also, usually, the low half of the MSRs (bits which must be 1) can
2173 * be set to 0, meaning that L1 may turn off any of these bits. The
2174 * reason is that if one of these bits is necessary, it will appear
2175 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2176 * fields of vmcs01 and vmcs02, will turn these bits off - and
2177 * nested_vmx_exit_handled() will not pass related exits to L1.
2178 * These rules have exceptions below.
2181 /* pin-based controls */
2182 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2183 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high);
2185 * According to the Intel spec, if bit 55 of VMX_BASIC is off (as it is
2186 * in our case), bits 1, 2 and 4 (i.e., 0x16) must be 1 in this MSR.
2188 nested_vmx_pinbased_ctls_low |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2189 nested_vmx_pinbased_ctls_high &= PIN_BASED_EXT_INTR_MASK |
2190 PIN_BASED_NMI_EXITING | PIN_BASED_VIRTUAL_NMIS |
2191 PIN_BASED_VMX_PREEMPTION_TIMER;
2192 nested_vmx_pinbased_ctls_high |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2196 * If bit 55 of VMX_BASIC is off, bits 0-8 and 10, 11, 13, 14, 16 and
2199 nested_vmx_exit_ctls_low = VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2200 /* Note that guest use of VM_EXIT_ACK_INTR_ON_EXIT is not supported. */
2201 #ifdef CONFIG_X86_64
2202 nested_vmx_exit_ctls_high = VM_EXIT_HOST_ADDR_SPACE_SIZE;
2204 nested_vmx_exit_ctls_high = 0;
2206 nested_vmx_exit_ctls_high |= (VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2207 VM_EXIT_LOAD_IA32_EFER);
2209 /* entry controls */
2210 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2211 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high);
2212 /* If bit 55 of VMX_BASIC is off, bits 0-8 and 12 must be 1. */
2213 nested_vmx_entry_ctls_low = VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2214 nested_vmx_entry_ctls_high &=
2215 VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_IA32E_MODE;
2216 nested_vmx_entry_ctls_high |= (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR |
2217 VM_ENTRY_LOAD_IA32_EFER);
2218 /* cpu-based controls */
2219 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2220 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high);
2221 nested_vmx_procbased_ctls_low = 0;
2222 nested_vmx_procbased_ctls_high &=
2223 CPU_BASED_VIRTUAL_INTR_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2224 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2225 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2226 CPU_BASED_CR3_STORE_EXITING |
2227 #ifdef CONFIG_X86_64
2228 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2230 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2231 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_EXITING |
2232 CPU_BASED_RDPMC_EXITING | CPU_BASED_RDTSC_EXITING |
2233 CPU_BASED_PAUSE_EXITING |
2234 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2236 * We can allow some features even when not supported by the
2237 * hardware. For example, L1 can specify an MSR bitmap - and we
2238 * can use it to avoid exits to L1 - even when L0 runs L2
2239 * without MSR bitmaps.
2241 nested_vmx_procbased_ctls_high |= CPU_BASED_USE_MSR_BITMAPS;
2243 /* secondary cpu-based controls */
2244 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2245 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high);
2246 nested_vmx_secondary_ctls_low = 0;
2247 nested_vmx_secondary_ctls_high &=
2248 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2249 SECONDARY_EXEC_WBINVD_EXITING;
2251 /* miscellaneous data */
2252 rdmsr(MSR_IA32_VMX_MISC, nested_vmx_misc_low, nested_vmx_misc_high);
2253 nested_vmx_misc_low &= VMX_MISC_PREEMPTION_TIMER_RATE_MASK |
2254 VMX_MISC_SAVE_EFER_LMA;
2255 nested_vmx_misc_high = 0;
2258 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2261 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2263 return ((control & high) | low) == control;
2266 static inline u64 vmx_control_msr(u32 low, u32 high)
2268 return low | ((u64)high << 32);
2272 * If we allow our guest to use VMX instructions (i.e., nested VMX), we should
2273 * also let it use VMX-specific MSRs.
2274 * vmx_get_vmx_msr() and vmx_set_vmx_msr() return 1 when we handled a
2275 * VMX-specific MSR, or 0 when we haven't (and the caller should handle it
2276 * like all other MSRs).
2278 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2280 if (!nested_vmx_allowed(vcpu) && msr_index >= MSR_IA32_VMX_BASIC &&
2281 msr_index <= MSR_IA32_VMX_TRUE_ENTRY_CTLS) {
2283 * According to the spec, processors which do not support VMX
2284 * should throw a #GP(0) when VMX capability MSRs are read.
2286 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
2290 switch (msr_index) {
2291 case MSR_IA32_FEATURE_CONTROL:
2292 if (nested_vmx_allowed(vcpu)) {
2293 *pdata = to_vmx(vcpu)->nested.msr_ia32_feature_control;
2297 case MSR_IA32_VMX_BASIC:
2299 * This MSR reports some information about VMX support. We
2300 * should return information about the VMX we emulate for the
2301 * guest, and the VMCS structure we give it - not about the
2302 * VMX support of the underlying hardware.
2304 *pdata = VMCS12_REVISION |
2305 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2306 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2308 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2309 case MSR_IA32_VMX_PINBASED_CTLS:
2310 *pdata = vmx_control_msr(nested_vmx_pinbased_ctls_low,
2311 nested_vmx_pinbased_ctls_high);
2313 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2314 case MSR_IA32_VMX_PROCBASED_CTLS:
2315 *pdata = vmx_control_msr(nested_vmx_procbased_ctls_low,
2316 nested_vmx_procbased_ctls_high);
2318 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2319 case MSR_IA32_VMX_EXIT_CTLS:
2320 *pdata = vmx_control_msr(nested_vmx_exit_ctls_low,
2321 nested_vmx_exit_ctls_high);
2323 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2324 case MSR_IA32_VMX_ENTRY_CTLS:
2325 *pdata = vmx_control_msr(nested_vmx_entry_ctls_low,
2326 nested_vmx_entry_ctls_high);
2328 case MSR_IA32_VMX_MISC:
2329 *pdata = vmx_control_msr(nested_vmx_misc_low,
2330 nested_vmx_misc_high);
2333 * These MSRs specify bits which the guest must keep fixed (on or off)
2334 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2335 * We picked the standard core2 setting.
2337 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2338 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2339 case MSR_IA32_VMX_CR0_FIXED0:
2340 *pdata = VMXON_CR0_ALWAYSON;
2342 case MSR_IA32_VMX_CR0_FIXED1:
2345 case MSR_IA32_VMX_CR4_FIXED0:
2346 *pdata = VMXON_CR4_ALWAYSON;
2348 case MSR_IA32_VMX_CR4_FIXED1:
2351 case MSR_IA32_VMX_VMCS_ENUM:
2354 case MSR_IA32_VMX_PROCBASED_CTLS2:
2355 *pdata = vmx_control_msr(nested_vmx_secondary_ctls_low,
2356 nested_vmx_secondary_ctls_high);
2358 case MSR_IA32_VMX_EPT_VPID_CAP:
2359 /* Currently, no nested ept or nested vpid */
2369 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2371 u32 msr_index = msr_info->index;
2372 u64 data = msr_info->data;
2373 bool host_initialized = msr_info->host_initiated;
2375 if (!nested_vmx_allowed(vcpu))
2378 if (msr_index == MSR_IA32_FEATURE_CONTROL) {
2379 if (!host_initialized &&
2380 to_vmx(vcpu)->nested.msr_ia32_feature_control
2381 & FEATURE_CONTROL_LOCKED)
2383 to_vmx(vcpu)->nested.msr_ia32_feature_control = data;
2388 * No need to treat VMX capability MSRs specially: If we don't handle
2389 * them, handle_wrmsr will #GP(0), which is correct (they are readonly)
2395 * Reads an msr value (of 'msr_index') into 'pdata'.
2396 * Returns 0 on success, non-0 otherwise.
2397 * Assumes vcpu_load() was already called.
2399 static int vmx_get_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2402 struct shared_msr_entry *msr;
2405 printk(KERN_ERR "BUG: get_msr called with NULL pdata\n");
2409 switch (msr_index) {
2410 #ifdef CONFIG_X86_64
2412 data = vmcs_readl(GUEST_FS_BASE);
2415 data = vmcs_readl(GUEST_GS_BASE);
2417 case MSR_KERNEL_GS_BASE:
2418 vmx_load_host_state(to_vmx(vcpu));
2419 data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2423 return kvm_get_msr_common(vcpu, msr_index, pdata);
2425 data = guest_read_tsc();
2427 case MSR_IA32_SYSENTER_CS:
2428 data = vmcs_read32(GUEST_SYSENTER_CS);
2430 case MSR_IA32_SYSENTER_EIP:
2431 data = vmcs_readl(GUEST_SYSENTER_EIP);
2433 case MSR_IA32_SYSENTER_ESP:
2434 data = vmcs_readl(GUEST_SYSENTER_ESP);
2437 if (!to_vmx(vcpu)->rdtscp_enabled)
2439 /* Otherwise falls through */
2441 if (vmx_get_vmx_msr(vcpu, msr_index, pdata))
2443 msr = find_msr_entry(to_vmx(vcpu), msr_index);
2448 return kvm_get_msr_common(vcpu, msr_index, pdata);
2456 * Writes msr value into into the appropriate "register".
2457 * Returns 0 on success, non-0 otherwise.
2458 * Assumes vcpu_load() was already called.
2460 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2462 struct vcpu_vmx *vmx = to_vmx(vcpu);
2463 struct shared_msr_entry *msr;
2465 u32 msr_index = msr_info->index;
2466 u64 data = msr_info->data;
2468 switch (msr_index) {
2470 ret = kvm_set_msr_common(vcpu, msr_info);
2472 #ifdef CONFIG_X86_64
2474 vmx_segment_cache_clear(vmx);
2475 vmcs_writel(GUEST_FS_BASE, data);
2478 vmx_segment_cache_clear(vmx);
2479 vmcs_writel(GUEST_GS_BASE, data);
2481 case MSR_KERNEL_GS_BASE:
2482 vmx_load_host_state(vmx);
2483 vmx->msr_guest_kernel_gs_base = data;
2486 case MSR_IA32_SYSENTER_CS:
2487 vmcs_write32(GUEST_SYSENTER_CS, data);
2489 case MSR_IA32_SYSENTER_EIP:
2490 vmcs_writel(GUEST_SYSENTER_EIP, data);
2492 case MSR_IA32_SYSENTER_ESP:
2493 vmcs_writel(GUEST_SYSENTER_ESP, data);
2496 kvm_write_tsc(vcpu, msr_info);
2498 case MSR_IA32_CR_PAT:
2499 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2500 vmcs_write64(GUEST_IA32_PAT, data);
2501 vcpu->arch.pat = data;
2504 ret = kvm_set_msr_common(vcpu, msr_info);
2506 case MSR_IA32_TSC_ADJUST:
2507 ret = kvm_set_msr_common(vcpu, msr_info);
2510 if (!vmx->rdtscp_enabled)
2512 /* Check reserved bit, higher 32 bits should be zero */
2513 if ((data >> 32) != 0)
2515 /* Otherwise falls through */
2517 if (vmx_set_vmx_msr(vcpu, msr_info))
2519 msr = find_msr_entry(vmx, msr_index);
2522 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2524 kvm_set_shared_msr(msr->index, msr->data,
2530 ret = kvm_set_msr_common(vcpu, msr_info);
2536 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2538 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2541 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2544 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2546 case VCPU_EXREG_PDPTR:
2548 ept_save_pdptrs(vcpu);
2555 static __init int cpu_has_kvm_support(void)
2557 return cpu_has_vmx();
2560 static __init int vmx_disabled_by_bios(void)
2564 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
2565 if (msr & FEATURE_CONTROL_LOCKED) {
2566 /* launched w/ TXT and VMX disabled */
2567 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2570 /* launched w/o TXT and VMX only enabled w/ TXT */
2571 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2572 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
2573 && !tboot_enabled()) {
2574 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
2575 "activate TXT before enabling KVM\n");
2578 /* launched w/o TXT and VMX disabled */
2579 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
2580 && !tboot_enabled())
2587 static void kvm_cpu_vmxon(u64 addr)
2589 asm volatile (ASM_VMX_VMXON_RAX
2590 : : "a"(&addr), "m"(addr)
2594 static int hardware_enable(void *garbage)
2596 int cpu = raw_smp_processor_id();
2597 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2600 if (read_cr4() & X86_CR4_VMXE)
2603 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
2606 * Now we can enable the vmclear operation in kdump
2607 * since the loaded_vmcss_on_cpu list on this cpu
2608 * has been initialized.
2610 * Though the cpu is not in VMX operation now, there
2611 * is no problem to enable the vmclear operation
2612 * for the loaded_vmcss_on_cpu list is empty!
2614 crash_enable_local_vmclear(cpu);
2616 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
2618 test_bits = FEATURE_CONTROL_LOCKED;
2619 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
2620 if (tboot_enabled())
2621 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
2623 if ((old & test_bits) != test_bits) {
2624 /* enable and lock */
2625 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
2627 write_cr4(read_cr4() | X86_CR4_VMXE); /* FIXME: not cpu hotplug safe */
2629 if (vmm_exclusive) {
2630 kvm_cpu_vmxon(phys_addr);
2634 native_store_gdt(&__get_cpu_var(host_gdt));
2639 static void vmclear_local_loaded_vmcss(void)
2641 int cpu = raw_smp_processor_id();
2642 struct loaded_vmcs *v, *n;
2644 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
2645 loaded_vmcss_on_cpu_link)
2646 __loaded_vmcs_clear(v);
2650 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
2653 static void kvm_cpu_vmxoff(void)
2655 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
2658 static void hardware_disable(void *garbage)
2660 if (vmm_exclusive) {
2661 vmclear_local_loaded_vmcss();
2664 write_cr4(read_cr4() & ~X86_CR4_VMXE);
2667 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
2668 u32 msr, u32 *result)
2670 u32 vmx_msr_low, vmx_msr_high;
2671 u32 ctl = ctl_min | ctl_opt;
2673 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2675 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
2676 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
2678 /* Ensure minimum (required) set of control bits are supported. */
2686 static __init bool allow_1_setting(u32 msr, u32 ctl)
2688 u32 vmx_msr_low, vmx_msr_high;
2690 rdmsr(msr, vmx_msr_low, vmx_msr_high);
2691 return vmx_msr_high & ctl;
2694 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
2696 u32 vmx_msr_low, vmx_msr_high;
2697 u32 min, opt, min2, opt2;
2698 u32 _pin_based_exec_control = 0;
2699 u32 _cpu_based_exec_control = 0;
2700 u32 _cpu_based_2nd_exec_control = 0;
2701 u32 _vmexit_control = 0;
2702 u32 _vmentry_control = 0;
2704 min = CPU_BASED_HLT_EXITING |
2705 #ifdef CONFIG_X86_64
2706 CPU_BASED_CR8_LOAD_EXITING |
2707 CPU_BASED_CR8_STORE_EXITING |
2709 CPU_BASED_CR3_LOAD_EXITING |
2710 CPU_BASED_CR3_STORE_EXITING |
2711 CPU_BASED_USE_IO_BITMAPS |
2712 CPU_BASED_MOV_DR_EXITING |
2713 CPU_BASED_USE_TSC_OFFSETING |
2714 CPU_BASED_MWAIT_EXITING |
2715 CPU_BASED_MONITOR_EXITING |
2716 CPU_BASED_INVLPG_EXITING |
2717 CPU_BASED_RDPMC_EXITING;
2719 opt = CPU_BASED_TPR_SHADOW |
2720 CPU_BASED_USE_MSR_BITMAPS |
2721 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2722 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
2723 &_cpu_based_exec_control) < 0)
2725 #ifdef CONFIG_X86_64
2726 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2727 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
2728 ~CPU_BASED_CR8_STORE_EXITING;
2730 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
2732 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2733 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2734 SECONDARY_EXEC_WBINVD_EXITING |
2735 SECONDARY_EXEC_ENABLE_VPID |
2736 SECONDARY_EXEC_ENABLE_EPT |
2737 SECONDARY_EXEC_UNRESTRICTED_GUEST |
2738 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
2739 SECONDARY_EXEC_RDTSCP |
2740 SECONDARY_EXEC_ENABLE_INVPCID |
2741 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2742 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2743 SECONDARY_EXEC_SHADOW_VMCS;
2744 if (adjust_vmx_controls(min2, opt2,
2745 MSR_IA32_VMX_PROCBASED_CTLS2,
2746 &_cpu_based_2nd_exec_control) < 0)
2749 #ifndef CONFIG_X86_64
2750 if (!(_cpu_based_2nd_exec_control &
2751 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
2752 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
2755 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
2756 _cpu_based_2nd_exec_control &= ~(
2757 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2758 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2759 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
2761 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
2762 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
2764 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
2765 CPU_BASED_CR3_STORE_EXITING |
2766 CPU_BASED_INVLPG_EXITING);
2767 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
2768 vmx_capability.ept, vmx_capability.vpid);
2772 #ifdef CONFIG_X86_64
2773 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
2775 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
2776 VM_EXIT_ACK_INTR_ON_EXIT;
2777 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
2778 &_vmexit_control) < 0)
2781 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
2782 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR;
2783 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
2784 &_pin_based_exec_control) < 0)
2787 if (!(_cpu_based_2nd_exec_control &
2788 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) ||
2789 !(_vmexit_control & VM_EXIT_ACK_INTR_ON_EXIT))
2790 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
2793 opt = VM_ENTRY_LOAD_IA32_PAT;
2794 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
2795 &_vmentry_control) < 0)
2798 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
2800 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
2801 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
2804 #ifdef CONFIG_X86_64
2805 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
2806 if (vmx_msr_high & (1u<<16))
2810 /* Require Write-Back (WB) memory type for VMCS accesses. */
2811 if (((vmx_msr_high >> 18) & 15) != 6)
2814 vmcs_conf->size = vmx_msr_high & 0x1fff;
2815 vmcs_conf->order = get_order(vmcs_config.size);
2816 vmcs_conf->revision_id = vmx_msr_low;
2818 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
2819 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
2820 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
2821 vmcs_conf->vmexit_ctrl = _vmexit_control;
2822 vmcs_conf->vmentry_ctrl = _vmentry_control;
2824 cpu_has_load_ia32_efer =
2825 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2826 VM_ENTRY_LOAD_IA32_EFER)
2827 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2828 VM_EXIT_LOAD_IA32_EFER);
2830 cpu_has_load_perf_global_ctrl =
2831 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
2832 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
2833 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
2834 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
2837 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
2838 * but due to arrata below it can't be used. Workaround is to use
2839 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
2841 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
2846 * BC86,AAY89,BD102 (model 44)
2850 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
2851 switch (boot_cpu_data.x86_model) {
2857 cpu_has_load_perf_global_ctrl = false;
2858 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
2859 "does not work properly. Using workaround\n");
2869 static struct vmcs *alloc_vmcs_cpu(int cpu)
2871 int node = cpu_to_node(cpu);
2875 pages = alloc_pages_exact_node(node, GFP_KERNEL, vmcs_config.order);
2878 vmcs = page_address(pages);
2879 memset(vmcs, 0, vmcs_config.size);
2880 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
2884 static struct vmcs *alloc_vmcs(void)
2886 return alloc_vmcs_cpu(raw_smp_processor_id());
2889 static void free_vmcs(struct vmcs *vmcs)
2891 free_pages((unsigned long)vmcs, vmcs_config.order);
2895 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
2897 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
2899 if (!loaded_vmcs->vmcs)
2901 loaded_vmcs_clear(loaded_vmcs);
2902 free_vmcs(loaded_vmcs->vmcs);
2903 loaded_vmcs->vmcs = NULL;
2906 static void free_kvm_area(void)
2910 for_each_possible_cpu(cpu) {
2911 free_vmcs(per_cpu(vmxarea, cpu));
2912 per_cpu(vmxarea, cpu) = NULL;
2916 static __init int alloc_kvm_area(void)
2920 for_each_possible_cpu(cpu) {
2923 vmcs = alloc_vmcs_cpu(cpu);
2929 per_cpu(vmxarea, cpu) = vmcs;
2934 static __init int hardware_setup(void)
2936 if (setup_vmcs_config(&vmcs_config) < 0)
2939 if (boot_cpu_has(X86_FEATURE_NX))
2940 kvm_enable_efer_bits(EFER_NX);
2942 if (!cpu_has_vmx_vpid())
2944 if (!cpu_has_vmx_shadow_vmcs())
2945 enable_shadow_vmcs = 0;
2947 if (!cpu_has_vmx_ept() ||
2948 !cpu_has_vmx_ept_4levels()) {
2950 enable_unrestricted_guest = 0;
2951 enable_ept_ad_bits = 0;
2954 if (!cpu_has_vmx_ept_ad_bits())
2955 enable_ept_ad_bits = 0;
2957 if (!cpu_has_vmx_unrestricted_guest())
2958 enable_unrestricted_guest = 0;
2960 if (!cpu_has_vmx_flexpriority())
2961 flexpriority_enabled = 0;
2963 if (!cpu_has_vmx_tpr_shadow())
2964 kvm_x86_ops->update_cr8_intercept = NULL;
2966 if (enable_ept && !cpu_has_vmx_ept_2m_page())
2967 kvm_disable_largepages();
2969 if (!cpu_has_vmx_ple())
2972 if (!cpu_has_vmx_apicv())
2976 kvm_x86_ops->update_cr8_intercept = NULL;
2978 kvm_x86_ops->hwapic_irr_update = NULL;
2979 kvm_x86_ops->deliver_posted_interrupt = NULL;
2980 kvm_x86_ops->sync_pir_to_irr = vmx_sync_pir_to_irr_dummy;
2984 nested_vmx_setup_ctls_msrs();
2986 return alloc_kvm_area();
2989 static __exit void hardware_unsetup(void)
2994 static bool emulation_required(struct kvm_vcpu *vcpu)
2996 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
2999 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3000 struct kvm_segment *save)
3002 if (!emulate_invalid_guest_state) {
3004 * CS and SS RPL should be equal during guest entry according
3005 * to VMX spec, but in reality it is not always so. Since vcpu
3006 * is in the middle of the transition from real mode to
3007 * protected mode it is safe to assume that RPL 0 is a good
3010 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3011 save->selector &= ~SELECTOR_RPL_MASK;
3012 save->dpl = save->selector & SELECTOR_RPL_MASK;
3015 vmx_set_segment(vcpu, save, seg);
3018 static void enter_pmode(struct kvm_vcpu *vcpu)
3020 unsigned long flags;
3021 struct vcpu_vmx *vmx = to_vmx(vcpu);
3024 * Update real mode segment cache. It may be not up-to-date if sement
3025 * register was written while vcpu was in a guest mode.
3027 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3028 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3029 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3030 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3031 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3032 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3034 vmx->rmode.vm86_active = 0;
3036 vmx_segment_cache_clear(vmx);
3038 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3040 flags = vmcs_readl(GUEST_RFLAGS);
3041 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3042 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3043 vmcs_writel(GUEST_RFLAGS, flags);
3045 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3046 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3048 update_exception_bitmap(vcpu);
3050 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3051 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3052 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3053 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3054 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3055 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3057 /* CPL is always 0 when CPU enters protected mode */
3058 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3062 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3064 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3065 struct kvm_segment var = *save;
3068 if (seg == VCPU_SREG_CS)
3071 if (!emulate_invalid_guest_state) {
3072 var.selector = var.base >> 4;
3073 var.base = var.base & 0xffff0;
3083 if (save->base & 0xf)
3084 printk_once(KERN_WARNING "kvm: segment base is not "
3085 "paragraph aligned when entering "
3086 "protected mode (seg=%d)", seg);
3089 vmcs_write16(sf->selector, var.selector);
3090 vmcs_write32(sf->base, var.base);
3091 vmcs_write32(sf->limit, var.limit);
3092 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3095 static void enter_rmode(struct kvm_vcpu *vcpu)
3097 unsigned long flags;
3098 struct vcpu_vmx *vmx = to_vmx(vcpu);
3100 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3101 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3102 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3103 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3104 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3105 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3106 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3108 vmx->rmode.vm86_active = 1;
3111 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3112 * vcpu. Warn the user that an update is overdue.
3114 if (!vcpu->kvm->arch.tss_addr)
3115 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3116 "called before entering vcpu\n");
3118 vmx_segment_cache_clear(vmx);
3120 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3121 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3122 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3124 flags = vmcs_readl(GUEST_RFLAGS);
3125 vmx->rmode.save_rflags = flags;
3127 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3129 vmcs_writel(GUEST_RFLAGS, flags);
3130 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3131 update_exception_bitmap(vcpu);
3133 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3134 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3135 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3136 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3137 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3138 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3140 kvm_mmu_reset_context(vcpu);
3143 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3145 struct vcpu_vmx *vmx = to_vmx(vcpu);
3146 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3152 * Force kernel_gs_base reloading before EFER changes, as control
3153 * of this msr depends on is_long_mode().
3155 vmx_load_host_state(to_vmx(vcpu));
3156 vcpu->arch.efer = efer;
3157 if (efer & EFER_LMA) {
3158 vmcs_write32(VM_ENTRY_CONTROLS,
3159 vmcs_read32(VM_ENTRY_CONTROLS) |
3160 VM_ENTRY_IA32E_MODE);
3163 vmcs_write32(VM_ENTRY_CONTROLS,
3164 vmcs_read32(VM_ENTRY_CONTROLS) &
3165 ~VM_ENTRY_IA32E_MODE);
3167 msr->data = efer & ~EFER_LME;
3172 #ifdef CONFIG_X86_64
3174 static void enter_lmode(struct kvm_vcpu *vcpu)
3178 vmx_segment_cache_clear(to_vmx(vcpu));
3180 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3181 if ((guest_tr_ar & AR_TYPE_MASK) != AR_TYPE_BUSY_64_TSS) {
3182 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3184 vmcs_write32(GUEST_TR_AR_BYTES,
3185 (guest_tr_ar & ~AR_TYPE_MASK)
3186 | AR_TYPE_BUSY_64_TSS);
3188 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3191 static void exit_lmode(struct kvm_vcpu *vcpu)
3193 vmcs_write32(VM_ENTRY_CONTROLS,
3194 vmcs_read32(VM_ENTRY_CONTROLS)
3195 & ~VM_ENTRY_IA32E_MODE);
3196 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3201 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3203 vpid_sync_context(to_vmx(vcpu));
3205 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3207 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3211 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3213 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3215 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3216 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3219 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3221 if (enable_ept && is_paging(vcpu))
3222 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3223 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3226 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3228 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3230 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3231 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3234 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3236 if (!test_bit(VCPU_EXREG_PDPTR,
3237 (unsigned long *)&vcpu->arch.regs_dirty))
3240 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3241 vmcs_write64(GUEST_PDPTR0, vcpu->arch.mmu.pdptrs[0]);
3242 vmcs_write64(GUEST_PDPTR1, vcpu->arch.mmu.pdptrs[1]);
3243 vmcs_write64(GUEST_PDPTR2, vcpu->arch.mmu.pdptrs[2]);
3244 vmcs_write64(GUEST_PDPTR3, vcpu->arch.mmu.pdptrs[3]);
3248 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3250 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3251 vcpu->arch.mmu.pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3252 vcpu->arch.mmu.pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3253 vcpu->arch.mmu.pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3254 vcpu->arch.mmu.pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3257 __set_bit(VCPU_EXREG_PDPTR,
3258 (unsigned long *)&vcpu->arch.regs_avail);
3259 __set_bit(VCPU_EXREG_PDPTR,
3260 (unsigned long *)&vcpu->arch.regs_dirty);
3263 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3265 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3267 struct kvm_vcpu *vcpu)
3269 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3270 vmx_decache_cr3(vcpu);
3271 if (!(cr0 & X86_CR0_PG)) {
3272 /* From paging/starting to nonpaging */
3273 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3274 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3275 (CPU_BASED_CR3_LOAD_EXITING |
3276 CPU_BASED_CR3_STORE_EXITING));
3277 vcpu->arch.cr0 = cr0;
3278 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3279 } else if (!is_paging(vcpu)) {
3280 /* From nonpaging to paging */
3281 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3282 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3283 ~(CPU_BASED_CR3_LOAD_EXITING |
3284 CPU_BASED_CR3_STORE_EXITING));
3285 vcpu->arch.cr0 = cr0;
3286 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3289 if (!(cr0 & X86_CR0_WP))
3290 *hw_cr0 &= ~X86_CR0_WP;
3293 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3295 struct vcpu_vmx *vmx = to_vmx(vcpu);
3296 unsigned long hw_cr0;
3298 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3299 if (enable_unrestricted_guest)
3300 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3302 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3304 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3307 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3311 #ifdef CONFIG_X86_64
3312 if (vcpu->arch.efer & EFER_LME) {
3313 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3315 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3321 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3323 if (!vcpu->fpu_active)
3324 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3326 vmcs_writel(CR0_READ_SHADOW, cr0);
3327 vmcs_writel(GUEST_CR0, hw_cr0);
3328 vcpu->arch.cr0 = cr0;
3330 /* depends on vcpu->arch.cr0 to be set to a new value */
3331 vmx->emulation_required = emulation_required(vcpu);
3334 static u64 construct_eptp(unsigned long root_hpa)
3338 /* TODO write the value reading from MSR */
3339 eptp = VMX_EPT_DEFAULT_MT |
3340 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3341 if (enable_ept_ad_bits)
3342 eptp |= VMX_EPT_AD_ENABLE_BIT;
3343 eptp |= (root_hpa & PAGE_MASK);
3348 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3350 unsigned long guest_cr3;
3355 eptp = construct_eptp(cr3);
3356 vmcs_write64(EPT_POINTER, eptp);
3357 guest_cr3 = is_paging(vcpu) ? kvm_read_cr3(vcpu) :
3358 vcpu->kvm->arch.ept_identity_map_addr;
3359 ept_load_pdptrs(vcpu);
3362 vmx_flush_tlb(vcpu);
3363 vmcs_writel(GUEST_CR3, guest_cr3);
3366 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3368 unsigned long hw_cr4 = cr4 | (to_vmx(vcpu)->rmode.vm86_active ?
3369 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3371 if (cr4 & X86_CR4_VMXE) {
3373 * To use VMXON (and later other VMX instructions), a guest
3374 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3375 * So basically the check on whether to allow nested VMX
3378 if (!nested_vmx_allowed(vcpu))
3381 if (to_vmx(vcpu)->nested.vmxon &&
3382 ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3385 vcpu->arch.cr4 = cr4;
3387 if (!is_paging(vcpu)) {
3388 hw_cr4 &= ~X86_CR4_PAE;
3389 hw_cr4 |= X86_CR4_PSE;
3391 * SMEP is disabled if CPU is in non-paging mode in
3392 * hardware. However KVM always uses paging mode to
3393 * emulate guest non-paging mode with TDP.
3394 * To emulate this behavior, SMEP needs to be manually
3395 * disabled when guest switches to non-paging mode.
3397 hw_cr4 &= ~X86_CR4_SMEP;
3398 } else if (!(cr4 & X86_CR4_PAE)) {
3399 hw_cr4 &= ~X86_CR4_PAE;
3403 vmcs_writel(CR4_READ_SHADOW, cr4);
3404 vmcs_writel(GUEST_CR4, hw_cr4);
3408 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3409 struct kvm_segment *var, int seg)
3411 struct vcpu_vmx *vmx = to_vmx(vcpu);
3414 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3415 *var = vmx->rmode.segs[seg];
3416 if (seg == VCPU_SREG_TR
3417 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3419 var->base = vmx_read_guest_seg_base(vmx, seg);
3420 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3423 var->base = vmx_read_guest_seg_base(vmx, seg);
3424 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3425 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3426 ar = vmx_read_guest_seg_ar(vmx, seg);
3427 var->unusable = (ar >> 16) & 1;
3428 var->type = ar & 15;
3429 var->s = (ar >> 4) & 1;
3430 var->dpl = (ar >> 5) & 3;
3432 * Some userspaces do not preserve unusable property. Since usable
3433 * segment has to be present according to VMX spec we can use present
3434 * property to amend userspace bug by making unusable segment always
3435 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3436 * segment as unusable.
3438 var->present = !var->unusable;
3439 var->avl = (ar >> 12) & 1;
3440 var->l = (ar >> 13) & 1;
3441 var->db = (ar >> 14) & 1;
3442 var->g = (ar >> 15) & 1;
3445 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3447 struct kvm_segment s;
3449 if (to_vmx(vcpu)->rmode.vm86_active) {
3450 vmx_get_segment(vcpu, &s, seg);
3453 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3456 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3458 struct vcpu_vmx *vmx = to_vmx(vcpu);
3460 if (!is_protmode(vcpu))
3463 if (!is_long_mode(vcpu)
3464 && (kvm_get_rflags(vcpu) & X86_EFLAGS_VM)) /* if virtual 8086 */
3467 if (!test_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail)) {
3468 __set_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3469 vmx->cpl = vmx_read_guest_seg_selector(vmx, VCPU_SREG_CS) & 3;
3476 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3480 if (var->unusable || !var->present)
3483 ar = var->type & 15;
3484 ar |= (var->s & 1) << 4;
3485 ar |= (var->dpl & 3) << 5;
3486 ar |= (var->present & 1) << 7;
3487 ar |= (var->avl & 1) << 12;
3488 ar |= (var->l & 1) << 13;
3489 ar |= (var->db & 1) << 14;
3490 ar |= (var->g & 1) << 15;
3496 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3497 struct kvm_segment *var, int seg)
3499 struct vcpu_vmx *vmx = to_vmx(vcpu);
3500 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3502 vmx_segment_cache_clear(vmx);
3503 if (seg == VCPU_SREG_CS)
3504 __clear_bit(VCPU_EXREG_CPL, (ulong *)&vcpu->arch.regs_avail);
3506 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3507 vmx->rmode.segs[seg] = *var;
3508 if (seg == VCPU_SREG_TR)
3509 vmcs_write16(sf->selector, var->selector);
3511 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3515 vmcs_writel(sf->base, var->base);
3516 vmcs_write32(sf->limit, var->limit);
3517 vmcs_write16(sf->selector, var->selector);
3520 * Fix the "Accessed" bit in AR field of segment registers for older
3522 * IA32 arch specifies that at the time of processor reset the
3523 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3524 * is setting it to 0 in the userland code. This causes invalid guest
3525 * state vmexit when "unrestricted guest" mode is turned on.
3526 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3527 * tree. Newer qemu binaries with that qemu fix would not need this
3530 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3531 var->type |= 0x1; /* Accessed */
3533 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3536 vmx->emulation_required |= emulation_required(vcpu);
3539 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3541 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3543 *db = (ar >> 14) & 1;
3544 *l = (ar >> 13) & 1;
3547 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3549 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3550 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3553 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3555 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3556 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3559 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3561 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3562 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3565 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3567 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3568 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3571 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3573 struct kvm_segment var;
3576 vmx_get_segment(vcpu, &var, seg);
3578 if (seg == VCPU_SREG_CS)
3580 ar = vmx_segment_access_rights(&var);
3582 if (var.base != (var.selector << 4))
3584 if (var.limit != 0xffff)
3592 static bool code_segment_valid(struct kvm_vcpu *vcpu)
3594 struct kvm_segment cs;
3595 unsigned int cs_rpl;
3597 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3598 cs_rpl = cs.selector & SELECTOR_RPL_MASK;
3602 if (~cs.type & (AR_TYPE_CODE_MASK|AR_TYPE_ACCESSES_MASK))
3606 if (cs.type & AR_TYPE_WRITEABLE_MASK) {
3607 if (cs.dpl > cs_rpl)
3610 if (cs.dpl != cs_rpl)
3616 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
3620 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
3622 struct kvm_segment ss;
3623 unsigned int ss_rpl;
3625 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3626 ss_rpl = ss.selector & SELECTOR_RPL_MASK;
3630 if (ss.type != 3 && ss.type != 7)
3634 if (ss.dpl != ss_rpl) /* DPL != RPL */
3642 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
3644 struct kvm_segment var;
3647 vmx_get_segment(vcpu, &var, seg);
3648 rpl = var.selector & SELECTOR_RPL_MASK;
3656 if (~var.type & (AR_TYPE_CODE_MASK|AR_TYPE_WRITEABLE_MASK)) {
3657 if (var.dpl < rpl) /* DPL < RPL */
3661 /* TODO: Add other members to kvm_segment_field to allow checking for other access
3667 static bool tr_valid(struct kvm_vcpu *vcpu)
3669 struct kvm_segment tr;
3671 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
3675 if (tr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3677 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
3685 static bool ldtr_valid(struct kvm_vcpu *vcpu)
3687 struct kvm_segment ldtr;
3689 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
3693 if (ldtr.selector & SELECTOR_TI_MASK) /* TI = 1 */
3703 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
3705 struct kvm_segment cs, ss;
3707 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
3708 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
3710 return ((cs.selector & SELECTOR_RPL_MASK) ==
3711 (ss.selector & SELECTOR_RPL_MASK));
3715 * Check if guest state is valid. Returns true if valid, false if
3717 * We assume that registers are always usable
3719 static bool guest_state_valid(struct kvm_vcpu *vcpu)
3721 if (enable_unrestricted_guest)
3724 /* real mode guest state checks */
3725 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
3726 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
3728 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
3730 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
3732 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
3734 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
3736 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
3739 /* protected mode guest state checks */
3740 if (!cs_ss_rpl_check(vcpu))
3742 if (!code_segment_valid(vcpu))
3744 if (!stack_segment_valid(vcpu))
3746 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
3748 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
3750 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
3752 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
3754 if (!tr_valid(vcpu))
3756 if (!ldtr_valid(vcpu))
3760 * - Add checks on RIP
3761 * - Add checks on RFLAGS
3767 static int init_rmode_tss(struct kvm *kvm)
3771 int r, idx, ret = 0;
3773 idx = srcu_read_lock(&kvm->srcu);
3774 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
3775 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3778 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
3779 r = kvm_write_guest_page(kvm, fn++, &data,
3780 TSS_IOPB_BASE_OFFSET, sizeof(u16));
3783 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
3786 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
3790 r = kvm_write_guest_page(kvm, fn, &data,
3791 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
3798 srcu_read_unlock(&kvm->srcu, idx);
3802 static int init_rmode_identity_map(struct kvm *kvm)
3805 pfn_t identity_map_pfn;
3810 if (unlikely(!kvm->arch.ept_identity_pagetable)) {
3811 printk(KERN_ERR "EPT: identity-mapping pagetable "
3812 "haven't been allocated!\n");
3815 if (likely(kvm->arch.ept_identity_pagetable_done))
3818 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
3819 idx = srcu_read_lock(&kvm->srcu);
3820 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
3823 /* Set up identity-mapping pagetable for EPT in real mode */
3824 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
3825 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
3826 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
3827 r = kvm_write_guest_page(kvm, identity_map_pfn,
3828 &tmp, i * sizeof(tmp), sizeof(tmp));
3832 kvm->arch.ept_identity_pagetable_done = true;
3835 srcu_read_unlock(&kvm->srcu, idx);
3839 static void seg_setup(int seg)
3841 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3844 vmcs_write16(sf->selector, 0);
3845 vmcs_writel(sf->base, 0);
3846 vmcs_write32(sf->limit, 0xffff);
3848 if (seg == VCPU_SREG_CS)
3849 ar |= 0x08; /* code segment */
3851 vmcs_write32(sf->ar_bytes, ar);
3854 static int alloc_apic_access_page(struct kvm *kvm)
3857 struct kvm_userspace_memory_region kvm_userspace_mem;
3860 mutex_lock(&kvm->slots_lock);
3861 if (kvm->arch.apic_access_page)
3863 kvm_userspace_mem.slot = APIC_ACCESS_PAGE_PRIVATE_MEMSLOT;
3864 kvm_userspace_mem.flags = 0;
3865 kvm_userspace_mem.guest_phys_addr = 0xfee00000ULL;
3866 kvm_userspace_mem.memory_size = PAGE_SIZE;
3867 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
3871 page = gfn_to_page(kvm, 0xfee00);
3872 if (is_error_page(page)) {
3877 kvm->arch.apic_access_page = page;
3879 mutex_unlock(&kvm->slots_lock);
3883 static int alloc_identity_pagetable(struct kvm *kvm)
3886 struct kvm_userspace_memory_region kvm_userspace_mem;
3889 mutex_lock(&kvm->slots_lock);
3890 if (kvm->arch.ept_identity_pagetable)
3892 kvm_userspace_mem.slot = IDENTITY_PAGETABLE_PRIVATE_MEMSLOT;
3893 kvm_userspace_mem.flags = 0;
3894 kvm_userspace_mem.guest_phys_addr =
3895 kvm->arch.ept_identity_map_addr;
3896 kvm_userspace_mem.memory_size = PAGE_SIZE;
3897 r = __kvm_set_memory_region(kvm, &kvm_userspace_mem);
3901 page = gfn_to_page(kvm, kvm->arch.ept_identity_map_addr >> PAGE_SHIFT);
3902 if (is_error_page(page)) {
3907 kvm->arch.ept_identity_pagetable = page;
3909 mutex_unlock(&kvm->slots_lock);
3913 static void allocate_vpid(struct vcpu_vmx *vmx)
3920 spin_lock(&vmx_vpid_lock);
3921 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
3922 if (vpid < VMX_NR_VPIDS) {
3924 __set_bit(vpid, vmx_vpid_bitmap);
3926 spin_unlock(&vmx_vpid_lock);
3929 static void free_vpid(struct vcpu_vmx *vmx)
3933 spin_lock(&vmx_vpid_lock);
3935 __clear_bit(vmx->vpid, vmx_vpid_bitmap);
3936 spin_unlock(&vmx_vpid_lock);
3939 #define MSR_TYPE_R 1
3940 #define MSR_TYPE_W 2
3941 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
3944 int f = sizeof(unsigned long);
3946 if (!cpu_has_vmx_msr_bitmap())
3950 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3951 * have the write-low and read-high bitmap offsets the wrong way round.
3952 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3954 if (msr <= 0x1fff) {
3955 if (type & MSR_TYPE_R)
3957 __clear_bit(msr, msr_bitmap + 0x000 / f);
3959 if (type & MSR_TYPE_W)
3961 __clear_bit(msr, msr_bitmap + 0x800 / f);
3963 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
3965 if (type & MSR_TYPE_R)
3967 __clear_bit(msr, msr_bitmap + 0x400 / f);
3969 if (type & MSR_TYPE_W)
3971 __clear_bit(msr, msr_bitmap + 0xc00 / f);
3976 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
3979 int f = sizeof(unsigned long);
3981 if (!cpu_has_vmx_msr_bitmap())
3985 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
3986 * have the write-low and read-high bitmap offsets the wrong way round.
3987 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
3989 if (msr <= 0x1fff) {
3990 if (type & MSR_TYPE_R)
3992 __set_bit(msr, msr_bitmap + 0x000 / f);
3994 if (type & MSR_TYPE_W)
3996 __set_bit(msr, msr_bitmap + 0x800 / f);
3998 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4000 if (type & MSR_TYPE_R)
4002 __set_bit(msr, msr_bitmap + 0x400 / f);
4004 if (type & MSR_TYPE_W)
4006 __set_bit(msr, msr_bitmap + 0xc00 / f);
4011 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4014 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4015 msr, MSR_TYPE_R | MSR_TYPE_W);
4016 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4017 msr, MSR_TYPE_R | MSR_TYPE_W);
4020 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4022 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4024 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4028 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4030 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4032 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4036 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4038 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4040 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4044 static int vmx_vm_has_apicv(struct kvm *kvm)
4046 return enable_apicv && irqchip_in_kernel(kvm);
4050 * Send interrupt to vcpu via posted interrupt way.
4051 * 1. If target vcpu is running(non-root mode), send posted interrupt
4052 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4053 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4054 * interrupt from PIR in next vmentry.
4056 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4058 struct vcpu_vmx *vmx = to_vmx(vcpu);
4061 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4064 r = pi_test_and_set_on(&vmx->pi_desc);
4065 kvm_make_request(KVM_REQ_EVENT, vcpu);
4067 if (!r && (vcpu->mode == IN_GUEST_MODE))
4068 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4069 POSTED_INTR_VECTOR);
4072 kvm_vcpu_kick(vcpu);
4075 static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4077 struct vcpu_vmx *vmx = to_vmx(vcpu);
4079 if (!pi_test_and_clear_on(&vmx->pi_desc))
4082 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4085 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu)
4091 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4092 * will not change in the lifetime of the guest.
4093 * Note that host-state that does change is set elsewhere. E.g., host-state
4094 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4096 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4102 vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */
4103 vmcs_writel(HOST_CR4, read_cr4()); /* 22.2.3, 22.2.5 */
4104 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
4106 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
4107 #ifdef CONFIG_X86_64
4109 * Load null selectors, so we can avoid reloading them in
4110 * __vmx_load_host_state(), in case userspace uses the null selectors
4111 * too (the expected case).
4113 vmcs_write16(HOST_DS_SELECTOR, 0);
4114 vmcs_write16(HOST_ES_SELECTOR, 0);
4116 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4117 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4119 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4120 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
4122 native_store_idt(&dt);
4123 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
4124 vmx->host_idt_base = dt.address;
4126 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4128 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4129 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4130 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4131 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4133 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4134 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4135 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4139 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4141 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4143 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4144 if (is_guest_mode(&vmx->vcpu))
4145 vmx->vcpu.arch.cr4_guest_owned_bits &=
4146 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4147 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4150 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4152 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4154 if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4155 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4156 return pin_based_exec_ctrl;
4159 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4161 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4162 if (!vm_need_tpr_shadow(vmx->vcpu.kvm)) {
4163 exec_control &= ~CPU_BASED_TPR_SHADOW;
4164 #ifdef CONFIG_X86_64
4165 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4166 CPU_BASED_CR8_LOAD_EXITING;
4170 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4171 CPU_BASED_CR3_LOAD_EXITING |
4172 CPU_BASED_INVLPG_EXITING;
4173 return exec_control;
4176 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4178 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4179 if (!vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
4180 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4182 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4184 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4185 enable_unrestricted_guest = 0;
4186 /* Enable INVPCID for non-ept guests may cause performance regression. */
4187 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4189 if (!enable_unrestricted_guest)
4190 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4192 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4193 if (!vmx_vm_has_apicv(vmx->vcpu.kvm))
4194 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4195 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4196 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4197 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4199 We can NOT enable shadow_vmcs here because we don't have yet
4202 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4203 return exec_control;
4206 static void ept_set_mmio_spte_mask(void)
4209 * EPT Misconfigurations can be generated if the value of bits 2:0
4210 * of an EPT paging-structure entry is 110b (write/execute).
4211 * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4214 kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4218 * Sets up the vmcs for emulated real mode.
4220 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4222 #ifdef CONFIG_X86_64
4228 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4229 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4231 if (enable_shadow_vmcs) {
4232 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4233 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4235 if (cpu_has_vmx_msr_bitmap())
4236 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4238 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4241 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4243 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4245 if (cpu_has_secondary_exec_ctrls()) {
4246 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4247 vmx_secondary_exec_control(vmx));
4250 if (vmx_vm_has_apicv(vmx->vcpu.kvm)) {
4251 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4252 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4253 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4254 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4256 vmcs_write16(GUEST_INTR_STATUS, 0);
4258 vmcs_write64(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4259 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4263 vmcs_write32(PLE_GAP, ple_gap);
4264 vmcs_write32(PLE_WINDOW, ple_window);
4267 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4268 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4269 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4271 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4272 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4273 vmx_set_constant_host_state(vmx);
4274 #ifdef CONFIG_X86_64
4275 rdmsrl(MSR_FS_BASE, a);
4276 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
4277 rdmsrl(MSR_GS_BASE, a);
4278 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
4280 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4281 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4284 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4285 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4286 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
4287 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4288 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
4290 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
4291 u32 msr_low, msr_high;
4293 rdmsr(MSR_IA32_CR_PAT, msr_low, msr_high);
4294 host_pat = msr_low | ((u64) msr_high << 32);
4295 /* Write the default value follow host pat */
4296 vmcs_write64(GUEST_IA32_PAT, host_pat);
4297 /* Keep arch.pat sync with GUEST_IA32_PAT */
4298 vmx->vcpu.arch.pat = host_pat;
4301 for (i = 0; i < NR_VMX_MSR; ++i) {
4302 u32 index = vmx_msr_index[i];
4303 u32 data_low, data_high;
4306 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4308 if (wrmsr_safe(index, data_low, data_high) < 0)
4310 vmx->guest_msrs[j].index = i;
4311 vmx->guest_msrs[j].data = 0;
4312 vmx->guest_msrs[j].mask = -1ull;
4316 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
4318 /* 22.2.1, 20.8.1 */
4319 vmcs_write32(VM_ENTRY_CONTROLS, vmcs_config.vmentry_ctrl);
4321 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4322 set_cr4_guest_host_mask(vmx);
4327 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu)
4329 struct vcpu_vmx *vmx = to_vmx(vcpu);
4332 vmx->rmode.vm86_active = 0;
4334 vmx->soft_vnmi_blocked = 0;
4336 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4337 kvm_set_cr8(&vmx->vcpu, 0);
4338 msr = 0xfee00000 | MSR_IA32_APICBASE_ENABLE;
4339 if (kvm_vcpu_is_bsp(&vmx->vcpu))
4340 msr |= MSR_IA32_APICBASE_BSP;
4341 kvm_set_apic_base(&vmx->vcpu, msr);
4343 vmx_segment_cache_clear(vmx);
4345 seg_setup(VCPU_SREG_CS);
4346 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4347 vmcs_write32(GUEST_CS_BASE, 0xffff0000);
4349 seg_setup(VCPU_SREG_DS);
4350 seg_setup(VCPU_SREG_ES);
4351 seg_setup(VCPU_SREG_FS);
4352 seg_setup(VCPU_SREG_GS);
4353 seg_setup(VCPU_SREG_SS);
4355 vmcs_write16(GUEST_TR_SELECTOR, 0);
4356 vmcs_writel(GUEST_TR_BASE, 0);
4357 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4358 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4360 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4361 vmcs_writel(GUEST_LDTR_BASE, 0);
4362 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4363 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4365 vmcs_write32(GUEST_SYSENTER_CS, 0);
4366 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4367 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4369 vmcs_writel(GUEST_RFLAGS, 0x02);
4370 kvm_rip_write(vcpu, 0xfff0);
4372 vmcs_writel(GUEST_GDTR_BASE, 0);
4373 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4375 vmcs_writel(GUEST_IDTR_BASE, 0);
4376 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4378 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4379 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4380 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4382 /* Special registers */
4383 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4387 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4389 if (cpu_has_vmx_tpr_shadow()) {
4390 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4391 if (vm_need_tpr_shadow(vmx->vcpu.kvm))
4392 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4393 __pa(vmx->vcpu.arch.apic->regs));
4394 vmcs_write32(TPR_THRESHOLD, 0);
4397 if (vm_need_virtualize_apic_accesses(vmx->vcpu.kvm))
4398 vmcs_write64(APIC_ACCESS_ADDR,
4399 page_to_phys(vmx->vcpu.kvm->arch.apic_access_page));
4401 if (vmx_vm_has_apicv(vcpu->kvm))
4402 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
4405 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4407 vmx->vcpu.arch.cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4408 vmx_set_cr0(&vmx->vcpu, kvm_read_cr0(vcpu)); /* enter rmode */
4409 vmx_set_cr4(&vmx->vcpu, 0);
4410 vmx_set_efer(&vmx->vcpu, 0);
4411 vmx_fpu_activate(&vmx->vcpu);
4412 update_exception_bitmap(&vmx->vcpu);
4414 vpid_sync_context(vmx);
4418 * In nested virtualization, check if L1 asked to exit on external interrupts.
4419 * For most existing hypervisors, this will always return true.
4421 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4423 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4424 PIN_BASED_EXT_INTR_MASK;
4427 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
4429 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4430 PIN_BASED_NMI_EXITING;
4433 static int enable_irq_window(struct kvm_vcpu *vcpu)
4435 u32 cpu_based_vm_exec_control;
4437 if (is_guest_mode(vcpu) && nested_exit_on_intr(vcpu))
4439 * We get here if vmx_interrupt_allowed() said we can't
4440 * inject to L1 now because L2 must run. The caller will have
4441 * to make L2 exit right after entry, so we can inject to L1
4446 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4447 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
4448 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4452 static int enable_nmi_window(struct kvm_vcpu *vcpu)
4454 u32 cpu_based_vm_exec_control;
4456 if (!cpu_has_virtual_nmis())
4457 return enable_irq_window(vcpu);
4459 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI)
4460 return enable_irq_window(vcpu);
4462 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
4463 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
4464 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
4468 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
4470 struct vcpu_vmx *vmx = to_vmx(vcpu);
4472 int irq = vcpu->arch.interrupt.nr;
4474 trace_kvm_inj_virq(irq);
4476 ++vcpu->stat.irq_injections;
4477 if (vmx->rmode.vm86_active) {
4479 if (vcpu->arch.interrupt.soft)
4480 inc_eip = vcpu->arch.event_exit_inst_len;
4481 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
4482 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4485 intr = irq | INTR_INFO_VALID_MASK;
4486 if (vcpu->arch.interrupt.soft) {
4487 intr |= INTR_TYPE_SOFT_INTR;
4488 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
4489 vmx->vcpu.arch.event_exit_inst_len);
4491 intr |= INTR_TYPE_EXT_INTR;
4492 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
4495 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
4497 struct vcpu_vmx *vmx = to_vmx(vcpu);
4499 if (is_guest_mode(vcpu))
4502 if (!cpu_has_virtual_nmis()) {
4504 * Tracking the NMI-blocked state in software is built upon
4505 * finding the next open IRQ window. This, in turn, depends on
4506 * well-behaving guests: They have to keep IRQs disabled at
4507 * least as long as the NMI handler runs. Otherwise we may
4508 * cause NMI nesting, maybe breaking the guest. But as this is
4509 * highly unlikely, we can live with the residual risk.
4511 vmx->soft_vnmi_blocked = 1;
4512 vmx->vnmi_blocked_time = 0;
4515 ++vcpu->stat.nmi_injections;
4516 vmx->nmi_known_unmasked = false;
4517 if (vmx->rmode.vm86_active) {
4518 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
4519 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
4522 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
4523 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
4526 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
4528 if (!cpu_has_virtual_nmis())
4529 return to_vmx(vcpu)->soft_vnmi_blocked;
4530 if (to_vmx(vcpu)->nmi_known_unmasked)
4532 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
4535 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
4537 struct vcpu_vmx *vmx = to_vmx(vcpu);
4539 if (!cpu_has_virtual_nmis()) {
4540 if (vmx->soft_vnmi_blocked != masked) {
4541 vmx->soft_vnmi_blocked = masked;
4542 vmx->vnmi_blocked_time = 0;
4545 vmx->nmi_known_unmasked = !masked;
4547 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
4548 GUEST_INTR_STATE_NMI);
4550 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
4551 GUEST_INTR_STATE_NMI);
4555 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
4557 if (is_guest_mode(vcpu)) {
4558 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4560 if (to_vmx(vcpu)->nested.nested_run_pending)
4562 if (nested_exit_on_nmi(vcpu)) {
4563 nested_vmx_vmexit(vcpu);
4564 vmcs12->vm_exit_reason = EXIT_REASON_EXCEPTION_NMI;
4565 vmcs12->vm_exit_intr_info = NMI_VECTOR |
4566 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK;
4568 * The NMI-triggered VM exit counts as injection:
4569 * clear this one and block further NMIs.
4571 vcpu->arch.nmi_pending = 0;
4572 vmx_set_nmi_mask(vcpu, true);
4577 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
4580 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4581 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
4582 | GUEST_INTR_STATE_NMI));
4585 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
4587 if (is_guest_mode(vcpu)) {
4588 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4590 if (to_vmx(vcpu)->nested.nested_run_pending)
4592 if (nested_exit_on_intr(vcpu)) {
4593 nested_vmx_vmexit(vcpu);
4594 vmcs12->vm_exit_reason =
4595 EXIT_REASON_EXTERNAL_INTERRUPT;
4596 vmcs12->vm_exit_intr_info = 0;
4598 * fall through to normal code, but now in L1, not L2
4603 return (vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
4604 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
4605 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
4608 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
4611 struct kvm_userspace_memory_region tss_mem = {
4612 .slot = TSS_PRIVATE_MEMSLOT,
4613 .guest_phys_addr = addr,
4614 .memory_size = PAGE_SIZE * 3,
4618 ret = kvm_set_memory_region(kvm, &tss_mem);
4621 kvm->arch.tss_addr = addr;
4622 if (!init_rmode_tss(kvm))
4628 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
4633 * Update instruction length as we may reinject the exception
4634 * from user space while in guest debugging mode.
4636 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
4637 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4638 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
4642 if (vcpu->guest_debug &
4643 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
4660 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
4661 int vec, u32 err_code)
4664 * Instruction with address size override prefix opcode 0x67
4665 * Cause the #SS fault with 0 error code in VM86 mode.
4667 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
4668 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
4669 if (vcpu->arch.halt_request) {
4670 vcpu->arch.halt_request = 0;
4671 return kvm_emulate_halt(vcpu);
4679 * Forward all other exceptions that are valid in real mode.
4680 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
4681 * the required debugging infrastructure rework.
4683 kvm_queue_exception(vcpu, vec);
4688 * Trigger machine check on the host. We assume all the MSRs are already set up
4689 * by the CPU and that we still run on the same CPU as the MCE occurred on.
4690 * We pass a fake environment to the machine check handler because we want
4691 * the guest to be always treated like user space, no matter what context
4692 * it used internally.
4694 static void kvm_machine_check(void)
4696 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
4697 struct pt_regs regs = {
4698 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
4699 .flags = X86_EFLAGS_IF,
4702 do_machine_check(®s, 0);
4706 static int handle_machine_check(struct kvm_vcpu *vcpu)
4708 /* already handled by vcpu_run */
4712 static int handle_exception(struct kvm_vcpu *vcpu)
4714 struct vcpu_vmx *vmx = to_vmx(vcpu);
4715 struct kvm_run *kvm_run = vcpu->run;
4716 u32 intr_info, ex_no, error_code;
4717 unsigned long cr2, rip, dr6;
4719 enum emulation_result er;
4721 vect_info = vmx->idt_vectoring_info;
4722 intr_info = vmx->exit_intr_info;
4724 if (is_machine_check(intr_info))
4725 return handle_machine_check(vcpu);
4727 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
4728 return 1; /* already handled by vmx_vcpu_run() */
4730 if (is_no_device(intr_info)) {
4731 vmx_fpu_activate(vcpu);
4735 if (is_invalid_opcode(intr_info)) {
4736 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
4737 if (er != EMULATE_DONE)
4738 kvm_queue_exception(vcpu, UD_VECTOR);
4743 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
4744 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
4747 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
4748 * MMIO, it is better to report an internal error.
4749 * See the comments in vmx_handle_exit.
4751 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
4752 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
4753 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
4754 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
4755 vcpu->run->internal.ndata = 2;
4756 vcpu->run->internal.data[0] = vect_info;
4757 vcpu->run->internal.data[1] = intr_info;
4761 if (is_page_fault(intr_info)) {
4762 /* EPT won't cause page fault directly */
4764 cr2 = vmcs_readl(EXIT_QUALIFICATION);
4765 trace_kvm_page_fault(cr2, error_code);
4767 if (kvm_event_needs_reinjection(vcpu))
4768 kvm_mmu_unprotect_page_virt(vcpu, cr2);
4769 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
4772 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
4774 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
4775 return handle_rmode_exception(vcpu, ex_no, error_code);
4779 dr6 = vmcs_readl(EXIT_QUALIFICATION);
4780 if (!(vcpu->guest_debug &
4781 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
4782 vcpu->arch.dr6 = dr6 | DR6_FIXED_1;
4783 kvm_queue_exception(vcpu, DB_VECTOR);
4786 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
4787 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
4791 * Update instruction length as we may reinject #BP from
4792 * user space while in guest debugging mode. Reading it for
4793 * #DB as well causes no harm, it is not used in that case.
4795 vmx->vcpu.arch.event_exit_inst_len =
4796 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
4797 kvm_run->exit_reason = KVM_EXIT_DEBUG;
4798 rip = kvm_rip_read(vcpu);
4799 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
4800 kvm_run->debug.arch.exception = ex_no;
4803 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
4804 kvm_run->ex.exception = ex_no;
4805 kvm_run->ex.error_code = error_code;
4811 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
4813 ++vcpu->stat.irq_exits;
4817 static int handle_triple_fault(struct kvm_vcpu *vcpu)
4819 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
4823 static int handle_io(struct kvm_vcpu *vcpu)
4825 unsigned long exit_qualification;
4826 int size, in, string;
4829 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4830 string = (exit_qualification & 16) != 0;
4831 in = (exit_qualification & 8) != 0;
4833 ++vcpu->stat.io_exits;
4836 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
4838 port = exit_qualification >> 16;
4839 size = (exit_qualification & 7) + 1;
4840 skip_emulated_instruction(vcpu);
4842 return kvm_fast_pio_out(vcpu, size, port);
4846 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
4849 * Patch in the VMCALL instruction:
4851 hypercall[0] = 0x0f;
4852 hypercall[1] = 0x01;
4853 hypercall[2] = 0xc1;
4856 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
4857 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
4859 if (is_guest_mode(vcpu)) {
4860 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4861 unsigned long orig_val = val;
4864 * We get here when L2 changed cr0 in a way that did not change
4865 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
4866 * but did change L0 shadowed bits. So we first calculate the
4867 * effective cr0 value that L1 would like to write into the
4868 * hardware. It consists of the L2-owned bits from the new
4869 * value combined with the L1-owned bits from L1's guest_cr0.
4871 val = (val & ~vmcs12->cr0_guest_host_mask) |
4872 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
4874 /* TODO: will have to take unrestricted guest mode into
4876 if ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON)
4879 if (kvm_set_cr0(vcpu, val))
4881 vmcs_writel(CR0_READ_SHADOW, orig_val);
4884 if (to_vmx(vcpu)->nested.vmxon &&
4885 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
4887 return kvm_set_cr0(vcpu, val);
4891 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
4893 if (is_guest_mode(vcpu)) {
4894 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4895 unsigned long orig_val = val;
4897 /* analogously to handle_set_cr0 */
4898 val = (val & ~vmcs12->cr4_guest_host_mask) |
4899 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
4900 if (kvm_set_cr4(vcpu, val))
4902 vmcs_writel(CR4_READ_SHADOW, orig_val);
4905 return kvm_set_cr4(vcpu, val);
4908 /* called to set cr0 as approriate for clts instruction exit. */
4909 static void handle_clts(struct kvm_vcpu *vcpu)
4911 if (is_guest_mode(vcpu)) {
4913 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
4914 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
4915 * just pretend it's off (also in arch.cr0 for fpu_activate).
4917 vmcs_writel(CR0_READ_SHADOW,
4918 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
4919 vcpu->arch.cr0 &= ~X86_CR0_TS;
4921 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
4924 static int handle_cr(struct kvm_vcpu *vcpu)
4926 unsigned long exit_qualification, val;
4931 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
4932 cr = exit_qualification & 15;
4933 reg = (exit_qualification >> 8) & 15;
4934 switch ((exit_qualification >> 4) & 3) {
4935 case 0: /* mov to cr */
4936 val = kvm_register_read(vcpu, reg);
4937 trace_kvm_cr_write(cr, val);
4940 err = handle_set_cr0(vcpu, val);
4941 kvm_complete_insn_gp(vcpu, err);
4944 err = kvm_set_cr3(vcpu, val);
4945 kvm_complete_insn_gp(vcpu, err);
4948 err = handle_set_cr4(vcpu, val);
4949 kvm_complete_insn_gp(vcpu, err);
4952 u8 cr8_prev = kvm_get_cr8(vcpu);
4953 u8 cr8 = kvm_register_read(vcpu, reg);
4954 err = kvm_set_cr8(vcpu, cr8);
4955 kvm_complete_insn_gp(vcpu, err);
4956 if (irqchip_in_kernel(vcpu->kvm))
4958 if (cr8_prev <= cr8)
4960 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
4967 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
4968 skip_emulated_instruction(vcpu);
4969 vmx_fpu_activate(vcpu);
4971 case 1: /*mov from cr*/
4974 val = kvm_read_cr3(vcpu);
4975 kvm_register_write(vcpu, reg, val);
4976 trace_kvm_cr_read(cr, val);
4977 skip_emulated_instruction(vcpu);
4980 val = kvm_get_cr8(vcpu);
4981 kvm_register_write(vcpu, reg, val);
4982 trace_kvm_cr_read(cr, val);
4983 skip_emulated_instruction(vcpu);
4988 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
4989 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
4990 kvm_lmsw(vcpu, val);
4992 skip_emulated_instruction(vcpu);
4997 vcpu->run->exit_reason = 0;
4998 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
4999 (int)(exit_qualification >> 4) & 3, cr);
5003 static int handle_dr(struct kvm_vcpu *vcpu)
5005 unsigned long exit_qualification;
5008 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5009 if (!kvm_require_cpl(vcpu, 0))
5011 dr = vmcs_readl(GUEST_DR7);
5014 * As the vm-exit takes precedence over the debug trap, we
5015 * need to emulate the latter, either for the host or the
5016 * guest debugging itself.
5018 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5019 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5020 vcpu->run->debug.arch.dr7 = dr;
5021 vcpu->run->debug.arch.pc =
5022 vmcs_readl(GUEST_CS_BASE) +
5023 vmcs_readl(GUEST_RIP);
5024 vcpu->run->debug.arch.exception = DB_VECTOR;
5025 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5028 vcpu->arch.dr7 &= ~DR7_GD;
5029 vcpu->arch.dr6 |= DR6_BD;
5030 vmcs_writel(GUEST_DR7, vcpu->arch.dr7);
5031 kvm_queue_exception(vcpu, DB_VECTOR);
5036 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5037 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5038 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5039 if (exit_qualification & TYPE_MOV_FROM_DR) {
5041 if (!kvm_get_dr(vcpu, dr, &val))
5042 kvm_register_write(vcpu, reg, val);
5044 kvm_set_dr(vcpu, dr, vcpu->arch.regs[reg]);
5045 skip_emulated_instruction(vcpu);
5049 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5051 vmcs_writel(GUEST_DR7, val);
5054 static int handle_cpuid(struct kvm_vcpu *vcpu)
5056 kvm_emulate_cpuid(vcpu);
5060 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5062 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5065 if (vmx_get_msr(vcpu, ecx, &data)) {
5066 trace_kvm_msr_read_ex(ecx);
5067 kvm_inject_gp(vcpu, 0);
5071 trace_kvm_msr_read(ecx, data);
5073 /* FIXME: handling of bits 32:63 of rax, rdx */
5074 vcpu->arch.regs[VCPU_REGS_RAX] = data & -1u;
5075 vcpu->arch.regs[VCPU_REGS_RDX] = (data >> 32) & -1u;
5076 skip_emulated_instruction(vcpu);
5080 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5082 struct msr_data msr;
5083 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5084 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5085 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5089 msr.host_initiated = false;
5090 if (vmx_set_msr(vcpu, &msr) != 0) {
5091 trace_kvm_msr_write_ex(ecx, data);
5092 kvm_inject_gp(vcpu, 0);
5096 trace_kvm_msr_write(ecx, data);
5097 skip_emulated_instruction(vcpu);
5101 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5103 kvm_make_request(KVM_REQ_EVENT, vcpu);
5107 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5109 u32 cpu_based_vm_exec_control;
5111 /* clear pending irq */
5112 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5113 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5114 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5116 kvm_make_request(KVM_REQ_EVENT, vcpu);
5118 ++vcpu->stat.irq_window_exits;
5121 * If the user space waits to inject interrupts, exit as soon as
5124 if (!irqchip_in_kernel(vcpu->kvm) &&
5125 vcpu->run->request_interrupt_window &&
5126 !kvm_cpu_has_interrupt(vcpu)) {
5127 vcpu->run->exit_reason = KVM_EXIT_IRQ_WINDOW_OPEN;
5133 static int handle_halt(struct kvm_vcpu *vcpu)
5135 skip_emulated_instruction(vcpu);
5136 return kvm_emulate_halt(vcpu);
5139 static int handle_vmcall(struct kvm_vcpu *vcpu)
5141 skip_emulated_instruction(vcpu);
5142 kvm_emulate_hypercall(vcpu);
5146 static int handle_invd(struct kvm_vcpu *vcpu)
5148 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5151 static int handle_invlpg(struct kvm_vcpu *vcpu)
5153 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5155 kvm_mmu_invlpg(vcpu, exit_qualification);
5156 skip_emulated_instruction(vcpu);
5160 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5164 err = kvm_rdpmc(vcpu);
5165 kvm_complete_insn_gp(vcpu, err);
5170 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5172 skip_emulated_instruction(vcpu);
5173 kvm_emulate_wbinvd(vcpu);
5177 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5179 u64 new_bv = kvm_read_edx_eax(vcpu);
5180 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5182 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5183 skip_emulated_instruction(vcpu);
5187 static int handle_apic_access(struct kvm_vcpu *vcpu)
5189 if (likely(fasteoi)) {
5190 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5191 int access_type, offset;
5193 access_type = exit_qualification & APIC_ACCESS_TYPE;
5194 offset = exit_qualification & APIC_ACCESS_OFFSET;
5196 * Sane guest uses MOV to write EOI, with written value
5197 * not cared. So make a short-circuit here by avoiding
5198 * heavy instruction emulation.
5200 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5201 (offset == APIC_EOI)) {
5202 kvm_lapic_set_eoi(vcpu);
5203 skip_emulated_instruction(vcpu);
5207 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5210 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5212 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5213 int vector = exit_qualification & 0xff;
5215 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5216 kvm_apic_set_eoi_accelerated(vcpu, vector);
5220 static int handle_apic_write(struct kvm_vcpu *vcpu)
5222 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5223 u32 offset = exit_qualification & 0xfff;
5225 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5226 kvm_apic_write_nodecode(vcpu, offset);
5230 static int handle_task_switch(struct kvm_vcpu *vcpu)
5232 struct vcpu_vmx *vmx = to_vmx(vcpu);
5233 unsigned long exit_qualification;
5234 bool has_error_code = false;
5237 int reason, type, idt_v, idt_index;
5239 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5240 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5241 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5243 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5245 reason = (u32)exit_qualification >> 30;
5246 if (reason == TASK_SWITCH_GATE && idt_v) {
5248 case INTR_TYPE_NMI_INTR:
5249 vcpu->arch.nmi_injected = false;
5250 vmx_set_nmi_mask(vcpu, true);
5252 case INTR_TYPE_EXT_INTR:
5253 case INTR_TYPE_SOFT_INTR:
5254 kvm_clear_interrupt_queue(vcpu);
5256 case INTR_TYPE_HARD_EXCEPTION:
5257 if (vmx->idt_vectoring_info &
5258 VECTORING_INFO_DELIVER_CODE_MASK) {
5259 has_error_code = true;
5261 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5264 case INTR_TYPE_SOFT_EXCEPTION:
5265 kvm_clear_exception_queue(vcpu);
5271 tss_selector = exit_qualification;
5273 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5274 type != INTR_TYPE_EXT_INTR &&
5275 type != INTR_TYPE_NMI_INTR))
5276 skip_emulated_instruction(vcpu);
5278 if (kvm_task_switch(vcpu, tss_selector,
5279 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5280 has_error_code, error_code) == EMULATE_FAIL) {
5281 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5282 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5283 vcpu->run->internal.ndata = 0;
5287 /* clear all local breakpoint enable flags */
5288 vmcs_writel(GUEST_DR7, vmcs_readl(GUEST_DR7) & ~55);
5291 * TODO: What about debug traps on tss switch?
5292 * Are we supposed to inject them and update dr6?
5298 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5300 unsigned long exit_qualification;
5305 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5307 gla_validity = (exit_qualification >> 7) & 0x3;
5308 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
5309 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
5310 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
5311 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
5312 vmcs_readl(GUEST_LINEAR_ADDRESS));
5313 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
5314 (long unsigned int)exit_qualification);
5315 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5316 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
5320 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5321 trace_kvm_page_fault(gpa, exit_qualification);
5323 /* It is a write fault? */
5324 error_code = exit_qualification & (1U << 1);
5325 /* It is a fetch fault? */
5326 error_code |= (exit_qualification & (1U << 2)) << 2;
5327 /* ept page table is present? */
5328 error_code |= (exit_qualification >> 3) & 0x1;
5330 vcpu->arch.exit_qualification = exit_qualification;
5332 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5335 static u64 ept_rsvd_mask(u64 spte, int level)
5340 for (i = 51; i > boot_cpu_data.x86_phys_bits; i--)
5341 mask |= (1ULL << i);
5344 /* bits 7:3 reserved */
5346 else if (level == 2) {
5347 if (spte & (1ULL << 7))
5348 /* 2MB ref, bits 20:12 reserved */
5351 /* bits 6:3 reserved */
5358 static void ept_misconfig_inspect_spte(struct kvm_vcpu *vcpu, u64 spte,
5361 printk(KERN_ERR "%s: spte 0x%llx level %d\n", __func__, spte, level);
5363 /* 010b (write-only) */
5364 WARN_ON((spte & 0x7) == 0x2);
5366 /* 110b (write/execute) */
5367 WARN_ON((spte & 0x7) == 0x6);
5369 /* 100b (execute-only) and value not supported by logical processor */
5370 if (!cpu_has_vmx_ept_execute_only())
5371 WARN_ON((spte & 0x7) == 0x4);
5375 u64 rsvd_bits = spte & ept_rsvd_mask(spte, level);
5377 if (rsvd_bits != 0) {
5378 printk(KERN_ERR "%s: rsvd_bits = 0x%llx\n",
5379 __func__, rsvd_bits);
5383 if (level == 1 || (level == 2 && (spte & (1ULL << 7)))) {
5384 u64 ept_mem_type = (spte & 0x38) >> 3;
5386 if (ept_mem_type == 2 || ept_mem_type == 3 ||
5387 ept_mem_type == 7) {
5388 printk(KERN_ERR "%s: ept_mem_type=0x%llx\n",
5389 __func__, ept_mem_type);
5396 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5399 int nr_sptes, i, ret;
5402 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5404 ret = handle_mmio_page_fault_common(vcpu, gpa, true);
5405 if (likely(ret == RET_MMIO_PF_EMULATE))
5406 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
5409 if (unlikely(ret == RET_MMIO_PF_INVALID))
5410 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
5412 if (unlikely(ret == RET_MMIO_PF_RETRY))
5415 /* It is the real ept misconfig */
5416 printk(KERN_ERR "EPT: Misconfiguration.\n");
5417 printk(KERN_ERR "EPT: GPA: 0x%llx\n", gpa);
5419 nr_sptes = kvm_mmu_get_spte_hierarchy(vcpu, gpa, sptes);
5421 for (i = PT64_ROOT_LEVEL; i > PT64_ROOT_LEVEL - nr_sptes; --i)
5422 ept_misconfig_inspect_spte(vcpu, sptes[i-1], i);
5424 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5425 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
5430 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5432 u32 cpu_based_vm_exec_control;
5434 /* clear pending NMI */
5435 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5436 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
5437 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5438 ++vcpu->stat.nmi_window_exits;
5439 kvm_make_request(KVM_REQ_EVENT, vcpu);
5444 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5446 struct vcpu_vmx *vmx = to_vmx(vcpu);
5447 enum emulation_result err = EMULATE_DONE;
5450 bool intr_window_requested;
5451 unsigned count = 130;
5453 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5454 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
5456 while (!guest_state_valid(vcpu) && count-- != 0) {
5457 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
5458 return handle_interrupt_window(&vmx->vcpu);
5460 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
5463 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
5465 if (err == EMULATE_USER_EXIT) {
5470 if (err != EMULATE_DONE) {
5471 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5472 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5473 vcpu->run->internal.ndata = 0;
5477 if (vcpu->arch.halt_request) {
5478 vcpu->arch.halt_request = 0;
5479 ret = kvm_emulate_halt(vcpu);
5483 if (signal_pending(current))
5489 vmx->emulation_required = emulation_required(vcpu);
5495 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
5496 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
5498 static int handle_pause(struct kvm_vcpu *vcpu)
5500 skip_emulated_instruction(vcpu);
5501 kvm_vcpu_on_spin(vcpu);
5506 static int handle_invalid_op(struct kvm_vcpu *vcpu)
5508 kvm_queue_exception(vcpu, UD_VECTOR);
5513 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
5514 * We could reuse a single VMCS for all the L2 guests, but we also want the
5515 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
5516 * allows keeping them loaded on the processor, and in the future will allow
5517 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
5518 * every entry if they never change.
5519 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
5520 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
5522 * The following functions allocate and free a vmcs02 in this pool.
5525 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
5526 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
5528 struct vmcs02_list *item;
5529 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5530 if (item->vmptr == vmx->nested.current_vmptr) {
5531 list_move(&item->list, &vmx->nested.vmcs02_pool);
5532 return &item->vmcs02;
5535 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
5536 /* Recycle the least recently used VMCS. */
5537 item = list_entry(vmx->nested.vmcs02_pool.prev,
5538 struct vmcs02_list, list);
5539 item->vmptr = vmx->nested.current_vmptr;
5540 list_move(&item->list, &vmx->nested.vmcs02_pool);
5541 return &item->vmcs02;
5544 /* Create a new VMCS */
5545 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
5548 item->vmcs02.vmcs = alloc_vmcs();
5549 if (!item->vmcs02.vmcs) {
5553 loaded_vmcs_init(&item->vmcs02);
5554 item->vmptr = vmx->nested.current_vmptr;
5555 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
5556 vmx->nested.vmcs02_num++;
5557 return &item->vmcs02;
5560 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
5561 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
5563 struct vmcs02_list *item;
5564 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
5565 if (item->vmptr == vmptr) {
5566 free_loaded_vmcs(&item->vmcs02);
5567 list_del(&item->list);
5569 vmx->nested.vmcs02_num--;
5575 * Free all VMCSs saved for this vcpu, except the one pointed by
5576 * vmx->loaded_vmcs. These include the VMCSs in vmcs02_pool (except the one
5577 * currently used, if running L2), and vmcs01 when running L2.
5579 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
5581 struct vmcs02_list *item, *n;
5582 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
5583 if (vmx->loaded_vmcs != &item->vmcs02)
5584 free_loaded_vmcs(&item->vmcs02);
5585 list_del(&item->list);
5588 vmx->nested.vmcs02_num = 0;
5590 if (vmx->loaded_vmcs != &vmx->vmcs01)
5591 free_loaded_vmcs(&vmx->vmcs01);
5595 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
5596 * set the success or error code of an emulated VMX instruction, as specified
5597 * by Vol 2B, VMX Instruction Reference, "Conventions".
5599 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
5601 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
5602 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5603 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
5606 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
5608 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5609 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
5610 X86_EFLAGS_SF | X86_EFLAGS_OF))
5614 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
5615 u32 vm_instruction_error)
5617 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
5619 * failValid writes the error number to the current VMCS, which
5620 * can't be done there isn't a current VMCS.
5622 nested_vmx_failInvalid(vcpu);
5625 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
5626 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
5627 X86_EFLAGS_SF | X86_EFLAGS_OF))
5629 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
5631 * We don't need to force a shadow sync because
5632 * VM_INSTRUCTION_ERROR is not shadowed
5637 * Emulate the VMXON instruction.
5638 * Currently, we just remember that VMX is active, and do not save or even
5639 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
5640 * do not currently need to store anything in that guest-allocated memory
5641 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
5642 * argument is different from the VMXON pointer (which the spec says they do).
5644 static int handle_vmon(struct kvm_vcpu *vcpu)
5646 struct kvm_segment cs;
5647 struct vcpu_vmx *vmx = to_vmx(vcpu);
5648 struct vmcs *shadow_vmcs;
5649 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
5650 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
5652 /* The Intel VMX Instruction Reference lists a bunch of bits that
5653 * are prerequisite to running VMXON, most notably cr4.VMXE must be
5654 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
5655 * Otherwise, we should fail with #UD. We test these now:
5657 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
5658 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
5659 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
5660 kvm_queue_exception(vcpu, UD_VECTOR);
5664 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5665 if (is_long_mode(vcpu) && !cs.l) {
5666 kvm_queue_exception(vcpu, UD_VECTOR);
5670 if (vmx_get_cpl(vcpu)) {
5671 kvm_inject_gp(vcpu, 0);
5674 if (vmx->nested.vmxon) {
5675 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
5676 skip_emulated_instruction(vcpu);
5680 if ((vmx->nested.msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
5681 != VMXON_NEEDED_FEATURES) {
5682 kvm_inject_gp(vcpu, 0);
5686 if (enable_shadow_vmcs) {
5687 shadow_vmcs = alloc_vmcs();
5690 /* mark vmcs as shadow */
5691 shadow_vmcs->revision_id |= (1u << 31);
5692 /* init shadow vmcs */
5693 vmcs_clear(shadow_vmcs);
5694 vmx->nested.current_shadow_vmcs = shadow_vmcs;
5697 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
5698 vmx->nested.vmcs02_num = 0;
5700 vmx->nested.vmxon = true;
5702 skip_emulated_instruction(vcpu);
5703 nested_vmx_succeed(vcpu);
5708 * Intel's VMX Instruction Reference specifies a common set of prerequisites
5709 * for running VMX instructions (except VMXON, whose prerequisites are
5710 * slightly different). It also specifies what exception to inject otherwise.
5712 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
5714 struct kvm_segment cs;
5715 struct vcpu_vmx *vmx = to_vmx(vcpu);
5717 if (!vmx->nested.vmxon) {
5718 kvm_queue_exception(vcpu, UD_VECTOR);
5722 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
5723 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
5724 (is_long_mode(vcpu) && !cs.l)) {
5725 kvm_queue_exception(vcpu, UD_VECTOR);
5729 if (vmx_get_cpl(vcpu)) {
5730 kvm_inject_gp(vcpu, 0);
5737 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
5740 if (enable_shadow_vmcs) {
5741 if (vmx->nested.current_vmcs12 != NULL) {
5742 /* copy to memory all shadowed fields in case
5743 they were modified */
5744 copy_shadow_to_vmcs12(vmx);
5745 vmx->nested.sync_shadow_vmcs = false;
5746 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
5747 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
5748 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
5749 vmcs_write64(VMCS_LINK_POINTER, -1ull);
5752 kunmap(vmx->nested.current_vmcs12_page);
5753 nested_release_page(vmx->nested.current_vmcs12_page);
5757 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
5758 * just stops using VMX.
5760 static void free_nested(struct vcpu_vmx *vmx)
5762 if (!vmx->nested.vmxon)
5764 vmx->nested.vmxon = false;
5765 if (vmx->nested.current_vmptr != -1ull) {
5766 nested_release_vmcs12(vmx);
5767 vmx->nested.current_vmptr = -1ull;
5768 vmx->nested.current_vmcs12 = NULL;
5770 if (enable_shadow_vmcs)
5771 free_vmcs(vmx->nested.current_shadow_vmcs);
5772 /* Unpin physical memory we referred to in current vmcs02 */
5773 if (vmx->nested.apic_access_page) {
5774 nested_release_page(vmx->nested.apic_access_page);
5775 vmx->nested.apic_access_page = 0;
5778 nested_free_all_saved_vmcss(vmx);
5781 /* Emulate the VMXOFF instruction */
5782 static int handle_vmoff(struct kvm_vcpu *vcpu)
5784 if (!nested_vmx_check_permission(vcpu))
5786 free_nested(to_vmx(vcpu));
5787 skip_emulated_instruction(vcpu);
5788 nested_vmx_succeed(vcpu);
5793 * Decode the memory-address operand of a vmx instruction, as recorded on an
5794 * exit caused by such an instruction (run by a guest hypervisor).
5795 * On success, returns 0. When the operand is invalid, returns 1 and throws
5798 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
5799 unsigned long exit_qualification,
5800 u32 vmx_instruction_info, gva_t *ret)
5803 * According to Vol. 3B, "Information for VM Exits Due to Instruction
5804 * Execution", on an exit, vmx_instruction_info holds most of the
5805 * addressing components of the operand. Only the displacement part
5806 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
5807 * For how an actual address is calculated from all these components,
5808 * refer to Vol. 1, "Operand Addressing".
5810 int scaling = vmx_instruction_info & 3;
5811 int addr_size = (vmx_instruction_info >> 7) & 7;
5812 bool is_reg = vmx_instruction_info & (1u << 10);
5813 int seg_reg = (vmx_instruction_info >> 15) & 7;
5814 int index_reg = (vmx_instruction_info >> 18) & 0xf;
5815 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
5816 int base_reg = (vmx_instruction_info >> 23) & 0xf;
5817 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
5820 kvm_queue_exception(vcpu, UD_VECTOR);
5824 /* Addr = segment_base + offset */
5825 /* offset = base + [index * scale] + displacement */
5826 *ret = vmx_get_segment_base(vcpu, seg_reg);
5828 *ret += kvm_register_read(vcpu, base_reg);
5830 *ret += kvm_register_read(vcpu, index_reg)<<scaling;
5831 *ret += exit_qualification; /* holds the displacement */
5833 if (addr_size == 1) /* 32 bit */
5837 * TODO: throw #GP (and return 1) in various cases that the VM*
5838 * instructions require it - e.g., offset beyond segment limit,
5839 * unusable or unreadable/unwritable segment, non-canonical 64-bit
5840 * address, and so on. Currently these are not checked.
5845 /* Emulate the VMCLEAR instruction */
5846 static int handle_vmclear(struct kvm_vcpu *vcpu)
5848 struct vcpu_vmx *vmx = to_vmx(vcpu);
5851 struct vmcs12 *vmcs12;
5853 struct x86_exception e;
5855 if (!nested_vmx_check_permission(vcpu))
5858 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
5859 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
5862 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
5863 sizeof(vmptr), &e)) {
5864 kvm_inject_page_fault(vcpu, &e);
5868 if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
5869 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
5870 skip_emulated_instruction(vcpu);
5874 if (vmptr == vmx->nested.current_vmptr) {
5875 nested_release_vmcs12(vmx);
5876 vmx->nested.current_vmptr = -1ull;
5877 vmx->nested.current_vmcs12 = NULL;
5880 page = nested_get_page(vcpu, vmptr);
5883 * For accurate processor emulation, VMCLEAR beyond available
5884 * physical memory should do nothing at all. However, it is
5885 * possible that a nested vmx bug, not a guest hypervisor bug,
5886 * resulted in this case, so let's shut down before doing any
5889 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5892 vmcs12 = kmap(page);
5893 vmcs12->launch_state = 0;
5895 nested_release_page(page);
5897 nested_free_vmcs02(vmx, vmptr);
5899 skip_emulated_instruction(vcpu);
5900 nested_vmx_succeed(vcpu);
5904 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
5906 /* Emulate the VMLAUNCH instruction */
5907 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
5909 return nested_vmx_run(vcpu, true);
5912 /* Emulate the VMRESUME instruction */
5913 static int handle_vmresume(struct kvm_vcpu *vcpu)
5916 return nested_vmx_run(vcpu, false);
5919 enum vmcs_field_type {
5920 VMCS_FIELD_TYPE_U16 = 0,
5921 VMCS_FIELD_TYPE_U64 = 1,
5922 VMCS_FIELD_TYPE_U32 = 2,
5923 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
5926 static inline int vmcs_field_type(unsigned long field)
5928 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
5929 return VMCS_FIELD_TYPE_U32;
5930 return (field >> 13) & 0x3 ;
5933 static inline int vmcs_field_readonly(unsigned long field)
5935 return (((field >> 10) & 0x3) == 1);
5939 * Read a vmcs12 field. Since these can have varying lengths and we return
5940 * one type, we chose the biggest type (u64) and zero-extend the return value
5941 * to that size. Note that the caller, handle_vmread, might need to use only
5942 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
5943 * 64-bit fields are to be returned).
5945 static inline bool vmcs12_read_any(struct kvm_vcpu *vcpu,
5946 unsigned long field, u64 *ret)
5948 short offset = vmcs_field_to_offset(field);
5954 p = ((char *)(get_vmcs12(vcpu))) + offset;
5956 switch (vmcs_field_type(field)) {
5957 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5958 *ret = *((natural_width *)p);
5960 case VMCS_FIELD_TYPE_U16:
5963 case VMCS_FIELD_TYPE_U32:
5966 case VMCS_FIELD_TYPE_U64:
5970 return 0; /* can never happen. */
5975 static inline bool vmcs12_write_any(struct kvm_vcpu *vcpu,
5976 unsigned long field, u64 field_value){
5977 short offset = vmcs_field_to_offset(field);
5978 char *p = ((char *) get_vmcs12(vcpu)) + offset;
5982 switch (vmcs_field_type(field)) {
5983 case VMCS_FIELD_TYPE_U16:
5984 *(u16 *)p = field_value;
5986 case VMCS_FIELD_TYPE_U32:
5987 *(u32 *)p = field_value;
5989 case VMCS_FIELD_TYPE_U64:
5990 *(u64 *)p = field_value;
5992 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
5993 *(natural_width *)p = field_value;
5996 return false; /* can never happen. */
6001 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
6004 unsigned long field;
6006 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6007 const unsigned long *fields = shadow_read_write_fields;
6008 const int num_fields = max_shadow_read_write_fields;
6010 vmcs_load(shadow_vmcs);
6012 for (i = 0; i < num_fields; i++) {
6014 switch (vmcs_field_type(field)) {
6015 case VMCS_FIELD_TYPE_U16:
6016 field_value = vmcs_read16(field);
6018 case VMCS_FIELD_TYPE_U32:
6019 field_value = vmcs_read32(field);
6021 case VMCS_FIELD_TYPE_U64:
6022 field_value = vmcs_read64(field);
6024 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6025 field_value = vmcs_readl(field);
6028 vmcs12_write_any(&vmx->vcpu, field, field_value);
6031 vmcs_clear(shadow_vmcs);
6032 vmcs_load(vmx->loaded_vmcs->vmcs);
6035 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
6037 const unsigned long *fields[] = {
6038 shadow_read_write_fields,
6039 shadow_read_only_fields
6041 const int max_fields[] = {
6042 max_shadow_read_write_fields,
6043 max_shadow_read_only_fields
6046 unsigned long field;
6047 u64 field_value = 0;
6048 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
6050 vmcs_load(shadow_vmcs);
6052 for (q = 0; q < ARRAY_SIZE(fields); q++) {
6053 for (i = 0; i < max_fields[q]; i++) {
6054 field = fields[q][i];
6055 vmcs12_read_any(&vmx->vcpu, field, &field_value);
6057 switch (vmcs_field_type(field)) {
6058 case VMCS_FIELD_TYPE_U16:
6059 vmcs_write16(field, (u16)field_value);
6061 case VMCS_FIELD_TYPE_U32:
6062 vmcs_write32(field, (u32)field_value);
6064 case VMCS_FIELD_TYPE_U64:
6065 vmcs_write64(field, (u64)field_value);
6067 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
6068 vmcs_writel(field, (long)field_value);
6074 vmcs_clear(shadow_vmcs);
6075 vmcs_load(vmx->loaded_vmcs->vmcs);
6079 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
6080 * used before) all generate the same failure when it is missing.
6082 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
6084 struct vcpu_vmx *vmx = to_vmx(vcpu);
6085 if (vmx->nested.current_vmptr == -1ull) {
6086 nested_vmx_failInvalid(vcpu);
6087 skip_emulated_instruction(vcpu);
6093 static int handle_vmread(struct kvm_vcpu *vcpu)
6095 unsigned long field;
6097 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6098 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6101 if (!nested_vmx_check_permission(vcpu) ||
6102 !nested_vmx_check_vmcs12(vcpu))
6105 /* Decode instruction info and find the field to read */
6106 field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6107 /* Read the field, zero-extended to a u64 field_value */
6108 if (!vmcs12_read_any(vcpu, field, &field_value)) {
6109 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6110 skip_emulated_instruction(vcpu);
6114 * Now copy part of this value to register or memory, as requested.
6115 * Note that the number of bits actually copied is 32 or 64 depending
6116 * on the guest's mode (32 or 64 bit), not on the given field's length.
6118 if (vmx_instruction_info & (1u << 10)) {
6119 kvm_register_write(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
6122 if (get_vmx_mem_address(vcpu, exit_qualification,
6123 vmx_instruction_info, &gva))
6125 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
6126 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
6127 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
6130 nested_vmx_succeed(vcpu);
6131 skip_emulated_instruction(vcpu);
6136 static int handle_vmwrite(struct kvm_vcpu *vcpu)
6138 unsigned long field;
6140 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6141 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6142 /* The value to write might be 32 or 64 bits, depending on L1's long
6143 * mode, and eventually we need to write that into a field of several
6144 * possible lengths. The code below first zero-extends the value to 64
6145 * bit (field_value), and then copies only the approriate number of
6146 * bits into the vmcs12 field.
6148 u64 field_value = 0;
6149 struct x86_exception e;
6151 if (!nested_vmx_check_permission(vcpu) ||
6152 !nested_vmx_check_vmcs12(vcpu))
6155 if (vmx_instruction_info & (1u << 10))
6156 field_value = kvm_register_read(vcpu,
6157 (((vmx_instruction_info) >> 3) & 0xf));
6159 if (get_vmx_mem_address(vcpu, exit_qualification,
6160 vmx_instruction_info, &gva))
6162 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
6163 &field_value, (is_long_mode(vcpu) ? 8 : 4), &e)) {
6164 kvm_inject_page_fault(vcpu, &e);
6170 field = kvm_register_read(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
6171 if (vmcs_field_readonly(field)) {
6172 nested_vmx_failValid(vcpu,
6173 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
6174 skip_emulated_instruction(vcpu);
6178 if (!vmcs12_write_any(vcpu, field, field_value)) {
6179 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
6180 skip_emulated_instruction(vcpu);
6184 nested_vmx_succeed(vcpu);
6185 skip_emulated_instruction(vcpu);
6189 /* Emulate the VMPTRLD instruction */
6190 static int handle_vmptrld(struct kvm_vcpu *vcpu)
6192 struct vcpu_vmx *vmx = to_vmx(vcpu);
6195 struct x86_exception e;
6198 if (!nested_vmx_check_permission(vcpu))
6201 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6202 vmcs_read32(VMX_INSTRUCTION_INFO), &gva))
6205 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6206 sizeof(vmptr), &e)) {
6207 kvm_inject_page_fault(vcpu, &e);
6211 if (!IS_ALIGNED(vmptr, PAGE_SIZE)) {
6212 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
6213 skip_emulated_instruction(vcpu);
6217 if (vmx->nested.current_vmptr != vmptr) {
6218 struct vmcs12 *new_vmcs12;
6220 page = nested_get_page(vcpu, vmptr);
6222 nested_vmx_failInvalid(vcpu);
6223 skip_emulated_instruction(vcpu);
6226 new_vmcs12 = kmap(page);
6227 if (new_vmcs12->revision_id != VMCS12_REVISION) {
6229 nested_release_page_clean(page);
6230 nested_vmx_failValid(vcpu,
6231 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
6232 skip_emulated_instruction(vcpu);
6235 if (vmx->nested.current_vmptr != -1ull)
6236 nested_release_vmcs12(vmx);
6238 vmx->nested.current_vmptr = vmptr;
6239 vmx->nested.current_vmcs12 = new_vmcs12;
6240 vmx->nested.current_vmcs12_page = page;
6241 if (enable_shadow_vmcs) {
6242 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6243 exec_control |= SECONDARY_EXEC_SHADOW_VMCS;
6244 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
6245 vmcs_write64(VMCS_LINK_POINTER,
6246 __pa(vmx->nested.current_shadow_vmcs));
6247 vmx->nested.sync_shadow_vmcs = true;
6251 nested_vmx_succeed(vcpu);
6252 skip_emulated_instruction(vcpu);
6256 /* Emulate the VMPTRST instruction */
6257 static int handle_vmptrst(struct kvm_vcpu *vcpu)
6259 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6260 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
6262 struct x86_exception e;
6264 if (!nested_vmx_check_permission(vcpu))
6267 if (get_vmx_mem_address(vcpu, exit_qualification,
6268 vmx_instruction_info, &vmcs_gva))
6270 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
6271 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
6272 (void *)&to_vmx(vcpu)->nested.current_vmptr,
6274 kvm_inject_page_fault(vcpu, &e);
6277 nested_vmx_succeed(vcpu);
6278 skip_emulated_instruction(vcpu);
6283 * The exit handlers return 1 if the exit was handled fully and guest execution
6284 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
6285 * to be done to userspace and return 0.
6287 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
6288 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
6289 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
6290 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
6291 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
6292 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
6293 [EXIT_REASON_CR_ACCESS] = handle_cr,
6294 [EXIT_REASON_DR_ACCESS] = handle_dr,
6295 [EXIT_REASON_CPUID] = handle_cpuid,
6296 [EXIT_REASON_MSR_READ] = handle_rdmsr,
6297 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
6298 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
6299 [EXIT_REASON_HLT] = handle_halt,
6300 [EXIT_REASON_INVD] = handle_invd,
6301 [EXIT_REASON_INVLPG] = handle_invlpg,
6302 [EXIT_REASON_RDPMC] = handle_rdpmc,
6303 [EXIT_REASON_VMCALL] = handle_vmcall,
6304 [EXIT_REASON_VMCLEAR] = handle_vmclear,
6305 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
6306 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
6307 [EXIT_REASON_VMPTRST] = handle_vmptrst,
6308 [EXIT_REASON_VMREAD] = handle_vmread,
6309 [EXIT_REASON_VMRESUME] = handle_vmresume,
6310 [EXIT_REASON_VMWRITE] = handle_vmwrite,
6311 [EXIT_REASON_VMOFF] = handle_vmoff,
6312 [EXIT_REASON_VMON] = handle_vmon,
6313 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
6314 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
6315 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
6316 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
6317 [EXIT_REASON_WBINVD] = handle_wbinvd,
6318 [EXIT_REASON_XSETBV] = handle_xsetbv,
6319 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
6320 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
6321 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
6322 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
6323 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
6324 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_invalid_op,
6325 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_invalid_op,
6328 static const int kvm_vmx_max_exit_handlers =
6329 ARRAY_SIZE(kvm_vmx_exit_handlers);
6331 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
6332 struct vmcs12 *vmcs12)
6334 unsigned long exit_qualification;
6335 gpa_t bitmap, last_bitmap;
6340 if (nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING))
6343 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
6346 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6348 port = exit_qualification >> 16;
6349 size = (exit_qualification & 7) + 1;
6351 last_bitmap = (gpa_t)-1;
6356 bitmap = vmcs12->io_bitmap_a;
6357 else if (port < 0x10000)
6358 bitmap = vmcs12->io_bitmap_b;
6361 bitmap += (port & 0x7fff) / 8;
6363 if (last_bitmap != bitmap)
6364 if (kvm_read_guest(vcpu->kvm, bitmap, &b, 1))
6366 if (b & (1 << (port & 7)))
6371 last_bitmap = bitmap;
6378 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
6379 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
6380 * disinterest in the current event (read or write a specific MSR) by using an
6381 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
6383 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
6384 struct vmcs12 *vmcs12, u32 exit_reason)
6386 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
6389 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
6393 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
6394 * for the four combinations of read/write and low/high MSR numbers.
6395 * First we need to figure out which of the four to use:
6397 bitmap = vmcs12->msr_bitmap;
6398 if (exit_reason == EXIT_REASON_MSR_WRITE)
6400 if (msr_index >= 0xc0000000) {
6401 msr_index -= 0xc0000000;
6405 /* Then read the msr_index'th bit from this bitmap: */
6406 if (msr_index < 1024*8) {
6408 if (kvm_read_guest(vcpu->kvm, bitmap + msr_index/8, &b, 1))
6410 return 1 & (b >> (msr_index & 7));
6412 return 1; /* let L1 handle the wrong parameter */
6416 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
6417 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
6418 * intercept (via guest_host_mask etc.) the current event.
6420 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
6421 struct vmcs12 *vmcs12)
6423 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6424 int cr = exit_qualification & 15;
6425 int reg = (exit_qualification >> 8) & 15;
6426 unsigned long val = kvm_register_read(vcpu, reg);
6428 switch ((exit_qualification >> 4) & 3) {
6429 case 0: /* mov to cr */
6432 if (vmcs12->cr0_guest_host_mask &
6433 (val ^ vmcs12->cr0_read_shadow))
6437 if ((vmcs12->cr3_target_count >= 1 &&
6438 vmcs12->cr3_target_value0 == val) ||
6439 (vmcs12->cr3_target_count >= 2 &&
6440 vmcs12->cr3_target_value1 == val) ||
6441 (vmcs12->cr3_target_count >= 3 &&
6442 vmcs12->cr3_target_value2 == val) ||
6443 (vmcs12->cr3_target_count >= 4 &&
6444 vmcs12->cr3_target_value3 == val))
6446 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
6450 if (vmcs12->cr4_guest_host_mask &
6451 (vmcs12->cr4_read_shadow ^ val))
6455 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
6461 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
6462 (vmcs12->cr0_read_shadow & X86_CR0_TS))
6465 case 1: /* mov from cr */
6468 if (vmcs12->cpu_based_vm_exec_control &
6469 CPU_BASED_CR3_STORE_EXITING)
6473 if (vmcs12->cpu_based_vm_exec_control &
6474 CPU_BASED_CR8_STORE_EXITING)
6481 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
6482 * cr0. Other attempted changes are ignored, with no exit.
6484 if (vmcs12->cr0_guest_host_mask & 0xe &
6485 (val ^ vmcs12->cr0_read_shadow))
6487 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
6488 !(vmcs12->cr0_read_shadow & 0x1) &&
6497 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
6498 * should handle it ourselves in L0 (and then continue L2). Only call this
6499 * when in is_guest_mode (L2).
6501 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
6503 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6504 struct vcpu_vmx *vmx = to_vmx(vcpu);
6505 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6506 u32 exit_reason = vmx->exit_reason;
6508 if (vmx->nested.nested_run_pending)
6511 if (unlikely(vmx->fail)) {
6512 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
6513 vmcs_read32(VM_INSTRUCTION_ERROR));
6517 switch (exit_reason) {
6518 case EXIT_REASON_EXCEPTION_NMI:
6519 if (!is_exception(intr_info))
6521 else if (is_page_fault(intr_info))
6523 return vmcs12->exception_bitmap &
6524 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
6525 case EXIT_REASON_EXTERNAL_INTERRUPT:
6527 case EXIT_REASON_TRIPLE_FAULT:
6529 case EXIT_REASON_PENDING_INTERRUPT:
6530 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
6531 case EXIT_REASON_NMI_WINDOW:
6532 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
6533 case EXIT_REASON_TASK_SWITCH:
6535 case EXIT_REASON_CPUID:
6537 case EXIT_REASON_HLT:
6538 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
6539 case EXIT_REASON_INVD:
6541 case EXIT_REASON_INVLPG:
6542 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
6543 case EXIT_REASON_RDPMC:
6544 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
6545 case EXIT_REASON_RDTSC:
6546 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
6547 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
6548 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
6549 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
6550 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
6551 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
6553 * VMX instructions trap unconditionally. This allows L1 to
6554 * emulate them for its L2 guest, i.e., allows 3-level nesting!
6557 case EXIT_REASON_CR_ACCESS:
6558 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
6559 case EXIT_REASON_DR_ACCESS:
6560 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
6561 case EXIT_REASON_IO_INSTRUCTION:
6562 return nested_vmx_exit_handled_io(vcpu, vmcs12);
6563 case EXIT_REASON_MSR_READ:
6564 case EXIT_REASON_MSR_WRITE:
6565 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
6566 case EXIT_REASON_INVALID_STATE:
6568 case EXIT_REASON_MWAIT_INSTRUCTION:
6569 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
6570 case EXIT_REASON_MONITOR_INSTRUCTION:
6571 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
6572 case EXIT_REASON_PAUSE_INSTRUCTION:
6573 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
6574 nested_cpu_has2(vmcs12,
6575 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
6576 case EXIT_REASON_MCE_DURING_VMENTRY:
6578 case EXIT_REASON_TPR_BELOW_THRESHOLD:
6580 case EXIT_REASON_APIC_ACCESS:
6581 return nested_cpu_has2(vmcs12,
6582 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
6583 case EXIT_REASON_EPT_VIOLATION:
6584 case EXIT_REASON_EPT_MISCONFIG:
6586 case EXIT_REASON_PREEMPTION_TIMER:
6587 return vmcs12->pin_based_vm_exec_control &
6588 PIN_BASED_VMX_PREEMPTION_TIMER;
6589 case EXIT_REASON_WBINVD:
6590 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
6591 case EXIT_REASON_XSETBV:
6598 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
6600 *info1 = vmcs_readl(EXIT_QUALIFICATION);
6601 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
6605 * The guest has exited. See if we can fix it or if we need userspace
6608 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
6610 struct vcpu_vmx *vmx = to_vmx(vcpu);
6611 u32 exit_reason = vmx->exit_reason;
6612 u32 vectoring_info = vmx->idt_vectoring_info;
6614 /* If guest state is invalid, start emulating */
6615 if (vmx->emulation_required)
6616 return handle_invalid_guest_state(vcpu);
6619 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
6620 * we did not inject a still-pending event to L1 now because of
6621 * nested_run_pending, we need to re-enable this bit.
6623 if (vmx->nested.nested_run_pending)
6624 kvm_make_request(KVM_REQ_EVENT, vcpu);
6626 if (!is_guest_mode(vcpu) && (exit_reason == EXIT_REASON_VMLAUNCH ||
6627 exit_reason == EXIT_REASON_VMRESUME))
6628 vmx->nested.nested_run_pending = 1;
6630 vmx->nested.nested_run_pending = 0;
6632 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
6633 nested_vmx_vmexit(vcpu);
6637 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
6638 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6639 vcpu->run->fail_entry.hardware_entry_failure_reason
6644 if (unlikely(vmx->fail)) {
6645 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
6646 vcpu->run->fail_entry.hardware_entry_failure_reason
6647 = vmcs_read32(VM_INSTRUCTION_ERROR);
6653 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
6654 * delivery event since it indicates guest is accessing MMIO.
6655 * The vm-exit can be triggered again after return to guest that
6656 * will cause infinite loop.
6658 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
6659 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
6660 exit_reason != EXIT_REASON_EPT_VIOLATION &&
6661 exit_reason != EXIT_REASON_TASK_SWITCH)) {
6662 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6663 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
6664 vcpu->run->internal.ndata = 2;
6665 vcpu->run->internal.data[0] = vectoring_info;
6666 vcpu->run->internal.data[1] = exit_reason;
6670 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
6671 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
6672 get_vmcs12(vcpu), vcpu)))) {
6673 if (vmx_interrupt_allowed(vcpu)) {
6674 vmx->soft_vnmi_blocked = 0;
6675 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
6676 vcpu->arch.nmi_pending) {
6678 * This CPU don't support us in finding the end of an
6679 * NMI-blocked window if the guest runs with IRQs
6680 * disabled. So we pull the trigger after 1 s of
6681 * futile waiting, but inform the user about this.
6683 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
6684 "state on VCPU %d after 1 s timeout\n",
6685 __func__, vcpu->vcpu_id);
6686 vmx->soft_vnmi_blocked = 0;
6690 if (exit_reason < kvm_vmx_max_exit_handlers
6691 && kvm_vmx_exit_handlers[exit_reason])
6692 return kvm_vmx_exit_handlers[exit_reason](vcpu);
6694 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6695 vcpu->run->hw.hardware_exit_reason = exit_reason;
6700 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
6702 if (irr == -1 || tpr < irr) {
6703 vmcs_write32(TPR_THRESHOLD, 0);
6707 vmcs_write32(TPR_THRESHOLD, irr);
6710 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
6712 u32 sec_exec_control;
6715 * There is not point to enable virtualize x2apic without enable
6718 if (!cpu_has_vmx_virtualize_x2apic_mode() ||
6719 !vmx_vm_has_apicv(vcpu->kvm))
6722 if (!vm_need_tpr_shadow(vcpu->kvm))
6725 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
6728 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6729 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6731 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
6732 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
6734 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
6736 vmx_set_msr_bitmap(vcpu);
6739 static void vmx_hwapic_isr_update(struct kvm *kvm, int isr)
6744 if (!vmx_vm_has_apicv(kvm))
6750 status = vmcs_read16(GUEST_INTR_STATUS);
6755 vmcs_write16(GUEST_INTR_STATUS, status);
6759 static void vmx_set_rvi(int vector)
6764 status = vmcs_read16(GUEST_INTR_STATUS);
6765 old = (u8)status & 0xff;
6766 if ((u8)vector != old) {
6768 status |= (u8)vector;
6769 vmcs_write16(GUEST_INTR_STATUS, status);
6773 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
6778 vmx_set_rvi(max_irr);
6781 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
6783 if (!vmx_vm_has_apicv(vcpu->kvm))
6786 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
6787 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
6788 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
6789 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
6792 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
6796 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
6797 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
6800 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6801 exit_intr_info = vmx->exit_intr_info;
6803 /* Handle machine checks before interrupts are enabled */
6804 if (is_machine_check(exit_intr_info))
6805 kvm_machine_check();
6807 /* We need to handle NMIs before interrupts are enabled */
6808 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
6809 (exit_intr_info & INTR_INFO_VALID_MASK)) {
6810 kvm_before_handle_nmi(&vmx->vcpu);
6812 kvm_after_handle_nmi(&vmx->vcpu);
6816 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
6818 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6821 * If external interrupt exists, IF bit is set in rflags/eflags on the
6822 * interrupt stack frame, and interrupt will be enabled on a return
6823 * from interrupt handler.
6825 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
6826 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
6827 unsigned int vector;
6828 unsigned long entry;
6830 struct vcpu_vmx *vmx = to_vmx(vcpu);
6831 #ifdef CONFIG_X86_64
6835 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6836 desc = (gate_desc *)vmx->host_idt_base + vector;
6837 entry = gate_offset(*desc);
6839 #ifdef CONFIG_X86_64
6840 "mov %%" _ASM_SP ", %[sp]\n\t"
6841 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
6846 "orl $0x200, (%%" _ASM_SP ")\n\t"
6847 __ASM_SIZE(push) " $%c[cs]\n\t"
6848 "call *%[entry]\n\t"
6850 #ifdef CONFIG_X86_64
6855 [ss]"i"(__KERNEL_DS),
6856 [cs]"i"(__KERNEL_CS)
6862 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
6867 bool idtv_info_valid;
6869 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6871 if (cpu_has_virtual_nmis()) {
6872 if (vmx->nmi_known_unmasked)
6875 * Can't use vmx->exit_intr_info since we're not sure what
6876 * the exit reason is.
6878 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
6879 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
6880 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
6882 * SDM 3: 27.7.1.2 (September 2008)
6883 * Re-set bit "block by NMI" before VM entry if vmexit caused by
6884 * a guest IRET fault.
6885 * SDM 3: 23.2.2 (September 2008)
6886 * Bit 12 is undefined in any of the following cases:
6887 * If the VM exit sets the valid bit in the IDT-vectoring
6888 * information field.
6889 * If the VM exit is due to a double fault.
6891 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
6892 vector != DF_VECTOR && !idtv_info_valid)
6893 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
6894 GUEST_INTR_STATE_NMI);
6896 vmx->nmi_known_unmasked =
6897 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
6898 & GUEST_INTR_STATE_NMI);
6899 } else if (unlikely(vmx->soft_vnmi_blocked))
6900 vmx->vnmi_blocked_time +=
6901 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
6904 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
6905 u32 idt_vectoring_info,
6906 int instr_len_field,
6907 int error_code_field)
6911 bool idtv_info_valid;
6913 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
6915 vcpu->arch.nmi_injected = false;
6916 kvm_clear_exception_queue(vcpu);
6917 kvm_clear_interrupt_queue(vcpu);
6919 if (!idtv_info_valid)
6922 kvm_make_request(KVM_REQ_EVENT, vcpu);
6924 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
6925 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
6928 case INTR_TYPE_NMI_INTR:
6929 vcpu->arch.nmi_injected = true;
6931 * SDM 3: 27.7.1.2 (September 2008)
6932 * Clear bit "block by NMI" before VM entry if a NMI
6935 vmx_set_nmi_mask(vcpu, false);
6937 case INTR_TYPE_SOFT_EXCEPTION:
6938 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6940 case INTR_TYPE_HARD_EXCEPTION:
6941 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
6942 u32 err = vmcs_read32(error_code_field);
6943 kvm_queue_exception_e(vcpu, vector, err);
6945 kvm_queue_exception(vcpu, vector);
6947 case INTR_TYPE_SOFT_INTR:
6948 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
6950 case INTR_TYPE_EXT_INTR:
6951 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
6958 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
6960 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
6961 VM_EXIT_INSTRUCTION_LEN,
6962 IDT_VECTORING_ERROR_CODE);
6965 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
6967 __vmx_complete_interrupts(vcpu,
6968 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
6969 VM_ENTRY_INSTRUCTION_LEN,
6970 VM_ENTRY_EXCEPTION_ERROR_CODE);
6972 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
6975 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
6978 struct perf_guest_switch_msr *msrs;
6980 msrs = perf_guest_get_msrs(&nr_msrs);
6985 for (i = 0; i < nr_msrs; i++)
6986 if (msrs[i].host == msrs[i].guest)
6987 clear_atomic_switch_msr(vmx, msrs[i].msr);
6989 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
6993 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
6995 struct vcpu_vmx *vmx = to_vmx(vcpu);
6996 unsigned long debugctlmsr;
6998 /* Record the guest's net vcpu time for enforced NMI injections. */
6999 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
7000 vmx->entry_time = ktime_get();
7002 /* Don't enter VMX if guest state is invalid, let the exit handler
7003 start emulation until we arrive back to a valid state */
7004 if (vmx->emulation_required)
7007 if (vmx->nested.sync_shadow_vmcs) {
7008 copy_vmcs12_to_shadow(vmx);
7009 vmx->nested.sync_shadow_vmcs = false;
7012 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
7013 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
7014 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
7015 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
7017 /* When single-stepping over STI and MOV SS, we must clear the
7018 * corresponding interruptibility bits in the guest state. Otherwise
7019 * vmentry fails as it then expects bit 14 (BS) in pending debug
7020 * exceptions being set, but that's not correct for the guest debugging
7022 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
7023 vmx_set_interrupt_shadow(vcpu, 0);
7025 atomic_switch_perf_msrs(vmx);
7026 debugctlmsr = get_debugctlmsr();
7028 vmx->__launched = vmx->loaded_vmcs->launched;
7030 /* Store host registers */
7031 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
7032 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
7033 "push %%" _ASM_CX " \n\t"
7034 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
7036 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
7037 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
7039 /* Reload cr2 if changed */
7040 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
7041 "mov %%cr2, %%" _ASM_DX " \n\t"
7042 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
7044 "mov %%" _ASM_AX", %%cr2 \n\t"
7046 /* Check if vmlaunch of vmresume is needed */
7047 "cmpl $0, %c[launched](%0) \n\t"
7048 /* Load guest registers. Don't clobber flags. */
7049 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
7050 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
7051 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
7052 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
7053 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
7054 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
7055 #ifdef CONFIG_X86_64
7056 "mov %c[r8](%0), %%r8 \n\t"
7057 "mov %c[r9](%0), %%r9 \n\t"
7058 "mov %c[r10](%0), %%r10 \n\t"
7059 "mov %c[r11](%0), %%r11 \n\t"
7060 "mov %c[r12](%0), %%r12 \n\t"
7061 "mov %c[r13](%0), %%r13 \n\t"
7062 "mov %c[r14](%0), %%r14 \n\t"
7063 "mov %c[r15](%0), %%r15 \n\t"
7065 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
7067 /* Enter guest mode */
7069 __ex(ASM_VMX_VMLAUNCH) "\n\t"
7071 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
7073 /* Save guest registers, load host registers, keep flags */
7074 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
7076 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
7077 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
7078 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
7079 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
7080 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
7081 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
7082 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
7083 #ifdef CONFIG_X86_64
7084 "mov %%r8, %c[r8](%0) \n\t"
7085 "mov %%r9, %c[r9](%0) \n\t"
7086 "mov %%r10, %c[r10](%0) \n\t"
7087 "mov %%r11, %c[r11](%0) \n\t"
7088 "mov %%r12, %c[r12](%0) \n\t"
7089 "mov %%r13, %c[r13](%0) \n\t"
7090 "mov %%r14, %c[r14](%0) \n\t"
7091 "mov %%r15, %c[r15](%0) \n\t"
7093 "mov %%cr2, %%" _ASM_AX " \n\t"
7094 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
7096 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
7097 "setbe %c[fail](%0) \n\t"
7098 ".pushsection .rodata \n\t"
7099 ".global vmx_return \n\t"
7100 "vmx_return: " _ASM_PTR " 2b \n\t"
7102 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
7103 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
7104 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
7105 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
7106 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
7107 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
7108 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
7109 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
7110 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
7111 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
7112 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
7113 #ifdef CONFIG_X86_64
7114 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
7115 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
7116 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
7117 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
7118 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
7119 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
7120 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
7121 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
7123 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
7124 [wordsize]"i"(sizeof(ulong))
7126 #ifdef CONFIG_X86_64
7127 , "rax", "rbx", "rdi", "rsi"
7128 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
7130 , "eax", "ebx", "edi", "esi"
7134 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
7136 update_debugctlmsr(debugctlmsr);
7138 #ifndef CONFIG_X86_64
7140 * The sysexit path does not restore ds/es, so we must set them to
7141 * a reasonable value ourselves.
7143 * We can't defer this to vmx_load_host_state() since that function
7144 * may be executed in interrupt context, which saves and restore segments
7145 * around it, nullifying its effect.
7147 loadsegment(ds, __USER_DS);
7148 loadsegment(es, __USER_DS);
7151 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
7152 | (1 << VCPU_EXREG_RFLAGS)
7153 | (1 << VCPU_EXREG_CPL)
7154 | (1 << VCPU_EXREG_PDPTR)
7155 | (1 << VCPU_EXREG_SEGMENTS)
7156 | (1 << VCPU_EXREG_CR3));
7157 vcpu->arch.regs_dirty = 0;
7159 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
7161 vmx->loaded_vmcs->launched = 1;
7163 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
7164 trace_kvm_exit(vmx->exit_reason, vcpu, KVM_ISA_VMX);
7166 vmx_complete_atomic_exit(vmx);
7167 vmx_recover_nmi_blocking(vmx);
7168 vmx_complete_interrupts(vmx);
7171 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
7173 struct vcpu_vmx *vmx = to_vmx(vcpu);
7177 free_loaded_vmcs(vmx->loaded_vmcs);
7178 kfree(vmx->guest_msrs);
7179 kvm_vcpu_uninit(vcpu);
7180 kmem_cache_free(kvm_vcpu_cache, vmx);
7183 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
7186 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
7190 return ERR_PTR(-ENOMEM);
7194 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
7198 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
7200 if (!vmx->guest_msrs) {
7204 vmx->loaded_vmcs = &vmx->vmcs01;
7205 vmx->loaded_vmcs->vmcs = alloc_vmcs();
7206 if (!vmx->loaded_vmcs->vmcs)
7209 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
7210 loaded_vmcs_init(vmx->loaded_vmcs);
7215 vmx_vcpu_load(&vmx->vcpu, cpu);
7216 vmx->vcpu.cpu = cpu;
7217 err = vmx_vcpu_setup(vmx);
7218 vmx_vcpu_put(&vmx->vcpu);
7222 if (vm_need_virtualize_apic_accesses(kvm)) {
7223 err = alloc_apic_access_page(kvm);
7229 if (!kvm->arch.ept_identity_map_addr)
7230 kvm->arch.ept_identity_map_addr =
7231 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
7233 if (alloc_identity_pagetable(kvm) != 0)
7235 if (!init_rmode_identity_map(kvm))
7239 vmx->nested.current_vmptr = -1ull;
7240 vmx->nested.current_vmcs12 = NULL;
7245 free_loaded_vmcs(vmx->loaded_vmcs);
7247 kfree(vmx->guest_msrs);
7249 kvm_vcpu_uninit(&vmx->vcpu);
7252 kmem_cache_free(kvm_vcpu_cache, vmx);
7253 return ERR_PTR(err);
7256 static void __init vmx_check_processor_compat(void *rtn)
7258 struct vmcs_config vmcs_conf;
7261 if (setup_vmcs_config(&vmcs_conf) < 0)
7263 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
7264 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
7265 smp_processor_id());
7270 static int get_ept_level(void)
7272 return VMX_EPT_DEFAULT_GAW + 1;
7275 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
7279 /* For VT-d and EPT combination
7280 * 1. MMIO: always map as UC
7282 * a. VT-d without snooping control feature: can't guarantee the
7283 * result, try to trust guest.
7284 * b. VT-d with snooping control feature: snooping control feature of
7285 * VT-d engine can guarantee the cache correctness. Just set it
7286 * to WB to keep consistent with host. So the same as item 3.
7287 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
7288 * consistent with host MTRR
7291 ret = MTRR_TYPE_UNCACHABLE << VMX_EPT_MT_EPTE_SHIFT;
7292 else if (vcpu->kvm->arch.iommu_domain &&
7293 !(vcpu->kvm->arch.iommu_flags & KVM_IOMMU_CACHE_COHERENCY))
7294 ret = kvm_get_guest_memory_type(vcpu, gfn) <<
7295 VMX_EPT_MT_EPTE_SHIFT;
7297 ret = (MTRR_TYPE_WRBACK << VMX_EPT_MT_EPTE_SHIFT)
7303 static int vmx_get_lpage_level(void)
7305 if (enable_ept && !cpu_has_vmx_ept_1g_page())
7306 return PT_DIRECTORY_LEVEL;
7308 /* For shadow and EPT supported 1GB page */
7309 return PT_PDPE_LEVEL;
7312 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
7314 struct kvm_cpuid_entry2 *best;
7315 struct vcpu_vmx *vmx = to_vmx(vcpu);
7318 vmx->rdtscp_enabled = false;
7319 if (vmx_rdtscp_supported()) {
7320 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7321 if (exec_control & SECONDARY_EXEC_RDTSCP) {
7322 best = kvm_find_cpuid_entry(vcpu, 0x80000001, 0);
7323 if (best && (best->edx & bit(X86_FEATURE_RDTSCP)))
7324 vmx->rdtscp_enabled = true;
7326 exec_control &= ~SECONDARY_EXEC_RDTSCP;
7327 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7333 /* Exposing INVPCID only when PCID is exposed */
7334 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
7335 if (vmx_invpcid_supported() &&
7336 best && (best->ebx & bit(X86_FEATURE_INVPCID)) &&
7337 guest_cpuid_has_pcid(vcpu)) {
7338 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7339 exec_control |= SECONDARY_EXEC_ENABLE_INVPCID;
7340 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7343 if (cpu_has_secondary_exec_ctrls()) {
7344 exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7345 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
7346 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
7350 best->ebx &= ~bit(X86_FEATURE_INVPCID);
7354 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
7356 if (func == 1 && nested)
7357 entry->ecx |= bit(X86_FEATURE_VMX);
7360 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
7361 struct x86_exception *fault)
7363 struct vmcs12 *vmcs12;
7364 nested_vmx_vmexit(vcpu);
7365 vmcs12 = get_vmcs12(vcpu);
7367 if (fault->error_code & PFERR_RSVD_MASK)
7368 vmcs12->vm_exit_reason = EXIT_REASON_EPT_MISCONFIG;
7370 vmcs12->vm_exit_reason = EXIT_REASON_EPT_VIOLATION;
7371 vmcs12->exit_qualification = vcpu->arch.exit_qualification;
7372 vmcs12->guest_physical_address = fault->address;
7375 /* Callbacks for nested_ept_init_mmu_context: */
7377 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
7379 /* return the page table to be shadowed - in our case, EPT12 */
7380 return get_vmcs12(vcpu)->ept_pointer;
7383 static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
7385 int r = kvm_init_shadow_ept_mmu(vcpu, &vcpu->arch.mmu,
7386 nested_vmx_ept_caps & VMX_EPT_EXECUTE_ONLY_BIT);
7388 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
7389 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
7390 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
7392 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
7397 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
7399 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
7403 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
7404 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
7405 * with L0's requirements for its guest (a.k.a. vmsc01), so we can run the L2
7406 * guest in a way that will both be appropriate to L1's requests, and our
7407 * needs. In addition to modifying the active vmcs (which is vmcs02), this
7408 * function also has additional necessary side-effects, like setting various
7409 * vcpu->arch fields.
7411 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7413 struct vcpu_vmx *vmx = to_vmx(vcpu);
7416 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
7417 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
7418 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
7419 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
7420 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
7421 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
7422 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
7423 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
7424 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
7425 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
7426 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
7427 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
7428 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
7429 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
7430 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
7431 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
7432 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
7433 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
7434 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
7435 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
7436 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
7437 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
7438 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
7439 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
7440 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
7441 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
7442 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
7443 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
7444 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
7445 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
7446 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
7447 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
7448 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
7449 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
7450 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
7451 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
7453 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
7454 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
7455 vmcs12->vm_entry_intr_info_field);
7456 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
7457 vmcs12->vm_entry_exception_error_code);
7458 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
7459 vmcs12->vm_entry_instruction_len);
7460 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
7461 vmcs12->guest_interruptibility_info);
7462 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
7463 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
7464 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
7465 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
7466 vmcs12->guest_pending_dbg_exceptions);
7467 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
7468 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
7470 vmcs_write64(VMCS_LINK_POINTER, -1ull);
7472 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL,
7473 (vmcs_config.pin_based_exec_ctrl |
7474 vmcs12->pin_based_vm_exec_control));
7476 if (vmcs12->pin_based_vm_exec_control & PIN_BASED_VMX_PREEMPTION_TIMER)
7477 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE,
7478 vmcs12->vmx_preemption_timer_value);
7481 * Whether page-faults are trapped is determined by a combination of
7482 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
7483 * If enable_ept, L0 doesn't care about page faults and we should
7484 * set all of these to L1's desires. However, if !enable_ept, L0 does
7485 * care about (at least some) page faults, and because it is not easy
7486 * (if at all possible?) to merge L0 and L1's desires, we simply ask
7487 * to exit on each and every L2 page fault. This is done by setting
7488 * MASK=MATCH=0 and (see below) EB.PF=1.
7489 * Note that below we don't need special code to set EB.PF beyond the
7490 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
7491 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
7492 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
7494 * A problem with this approach (when !enable_ept) is that L1 may be
7495 * injected with more page faults than it asked for. This could have
7496 * caused problems, but in practice existing hypervisors don't care.
7497 * To fix this, we will need to emulate the PFEC checking (on the L1
7498 * page tables), using walk_addr(), when injecting PFs to L1.
7500 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
7501 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
7502 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
7503 enable_ept ? vmcs12->page_fault_error_code_match : 0);
7505 if (cpu_has_secondary_exec_ctrls()) {
7506 u32 exec_control = vmx_secondary_exec_control(vmx);
7507 if (!vmx->rdtscp_enabled)
7508 exec_control &= ~SECONDARY_EXEC_RDTSCP;
7509 /* Take the following fields only from vmcs12 */
7510 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7511 if (nested_cpu_has(vmcs12,
7512 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
7513 exec_control |= vmcs12->secondary_vm_exec_control;
7515 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
7517 * Translate L1 physical address to host physical
7518 * address for vmcs02. Keep the page pinned, so this
7519 * physical address remains valid. We keep a reference
7520 * to it so we can release it later.
7522 if (vmx->nested.apic_access_page) /* shouldn't happen */
7523 nested_release_page(vmx->nested.apic_access_page);
7524 vmx->nested.apic_access_page =
7525 nested_get_page(vcpu, vmcs12->apic_access_addr);
7527 * If translation failed, no matter: This feature asks
7528 * to exit when accessing the given address, and if it
7529 * can never be accessed, this feature won't do
7532 if (!vmx->nested.apic_access_page)
7534 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
7536 vmcs_write64(APIC_ACCESS_ADDR,
7537 page_to_phys(vmx->nested.apic_access_page));
7540 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
7545 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
7546 * Some constant fields are set here by vmx_set_constant_host_state().
7547 * Other fields are different per CPU, and will be set later when
7548 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
7550 vmx_set_constant_host_state(vmx);
7553 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
7554 * entry, but only if the current (host) sp changed from the value
7555 * we wrote last (vmx->host_rsp). This cache is no longer relevant
7556 * if we switch vmcs, and rather than hold a separate cache per vmcs,
7557 * here we just force the write to happen on entry.
7561 exec_control = vmx_exec_control(vmx); /* L0's desires */
7562 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
7563 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
7564 exec_control &= ~CPU_BASED_TPR_SHADOW;
7565 exec_control |= vmcs12->cpu_based_vm_exec_control;
7567 * Merging of IO and MSR bitmaps not currently supported.
7568 * Rather, exit every time.
7570 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
7571 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
7572 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
7574 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
7576 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
7577 * bitwise-or of what L1 wants to trap for L2, and what we want to
7578 * trap. Note that CR0.TS also needs updating - we do this later.
7580 update_exception_bitmap(vcpu);
7581 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
7582 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
7584 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
7585 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
7586 * bits are further modified by vmx_set_efer() below.
7588 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
7590 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
7591 * emulated by vmx_set_efer(), below.
7593 vmcs_write32(VM_ENTRY_CONTROLS,
7594 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
7595 ~VM_ENTRY_IA32E_MODE) |
7596 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
7598 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)
7599 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
7600 else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
7601 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
7604 set_cr4_guest_host_mask(vmx);
7606 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
7607 vmcs_write64(TSC_OFFSET,
7608 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
7610 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
7614 * Trivially support vpid by letting L2s share their parent
7615 * L1's vpid. TODO: move to a more elaborate solution, giving
7616 * each L2 its own vpid and exposing the vpid feature to L1.
7618 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
7619 vmx_flush_tlb(vcpu);
7622 if (nested_cpu_has_ept(vmcs12)) {
7623 kvm_mmu_unload(vcpu);
7624 nested_ept_init_mmu_context(vcpu);
7627 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
7628 vcpu->arch.efer = vmcs12->guest_ia32_efer;
7629 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
7630 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
7632 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
7633 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
7634 vmx_set_efer(vcpu, vcpu->arch.efer);
7637 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
7638 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
7639 * The CR0_READ_SHADOW is what L2 should have expected to read given
7640 * the specifications by L1; It's not enough to take
7641 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
7642 * have more bits than L1 expected.
7644 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
7645 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
7647 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
7648 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
7650 /* shadow page tables on either EPT or shadow page tables */
7651 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
7652 kvm_mmu_reset_context(vcpu);
7655 * L1 may access the L2's PDPTR, so save them to construct vmcs12
7658 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
7659 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
7660 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
7661 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
7664 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
7665 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
7669 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
7670 * for running an L2 nested guest.
7672 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
7674 struct vmcs12 *vmcs12;
7675 struct vcpu_vmx *vmx = to_vmx(vcpu);
7677 struct loaded_vmcs *vmcs02;
7680 if (!nested_vmx_check_permission(vcpu) ||
7681 !nested_vmx_check_vmcs12(vcpu))
7684 skip_emulated_instruction(vcpu);
7685 vmcs12 = get_vmcs12(vcpu);
7687 if (enable_shadow_vmcs)
7688 copy_shadow_to_vmcs12(vmx);
7691 * The nested entry process starts with enforcing various prerequisites
7692 * on vmcs12 as required by the Intel SDM, and act appropriately when
7693 * they fail: As the SDM explains, some conditions should cause the
7694 * instruction to fail, while others will cause the instruction to seem
7695 * to succeed, but return an EXIT_REASON_INVALID_STATE.
7696 * To speed up the normal (success) code path, we should avoid checking
7697 * for misconfigurations which will anyway be caught by the processor
7698 * when using the merged vmcs02.
7700 if (vmcs12->launch_state == launch) {
7701 nested_vmx_failValid(vcpu,
7702 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
7703 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
7707 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE) {
7708 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7712 if ((vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_MSR_BITMAPS) &&
7713 !IS_ALIGNED(vmcs12->msr_bitmap, PAGE_SIZE)) {
7714 /*TODO: Also verify bits beyond physical address width are 0*/
7715 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7719 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) &&
7720 !IS_ALIGNED(vmcs12->apic_access_addr, PAGE_SIZE)) {
7721 /*TODO: Also verify bits beyond physical address width are 0*/
7722 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7726 if (vmcs12->vm_entry_msr_load_count > 0 ||
7727 vmcs12->vm_exit_msr_load_count > 0 ||
7728 vmcs12->vm_exit_msr_store_count > 0) {
7729 pr_warn_ratelimited("%s: VMCS MSR_{LOAD,STORE} unsupported\n",
7731 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7735 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
7736 nested_vmx_procbased_ctls_low, nested_vmx_procbased_ctls_high) ||
7737 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
7738 nested_vmx_secondary_ctls_low, nested_vmx_secondary_ctls_high) ||
7739 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
7740 nested_vmx_pinbased_ctls_low, nested_vmx_pinbased_ctls_high) ||
7741 !vmx_control_verify(vmcs12->vm_exit_controls,
7742 nested_vmx_exit_ctls_low, nested_vmx_exit_ctls_high) ||
7743 !vmx_control_verify(vmcs12->vm_entry_controls,
7744 nested_vmx_entry_ctls_low, nested_vmx_entry_ctls_high))
7746 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
7750 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
7751 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
7752 nested_vmx_failValid(vcpu,
7753 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
7757 if (((vmcs12->guest_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
7758 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
7759 nested_vmx_entry_failure(vcpu, vmcs12,
7760 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
7763 if (vmcs12->vmcs_link_pointer != -1ull) {
7764 nested_vmx_entry_failure(vcpu, vmcs12,
7765 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
7770 * If the load IA32_EFER VM-entry control is 1, the following checks
7771 * are performed on the field for the IA32_EFER MSR:
7772 * - Bits reserved in the IA32_EFER MSR must be 0.
7773 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
7774 * the IA-32e mode guest VM-exit control. It must also be identical
7775 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
7778 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
7779 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
7780 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
7781 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
7782 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
7783 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
7784 nested_vmx_entry_failure(vcpu, vmcs12,
7785 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
7791 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
7792 * IA32_EFER MSR must be 0 in the field for that register. In addition,
7793 * the values of the LMA and LME bits in the field must each be that of
7794 * the host address-space size VM-exit control.
7796 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
7797 ia32e = (vmcs12->vm_exit_controls &
7798 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
7799 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
7800 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
7801 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
7802 nested_vmx_entry_failure(vcpu, vmcs12,
7803 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
7809 * We're finally done with prerequisite checking, and can start with
7813 vmcs02 = nested_get_current_vmcs02(vmx);
7817 enter_guest_mode(vcpu);
7819 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
7822 vmx->loaded_vmcs = vmcs02;
7824 vmx_vcpu_load(vcpu, cpu);
7828 vmx_segment_cache_clear(vmx);
7830 vmcs12->launch_state = 1;
7832 prepare_vmcs02(vcpu, vmcs12);
7835 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
7836 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
7837 * returned as far as L1 is concerned. It will only return (and set
7838 * the success flag) when L2 exits (see nested_vmx_vmexit()).
7844 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
7845 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
7846 * This function returns the new value we should put in vmcs12.guest_cr0.
7847 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
7848 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
7849 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
7850 * didn't trap the bit, because if L1 did, so would L0).
7851 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
7852 * been modified by L2, and L1 knows it. So just leave the old value of
7853 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
7854 * isn't relevant, because if L0 traps this bit it can set it to anything.
7855 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
7856 * changed these bits, and therefore they need to be updated, but L0
7857 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
7858 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
7860 static inline unsigned long
7861 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7864 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
7865 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
7866 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
7867 vcpu->arch.cr0_guest_owned_bits));
7870 static inline unsigned long
7871 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7874 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
7875 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
7876 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
7877 vcpu->arch.cr4_guest_owned_bits));
7880 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
7881 struct vmcs12 *vmcs12)
7886 if (vcpu->arch.exception.pending) {
7887 nr = vcpu->arch.exception.nr;
7888 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
7890 if (kvm_exception_is_soft(nr)) {
7891 vmcs12->vm_exit_instruction_len =
7892 vcpu->arch.event_exit_inst_len;
7893 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
7895 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
7897 if (vcpu->arch.exception.has_error_code) {
7898 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
7899 vmcs12->idt_vectoring_error_code =
7900 vcpu->arch.exception.error_code;
7903 vmcs12->idt_vectoring_info_field = idt_vectoring;
7904 } else if (vcpu->arch.nmi_pending) {
7905 vmcs12->idt_vectoring_info_field =
7906 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
7907 } else if (vcpu->arch.interrupt.pending) {
7908 nr = vcpu->arch.interrupt.nr;
7909 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
7911 if (vcpu->arch.interrupt.soft) {
7912 idt_vectoring |= INTR_TYPE_SOFT_INTR;
7913 vmcs12->vm_entry_instruction_len =
7914 vcpu->arch.event_exit_inst_len;
7916 idt_vectoring |= INTR_TYPE_EXT_INTR;
7918 vmcs12->idt_vectoring_info_field = idt_vectoring;
7923 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
7924 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
7925 * and this function updates it to reflect the changes to the guest state while
7926 * L2 was running (and perhaps made some exits which were handled directly by L0
7927 * without going back to L1), and to reflect the exit reason.
7928 * Note that we do not have to copy here all VMCS fields, just those that
7929 * could have changed by the L2 guest or the exit - i.e., the guest-state and
7930 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
7931 * which already writes to vmcs12 directly.
7933 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
7935 /* update guest state fields: */
7936 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
7937 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
7939 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
7940 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
7941 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
7942 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
7944 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
7945 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
7946 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
7947 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
7948 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
7949 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
7950 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
7951 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
7952 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
7953 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
7954 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
7955 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
7956 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
7957 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
7958 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
7959 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
7960 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
7961 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
7962 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
7963 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
7964 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
7965 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
7966 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
7967 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
7968 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
7969 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
7970 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
7971 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
7972 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
7973 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
7974 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
7975 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
7976 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
7977 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
7978 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
7979 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
7981 vmcs12->guest_interruptibility_info =
7982 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
7983 vmcs12->guest_pending_dbg_exceptions =
7984 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
7987 * In some cases (usually, nested EPT), L2 is allowed to change its
7988 * own CR3 without exiting. If it has changed it, we must keep it.
7989 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
7990 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
7992 * Additionally, restore L2's PDPTR to vmcs12.
7995 vmcs12->guest_cr3 = vmcs_read64(GUEST_CR3);
7996 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
7997 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
7998 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
7999 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
8002 vmcs12->vm_entry_controls =
8003 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
8004 (vmcs_read32(VM_ENTRY_CONTROLS) & VM_ENTRY_IA32E_MODE);
8006 /* TODO: These cannot have changed unless we have MSR bitmaps and
8007 * the relevant bit asks not to trap the change */
8008 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
8009 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
8010 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
8011 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
8012 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
8013 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
8015 /* update exit information fields: */
8017 vmcs12->vm_exit_reason = to_vmx(vcpu)->exit_reason;
8018 vmcs12->exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8020 vmcs12->vm_exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8021 if ((vmcs12->vm_exit_intr_info &
8022 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
8023 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
8024 vmcs12->vm_exit_intr_error_code =
8025 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
8026 vmcs12->idt_vectoring_info_field = 0;
8027 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
8028 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
8030 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
8031 /* vm_entry_intr_info_field is cleared on exit. Emulate this
8032 * instead of reading the real value. */
8033 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
8036 * Transfer the event that L0 or L1 may wanted to inject into
8037 * L2 to IDT_VECTORING_INFO_FIELD.
8039 vmcs12_save_pending_event(vcpu, vmcs12);
8043 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
8044 * preserved above and would only end up incorrectly in L1.
8046 vcpu->arch.nmi_injected = false;
8047 kvm_clear_exception_queue(vcpu);
8048 kvm_clear_interrupt_queue(vcpu);
8052 * A part of what we need to when the nested L2 guest exits and we want to
8053 * run its L1 parent, is to reset L1's guest state to the host state specified
8055 * This function is to be called not only on normal nested exit, but also on
8056 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
8057 * Failures During or After Loading Guest State").
8058 * This function should be called when the active VMCS is L1's (vmcs01).
8060 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
8061 struct vmcs12 *vmcs12)
8063 struct kvm_segment seg;
8065 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
8066 vcpu->arch.efer = vmcs12->host_ia32_efer;
8067 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
8068 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
8070 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
8071 vmx_set_efer(vcpu, vcpu->arch.efer);
8073 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
8074 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
8075 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
8077 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
8078 * actually changed, because it depends on the current state of
8079 * fpu_active (which may have changed).
8080 * Note that vmx_set_cr0 refers to efer set above.
8082 kvm_set_cr0(vcpu, vmcs12->host_cr0);
8084 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
8085 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
8086 * but we also need to update cr0_guest_host_mask and exception_bitmap.
8088 update_exception_bitmap(vcpu);
8089 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
8090 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
8093 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
8094 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
8096 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
8097 kvm_set_cr4(vcpu, vmcs12->host_cr4);
8099 if (nested_cpu_has_ept(vmcs12))
8100 nested_ept_uninit_mmu_context(vcpu);
8102 kvm_set_cr3(vcpu, vmcs12->host_cr3);
8103 kvm_mmu_reset_context(vcpu);
8107 * Trivially support vpid by letting L2s share their parent
8108 * L1's vpid. TODO: move to a more elaborate solution, giving
8109 * each L2 its own vpid and exposing the vpid feature to L1.
8111 vmx_flush_tlb(vcpu);
8115 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
8116 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
8117 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
8118 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
8119 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
8121 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT)
8122 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
8123 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8124 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
8125 vmcs12->host_ia32_perf_global_ctrl);
8127 /* Set L1 segment info according to Intel SDM
8128 27.5.2 Loading Host Segment and Descriptor-Table Registers */
8129 seg = (struct kvm_segment) {
8131 .limit = 0xFFFFFFFF,
8132 .selector = vmcs12->host_cs_selector,
8138 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
8142 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
8143 seg = (struct kvm_segment) {
8145 .limit = 0xFFFFFFFF,
8152 seg.selector = vmcs12->host_ds_selector;
8153 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
8154 seg.selector = vmcs12->host_es_selector;
8155 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
8156 seg.selector = vmcs12->host_ss_selector;
8157 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
8158 seg.selector = vmcs12->host_fs_selector;
8159 seg.base = vmcs12->host_fs_base;
8160 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
8161 seg.selector = vmcs12->host_gs_selector;
8162 seg.base = vmcs12->host_gs_base;
8163 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
8164 seg = (struct kvm_segment) {
8165 .base = vmcs12->host_tr_base,
8167 .selector = vmcs12->host_tr_selector,
8171 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
8173 kvm_set_dr(vcpu, 7, 0x400);
8174 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
8178 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
8179 * and modify vmcs12 to make it see what it would expect to see there if
8180 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
8182 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu)
8184 struct vcpu_vmx *vmx = to_vmx(vcpu);
8186 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8188 /* trying to cancel vmlaunch/vmresume is a bug */
8189 WARN_ON_ONCE(vmx->nested.nested_run_pending);
8191 leave_guest_mode(vcpu);
8192 prepare_vmcs12(vcpu, vmcs12);
8195 vmx->loaded_vmcs = &vmx->vmcs01;
8197 vmx_vcpu_load(vcpu, cpu);
8201 vmx_segment_cache_clear(vmx);
8203 /* if no vmcs02 cache requested, remove the one we used */
8204 if (VMCS02_POOL_SIZE == 0)
8205 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
8207 load_vmcs12_host_state(vcpu, vmcs12);
8209 /* Update TSC_OFFSET if TSC was changed while L2 ran */
8210 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
8212 /* This is needed for same reason as it was needed in prepare_vmcs02 */
8215 /* Unpin physical memory we referred to in vmcs02 */
8216 if (vmx->nested.apic_access_page) {
8217 nested_release_page(vmx->nested.apic_access_page);
8218 vmx->nested.apic_access_page = 0;
8222 * Exiting from L2 to L1, we're now back to L1 which thinks it just
8223 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
8224 * success or failure flag accordingly.
8226 if (unlikely(vmx->fail)) {
8228 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
8230 nested_vmx_succeed(vcpu);
8231 if (enable_shadow_vmcs)
8232 vmx->nested.sync_shadow_vmcs = true;
8236 * L1's failure to enter L2 is a subset of a normal exit, as explained in
8237 * 23.7 "VM-entry failures during or after loading guest state" (this also
8238 * lists the acceptable exit-reason and exit-qualification parameters).
8239 * It should only be called before L2 actually succeeded to run, and when
8240 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
8242 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
8243 struct vmcs12 *vmcs12,
8244 u32 reason, unsigned long qualification)
8246 load_vmcs12_host_state(vcpu, vmcs12);
8247 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
8248 vmcs12->exit_qualification = qualification;
8249 nested_vmx_succeed(vcpu);
8250 if (enable_shadow_vmcs)
8251 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
8254 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
8255 struct x86_instruction_info *info,
8256 enum x86_intercept_stage stage)
8258 return X86EMUL_CONTINUE;
8261 static struct kvm_x86_ops vmx_x86_ops = {
8262 .cpu_has_kvm_support = cpu_has_kvm_support,
8263 .disabled_by_bios = vmx_disabled_by_bios,
8264 .hardware_setup = hardware_setup,
8265 .hardware_unsetup = hardware_unsetup,
8266 .check_processor_compatibility = vmx_check_processor_compat,
8267 .hardware_enable = hardware_enable,
8268 .hardware_disable = hardware_disable,
8269 .cpu_has_accelerated_tpr = report_flexpriority,
8271 .vcpu_create = vmx_create_vcpu,
8272 .vcpu_free = vmx_free_vcpu,
8273 .vcpu_reset = vmx_vcpu_reset,
8275 .prepare_guest_switch = vmx_save_host_state,
8276 .vcpu_load = vmx_vcpu_load,
8277 .vcpu_put = vmx_vcpu_put,
8279 .update_db_bp_intercept = update_exception_bitmap,
8280 .get_msr = vmx_get_msr,
8281 .set_msr = vmx_set_msr,
8282 .get_segment_base = vmx_get_segment_base,
8283 .get_segment = vmx_get_segment,
8284 .set_segment = vmx_set_segment,
8285 .get_cpl = vmx_get_cpl,
8286 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
8287 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
8288 .decache_cr3 = vmx_decache_cr3,
8289 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
8290 .set_cr0 = vmx_set_cr0,
8291 .set_cr3 = vmx_set_cr3,
8292 .set_cr4 = vmx_set_cr4,
8293 .set_efer = vmx_set_efer,
8294 .get_idt = vmx_get_idt,
8295 .set_idt = vmx_set_idt,
8296 .get_gdt = vmx_get_gdt,
8297 .set_gdt = vmx_set_gdt,
8298 .set_dr7 = vmx_set_dr7,
8299 .cache_reg = vmx_cache_reg,
8300 .get_rflags = vmx_get_rflags,
8301 .set_rflags = vmx_set_rflags,
8302 .fpu_activate = vmx_fpu_activate,
8303 .fpu_deactivate = vmx_fpu_deactivate,
8305 .tlb_flush = vmx_flush_tlb,
8307 .run = vmx_vcpu_run,
8308 .handle_exit = vmx_handle_exit,
8309 .skip_emulated_instruction = skip_emulated_instruction,
8310 .set_interrupt_shadow = vmx_set_interrupt_shadow,
8311 .get_interrupt_shadow = vmx_get_interrupt_shadow,
8312 .patch_hypercall = vmx_patch_hypercall,
8313 .set_irq = vmx_inject_irq,
8314 .set_nmi = vmx_inject_nmi,
8315 .queue_exception = vmx_queue_exception,
8316 .cancel_injection = vmx_cancel_injection,
8317 .interrupt_allowed = vmx_interrupt_allowed,
8318 .nmi_allowed = vmx_nmi_allowed,
8319 .get_nmi_mask = vmx_get_nmi_mask,
8320 .set_nmi_mask = vmx_set_nmi_mask,
8321 .enable_nmi_window = enable_nmi_window,
8322 .enable_irq_window = enable_irq_window,
8323 .update_cr8_intercept = update_cr8_intercept,
8324 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
8325 .vm_has_apicv = vmx_vm_has_apicv,
8326 .load_eoi_exitmap = vmx_load_eoi_exitmap,
8327 .hwapic_irr_update = vmx_hwapic_irr_update,
8328 .hwapic_isr_update = vmx_hwapic_isr_update,
8329 .sync_pir_to_irr = vmx_sync_pir_to_irr,
8330 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
8332 .set_tss_addr = vmx_set_tss_addr,
8333 .get_tdp_level = get_ept_level,
8334 .get_mt_mask = vmx_get_mt_mask,
8336 .get_exit_info = vmx_get_exit_info,
8338 .get_lpage_level = vmx_get_lpage_level,
8340 .cpuid_update = vmx_cpuid_update,
8342 .rdtscp_supported = vmx_rdtscp_supported,
8343 .invpcid_supported = vmx_invpcid_supported,
8345 .set_supported_cpuid = vmx_set_supported_cpuid,
8347 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
8349 .set_tsc_khz = vmx_set_tsc_khz,
8350 .read_tsc_offset = vmx_read_tsc_offset,
8351 .write_tsc_offset = vmx_write_tsc_offset,
8352 .adjust_tsc_offset = vmx_adjust_tsc_offset,
8353 .compute_tsc_offset = vmx_compute_tsc_offset,
8354 .read_l1_tsc = vmx_read_l1_tsc,
8356 .set_tdp_cr3 = vmx_set_cr3,
8358 .check_intercept = vmx_check_intercept,
8359 .handle_external_intr = vmx_handle_external_intr,
8362 static int __init vmx_init(void)
8366 rdmsrl_safe(MSR_EFER, &host_efer);
8368 for (i = 0; i < NR_VMX_MSR; ++i)
8369 kvm_define_shared_msr(i, vmx_msr_index[i]);
8371 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
8372 if (!vmx_io_bitmap_a)
8377 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
8378 if (!vmx_io_bitmap_b)
8381 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
8382 if (!vmx_msr_bitmap_legacy)
8385 vmx_msr_bitmap_legacy_x2apic =
8386 (unsigned long *)__get_free_page(GFP_KERNEL);
8387 if (!vmx_msr_bitmap_legacy_x2apic)
8390 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
8391 if (!vmx_msr_bitmap_longmode)
8394 vmx_msr_bitmap_longmode_x2apic =
8395 (unsigned long *)__get_free_page(GFP_KERNEL);
8396 if (!vmx_msr_bitmap_longmode_x2apic)
8398 vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
8399 if (!vmx_vmread_bitmap)
8402 vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
8403 if (!vmx_vmwrite_bitmap)
8406 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
8407 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
8408 /* shadowed read/write fields */
8409 for (i = 0; i < max_shadow_read_write_fields; i++) {
8410 clear_bit(shadow_read_write_fields[i], vmx_vmwrite_bitmap);
8411 clear_bit(shadow_read_write_fields[i], vmx_vmread_bitmap);
8413 /* shadowed read only fields */
8414 for (i = 0; i < max_shadow_read_only_fields; i++)
8415 clear_bit(shadow_read_only_fields[i], vmx_vmread_bitmap);
8418 * Allow direct access to the PC debug port (it is often used for I/O
8419 * delays, but the vmexits simply slow things down).
8421 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
8422 clear_bit(0x80, vmx_io_bitmap_a);
8424 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
8426 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
8427 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
8429 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
8431 r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
8432 __alignof__(struct vcpu_vmx), THIS_MODULE);
8437 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
8438 crash_vmclear_local_loaded_vmcss);
8441 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
8442 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
8443 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
8444 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
8445 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
8446 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
8447 memcpy(vmx_msr_bitmap_legacy_x2apic,
8448 vmx_msr_bitmap_legacy, PAGE_SIZE);
8449 memcpy(vmx_msr_bitmap_longmode_x2apic,
8450 vmx_msr_bitmap_longmode, PAGE_SIZE);
8453 for (msr = 0x800; msr <= 0x8ff; msr++)
8454 vmx_disable_intercept_msr_read_x2apic(msr);
8456 /* According SDM, in x2apic mode, the whole id reg is used.
8457 * But in KVM, it only use the highest eight bits. Need to
8459 vmx_enable_intercept_msr_read_x2apic(0x802);
8461 vmx_enable_intercept_msr_read_x2apic(0x839);
8463 vmx_disable_intercept_msr_write_x2apic(0x808);
8465 vmx_disable_intercept_msr_write_x2apic(0x80b);
8467 vmx_disable_intercept_msr_write_x2apic(0x83f);
8471 kvm_mmu_set_mask_ptes(0ull,
8472 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
8473 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
8474 0ull, VMX_EPT_EXECUTABLE_MASK);
8475 ept_set_mmio_spte_mask();
8483 free_page((unsigned long)vmx_vmwrite_bitmap);
8485 free_page((unsigned long)vmx_vmread_bitmap);
8487 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
8489 free_page((unsigned long)vmx_msr_bitmap_longmode);
8491 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
8493 free_page((unsigned long)vmx_msr_bitmap_legacy);
8495 free_page((unsigned long)vmx_io_bitmap_b);
8497 free_page((unsigned long)vmx_io_bitmap_a);
8501 static void __exit vmx_exit(void)
8503 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
8504 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
8505 free_page((unsigned long)vmx_msr_bitmap_legacy);
8506 free_page((unsigned long)vmx_msr_bitmap_longmode);
8507 free_page((unsigned long)vmx_io_bitmap_b);
8508 free_page((unsigned long)vmx_io_bitmap_a);
8509 free_page((unsigned long)vmx_vmwrite_bitmap);
8510 free_page((unsigned long)vmx_vmread_bitmap);
8513 rcu_assign_pointer(crash_vmclear_loaded_vmcss, NULL);
8520 module_init(vmx_init)
8521 module_exit(vmx_exit)