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.
24 #include <linux/kvm_host.h>
25 #include <linux/module.h>
26 #include <linux/kernel.h>
28 #include <linux/highmem.h>
29 #include <linux/sched.h>
30 #include <linux/moduleparam.h>
31 #include <linux/mod_devicetable.h>
32 #include <linux/trace_events.h>
33 #include <linux/slab.h>
34 #include <linux/tboot.h>
35 #include <linux/hrtimer.h>
36 #include "kvm_cache_regs.h"
43 #include <asm/virtext.h>
45 #include <asm/fpu/internal.h>
46 #include <asm/perf_event.h>
47 #include <asm/debugreg.h>
48 #include <asm/kexec.h>
50 #include <asm/irq_remapping.h>
55 #define __ex(x) __kvm_handle_fault_on_reboot(x)
56 #define __ex_clear(x, reg) \
57 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
59 MODULE_AUTHOR("Qumranet");
60 MODULE_LICENSE("GPL");
62 static const struct x86_cpu_id vmx_cpu_id[] = {
63 X86_FEATURE_MATCH(X86_FEATURE_VMX),
66 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
68 static bool __read_mostly enable_vpid = 1;
69 module_param_named(vpid, enable_vpid, bool, 0444);
71 static bool __read_mostly flexpriority_enabled = 1;
72 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
74 static bool __read_mostly enable_ept = 1;
75 module_param_named(ept, enable_ept, bool, S_IRUGO);
77 static bool __read_mostly enable_unrestricted_guest = 1;
78 module_param_named(unrestricted_guest,
79 enable_unrestricted_guest, bool, S_IRUGO);
81 static bool __read_mostly enable_ept_ad_bits = 1;
82 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
84 static bool __read_mostly emulate_invalid_guest_state = true;
85 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
87 static bool __read_mostly vmm_exclusive = 1;
88 module_param(vmm_exclusive, bool, S_IRUGO);
90 static bool __read_mostly fasteoi = 1;
91 module_param(fasteoi, bool, S_IRUGO);
93 static bool __read_mostly enable_apicv = 1;
94 module_param(enable_apicv, bool, S_IRUGO);
96 static bool __read_mostly enable_shadow_vmcs = 1;
97 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
99 * If nested=1, nested virtualization is supported, i.e., guests may use
100 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
101 * use VMX instructions.
103 static bool __read_mostly nested = 0;
104 module_param(nested, bool, S_IRUGO);
106 static u64 __read_mostly host_xss;
108 static bool __read_mostly enable_pml = 1;
109 module_param_named(pml, enable_pml, bool, S_IRUGO);
111 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
113 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
114 static int __read_mostly cpu_preemption_timer_multi;
115 static bool __read_mostly enable_preemption_timer = 1;
117 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
120 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
121 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
122 #define KVM_VM_CR0_ALWAYS_ON \
123 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
124 #define KVM_CR4_GUEST_OWNED_BITS \
125 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
126 | X86_CR4_OSXMMEXCPT | X86_CR4_TSD)
128 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
129 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
131 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
133 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
136 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
137 * ple_gap: upper bound on the amount of time between two successive
138 * executions of PAUSE in a loop. Also indicate if ple enabled.
139 * According to test, this time is usually smaller than 128 cycles.
140 * ple_window: upper bound on the amount of time a guest is allowed to execute
141 * in a PAUSE loop. Tests indicate that most spinlocks are held for
142 * less than 2^12 cycles
143 * Time is measured based on a counter that runs at the same rate as the TSC,
144 * refer SDM volume 3b section 21.6.13 & 22.1.3.
146 #define KVM_VMX_DEFAULT_PLE_GAP 128
147 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
148 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
149 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
150 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
151 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
153 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
154 module_param(ple_gap, int, S_IRUGO);
156 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
157 module_param(ple_window, int, S_IRUGO);
159 /* Default doubles per-vcpu window every exit. */
160 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
161 module_param(ple_window_grow, int, S_IRUGO);
163 /* Default resets per-vcpu window every exit to ple_window. */
164 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
165 module_param(ple_window_shrink, int, S_IRUGO);
167 /* Default is to compute the maximum so we can never overflow. */
168 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
169 static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
170 module_param(ple_window_max, int, S_IRUGO);
172 extern const ulong vmx_return;
174 #define NR_AUTOLOAD_MSRS 8
175 #define VMCS02_POOL_SIZE 1
184 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
185 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
186 * loaded on this CPU (so we can clear them if the CPU goes down).
192 struct list_head loaded_vmcss_on_cpu_link;
195 struct shared_msr_entry {
202 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
203 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
204 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
205 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
206 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
207 * More than one of these structures may exist, if L1 runs multiple L2 guests.
208 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
209 * underlying hardware which will be used to run L2.
210 * This structure is packed to ensure that its layout is identical across
211 * machines (necessary for live migration).
212 * If there are changes in this struct, VMCS12_REVISION must be changed.
214 typedef u64 natural_width;
215 struct __packed vmcs12 {
216 /* According to the Intel spec, a VMCS region must start with the
217 * following two fields. Then follow implementation-specific data.
222 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
223 u32 padding[7]; /* room for future expansion */
228 u64 vm_exit_msr_store_addr;
229 u64 vm_exit_msr_load_addr;
230 u64 vm_entry_msr_load_addr;
232 u64 virtual_apic_page_addr;
233 u64 apic_access_addr;
234 u64 posted_intr_desc_addr;
236 u64 eoi_exit_bitmap0;
237 u64 eoi_exit_bitmap1;
238 u64 eoi_exit_bitmap2;
239 u64 eoi_exit_bitmap3;
241 u64 guest_physical_address;
242 u64 vmcs_link_pointer;
243 u64 guest_ia32_debugctl;
246 u64 guest_ia32_perf_global_ctrl;
254 u64 host_ia32_perf_global_ctrl;
255 u64 padding64[8]; /* room for future expansion */
257 * To allow migration of L1 (complete with its L2 guests) between
258 * machines of different natural widths (32 or 64 bit), we cannot have
259 * unsigned long fields with no explict size. We use u64 (aliased
260 * natural_width) instead. Luckily, x86 is little-endian.
262 natural_width cr0_guest_host_mask;
263 natural_width cr4_guest_host_mask;
264 natural_width cr0_read_shadow;
265 natural_width cr4_read_shadow;
266 natural_width cr3_target_value0;
267 natural_width cr3_target_value1;
268 natural_width cr3_target_value2;
269 natural_width cr3_target_value3;
270 natural_width exit_qualification;
271 natural_width guest_linear_address;
272 natural_width guest_cr0;
273 natural_width guest_cr3;
274 natural_width guest_cr4;
275 natural_width guest_es_base;
276 natural_width guest_cs_base;
277 natural_width guest_ss_base;
278 natural_width guest_ds_base;
279 natural_width guest_fs_base;
280 natural_width guest_gs_base;
281 natural_width guest_ldtr_base;
282 natural_width guest_tr_base;
283 natural_width guest_gdtr_base;
284 natural_width guest_idtr_base;
285 natural_width guest_dr7;
286 natural_width guest_rsp;
287 natural_width guest_rip;
288 natural_width guest_rflags;
289 natural_width guest_pending_dbg_exceptions;
290 natural_width guest_sysenter_esp;
291 natural_width guest_sysenter_eip;
292 natural_width host_cr0;
293 natural_width host_cr3;
294 natural_width host_cr4;
295 natural_width host_fs_base;
296 natural_width host_gs_base;
297 natural_width host_tr_base;
298 natural_width host_gdtr_base;
299 natural_width host_idtr_base;
300 natural_width host_ia32_sysenter_esp;
301 natural_width host_ia32_sysenter_eip;
302 natural_width host_rsp;
303 natural_width host_rip;
304 natural_width paddingl[8]; /* room for future expansion */
305 u32 pin_based_vm_exec_control;
306 u32 cpu_based_vm_exec_control;
307 u32 exception_bitmap;
308 u32 page_fault_error_code_mask;
309 u32 page_fault_error_code_match;
310 u32 cr3_target_count;
311 u32 vm_exit_controls;
312 u32 vm_exit_msr_store_count;
313 u32 vm_exit_msr_load_count;
314 u32 vm_entry_controls;
315 u32 vm_entry_msr_load_count;
316 u32 vm_entry_intr_info_field;
317 u32 vm_entry_exception_error_code;
318 u32 vm_entry_instruction_len;
320 u32 secondary_vm_exec_control;
321 u32 vm_instruction_error;
323 u32 vm_exit_intr_info;
324 u32 vm_exit_intr_error_code;
325 u32 idt_vectoring_info_field;
326 u32 idt_vectoring_error_code;
327 u32 vm_exit_instruction_len;
328 u32 vmx_instruction_info;
335 u32 guest_ldtr_limit;
337 u32 guest_gdtr_limit;
338 u32 guest_idtr_limit;
339 u32 guest_es_ar_bytes;
340 u32 guest_cs_ar_bytes;
341 u32 guest_ss_ar_bytes;
342 u32 guest_ds_ar_bytes;
343 u32 guest_fs_ar_bytes;
344 u32 guest_gs_ar_bytes;
345 u32 guest_ldtr_ar_bytes;
346 u32 guest_tr_ar_bytes;
347 u32 guest_interruptibility_info;
348 u32 guest_activity_state;
349 u32 guest_sysenter_cs;
350 u32 host_ia32_sysenter_cs;
351 u32 vmx_preemption_timer_value;
352 u32 padding32[7]; /* room for future expansion */
353 u16 virtual_processor_id;
355 u16 guest_es_selector;
356 u16 guest_cs_selector;
357 u16 guest_ss_selector;
358 u16 guest_ds_selector;
359 u16 guest_fs_selector;
360 u16 guest_gs_selector;
361 u16 guest_ldtr_selector;
362 u16 guest_tr_selector;
363 u16 guest_intr_status;
364 u16 host_es_selector;
365 u16 host_cs_selector;
366 u16 host_ss_selector;
367 u16 host_ds_selector;
368 u16 host_fs_selector;
369 u16 host_gs_selector;
370 u16 host_tr_selector;
374 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
375 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
376 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
378 #define VMCS12_REVISION 0x11e57ed0
381 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
382 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
383 * current implementation, 4K are reserved to avoid future complications.
385 #define VMCS12_SIZE 0x1000
387 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
389 struct list_head list;
391 struct loaded_vmcs vmcs02;
395 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
396 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
399 /* Has the level1 guest done vmxon? */
403 /* The guest-physical address of the current VMCS L1 keeps for L2 */
405 /* The host-usable pointer to the above */
406 struct page *current_vmcs12_page;
407 struct vmcs12 *current_vmcs12;
408 struct vmcs *current_shadow_vmcs;
410 * Indicates if the shadow vmcs must be updated with the
411 * data hold by vmcs12
413 bool sync_shadow_vmcs;
415 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
416 struct list_head vmcs02_pool;
418 u64 vmcs01_tsc_offset;
419 /* L2 must run next, and mustn't decide to exit to L1. */
420 bool nested_run_pending;
422 * Guest pages referred to in vmcs02 with host-physical pointers, so
423 * we must keep them pinned while L2 runs.
425 struct page *apic_access_page;
426 struct page *virtual_apic_page;
427 struct page *pi_desc_page;
428 struct pi_desc *pi_desc;
432 struct hrtimer preemption_timer;
433 bool preemption_timer_expired;
435 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
441 u32 nested_vmx_procbased_ctls_low;
442 u32 nested_vmx_procbased_ctls_high;
443 u32 nested_vmx_true_procbased_ctls_low;
444 u32 nested_vmx_secondary_ctls_low;
445 u32 nested_vmx_secondary_ctls_high;
446 u32 nested_vmx_pinbased_ctls_low;
447 u32 nested_vmx_pinbased_ctls_high;
448 u32 nested_vmx_exit_ctls_low;
449 u32 nested_vmx_exit_ctls_high;
450 u32 nested_vmx_true_exit_ctls_low;
451 u32 nested_vmx_entry_ctls_low;
452 u32 nested_vmx_entry_ctls_high;
453 u32 nested_vmx_true_entry_ctls_low;
454 u32 nested_vmx_misc_low;
455 u32 nested_vmx_misc_high;
456 u32 nested_vmx_ept_caps;
457 u32 nested_vmx_vpid_caps;
460 #define POSTED_INTR_ON 0
461 #define POSTED_INTR_SN 1
463 /* Posted-Interrupt Descriptor */
465 u32 pir[8]; /* Posted interrupt requested */
468 /* bit 256 - Outstanding Notification */
470 /* bit 257 - Suppress Notification */
472 /* bit 271:258 - Reserved */
474 /* bit 279:272 - Notification Vector */
476 /* bit 287:280 - Reserved */
478 /* bit 319:288 - Notification Destination */
486 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
488 return test_and_set_bit(POSTED_INTR_ON,
489 (unsigned long *)&pi_desc->control);
492 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
494 return test_and_clear_bit(POSTED_INTR_ON,
495 (unsigned long *)&pi_desc->control);
498 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
500 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
503 static inline void pi_clear_sn(struct pi_desc *pi_desc)
505 return clear_bit(POSTED_INTR_SN,
506 (unsigned long *)&pi_desc->control);
509 static inline void pi_set_sn(struct pi_desc *pi_desc)
511 return set_bit(POSTED_INTR_SN,
512 (unsigned long *)&pi_desc->control);
515 static inline int pi_test_on(struct pi_desc *pi_desc)
517 return test_bit(POSTED_INTR_ON,
518 (unsigned long *)&pi_desc->control);
521 static inline int pi_test_sn(struct pi_desc *pi_desc)
523 return test_bit(POSTED_INTR_SN,
524 (unsigned long *)&pi_desc->control);
528 struct kvm_vcpu vcpu;
529 unsigned long host_rsp;
531 bool nmi_known_unmasked;
533 u32 idt_vectoring_info;
535 struct shared_msr_entry *guest_msrs;
538 unsigned long host_idt_base;
540 u64 msr_host_kernel_gs_base;
541 u64 msr_guest_kernel_gs_base;
543 u32 vm_entry_controls_shadow;
544 u32 vm_exit_controls_shadow;
546 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
547 * non-nested (L1) guest, it always points to vmcs01. For a nested
548 * guest (L2), it points to a different VMCS.
550 struct loaded_vmcs vmcs01;
551 struct loaded_vmcs *loaded_vmcs;
552 bool __launched; /* temporary, used in vmx_vcpu_run */
553 struct msr_autoload {
555 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
556 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
560 u16 fs_sel, gs_sel, ldt_sel;
564 int gs_ldt_reload_needed;
565 int fs_reload_needed;
566 u64 msr_host_bndcfgs;
567 unsigned long vmcs_host_cr4; /* May not match real cr4 */
572 struct kvm_segment segs[8];
575 u32 bitmask; /* 4 bits per segment (1 bit per field) */
576 struct kvm_save_segment {
584 bool emulation_required;
586 /* Support for vnmi-less CPUs */
587 int soft_vnmi_blocked;
589 s64 vnmi_blocked_time;
592 /* Posted interrupt descriptor */
593 struct pi_desc pi_desc;
595 /* Support for a guest hypervisor (nested VMX) */
596 struct nested_vmx nested;
598 /* Dynamic PLE window. */
600 bool ple_window_dirty;
602 /* Support for PML */
603 #define PML_ENTITY_NUM 512
606 /* apic deadline value in host tsc */
609 u64 current_tsc_ratio;
611 bool guest_pkru_valid;
616 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
617 * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
618 * in msr_ia32_feature_control_valid_bits.
620 u64 msr_ia32_feature_control;
621 u64 msr_ia32_feature_control_valid_bits;
624 enum segment_cache_field {
633 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
635 return container_of(vcpu, struct vcpu_vmx, vcpu);
638 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
640 return &(to_vmx(vcpu)->pi_desc);
643 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
644 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
645 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
646 [number##_HIGH] = VMCS12_OFFSET(name)+4
649 static unsigned long shadow_read_only_fields[] = {
651 * We do NOT shadow fields that are modified when L0
652 * traps and emulates any vmx instruction (e.g. VMPTRLD,
653 * VMXON...) executed by L1.
654 * For example, VM_INSTRUCTION_ERROR is read
655 * by L1 if a vmx instruction fails (part of the error path).
656 * Note the code assumes this logic. If for some reason
657 * we start shadowing these fields then we need to
658 * force a shadow sync when L0 emulates vmx instructions
659 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
660 * by nested_vmx_failValid)
664 VM_EXIT_INSTRUCTION_LEN,
665 IDT_VECTORING_INFO_FIELD,
666 IDT_VECTORING_ERROR_CODE,
667 VM_EXIT_INTR_ERROR_CODE,
669 GUEST_LINEAR_ADDRESS,
670 GUEST_PHYSICAL_ADDRESS
672 static int max_shadow_read_only_fields =
673 ARRAY_SIZE(shadow_read_only_fields);
675 static unsigned long shadow_read_write_fields[] = {
682 GUEST_INTERRUPTIBILITY_INFO,
695 CPU_BASED_VM_EXEC_CONTROL,
696 VM_ENTRY_EXCEPTION_ERROR_CODE,
697 VM_ENTRY_INTR_INFO_FIELD,
698 VM_ENTRY_INSTRUCTION_LEN,
699 VM_ENTRY_EXCEPTION_ERROR_CODE,
705 static int max_shadow_read_write_fields =
706 ARRAY_SIZE(shadow_read_write_fields);
708 static const unsigned short vmcs_field_to_offset_table[] = {
709 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
710 FIELD(POSTED_INTR_NV, posted_intr_nv),
711 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
712 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
713 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
714 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
715 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
716 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
717 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
718 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
719 FIELD(GUEST_INTR_STATUS, guest_intr_status),
720 FIELD(HOST_ES_SELECTOR, host_es_selector),
721 FIELD(HOST_CS_SELECTOR, host_cs_selector),
722 FIELD(HOST_SS_SELECTOR, host_ss_selector),
723 FIELD(HOST_DS_SELECTOR, host_ds_selector),
724 FIELD(HOST_FS_SELECTOR, host_fs_selector),
725 FIELD(HOST_GS_SELECTOR, host_gs_selector),
726 FIELD(HOST_TR_SELECTOR, host_tr_selector),
727 FIELD64(IO_BITMAP_A, io_bitmap_a),
728 FIELD64(IO_BITMAP_B, io_bitmap_b),
729 FIELD64(MSR_BITMAP, msr_bitmap),
730 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
731 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
732 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
733 FIELD64(TSC_OFFSET, tsc_offset),
734 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
735 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
736 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
737 FIELD64(EPT_POINTER, ept_pointer),
738 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
739 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
740 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
741 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
742 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
743 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
744 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
745 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
746 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
747 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
748 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
749 FIELD64(GUEST_PDPTR0, guest_pdptr0),
750 FIELD64(GUEST_PDPTR1, guest_pdptr1),
751 FIELD64(GUEST_PDPTR2, guest_pdptr2),
752 FIELD64(GUEST_PDPTR3, guest_pdptr3),
753 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
754 FIELD64(HOST_IA32_PAT, host_ia32_pat),
755 FIELD64(HOST_IA32_EFER, host_ia32_efer),
756 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
757 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
758 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
759 FIELD(EXCEPTION_BITMAP, exception_bitmap),
760 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
761 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
762 FIELD(CR3_TARGET_COUNT, cr3_target_count),
763 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
764 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
765 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
766 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
767 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
768 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
769 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
770 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
771 FIELD(TPR_THRESHOLD, tpr_threshold),
772 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
773 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
774 FIELD(VM_EXIT_REASON, vm_exit_reason),
775 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
776 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
777 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
778 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
779 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
780 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
781 FIELD(GUEST_ES_LIMIT, guest_es_limit),
782 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
783 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
784 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
785 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
786 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
787 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
788 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
789 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
790 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
791 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
792 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
793 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
794 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
795 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
796 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
797 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
798 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
799 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
800 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
801 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
802 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
803 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
804 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
805 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
806 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
807 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
808 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
809 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
810 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
811 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
812 FIELD(EXIT_QUALIFICATION, exit_qualification),
813 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
814 FIELD(GUEST_CR0, guest_cr0),
815 FIELD(GUEST_CR3, guest_cr3),
816 FIELD(GUEST_CR4, guest_cr4),
817 FIELD(GUEST_ES_BASE, guest_es_base),
818 FIELD(GUEST_CS_BASE, guest_cs_base),
819 FIELD(GUEST_SS_BASE, guest_ss_base),
820 FIELD(GUEST_DS_BASE, guest_ds_base),
821 FIELD(GUEST_FS_BASE, guest_fs_base),
822 FIELD(GUEST_GS_BASE, guest_gs_base),
823 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
824 FIELD(GUEST_TR_BASE, guest_tr_base),
825 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
826 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
827 FIELD(GUEST_DR7, guest_dr7),
828 FIELD(GUEST_RSP, guest_rsp),
829 FIELD(GUEST_RIP, guest_rip),
830 FIELD(GUEST_RFLAGS, guest_rflags),
831 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
832 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
833 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
834 FIELD(HOST_CR0, host_cr0),
835 FIELD(HOST_CR3, host_cr3),
836 FIELD(HOST_CR4, host_cr4),
837 FIELD(HOST_FS_BASE, host_fs_base),
838 FIELD(HOST_GS_BASE, host_gs_base),
839 FIELD(HOST_TR_BASE, host_tr_base),
840 FIELD(HOST_GDTR_BASE, host_gdtr_base),
841 FIELD(HOST_IDTR_BASE, host_idtr_base),
842 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
843 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
844 FIELD(HOST_RSP, host_rsp),
845 FIELD(HOST_RIP, host_rip),
848 static inline short vmcs_field_to_offset(unsigned long field)
850 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
852 if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
853 vmcs_field_to_offset_table[field] == 0)
856 return vmcs_field_to_offset_table[field];
859 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
861 return to_vmx(vcpu)->nested.current_vmcs12;
864 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
866 struct page *page = kvm_vcpu_gfn_to_page(vcpu, addr >> PAGE_SHIFT);
867 if (is_error_page(page))
873 static void nested_release_page(struct page *page)
875 kvm_release_page_dirty(page);
878 static void nested_release_page_clean(struct page *page)
880 kvm_release_page_clean(page);
883 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
884 static u64 construct_eptp(unsigned long root_hpa);
885 static void kvm_cpu_vmxon(u64 addr);
886 static void kvm_cpu_vmxoff(void);
887 static bool vmx_xsaves_supported(void);
888 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
889 static void vmx_set_segment(struct kvm_vcpu *vcpu,
890 struct kvm_segment *var, int seg);
891 static void vmx_get_segment(struct kvm_vcpu *vcpu,
892 struct kvm_segment *var, int seg);
893 static bool guest_state_valid(struct kvm_vcpu *vcpu);
894 static u32 vmx_segment_access_rights(struct kvm_segment *var);
895 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
896 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
897 static int alloc_identity_pagetable(struct kvm *kvm);
899 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
900 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
902 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
903 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
905 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
906 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
909 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
910 * can find which vCPU should be waken up.
912 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
913 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
915 static unsigned long *vmx_io_bitmap_a;
916 static unsigned long *vmx_io_bitmap_b;
917 static unsigned long *vmx_msr_bitmap_legacy;
918 static unsigned long *vmx_msr_bitmap_longmode;
919 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
920 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
921 static unsigned long *vmx_msr_bitmap_nested;
922 static unsigned long *vmx_vmread_bitmap;
923 static unsigned long *vmx_vmwrite_bitmap;
925 static bool cpu_has_load_ia32_efer;
926 static bool cpu_has_load_perf_global_ctrl;
928 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
929 static DEFINE_SPINLOCK(vmx_vpid_lock);
931 static struct vmcs_config {
935 u32 pin_based_exec_ctrl;
936 u32 cpu_based_exec_ctrl;
937 u32 cpu_based_2nd_exec_ctrl;
942 static struct vmx_capability {
947 #define VMX_SEGMENT_FIELD(seg) \
948 [VCPU_SREG_##seg] = { \
949 .selector = GUEST_##seg##_SELECTOR, \
950 .base = GUEST_##seg##_BASE, \
951 .limit = GUEST_##seg##_LIMIT, \
952 .ar_bytes = GUEST_##seg##_AR_BYTES, \
955 static const struct kvm_vmx_segment_field {
960 } kvm_vmx_segment_fields[] = {
961 VMX_SEGMENT_FIELD(CS),
962 VMX_SEGMENT_FIELD(DS),
963 VMX_SEGMENT_FIELD(ES),
964 VMX_SEGMENT_FIELD(FS),
965 VMX_SEGMENT_FIELD(GS),
966 VMX_SEGMENT_FIELD(SS),
967 VMX_SEGMENT_FIELD(TR),
968 VMX_SEGMENT_FIELD(LDTR),
971 static u64 host_efer;
973 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
976 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
977 * away by decrementing the array size.
979 static const u32 vmx_msr_index[] = {
981 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
983 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
986 static inline bool is_exception_n(u32 intr_info, u8 vector)
988 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
989 INTR_INFO_VALID_MASK)) ==
990 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
993 static inline bool is_debug(u32 intr_info)
995 return is_exception_n(intr_info, DB_VECTOR);
998 static inline bool is_breakpoint(u32 intr_info)
1000 return is_exception_n(intr_info, BP_VECTOR);
1003 static inline bool is_page_fault(u32 intr_info)
1005 return is_exception_n(intr_info, PF_VECTOR);
1008 static inline bool is_no_device(u32 intr_info)
1010 return is_exception_n(intr_info, NM_VECTOR);
1013 static inline bool is_invalid_opcode(u32 intr_info)
1015 return is_exception_n(intr_info, UD_VECTOR);
1018 static inline bool is_external_interrupt(u32 intr_info)
1020 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1021 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1024 static inline bool is_machine_check(u32 intr_info)
1026 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1027 INTR_INFO_VALID_MASK)) ==
1028 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1031 static inline bool cpu_has_vmx_msr_bitmap(void)
1033 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1036 static inline bool cpu_has_vmx_tpr_shadow(void)
1038 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1041 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1043 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1046 static inline bool cpu_has_secondary_exec_ctrls(void)
1048 return vmcs_config.cpu_based_exec_ctrl &
1049 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1052 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1054 return vmcs_config.cpu_based_2nd_exec_ctrl &
1055 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1058 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1060 return vmcs_config.cpu_based_2nd_exec_ctrl &
1061 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1064 static inline bool cpu_has_vmx_apic_register_virt(void)
1066 return vmcs_config.cpu_based_2nd_exec_ctrl &
1067 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1070 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1072 return vmcs_config.cpu_based_2nd_exec_ctrl &
1073 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1077 * Comment's format: document - errata name - stepping - processor name.
1079 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1081 static u32 vmx_preemption_cpu_tfms[] = {
1082 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
1084 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
1085 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1086 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1088 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1090 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
1091 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
1093 * 320767.pdf - AAP86 - B1 -
1094 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1097 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1099 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1101 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1103 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1104 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1105 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1109 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1111 u32 eax = cpuid_eax(0x00000001), i;
1113 /* Clear the reserved bits */
1114 eax &= ~(0x3U << 14 | 0xfU << 28);
1115 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
1116 if (eax == vmx_preemption_cpu_tfms[i])
1122 static inline bool cpu_has_vmx_preemption_timer(void)
1124 return vmcs_config.pin_based_exec_ctrl &
1125 PIN_BASED_VMX_PREEMPTION_TIMER;
1128 static inline bool cpu_has_vmx_posted_intr(void)
1130 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1131 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1134 static inline bool cpu_has_vmx_apicv(void)
1136 return cpu_has_vmx_apic_register_virt() &&
1137 cpu_has_vmx_virtual_intr_delivery() &&
1138 cpu_has_vmx_posted_intr();
1141 static inline bool cpu_has_vmx_flexpriority(void)
1143 return cpu_has_vmx_tpr_shadow() &&
1144 cpu_has_vmx_virtualize_apic_accesses();
1147 static inline bool cpu_has_vmx_ept_execute_only(void)
1149 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1152 static inline bool cpu_has_vmx_ept_2m_page(void)
1154 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1157 static inline bool cpu_has_vmx_ept_1g_page(void)
1159 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1162 static inline bool cpu_has_vmx_ept_4levels(void)
1164 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1167 static inline bool cpu_has_vmx_ept_ad_bits(void)
1169 return vmx_capability.ept & VMX_EPT_AD_BIT;
1172 static inline bool cpu_has_vmx_invept_context(void)
1174 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1177 static inline bool cpu_has_vmx_invept_global(void)
1179 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1182 static inline bool cpu_has_vmx_invvpid_single(void)
1184 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1187 static inline bool cpu_has_vmx_invvpid_global(void)
1189 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1192 static inline bool cpu_has_vmx_ept(void)
1194 return vmcs_config.cpu_based_2nd_exec_ctrl &
1195 SECONDARY_EXEC_ENABLE_EPT;
1198 static inline bool cpu_has_vmx_unrestricted_guest(void)
1200 return vmcs_config.cpu_based_2nd_exec_ctrl &
1201 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1204 static inline bool cpu_has_vmx_ple(void)
1206 return vmcs_config.cpu_based_2nd_exec_ctrl &
1207 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1210 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1212 return flexpriority_enabled && lapic_in_kernel(vcpu);
1215 static inline bool cpu_has_vmx_vpid(void)
1217 return vmcs_config.cpu_based_2nd_exec_ctrl &
1218 SECONDARY_EXEC_ENABLE_VPID;
1221 static inline bool cpu_has_vmx_rdtscp(void)
1223 return vmcs_config.cpu_based_2nd_exec_ctrl &
1224 SECONDARY_EXEC_RDTSCP;
1227 static inline bool cpu_has_vmx_invpcid(void)
1229 return vmcs_config.cpu_based_2nd_exec_ctrl &
1230 SECONDARY_EXEC_ENABLE_INVPCID;
1233 static inline bool cpu_has_virtual_nmis(void)
1235 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1238 static inline bool cpu_has_vmx_wbinvd_exit(void)
1240 return vmcs_config.cpu_based_2nd_exec_ctrl &
1241 SECONDARY_EXEC_WBINVD_EXITING;
1244 static inline bool cpu_has_vmx_shadow_vmcs(void)
1247 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1248 /* check if the cpu supports writing r/o exit information fields */
1249 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1252 return vmcs_config.cpu_based_2nd_exec_ctrl &
1253 SECONDARY_EXEC_SHADOW_VMCS;
1256 static inline bool cpu_has_vmx_pml(void)
1258 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1261 static inline bool cpu_has_vmx_tsc_scaling(void)
1263 return vmcs_config.cpu_based_2nd_exec_ctrl &
1264 SECONDARY_EXEC_TSC_SCALING;
1267 static inline bool report_flexpriority(void)
1269 return flexpriority_enabled;
1272 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1274 return vmcs12->cpu_based_vm_exec_control & bit;
1277 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1279 return (vmcs12->cpu_based_vm_exec_control &
1280 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1281 (vmcs12->secondary_vm_exec_control & bit);
1284 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1286 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1289 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1291 return vmcs12->pin_based_vm_exec_control &
1292 PIN_BASED_VMX_PREEMPTION_TIMER;
1295 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1297 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1300 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1302 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) &&
1303 vmx_xsaves_supported();
1306 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1308 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1311 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1313 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1316 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1318 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1321 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1323 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1326 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1328 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1331 static inline bool is_exception(u32 intr_info)
1333 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1334 == (INTR_TYPE_HARD_EXCEPTION | INTR_INFO_VALID_MASK);
1337 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1339 unsigned long exit_qualification);
1340 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1341 struct vmcs12 *vmcs12,
1342 u32 reason, unsigned long qualification);
1344 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1348 for (i = 0; i < vmx->nmsrs; ++i)
1349 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1354 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1360 } operand = { vpid, 0, gva };
1362 asm volatile (__ex(ASM_VMX_INVVPID)
1363 /* CF==1 or ZF==1 --> rc = -1 */
1364 "; ja 1f ; ud2 ; 1:"
1365 : : "a"(&operand), "c"(ext) : "cc", "memory");
1368 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1372 } operand = {eptp, gpa};
1374 asm volatile (__ex(ASM_VMX_INVEPT)
1375 /* CF==1 or ZF==1 --> rc = -1 */
1376 "; ja 1f ; ud2 ; 1:\n"
1377 : : "a" (&operand), "c" (ext) : "cc", "memory");
1380 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1384 i = __find_msr_index(vmx, msr);
1386 return &vmx->guest_msrs[i];
1390 static void vmcs_clear(struct vmcs *vmcs)
1392 u64 phys_addr = __pa(vmcs);
1395 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1396 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1399 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1403 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1405 vmcs_clear(loaded_vmcs->vmcs);
1406 loaded_vmcs->cpu = -1;
1407 loaded_vmcs->launched = 0;
1410 static void vmcs_load(struct vmcs *vmcs)
1412 u64 phys_addr = __pa(vmcs);
1415 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1416 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1419 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1423 #ifdef CONFIG_KEXEC_CORE
1425 * This bitmap is used to indicate whether the vmclear
1426 * operation is enabled on all cpus. All disabled by
1429 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1431 static inline void crash_enable_local_vmclear(int cpu)
1433 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1436 static inline void crash_disable_local_vmclear(int cpu)
1438 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1441 static inline int crash_local_vmclear_enabled(int cpu)
1443 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1446 static void crash_vmclear_local_loaded_vmcss(void)
1448 int cpu = raw_smp_processor_id();
1449 struct loaded_vmcs *v;
1451 if (!crash_local_vmclear_enabled(cpu))
1454 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1455 loaded_vmcss_on_cpu_link)
1456 vmcs_clear(v->vmcs);
1459 static inline void crash_enable_local_vmclear(int cpu) { }
1460 static inline void crash_disable_local_vmclear(int cpu) { }
1461 #endif /* CONFIG_KEXEC_CORE */
1463 static void __loaded_vmcs_clear(void *arg)
1465 struct loaded_vmcs *loaded_vmcs = arg;
1466 int cpu = raw_smp_processor_id();
1468 if (loaded_vmcs->cpu != cpu)
1469 return; /* vcpu migration can race with cpu offline */
1470 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1471 per_cpu(current_vmcs, cpu) = NULL;
1472 crash_disable_local_vmclear(cpu);
1473 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1476 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1477 * is before setting loaded_vmcs->vcpu to -1 which is done in
1478 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1479 * then adds the vmcs into percpu list before it is deleted.
1483 loaded_vmcs_init(loaded_vmcs);
1484 crash_enable_local_vmclear(cpu);
1487 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1489 int cpu = loaded_vmcs->cpu;
1492 smp_call_function_single(cpu,
1493 __loaded_vmcs_clear, loaded_vmcs, 1);
1496 static inline void vpid_sync_vcpu_single(int vpid)
1501 if (cpu_has_vmx_invvpid_single())
1502 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1505 static inline void vpid_sync_vcpu_global(void)
1507 if (cpu_has_vmx_invvpid_global())
1508 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1511 static inline void vpid_sync_context(int vpid)
1513 if (cpu_has_vmx_invvpid_single())
1514 vpid_sync_vcpu_single(vpid);
1516 vpid_sync_vcpu_global();
1519 static inline void ept_sync_global(void)
1521 if (cpu_has_vmx_invept_global())
1522 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1525 static inline void ept_sync_context(u64 eptp)
1528 if (cpu_has_vmx_invept_context())
1529 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1535 static __always_inline void vmcs_check16(unsigned long field)
1537 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1538 "16-bit accessor invalid for 64-bit field");
1539 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1540 "16-bit accessor invalid for 64-bit high field");
1541 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1542 "16-bit accessor invalid for 32-bit high field");
1543 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1544 "16-bit accessor invalid for natural width field");
1547 static __always_inline void vmcs_check32(unsigned long field)
1549 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1550 "32-bit accessor invalid for 16-bit field");
1551 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1552 "32-bit accessor invalid for natural width field");
1555 static __always_inline void vmcs_check64(unsigned long field)
1557 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1558 "64-bit accessor invalid for 16-bit field");
1559 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1560 "64-bit accessor invalid for 64-bit high field");
1561 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1562 "64-bit accessor invalid for 32-bit field");
1563 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1564 "64-bit accessor invalid for natural width field");
1567 static __always_inline void vmcs_checkl(unsigned long field)
1569 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1570 "Natural width accessor invalid for 16-bit field");
1571 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1572 "Natural width accessor invalid for 64-bit field");
1573 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1574 "Natural width accessor invalid for 64-bit high field");
1575 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1576 "Natural width accessor invalid for 32-bit field");
1579 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1581 unsigned long value;
1583 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1584 : "=a"(value) : "d"(field) : "cc");
1588 static __always_inline u16 vmcs_read16(unsigned long field)
1590 vmcs_check16(field);
1591 return __vmcs_readl(field);
1594 static __always_inline u32 vmcs_read32(unsigned long field)
1596 vmcs_check32(field);
1597 return __vmcs_readl(field);
1600 static __always_inline u64 vmcs_read64(unsigned long field)
1602 vmcs_check64(field);
1603 #ifdef CONFIG_X86_64
1604 return __vmcs_readl(field);
1606 return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1610 static __always_inline unsigned long vmcs_readl(unsigned long field)
1613 return __vmcs_readl(field);
1616 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1618 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1619 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1623 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1627 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1628 : "=q"(error) : "a"(value), "d"(field) : "cc");
1629 if (unlikely(error))
1630 vmwrite_error(field, value);
1633 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1635 vmcs_check16(field);
1636 __vmcs_writel(field, value);
1639 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1641 vmcs_check32(field);
1642 __vmcs_writel(field, value);
1645 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1647 vmcs_check64(field);
1648 __vmcs_writel(field, value);
1649 #ifndef CONFIG_X86_64
1651 __vmcs_writel(field+1, value >> 32);
1655 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1658 __vmcs_writel(field, value);
1661 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1663 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1664 "vmcs_clear_bits does not support 64-bit fields");
1665 __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1668 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1670 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1671 "vmcs_set_bits does not support 64-bit fields");
1672 __vmcs_writel(field, __vmcs_readl(field) | mask);
1675 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1677 vmcs_write32(VM_ENTRY_CONTROLS, val);
1678 vmx->vm_entry_controls_shadow = val;
1681 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1683 if (vmx->vm_entry_controls_shadow != val)
1684 vm_entry_controls_init(vmx, val);
1687 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1689 return vmx->vm_entry_controls_shadow;
1693 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1695 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1698 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1700 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1703 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1705 vmcs_write32(VM_EXIT_CONTROLS, val);
1706 vmx->vm_exit_controls_shadow = val;
1709 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1711 if (vmx->vm_exit_controls_shadow != val)
1712 vm_exit_controls_init(vmx, val);
1715 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1717 return vmx->vm_exit_controls_shadow;
1721 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1723 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1726 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1728 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1731 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1733 vmx->segment_cache.bitmask = 0;
1736 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1740 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1742 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1743 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1744 vmx->segment_cache.bitmask = 0;
1746 ret = vmx->segment_cache.bitmask & mask;
1747 vmx->segment_cache.bitmask |= mask;
1751 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1753 u16 *p = &vmx->segment_cache.seg[seg].selector;
1755 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1756 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1760 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1762 ulong *p = &vmx->segment_cache.seg[seg].base;
1764 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1765 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1769 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1771 u32 *p = &vmx->segment_cache.seg[seg].limit;
1773 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1774 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1778 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1780 u32 *p = &vmx->segment_cache.seg[seg].ar;
1782 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1783 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1787 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1791 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1792 (1u << NM_VECTOR) | (1u << DB_VECTOR) | (1u << AC_VECTOR);
1793 if ((vcpu->guest_debug &
1794 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1795 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1796 eb |= 1u << BP_VECTOR;
1797 if (to_vmx(vcpu)->rmode.vm86_active)
1800 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1801 if (vcpu->fpu_active)
1802 eb &= ~(1u << NM_VECTOR);
1804 /* When we are running a nested L2 guest and L1 specified for it a
1805 * certain exception bitmap, we must trap the same exceptions and pass
1806 * them to L1. When running L2, we will only handle the exceptions
1807 * specified above if L1 did not want them.
1809 if (is_guest_mode(vcpu))
1810 eb |= get_vmcs12(vcpu)->exception_bitmap;
1812 vmcs_write32(EXCEPTION_BITMAP, eb);
1815 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1816 unsigned long entry, unsigned long exit)
1818 vm_entry_controls_clearbit(vmx, entry);
1819 vm_exit_controls_clearbit(vmx, exit);
1822 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1825 struct msr_autoload *m = &vmx->msr_autoload;
1829 if (cpu_has_load_ia32_efer) {
1830 clear_atomic_switch_msr_special(vmx,
1831 VM_ENTRY_LOAD_IA32_EFER,
1832 VM_EXIT_LOAD_IA32_EFER);
1836 case MSR_CORE_PERF_GLOBAL_CTRL:
1837 if (cpu_has_load_perf_global_ctrl) {
1838 clear_atomic_switch_msr_special(vmx,
1839 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1840 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1846 for (i = 0; i < m->nr; ++i)
1847 if (m->guest[i].index == msr)
1853 m->guest[i] = m->guest[m->nr];
1854 m->host[i] = m->host[m->nr];
1855 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1856 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1859 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1860 unsigned long entry, unsigned long exit,
1861 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1862 u64 guest_val, u64 host_val)
1864 vmcs_write64(guest_val_vmcs, guest_val);
1865 vmcs_write64(host_val_vmcs, host_val);
1866 vm_entry_controls_setbit(vmx, entry);
1867 vm_exit_controls_setbit(vmx, exit);
1870 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1871 u64 guest_val, u64 host_val)
1874 struct msr_autoload *m = &vmx->msr_autoload;
1878 if (cpu_has_load_ia32_efer) {
1879 add_atomic_switch_msr_special(vmx,
1880 VM_ENTRY_LOAD_IA32_EFER,
1881 VM_EXIT_LOAD_IA32_EFER,
1884 guest_val, host_val);
1888 case MSR_CORE_PERF_GLOBAL_CTRL:
1889 if (cpu_has_load_perf_global_ctrl) {
1890 add_atomic_switch_msr_special(vmx,
1891 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1892 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1893 GUEST_IA32_PERF_GLOBAL_CTRL,
1894 HOST_IA32_PERF_GLOBAL_CTRL,
1895 guest_val, host_val);
1899 case MSR_IA32_PEBS_ENABLE:
1900 /* PEBS needs a quiescent period after being disabled (to write
1901 * a record). Disabling PEBS through VMX MSR swapping doesn't
1902 * provide that period, so a CPU could write host's record into
1905 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1908 for (i = 0; i < m->nr; ++i)
1909 if (m->guest[i].index == msr)
1912 if (i == NR_AUTOLOAD_MSRS) {
1913 printk_once(KERN_WARNING "Not enough msr switch entries. "
1914 "Can't add msr %x\n", msr);
1916 } else if (i == m->nr) {
1918 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1919 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1922 m->guest[i].index = msr;
1923 m->guest[i].value = guest_val;
1924 m->host[i].index = msr;
1925 m->host[i].value = host_val;
1928 static void reload_tss(void)
1931 * VT restores TR but not its size. Useless.
1933 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1934 struct desc_struct *descs;
1936 descs = (void *)gdt->address;
1937 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1941 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1943 u64 guest_efer = vmx->vcpu.arch.efer;
1944 u64 ignore_bits = 0;
1948 * NX is needed to handle CR0.WP=1, CR4.SMEP=1. Testing
1949 * host CPUID is more efficient than testing guest CPUID
1950 * or CR4. Host SMEP is anyway a requirement for guest SMEP.
1952 if (boot_cpu_has(X86_FEATURE_SMEP))
1953 guest_efer |= EFER_NX;
1954 else if (!(guest_efer & EFER_NX))
1955 ignore_bits |= EFER_NX;
1959 * LMA and LME handled by hardware; SCE meaningless outside long mode.
1961 ignore_bits |= EFER_SCE;
1962 #ifdef CONFIG_X86_64
1963 ignore_bits |= EFER_LMA | EFER_LME;
1964 /* SCE is meaningful only in long mode on Intel */
1965 if (guest_efer & EFER_LMA)
1966 ignore_bits &= ~(u64)EFER_SCE;
1969 clear_atomic_switch_msr(vmx, MSR_EFER);
1972 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1973 * On CPUs that support "load IA32_EFER", always switch EFER
1974 * atomically, since it's faster than switching it manually.
1976 if (cpu_has_load_ia32_efer ||
1977 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1978 if (!(guest_efer & EFER_LMA))
1979 guest_efer &= ~EFER_LME;
1980 if (guest_efer != host_efer)
1981 add_atomic_switch_msr(vmx, MSR_EFER,
1982 guest_efer, host_efer);
1985 guest_efer &= ~ignore_bits;
1986 guest_efer |= host_efer & ignore_bits;
1988 vmx->guest_msrs[efer_offset].data = guest_efer;
1989 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1995 static unsigned long segment_base(u16 selector)
1997 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1998 struct desc_struct *d;
1999 unsigned long table_base;
2002 if (!(selector & ~3))
2005 table_base = gdt->address;
2007 if (selector & 4) { /* from ldt */
2008 u16 ldt_selector = kvm_read_ldt();
2010 if (!(ldt_selector & ~3))
2013 table_base = segment_base(ldt_selector);
2015 d = (struct desc_struct *)(table_base + (selector & ~7));
2016 v = get_desc_base(d);
2017 #ifdef CONFIG_X86_64
2018 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
2019 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
2024 static inline unsigned long kvm_read_tr_base(void)
2027 asm("str %0" : "=g"(tr));
2028 return segment_base(tr);
2031 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
2033 struct vcpu_vmx *vmx = to_vmx(vcpu);
2036 if (vmx->host_state.loaded)
2039 vmx->host_state.loaded = 1;
2041 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
2042 * allow segment selectors with cpl > 0 or ti == 1.
2044 vmx->host_state.ldt_sel = kvm_read_ldt();
2045 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
2046 savesegment(fs, vmx->host_state.fs_sel);
2047 if (!(vmx->host_state.fs_sel & 7)) {
2048 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
2049 vmx->host_state.fs_reload_needed = 0;
2051 vmcs_write16(HOST_FS_SELECTOR, 0);
2052 vmx->host_state.fs_reload_needed = 1;
2054 savesegment(gs, vmx->host_state.gs_sel);
2055 if (!(vmx->host_state.gs_sel & 7))
2056 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
2058 vmcs_write16(HOST_GS_SELECTOR, 0);
2059 vmx->host_state.gs_ldt_reload_needed = 1;
2062 #ifdef CONFIG_X86_64
2063 savesegment(ds, vmx->host_state.ds_sel);
2064 savesegment(es, vmx->host_state.es_sel);
2067 #ifdef CONFIG_X86_64
2068 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
2069 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
2071 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
2072 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
2075 #ifdef CONFIG_X86_64
2076 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2077 if (is_long_mode(&vmx->vcpu))
2078 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2080 if (boot_cpu_has(X86_FEATURE_MPX))
2081 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2082 for (i = 0; i < vmx->save_nmsrs; ++i)
2083 kvm_set_shared_msr(vmx->guest_msrs[i].index,
2084 vmx->guest_msrs[i].data,
2085 vmx->guest_msrs[i].mask);
2088 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
2090 if (!vmx->host_state.loaded)
2093 ++vmx->vcpu.stat.host_state_reload;
2094 vmx->host_state.loaded = 0;
2095 #ifdef CONFIG_X86_64
2096 if (is_long_mode(&vmx->vcpu))
2097 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2099 if (vmx->host_state.gs_ldt_reload_needed) {
2100 kvm_load_ldt(vmx->host_state.ldt_sel);
2101 #ifdef CONFIG_X86_64
2102 load_gs_index(vmx->host_state.gs_sel);
2104 loadsegment(gs, vmx->host_state.gs_sel);
2107 if (vmx->host_state.fs_reload_needed)
2108 loadsegment(fs, vmx->host_state.fs_sel);
2109 #ifdef CONFIG_X86_64
2110 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2111 loadsegment(ds, vmx->host_state.ds_sel);
2112 loadsegment(es, vmx->host_state.es_sel);
2116 #ifdef CONFIG_X86_64
2117 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2119 if (vmx->host_state.msr_host_bndcfgs)
2120 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2122 * If the FPU is not active (through the host task or
2123 * the guest vcpu), then restore the cr0.TS bit.
2125 if (!fpregs_active() && !vmx->vcpu.guest_fpu_loaded)
2127 load_gdt(this_cpu_ptr(&host_gdt));
2130 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2133 __vmx_load_host_state(vmx);
2137 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2139 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2140 struct pi_desc old, new;
2143 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2144 !irq_remapping_cap(IRQ_POSTING_CAP))
2148 old.control = new.control = pi_desc->control;
2151 * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
2152 * are two possible cases:
2153 * 1. After running 'pre_block', context switch
2154 * happened. For this case, 'sn' was set in
2155 * vmx_vcpu_put(), so we need to clear it here.
2156 * 2. After running 'pre_block', we were blocked,
2157 * and woken up by some other guy. For this case,
2158 * we don't need to do anything, 'pi_post_block'
2159 * will do everything for us. However, we cannot
2160 * check whether it is case #1 or case #2 here
2161 * (maybe, not needed), so we also clear sn here,
2162 * I think it is not a big deal.
2164 if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) {
2165 if (vcpu->cpu != cpu) {
2166 dest = cpu_physical_id(cpu);
2168 if (x2apic_enabled())
2171 new.ndst = (dest << 8) & 0xFF00;
2174 /* set 'NV' to 'notification vector' */
2175 new.nv = POSTED_INTR_VECTOR;
2178 /* Allow posting non-urgent interrupts */
2180 } while (cmpxchg(&pi_desc->control, old.control,
2181 new.control) != old.control);
2185 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2186 * vcpu mutex is already taken.
2188 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2190 struct vcpu_vmx *vmx = to_vmx(vcpu);
2191 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2194 kvm_cpu_vmxon(phys_addr);
2195 else if (vmx->loaded_vmcs->cpu != cpu)
2196 loaded_vmcs_clear(vmx->loaded_vmcs);
2198 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2199 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2200 vmcs_load(vmx->loaded_vmcs->vmcs);
2203 if (vmx->loaded_vmcs->cpu != cpu) {
2204 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2205 unsigned long sysenter_esp;
2207 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2208 local_irq_disable();
2209 crash_disable_local_vmclear(cpu);
2212 * Read loaded_vmcs->cpu should be before fetching
2213 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2214 * See the comments in __loaded_vmcs_clear().
2218 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2219 &per_cpu(loaded_vmcss_on_cpu, cpu));
2220 crash_enable_local_vmclear(cpu);
2224 * Linux uses per-cpu TSS and GDT, so set these when switching
2227 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
2228 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
2230 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2231 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2233 vmx->loaded_vmcs->cpu = cpu;
2236 /* Setup TSC multiplier */
2237 if (kvm_has_tsc_control &&
2238 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio) {
2239 vmx->current_tsc_ratio = vcpu->arch.tsc_scaling_ratio;
2240 vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2243 vmx_vcpu_pi_load(vcpu, cpu);
2244 vmx->host_pkru = read_pkru();
2247 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2249 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2251 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2252 !irq_remapping_cap(IRQ_POSTING_CAP))
2255 /* Set SN when the vCPU is preempted */
2256 if (vcpu->preempted)
2260 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2262 vmx_vcpu_pi_put(vcpu);
2264 __vmx_load_host_state(to_vmx(vcpu));
2265 if (!vmm_exclusive) {
2266 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
2272 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
2276 if (vcpu->fpu_active)
2278 vcpu->fpu_active = 1;
2279 cr0 = vmcs_readl(GUEST_CR0);
2280 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
2281 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
2282 vmcs_writel(GUEST_CR0, cr0);
2283 update_exception_bitmap(vcpu);
2284 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
2285 if (is_guest_mode(vcpu))
2286 vcpu->arch.cr0_guest_owned_bits &=
2287 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
2288 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2291 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2294 * Return the cr0 value that a nested guest would read. This is a combination
2295 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2296 * its hypervisor (cr0_read_shadow).
2298 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2300 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2301 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2303 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2305 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2306 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2309 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
2311 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
2312 * set this *before* calling this function.
2314 vmx_decache_cr0_guest_bits(vcpu);
2315 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
2316 update_exception_bitmap(vcpu);
2317 vcpu->arch.cr0_guest_owned_bits = 0;
2318 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2319 if (is_guest_mode(vcpu)) {
2321 * L1's specified read shadow might not contain the TS bit,
2322 * so now that we turned on shadowing of this bit, we need to
2323 * set this bit of the shadow. Like in nested_vmx_run we need
2324 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
2325 * up-to-date here because we just decached cr0.TS (and we'll
2326 * only update vmcs12->guest_cr0 on nested exit).
2328 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2329 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
2330 (vcpu->arch.cr0 & X86_CR0_TS);
2331 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2333 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2336 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2338 unsigned long rflags, save_rflags;
2340 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2341 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2342 rflags = vmcs_readl(GUEST_RFLAGS);
2343 if (to_vmx(vcpu)->rmode.vm86_active) {
2344 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2345 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2346 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2348 to_vmx(vcpu)->rflags = rflags;
2350 return to_vmx(vcpu)->rflags;
2353 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2355 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2356 to_vmx(vcpu)->rflags = rflags;
2357 if (to_vmx(vcpu)->rmode.vm86_active) {
2358 to_vmx(vcpu)->rmode.save_rflags = rflags;
2359 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2361 vmcs_writel(GUEST_RFLAGS, rflags);
2364 static u32 vmx_get_pkru(struct kvm_vcpu *vcpu)
2366 return to_vmx(vcpu)->guest_pkru;
2369 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2371 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2374 if (interruptibility & GUEST_INTR_STATE_STI)
2375 ret |= KVM_X86_SHADOW_INT_STI;
2376 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2377 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2382 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2384 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2385 u32 interruptibility = interruptibility_old;
2387 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2389 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2390 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2391 else if (mask & KVM_X86_SHADOW_INT_STI)
2392 interruptibility |= GUEST_INTR_STATE_STI;
2394 if ((interruptibility != interruptibility_old))
2395 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2398 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2402 rip = kvm_rip_read(vcpu);
2403 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2404 kvm_rip_write(vcpu, rip);
2406 /* skipping an emulated instruction also counts */
2407 vmx_set_interrupt_shadow(vcpu, 0);
2411 * KVM wants to inject page-faults which it got to the guest. This function
2412 * checks whether in a nested guest, we need to inject them to L1 or L2.
2414 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
2416 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2418 if (!(vmcs12->exception_bitmap & (1u << nr)))
2421 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
2422 vmcs_read32(VM_EXIT_INTR_INFO),
2423 vmcs_readl(EXIT_QUALIFICATION));
2427 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2428 bool has_error_code, u32 error_code,
2431 struct vcpu_vmx *vmx = to_vmx(vcpu);
2432 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2434 if (!reinject && is_guest_mode(vcpu) &&
2435 nested_vmx_check_exception(vcpu, nr))
2438 if (has_error_code) {
2439 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2440 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2443 if (vmx->rmode.vm86_active) {
2445 if (kvm_exception_is_soft(nr))
2446 inc_eip = vcpu->arch.event_exit_inst_len;
2447 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2448 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2452 if (kvm_exception_is_soft(nr)) {
2453 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2454 vmx->vcpu.arch.event_exit_inst_len);
2455 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2457 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2459 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2462 static bool vmx_rdtscp_supported(void)
2464 return cpu_has_vmx_rdtscp();
2467 static bool vmx_invpcid_supported(void)
2469 return cpu_has_vmx_invpcid() && enable_ept;
2473 * Swap MSR entry in host/guest MSR entry array.
2475 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2477 struct shared_msr_entry tmp;
2479 tmp = vmx->guest_msrs[to];
2480 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2481 vmx->guest_msrs[from] = tmp;
2484 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2486 unsigned long *msr_bitmap;
2488 if (is_guest_mode(vcpu))
2489 msr_bitmap = vmx_msr_bitmap_nested;
2490 else if (cpu_has_secondary_exec_ctrls() &&
2491 (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
2492 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
2493 if (is_long_mode(vcpu))
2494 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2496 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2498 if (is_long_mode(vcpu))
2499 msr_bitmap = vmx_msr_bitmap_longmode;
2501 msr_bitmap = vmx_msr_bitmap_legacy;
2504 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2508 * Set up the vmcs to automatically save and restore system
2509 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2510 * mode, as fiddling with msrs is very expensive.
2512 static void setup_msrs(struct vcpu_vmx *vmx)
2514 int save_nmsrs, index;
2517 #ifdef CONFIG_X86_64
2518 if (is_long_mode(&vmx->vcpu)) {
2519 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2521 move_msr_up(vmx, index, save_nmsrs++);
2522 index = __find_msr_index(vmx, MSR_LSTAR);
2524 move_msr_up(vmx, index, save_nmsrs++);
2525 index = __find_msr_index(vmx, MSR_CSTAR);
2527 move_msr_up(vmx, index, save_nmsrs++);
2528 index = __find_msr_index(vmx, MSR_TSC_AUX);
2529 if (index >= 0 && guest_cpuid_has_rdtscp(&vmx->vcpu))
2530 move_msr_up(vmx, index, save_nmsrs++);
2532 * MSR_STAR is only needed on long mode guests, and only
2533 * if efer.sce is enabled.
2535 index = __find_msr_index(vmx, MSR_STAR);
2536 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2537 move_msr_up(vmx, index, save_nmsrs++);
2540 index = __find_msr_index(vmx, MSR_EFER);
2541 if (index >= 0 && update_transition_efer(vmx, index))
2542 move_msr_up(vmx, index, save_nmsrs++);
2544 vmx->save_nmsrs = save_nmsrs;
2546 if (cpu_has_vmx_msr_bitmap())
2547 vmx_set_msr_bitmap(&vmx->vcpu);
2551 * reads and returns guest's timestamp counter "register"
2552 * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2553 * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2555 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2557 u64 host_tsc, tsc_offset;
2560 tsc_offset = vmcs_read64(TSC_OFFSET);
2561 return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2565 * Like guest_read_tsc, but always returns L1's notion of the timestamp
2566 * counter, even if a nested guest (L2) is currently running.
2568 static u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2572 tsc_offset = is_guest_mode(vcpu) ?
2573 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2574 vmcs_read64(TSC_OFFSET);
2575 return host_tsc + tsc_offset;
2578 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2580 return vmcs_read64(TSC_OFFSET);
2584 * writes 'offset' into guest's timestamp counter offset register
2586 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2588 if (is_guest_mode(vcpu)) {
2590 * We're here if L1 chose not to trap WRMSR to TSC. According
2591 * to the spec, this should set L1's TSC; The offset that L1
2592 * set for L2 remains unchanged, and still needs to be added
2593 * to the newly set TSC to get L2's TSC.
2595 struct vmcs12 *vmcs12;
2596 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2597 /* recalculate vmcs02.TSC_OFFSET: */
2598 vmcs12 = get_vmcs12(vcpu);
2599 vmcs_write64(TSC_OFFSET, offset +
2600 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2601 vmcs12->tsc_offset : 0));
2603 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2604 vmcs_read64(TSC_OFFSET), offset);
2605 vmcs_write64(TSC_OFFSET, offset);
2609 static void vmx_adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, s64 adjustment)
2611 u64 offset = vmcs_read64(TSC_OFFSET);
2613 vmcs_write64(TSC_OFFSET, offset + adjustment);
2614 if (is_guest_mode(vcpu)) {
2615 /* Even when running L2, the adjustment needs to apply to L1 */
2616 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2618 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2619 offset + adjustment);
2622 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2624 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2625 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2629 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2630 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2631 * all guests if the "nested" module option is off, and can also be disabled
2632 * for a single guest by disabling its VMX cpuid bit.
2634 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2636 return nested && guest_cpuid_has_vmx(vcpu);
2640 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2641 * returned for the various VMX controls MSRs when nested VMX is enabled.
2642 * The same values should also be used to verify that vmcs12 control fields are
2643 * valid during nested entry from L1 to L2.
2644 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2645 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2646 * bit in the high half is on if the corresponding bit in the control field
2647 * may be on. See also vmx_control_verify().
2649 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2652 * Note that as a general rule, the high half of the MSRs (bits in
2653 * the control fields which may be 1) should be initialized by the
2654 * intersection of the underlying hardware's MSR (i.e., features which
2655 * can be supported) and the list of features we want to expose -
2656 * because they are known to be properly supported in our code.
2657 * Also, usually, the low half of the MSRs (bits which must be 1) can
2658 * be set to 0, meaning that L1 may turn off any of these bits. The
2659 * reason is that if one of these bits is necessary, it will appear
2660 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2661 * fields of vmcs01 and vmcs02, will turn these bits off - and
2662 * nested_vmx_exit_handled() will not pass related exits to L1.
2663 * These rules have exceptions below.
2666 /* pin-based controls */
2667 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2668 vmx->nested.nested_vmx_pinbased_ctls_low,
2669 vmx->nested.nested_vmx_pinbased_ctls_high);
2670 vmx->nested.nested_vmx_pinbased_ctls_low |=
2671 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2672 vmx->nested.nested_vmx_pinbased_ctls_high &=
2673 PIN_BASED_EXT_INTR_MASK |
2674 PIN_BASED_NMI_EXITING |
2675 PIN_BASED_VIRTUAL_NMIS;
2676 vmx->nested.nested_vmx_pinbased_ctls_high |=
2677 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2678 PIN_BASED_VMX_PREEMPTION_TIMER;
2679 if (kvm_vcpu_apicv_active(&vmx->vcpu))
2680 vmx->nested.nested_vmx_pinbased_ctls_high |=
2681 PIN_BASED_POSTED_INTR;
2684 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2685 vmx->nested.nested_vmx_exit_ctls_low,
2686 vmx->nested.nested_vmx_exit_ctls_high);
2687 vmx->nested.nested_vmx_exit_ctls_low =
2688 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2690 vmx->nested.nested_vmx_exit_ctls_high &=
2691 #ifdef CONFIG_X86_64
2692 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2694 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2695 vmx->nested.nested_vmx_exit_ctls_high |=
2696 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2697 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2698 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2700 if (kvm_mpx_supported())
2701 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2703 /* We support free control of debug control saving. */
2704 vmx->nested.nested_vmx_true_exit_ctls_low =
2705 vmx->nested.nested_vmx_exit_ctls_low &
2706 ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2708 /* entry controls */
2709 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2710 vmx->nested.nested_vmx_entry_ctls_low,
2711 vmx->nested.nested_vmx_entry_ctls_high);
2712 vmx->nested.nested_vmx_entry_ctls_low =
2713 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2714 vmx->nested.nested_vmx_entry_ctls_high &=
2715 #ifdef CONFIG_X86_64
2716 VM_ENTRY_IA32E_MODE |
2718 VM_ENTRY_LOAD_IA32_PAT;
2719 vmx->nested.nested_vmx_entry_ctls_high |=
2720 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2721 if (kvm_mpx_supported())
2722 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2724 /* We support free control of debug control loading. */
2725 vmx->nested.nested_vmx_true_entry_ctls_low =
2726 vmx->nested.nested_vmx_entry_ctls_low &
2727 ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2729 /* cpu-based controls */
2730 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2731 vmx->nested.nested_vmx_procbased_ctls_low,
2732 vmx->nested.nested_vmx_procbased_ctls_high);
2733 vmx->nested.nested_vmx_procbased_ctls_low =
2734 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2735 vmx->nested.nested_vmx_procbased_ctls_high &=
2736 CPU_BASED_VIRTUAL_INTR_PENDING |
2737 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2738 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2739 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2740 CPU_BASED_CR3_STORE_EXITING |
2741 #ifdef CONFIG_X86_64
2742 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2744 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2745 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2746 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2747 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2748 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2750 * We can allow some features even when not supported by the
2751 * hardware. For example, L1 can specify an MSR bitmap - and we
2752 * can use it to avoid exits to L1 - even when L0 runs L2
2753 * without MSR bitmaps.
2755 vmx->nested.nested_vmx_procbased_ctls_high |=
2756 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2757 CPU_BASED_USE_MSR_BITMAPS;
2759 /* We support free control of CR3 access interception. */
2760 vmx->nested.nested_vmx_true_procbased_ctls_low =
2761 vmx->nested.nested_vmx_procbased_ctls_low &
2762 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2764 /* secondary cpu-based controls */
2765 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2766 vmx->nested.nested_vmx_secondary_ctls_low,
2767 vmx->nested.nested_vmx_secondary_ctls_high);
2768 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2769 vmx->nested.nested_vmx_secondary_ctls_high &=
2770 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2771 SECONDARY_EXEC_RDTSCP |
2772 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2773 SECONDARY_EXEC_ENABLE_VPID |
2774 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2775 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2776 SECONDARY_EXEC_WBINVD_EXITING |
2777 SECONDARY_EXEC_XSAVES |
2778 SECONDARY_EXEC_PCOMMIT;
2781 /* nested EPT: emulate EPT also to L1 */
2782 vmx->nested.nested_vmx_secondary_ctls_high |=
2783 SECONDARY_EXEC_ENABLE_EPT;
2784 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2785 VMX_EPTP_WB_BIT | VMX_EPT_2MB_PAGE_BIT |
2787 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2789 * For nested guests, we don't do anything specific
2790 * for single context invalidation. Hence, only advertise
2791 * support for global context invalidation.
2793 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT;
2795 vmx->nested.nested_vmx_ept_caps = 0;
2798 * Old versions of KVM use the single-context version without
2799 * checking for support, so declare that it is supported even
2800 * though it is treated as global context. The alternative is
2801 * not failing the single-context invvpid, and it is worse.
2804 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2805 VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |
2806 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
2808 vmx->nested.nested_vmx_vpid_caps = 0;
2810 if (enable_unrestricted_guest)
2811 vmx->nested.nested_vmx_secondary_ctls_high |=
2812 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2814 /* miscellaneous data */
2815 rdmsr(MSR_IA32_VMX_MISC,
2816 vmx->nested.nested_vmx_misc_low,
2817 vmx->nested.nested_vmx_misc_high);
2818 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2819 vmx->nested.nested_vmx_misc_low |=
2820 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2821 VMX_MISC_ACTIVITY_HLT;
2822 vmx->nested.nested_vmx_misc_high = 0;
2825 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2828 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2830 return ((control & high) | low) == control;
2833 static inline u64 vmx_control_msr(u32 low, u32 high)
2835 return low | ((u64)high << 32);
2838 /* Returns 0 on success, non-0 otherwise. */
2839 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2841 struct vcpu_vmx *vmx = to_vmx(vcpu);
2843 switch (msr_index) {
2844 case MSR_IA32_VMX_BASIC:
2846 * This MSR reports some information about VMX support. We
2847 * should return information about the VMX we emulate for the
2848 * guest, and the VMCS structure we give it - not about the
2849 * VMX support of the underlying hardware.
2851 *pdata = VMCS12_REVISION | VMX_BASIC_TRUE_CTLS |
2852 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2853 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2855 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2856 case MSR_IA32_VMX_PINBASED_CTLS:
2857 *pdata = vmx_control_msr(
2858 vmx->nested.nested_vmx_pinbased_ctls_low,
2859 vmx->nested.nested_vmx_pinbased_ctls_high);
2861 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2862 *pdata = vmx_control_msr(
2863 vmx->nested.nested_vmx_true_procbased_ctls_low,
2864 vmx->nested.nested_vmx_procbased_ctls_high);
2866 case MSR_IA32_VMX_PROCBASED_CTLS:
2867 *pdata = vmx_control_msr(
2868 vmx->nested.nested_vmx_procbased_ctls_low,
2869 vmx->nested.nested_vmx_procbased_ctls_high);
2871 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2872 *pdata = vmx_control_msr(
2873 vmx->nested.nested_vmx_true_exit_ctls_low,
2874 vmx->nested.nested_vmx_exit_ctls_high);
2876 case MSR_IA32_VMX_EXIT_CTLS:
2877 *pdata = vmx_control_msr(
2878 vmx->nested.nested_vmx_exit_ctls_low,
2879 vmx->nested.nested_vmx_exit_ctls_high);
2881 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2882 *pdata = vmx_control_msr(
2883 vmx->nested.nested_vmx_true_entry_ctls_low,
2884 vmx->nested.nested_vmx_entry_ctls_high);
2886 case MSR_IA32_VMX_ENTRY_CTLS:
2887 *pdata = vmx_control_msr(
2888 vmx->nested.nested_vmx_entry_ctls_low,
2889 vmx->nested.nested_vmx_entry_ctls_high);
2891 case MSR_IA32_VMX_MISC:
2892 *pdata = vmx_control_msr(
2893 vmx->nested.nested_vmx_misc_low,
2894 vmx->nested.nested_vmx_misc_high);
2897 * These MSRs specify bits which the guest must keep fixed (on or off)
2898 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2899 * We picked the standard core2 setting.
2901 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2902 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2903 case MSR_IA32_VMX_CR0_FIXED0:
2904 *pdata = VMXON_CR0_ALWAYSON;
2906 case MSR_IA32_VMX_CR0_FIXED1:
2909 case MSR_IA32_VMX_CR4_FIXED0:
2910 *pdata = VMXON_CR4_ALWAYSON;
2912 case MSR_IA32_VMX_CR4_FIXED1:
2915 case MSR_IA32_VMX_VMCS_ENUM:
2916 *pdata = 0x2e; /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2918 case MSR_IA32_VMX_PROCBASED_CTLS2:
2919 *pdata = vmx_control_msr(
2920 vmx->nested.nested_vmx_secondary_ctls_low,
2921 vmx->nested.nested_vmx_secondary_ctls_high);
2923 case MSR_IA32_VMX_EPT_VPID_CAP:
2924 /* Currently, no nested vpid support */
2925 *pdata = vmx->nested.nested_vmx_ept_caps |
2926 ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
2935 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
2938 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
2940 return !(val & ~valid_bits);
2944 * Reads an msr value (of 'msr_index') into 'pdata'.
2945 * Returns 0 on success, non-0 otherwise.
2946 * Assumes vcpu_load() was already called.
2948 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2950 struct shared_msr_entry *msr;
2952 switch (msr_info->index) {
2953 #ifdef CONFIG_X86_64
2955 msr_info->data = vmcs_readl(GUEST_FS_BASE);
2958 msr_info->data = vmcs_readl(GUEST_GS_BASE);
2960 case MSR_KERNEL_GS_BASE:
2961 vmx_load_host_state(to_vmx(vcpu));
2962 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2966 return kvm_get_msr_common(vcpu, msr_info);
2968 msr_info->data = guest_read_tsc(vcpu);
2970 case MSR_IA32_SYSENTER_CS:
2971 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
2973 case MSR_IA32_SYSENTER_EIP:
2974 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
2976 case MSR_IA32_SYSENTER_ESP:
2977 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
2979 case MSR_IA32_BNDCFGS:
2980 if (!kvm_mpx_supported())
2982 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
2984 case MSR_IA32_MCG_EXT_CTL:
2985 if (!msr_info->host_initiated &&
2986 !(to_vmx(vcpu)->msr_ia32_feature_control &
2987 FEATURE_CONTROL_LMCE))
2989 msr_info->data = vcpu->arch.mcg_ext_ctl;
2991 case MSR_IA32_FEATURE_CONTROL:
2992 msr_info->data = to_vmx(vcpu)->msr_ia32_feature_control;
2994 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2995 if (!nested_vmx_allowed(vcpu))
2997 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
2999 if (!vmx_xsaves_supported())
3001 msr_info->data = vcpu->arch.ia32_xss;
3004 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3006 /* Otherwise falls through */
3008 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
3010 msr_info->data = msr->data;
3013 return kvm_get_msr_common(vcpu, msr_info);
3019 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3022 * Writes msr value into into the appropriate "register".
3023 * Returns 0 on success, non-0 otherwise.
3024 * Assumes vcpu_load() was already called.
3026 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3028 struct vcpu_vmx *vmx = to_vmx(vcpu);
3029 struct shared_msr_entry *msr;
3031 u32 msr_index = msr_info->index;
3032 u64 data = msr_info->data;
3034 switch (msr_index) {
3036 ret = kvm_set_msr_common(vcpu, msr_info);
3038 #ifdef CONFIG_X86_64
3040 vmx_segment_cache_clear(vmx);
3041 vmcs_writel(GUEST_FS_BASE, data);
3044 vmx_segment_cache_clear(vmx);
3045 vmcs_writel(GUEST_GS_BASE, data);
3047 case MSR_KERNEL_GS_BASE:
3048 vmx_load_host_state(vmx);
3049 vmx->msr_guest_kernel_gs_base = data;
3052 case MSR_IA32_SYSENTER_CS:
3053 vmcs_write32(GUEST_SYSENTER_CS, data);
3055 case MSR_IA32_SYSENTER_EIP:
3056 vmcs_writel(GUEST_SYSENTER_EIP, data);
3058 case MSR_IA32_SYSENTER_ESP:
3059 vmcs_writel(GUEST_SYSENTER_ESP, data);
3061 case MSR_IA32_BNDCFGS:
3062 if (!kvm_mpx_supported())
3064 vmcs_write64(GUEST_BNDCFGS, data);
3067 kvm_write_tsc(vcpu, msr_info);
3069 case MSR_IA32_CR_PAT:
3070 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3071 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3073 vmcs_write64(GUEST_IA32_PAT, data);
3074 vcpu->arch.pat = data;
3077 ret = kvm_set_msr_common(vcpu, msr_info);
3079 case MSR_IA32_TSC_ADJUST:
3080 ret = kvm_set_msr_common(vcpu, msr_info);
3082 case MSR_IA32_MCG_EXT_CTL:
3083 if ((!msr_info->host_initiated &&
3084 !(to_vmx(vcpu)->msr_ia32_feature_control &
3085 FEATURE_CONTROL_LMCE)) ||
3086 (data & ~MCG_EXT_CTL_LMCE_EN))
3088 vcpu->arch.mcg_ext_ctl = data;
3090 case MSR_IA32_FEATURE_CONTROL:
3091 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3092 (to_vmx(vcpu)->msr_ia32_feature_control &
3093 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3095 vmx->msr_ia32_feature_control = data;
3096 if (msr_info->host_initiated && data == 0)
3097 vmx_leave_nested(vcpu);
3099 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3100 return 1; /* they are read-only */
3102 if (!vmx_xsaves_supported())
3105 * The only supported bit as of Skylake is bit 8, but
3106 * it is not supported on KVM.
3110 vcpu->arch.ia32_xss = data;
3111 if (vcpu->arch.ia32_xss != host_xss)
3112 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3113 vcpu->arch.ia32_xss, host_xss);
3115 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3118 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3120 /* Check reserved bit, higher 32 bits should be zero */
3121 if ((data >> 32) != 0)
3123 /* Otherwise falls through */
3125 msr = find_msr_entry(vmx, msr_index);
3127 u64 old_msr_data = msr->data;
3129 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3131 ret = kvm_set_shared_msr(msr->index, msr->data,
3135 msr->data = old_msr_data;
3139 ret = kvm_set_msr_common(vcpu, msr_info);
3145 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3147 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3150 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3153 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3155 case VCPU_EXREG_PDPTR:
3157 ept_save_pdptrs(vcpu);
3164 static __init int cpu_has_kvm_support(void)
3166 return cpu_has_vmx();
3169 static __init int vmx_disabled_by_bios(void)
3173 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3174 if (msr & FEATURE_CONTROL_LOCKED) {
3175 /* launched w/ TXT and VMX disabled */
3176 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3179 /* launched w/o TXT and VMX only enabled w/ TXT */
3180 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3181 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3182 && !tboot_enabled()) {
3183 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3184 "activate TXT before enabling KVM\n");
3187 /* launched w/o TXT and VMX disabled */
3188 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3189 && !tboot_enabled())
3196 static void kvm_cpu_vmxon(u64 addr)
3198 intel_pt_handle_vmx(1);
3200 asm volatile (ASM_VMX_VMXON_RAX
3201 : : "a"(&addr), "m"(addr)
3205 static int hardware_enable(void)
3207 int cpu = raw_smp_processor_id();
3208 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3211 if (cr4_read_shadow() & X86_CR4_VMXE)
3214 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3215 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3216 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3219 * Now we can enable the vmclear operation in kdump
3220 * since the loaded_vmcss_on_cpu list on this cpu
3221 * has been initialized.
3223 * Though the cpu is not in VMX operation now, there
3224 * is no problem to enable the vmclear operation
3225 * for the loaded_vmcss_on_cpu list is empty!
3227 crash_enable_local_vmclear(cpu);
3229 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3231 test_bits = FEATURE_CONTROL_LOCKED;
3232 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3233 if (tboot_enabled())
3234 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3236 if ((old & test_bits) != test_bits) {
3237 /* enable and lock */
3238 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3240 cr4_set_bits(X86_CR4_VMXE);
3242 if (vmm_exclusive) {
3243 kvm_cpu_vmxon(phys_addr);
3247 native_store_gdt(this_cpu_ptr(&host_gdt));
3252 static void vmclear_local_loaded_vmcss(void)
3254 int cpu = raw_smp_processor_id();
3255 struct loaded_vmcs *v, *n;
3257 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3258 loaded_vmcss_on_cpu_link)
3259 __loaded_vmcs_clear(v);
3263 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3266 static void kvm_cpu_vmxoff(void)
3268 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3270 intel_pt_handle_vmx(0);
3273 static void hardware_disable(void)
3275 if (vmm_exclusive) {
3276 vmclear_local_loaded_vmcss();
3279 cr4_clear_bits(X86_CR4_VMXE);
3282 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3283 u32 msr, u32 *result)
3285 u32 vmx_msr_low, vmx_msr_high;
3286 u32 ctl = ctl_min | ctl_opt;
3288 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3290 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3291 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3293 /* Ensure minimum (required) set of control bits are supported. */
3301 static __init bool allow_1_setting(u32 msr, u32 ctl)
3303 u32 vmx_msr_low, vmx_msr_high;
3305 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3306 return vmx_msr_high & ctl;
3309 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3311 u32 vmx_msr_low, vmx_msr_high;
3312 u32 min, opt, min2, opt2;
3313 u32 _pin_based_exec_control = 0;
3314 u32 _cpu_based_exec_control = 0;
3315 u32 _cpu_based_2nd_exec_control = 0;
3316 u32 _vmexit_control = 0;
3317 u32 _vmentry_control = 0;
3319 min = CPU_BASED_HLT_EXITING |
3320 #ifdef CONFIG_X86_64
3321 CPU_BASED_CR8_LOAD_EXITING |
3322 CPU_BASED_CR8_STORE_EXITING |
3324 CPU_BASED_CR3_LOAD_EXITING |
3325 CPU_BASED_CR3_STORE_EXITING |
3326 CPU_BASED_USE_IO_BITMAPS |
3327 CPU_BASED_MOV_DR_EXITING |
3328 CPU_BASED_USE_TSC_OFFSETING |
3329 CPU_BASED_MWAIT_EXITING |
3330 CPU_BASED_MONITOR_EXITING |
3331 CPU_BASED_INVLPG_EXITING |
3332 CPU_BASED_RDPMC_EXITING;
3334 opt = CPU_BASED_TPR_SHADOW |
3335 CPU_BASED_USE_MSR_BITMAPS |
3336 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3337 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3338 &_cpu_based_exec_control) < 0)
3340 #ifdef CONFIG_X86_64
3341 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3342 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3343 ~CPU_BASED_CR8_STORE_EXITING;
3345 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3347 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3348 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3349 SECONDARY_EXEC_WBINVD_EXITING |
3350 SECONDARY_EXEC_ENABLE_VPID |
3351 SECONDARY_EXEC_ENABLE_EPT |
3352 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3353 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3354 SECONDARY_EXEC_RDTSCP |
3355 SECONDARY_EXEC_ENABLE_INVPCID |
3356 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3357 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3358 SECONDARY_EXEC_SHADOW_VMCS |
3359 SECONDARY_EXEC_XSAVES |
3360 SECONDARY_EXEC_ENABLE_PML |
3361 SECONDARY_EXEC_PCOMMIT |
3362 SECONDARY_EXEC_TSC_SCALING;
3363 if (adjust_vmx_controls(min2, opt2,
3364 MSR_IA32_VMX_PROCBASED_CTLS2,
3365 &_cpu_based_2nd_exec_control) < 0)
3368 #ifndef CONFIG_X86_64
3369 if (!(_cpu_based_2nd_exec_control &
3370 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3371 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3374 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3375 _cpu_based_2nd_exec_control &= ~(
3376 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3377 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3378 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3380 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3381 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3383 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3384 CPU_BASED_CR3_STORE_EXITING |
3385 CPU_BASED_INVLPG_EXITING);
3386 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3387 vmx_capability.ept, vmx_capability.vpid);
3390 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
3391 #ifdef CONFIG_X86_64
3392 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3394 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3395 VM_EXIT_CLEAR_BNDCFGS;
3396 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3397 &_vmexit_control) < 0)
3400 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
3401 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
3402 PIN_BASED_VMX_PREEMPTION_TIMER;
3403 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3404 &_pin_based_exec_control) < 0)
3407 if (cpu_has_broken_vmx_preemption_timer())
3408 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3409 if (!(_cpu_based_2nd_exec_control &
3410 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
3411 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3413 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3414 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3415 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3416 &_vmentry_control) < 0)
3419 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3421 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3422 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3425 #ifdef CONFIG_X86_64
3426 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3427 if (vmx_msr_high & (1u<<16))
3431 /* Require Write-Back (WB) memory type for VMCS accesses. */
3432 if (((vmx_msr_high >> 18) & 15) != 6)
3435 vmcs_conf->size = vmx_msr_high & 0x1fff;
3436 vmcs_conf->order = get_order(vmcs_config.size);
3437 vmcs_conf->revision_id = vmx_msr_low;
3439 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3440 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3441 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3442 vmcs_conf->vmexit_ctrl = _vmexit_control;
3443 vmcs_conf->vmentry_ctrl = _vmentry_control;
3445 cpu_has_load_ia32_efer =
3446 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3447 VM_ENTRY_LOAD_IA32_EFER)
3448 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3449 VM_EXIT_LOAD_IA32_EFER);
3451 cpu_has_load_perf_global_ctrl =
3452 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3453 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3454 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3455 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3458 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3459 * but due to errata below it can't be used. Workaround is to use
3460 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3462 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3467 * BC86,AAY89,BD102 (model 44)
3471 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3472 switch (boot_cpu_data.x86_model) {
3478 cpu_has_load_perf_global_ctrl = false;
3479 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3480 "does not work properly. Using workaround\n");
3487 if (boot_cpu_has(X86_FEATURE_XSAVES))
3488 rdmsrl(MSR_IA32_XSS, host_xss);
3493 static struct vmcs *alloc_vmcs_cpu(int cpu)
3495 int node = cpu_to_node(cpu);
3499 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3502 vmcs = page_address(pages);
3503 memset(vmcs, 0, vmcs_config.size);
3504 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3508 static struct vmcs *alloc_vmcs(void)
3510 return alloc_vmcs_cpu(raw_smp_processor_id());
3513 static void free_vmcs(struct vmcs *vmcs)
3515 free_pages((unsigned long)vmcs, vmcs_config.order);
3519 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3521 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3523 if (!loaded_vmcs->vmcs)
3525 loaded_vmcs_clear(loaded_vmcs);
3526 free_vmcs(loaded_vmcs->vmcs);
3527 loaded_vmcs->vmcs = NULL;
3530 static void free_kvm_area(void)
3534 for_each_possible_cpu(cpu) {
3535 free_vmcs(per_cpu(vmxarea, cpu));
3536 per_cpu(vmxarea, cpu) = NULL;
3540 static void init_vmcs_shadow_fields(void)
3544 /* No checks for read only fields yet */
3546 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3547 switch (shadow_read_write_fields[i]) {
3549 if (!kvm_mpx_supported())
3557 shadow_read_write_fields[j] =
3558 shadow_read_write_fields[i];
3561 max_shadow_read_write_fields = j;
3563 /* shadowed fields guest access without vmexit */
3564 for (i = 0; i < max_shadow_read_write_fields; i++) {
3565 clear_bit(shadow_read_write_fields[i],
3566 vmx_vmwrite_bitmap);
3567 clear_bit(shadow_read_write_fields[i],
3570 for (i = 0; i < max_shadow_read_only_fields; i++)
3571 clear_bit(shadow_read_only_fields[i],
3575 static __init int alloc_kvm_area(void)
3579 for_each_possible_cpu(cpu) {
3582 vmcs = alloc_vmcs_cpu(cpu);
3588 per_cpu(vmxarea, cpu) = vmcs;
3593 static bool emulation_required(struct kvm_vcpu *vcpu)
3595 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3598 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3599 struct kvm_segment *save)
3601 if (!emulate_invalid_guest_state) {
3603 * CS and SS RPL should be equal during guest entry according
3604 * to VMX spec, but in reality it is not always so. Since vcpu
3605 * is in the middle of the transition from real mode to
3606 * protected mode it is safe to assume that RPL 0 is a good
3609 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3610 save->selector &= ~SEGMENT_RPL_MASK;
3611 save->dpl = save->selector & SEGMENT_RPL_MASK;
3614 vmx_set_segment(vcpu, save, seg);
3617 static void enter_pmode(struct kvm_vcpu *vcpu)
3619 unsigned long flags;
3620 struct vcpu_vmx *vmx = to_vmx(vcpu);
3623 * Update real mode segment cache. It may be not up-to-date if sement
3624 * register was written while vcpu was in a guest mode.
3626 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3627 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3628 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3629 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3630 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3631 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3633 vmx->rmode.vm86_active = 0;
3635 vmx_segment_cache_clear(vmx);
3637 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3639 flags = vmcs_readl(GUEST_RFLAGS);
3640 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3641 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3642 vmcs_writel(GUEST_RFLAGS, flags);
3644 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3645 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3647 update_exception_bitmap(vcpu);
3649 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3650 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3651 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3652 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3653 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3654 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3657 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3659 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3660 struct kvm_segment var = *save;
3663 if (seg == VCPU_SREG_CS)
3666 if (!emulate_invalid_guest_state) {
3667 var.selector = var.base >> 4;
3668 var.base = var.base & 0xffff0;
3678 if (save->base & 0xf)
3679 printk_once(KERN_WARNING "kvm: segment base is not "
3680 "paragraph aligned when entering "
3681 "protected mode (seg=%d)", seg);
3684 vmcs_write16(sf->selector, var.selector);
3685 vmcs_write32(sf->base, var.base);
3686 vmcs_write32(sf->limit, var.limit);
3687 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3690 static void enter_rmode(struct kvm_vcpu *vcpu)
3692 unsigned long flags;
3693 struct vcpu_vmx *vmx = to_vmx(vcpu);
3695 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3696 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3697 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3698 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3699 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3700 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3701 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3703 vmx->rmode.vm86_active = 1;
3706 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3707 * vcpu. Warn the user that an update is overdue.
3709 if (!vcpu->kvm->arch.tss_addr)
3710 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3711 "called before entering vcpu\n");
3713 vmx_segment_cache_clear(vmx);
3715 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3716 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3717 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3719 flags = vmcs_readl(GUEST_RFLAGS);
3720 vmx->rmode.save_rflags = flags;
3722 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3724 vmcs_writel(GUEST_RFLAGS, flags);
3725 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3726 update_exception_bitmap(vcpu);
3728 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3729 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3730 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3731 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3732 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3733 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3735 kvm_mmu_reset_context(vcpu);
3738 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3740 struct vcpu_vmx *vmx = to_vmx(vcpu);
3741 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3747 * Force kernel_gs_base reloading before EFER changes, as control
3748 * of this msr depends on is_long_mode().
3750 vmx_load_host_state(to_vmx(vcpu));
3751 vcpu->arch.efer = efer;
3752 if (efer & EFER_LMA) {
3753 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3756 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3758 msr->data = efer & ~EFER_LME;
3763 #ifdef CONFIG_X86_64
3765 static void enter_lmode(struct kvm_vcpu *vcpu)
3769 vmx_segment_cache_clear(to_vmx(vcpu));
3771 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3772 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
3773 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3775 vmcs_write32(GUEST_TR_AR_BYTES,
3776 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
3777 | VMX_AR_TYPE_BUSY_64_TSS);
3779 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3782 static void exit_lmode(struct kvm_vcpu *vcpu)
3784 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3785 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3790 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
3792 vpid_sync_context(vpid);
3794 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3796 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3800 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3802 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
3805 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3807 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3809 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3810 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3813 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3815 if (enable_ept && is_paging(vcpu))
3816 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3817 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3820 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3822 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3824 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3825 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3828 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3830 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3832 if (!test_bit(VCPU_EXREG_PDPTR,
3833 (unsigned long *)&vcpu->arch.regs_dirty))
3836 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3837 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3838 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3839 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3840 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3844 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3846 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3848 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3849 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3850 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3851 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3852 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3855 __set_bit(VCPU_EXREG_PDPTR,
3856 (unsigned long *)&vcpu->arch.regs_avail);
3857 __set_bit(VCPU_EXREG_PDPTR,
3858 (unsigned long *)&vcpu->arch.regs_dirty);
3861 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3863 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3865 struct kvm_vcpu *vcpu)
3867 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3868 vmx_decache_cr3(vcpu);
3869 if (!(cr0 & X86_CR0_PG)) {
3870 /* From paging/starting to nonpaging */
3871 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3872 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3873 (CPU_BASED_CR3_LOAD_EXITING |
3874 CPU_BASED_CR3_STORE_EXITING));
3875 vcpu->arch.cr0 = cr0;
3876 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3877 } else if (!is_paging(vcpu)) {
3878 /* From nonpaging to paging */
3879 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3880 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3881 ~(CPU_BASED_CR3_LOAD_EXITING |
3882 CPU_BASED_CR3_STORE_EXITING));
3883 vcpu->arch.cr0 = cr0;
3884 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3887 if (!(cr0 & X86_CR0_WP))
3888 *hw_cr0 &= ~X86_CR0_WP;
3891 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3893 struct vcpu_vmx *vmx = to_vmx(vcpu);
3894 unsigned long hw_cr0;
3896 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3897 if (enable_unrestricted_guest)
3898 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3900 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3902 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3905 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3909 #ifdef CONFIG_X86_64
3910 if (vcpu->arch.efer & EFER_LME) {
3911 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3913 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3919 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3921 if (!vcpu->fpu_active)
3922 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3924 vmcs_writel(CR0_READ_SHADOW, cr0);
3925 vmcs_writel(GUEST_CR0, hw_cr0);
3926 vcpu->arch.cr0 = cr0;
3928 /* depends on vcpu->arch.cr0 to be set to a new value */
3929 vmx->emulation_required = emulation_required(vcpu);
3932 static u64 construct_eptp(unsigned long root_hpa)
3936 /* TODO write the value reading from MSR */
3937 eptp = VMX_EPT_DEFAULT_MT |
3938 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3939 if (enable_ept_ad_bits)
3940 eptp |= VMX_EPT_AD_ENABLE_BIT;
3941 eptp |= (root_hpa & PAGE_MASK);
3946 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3948 unsigned long guest_cr3;
3953 eptp = construct_eptp(cr3);
3954 vmcs_write64(EPT_POINTER, eptp);
3955 if (is_paging(vcpu) || is_guest_mode(vcpu))
3956 guest_cr3 = kvm_read_cr3(vcpu);
3958 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
3959 ept_load_pdptrs(vcpu);
3962 vmx_flush_tlb(vcpu);
3963 vmcs_writel(GUEST_CR3, guest_cr3);
3966 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3969 * Pass through host's Machine Check Enable value to hw_cr4, which
3970 * is in force while we are in guest mode. Do not let guests control
3971 * this bit, even if host CR4.MCE == 0.
3973 unsigned long hw_cr4 =
3974 (cr4_read_shadow() & X86_CR4_MCE) |
3975 (cr4 & ~X86_CR4_MCE) |
3976 (to_vmx(vcpu)->rmode.vm86_active ?
3977 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3979 if (cr4 & X86_CR4_VMXE) {
3981 * To use VMXON (and later other VMX instructions), a guest
3982 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3983 * So basically the check on whether to allow nested VMX
3986 if (!nested_vmx_allowed(vcpu))
3989 if (to_vmx(vcpu)->nested.vmxon &&
3990 ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3993 vcpu->arch.cr4 = cr4;
3995 if (!is_paging(vcpu)) {
3996 hw_cr4 &= ~X86_CR4_PAE;
3997 hw_cr4 |= X86_CR4_PSE;
3998 } else if (!(cr4 & X86_CR4_PAE)) {
3999 hw_cr4 &= ~X86_CR4_PAE;
4003 if (!enable_unrestricted_guest && !is_paging(vcpu))
4005 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4006 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
4007 * to be manually disabled when guest switches to non-paging
4010 * If !enable_unrestricted_guest, the CPU is always running
4011 * with CR0.PG=1 and CR4 needs to be modified.
4012 * If enable_unrestricted_guest, the CPU automatically
4013 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4015 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
4017 vmcs_writel(CR4_READ_SHADOW, cr4);
4018 vmcs_writel(GUEST_CR4, hw_cr4);
4022 static void vmx_get_segment(struct kvm_vcpu *vcpu,
4023 struct kvm_segment *var, int seg)
4025 struct vcpu_vmx *vmx = to_vmx(vcpu);
4028 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4029 *var = vmx->rmode.segs[seg];
4030 if (seg == VCPU_SREG_TR
4031 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
4033 var->base = vmx_read_guest_seg_base(vmx, seg);
4034 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4037 var->base = vmx_read_guest_seg_base(vmx, seg);
4038 var->limit = vmx_read_guest_seg_limit(vmx, seg);
4039 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4040 ar = vmx_read_guest_seg_ar(vmx, seg);
4041 var->unusable = (ar >> 16) & 1;
4042 var->type = ar & 15;
4043 var->s = (ar >> 4) & 1;
4044 var->dpl = (ar >> 5) & 3;
4046 * Some userspaces do not preserve unusable property. Since usable
4047 * segment has to be present according to VMX spec we can use present
4048 * property to amend userspace bug by making unusable segment always
4049 * nonpresent. vmx_segment_access_rights() already marks nonpresent
4050 * segment as unusable.
4052 var->present = !var->unusable;
4053 var->avl = (ar >> 12) & 1;
4054 var->l = (ar >> 13) & 1;
4055 var->db = (ar >> 14) & 1;
4056 var->g = (ar >> 15) & 1;
4059 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
4061 struct kvm_segment s;
4063 if (to_vmx(vcpu)->rmode.vm86_active) {
4064 vmx_get_segment(vcpu, &s, seg);
4067 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
4070 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
4072 struct vcpu_vmx *vmx = to_vmx(vcpu);
4074 if (unlikely(vmx->rmode.vm86_active))
4077 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
4078 return VMX_AR_DPL(ar);
4082 static u32 vmx_segment_access_rights(struct kvm_segment *var)
4086 if (var->unusable || !var->present)
4089 ar = var->type & 15;
4090 ar |= (var->s & 1) << 4;
4091 ar |= (var->dpl & 3) << 5;
4092 ar |= (var->present & 1) << 7;
4093 ar |= (var->avl & 1) << 12;
4094 ar |= (var->l & 1) << 13;
4095 ar |= (var->db & 1) << 14;
4096 ar |= (var->g & 1) << 15;
4102 static void vmx_set_segment(struct kvm_vcpu *vcpu,
4103 struct kvm_segment *var, int seg)
4105 struct vcpu_vmx *vmx = to_vmx(vcpu);
4106 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4108 vmx_segment_cache_clear(vmx);
4110 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4111 vmx->rmode.segs[seg] = *var;
4112 if (seg == VCPU_SREG_TR)
4113 vmcs_write16(sf->selector, var->selector);
4115 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
4119 vmcs_writel(sf->base, var->base);
4120 vmcs_write32(sf->limit, var->limit);
4121 vmcs_write16(sf->selector, var->selector);
4124 * Fix the "Accessed" bit in AR field of segment registers for older
4126 * IA32 arch specifies that at the time of processor reset the
4127 * "Accessed" bit in the AR field of segment registers is 1. And qemu
4128 * is setting it to 0 in the userland code. This causes invalid guest
4129 * state vmexit when "unrestricted guest" mode is turned on.
4130 * Fix for this setup issue in cpu_reset is being pushed in the qemu
4131 * tree. Newer qemu binaries with that qemu fix would not need this
4134 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
4135 var->type |= 0x1; /* Accessed */
4137 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4140 vmx->emulation_required = emulation_required(vcpu);
4143 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4145 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4147 *db = (ar >> 14) & 1;
4148 *l = (ar >> 13) & 1;
4151 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4153 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4154 dt->address = vmcs_readl(GUEST_IDTR_BASE);
4157 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4159 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4160 vmcs_writel(GUEST_IDTR_BASE, dt->address);
4163 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4165 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4166 dt->address = vmcs_readl(GUEST_GDTR_BASE);
4169 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4171 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4172 vmcs_writel(GUEST_GDTR_BASE, dt->address);
4175 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4177 struct kvm_segment var;
4180 vmx_get_segment(vcpu, &var, seg);
4182 if (seg == VCPU_SREG_CS)
4184 ar = vmx_segment_access_rights(&var);
4186 if (var.base != (var.selector << 4))
4188 if (var.limit != 0xffff)
4196 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4198 struct kvm_segment cs;
4199 unsigned int cs_rpl;
4201 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4202 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4206 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4210 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4211 if (cs.dpl > cs_rpl)
4214 if (cs.dpl != cs_rpl)
4220 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4224 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4226 struct kvm_segment ss;
4227 unsigned int ss_rpl;
4229 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4230 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4234 if (ss.type != 3 && ss.type != 7)
4238 if (ss.dpl != ss_rpl) /* DPL != RPL */
4246 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4248 struct kvm_segment var;
4251 vmx_get_segment(vcpu, &var, seg);
4252 rpl = var.selector & SEGMENT_RPL_MASK;
4260 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4261 if (var.dpl < rpl) /* DPL < RPL */
4265 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4271 static bool tr_valid(struct kvm_vcpu *vcpu)
4273 struct kvm_segment tr;
4275 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4279 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4281 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4289 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4291 struct kvm_segment ldtr;
4293 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4297 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4307 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4309 struct kvm_segment cs, ss;
4311 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4312 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4314 return ((cs.selector & SEGMENT_RPL_MASK) ==
4315 (ss.selector & SEGMENT_RPL_MASK));
4319 * Check if guest state is valid. Returns true if valid, false if
4321 * We assume that registers are always usable
4323 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4325 if (enable_unrestricted_guest)
4328 /* real mode guest state checks */
4329 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4330 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4332 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4334 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4336 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4338 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4340 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4343 /* protected mode guest state checks */
4344 if (!cs_ss_rpl_check(vcpu))
4346 if (!code_segment_valid(vcpu))
4348 if (!stack_segment_valid(vcpu))
4350 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4352 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4354 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4356 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4358 if (!tr_valid(vcpu))
4360 if (!ldtr_valid(vcpu))
4364 * - Add checks on RIP
4365 * - Add checks on RFLAGS
4371 static int init_rmode_tss(struct kvm *kvm)
4377 idx = srcu_read_lock(&kvm->srcu);
4378 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4379 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4382 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4383 r = kvm_write_guest_page(kvm, fn++, &data,
4384 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4387 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4390 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4394 r = kvm_write_guest_page(kvm, fn, &data,
4395 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4398 srcu_read_unlock(&kvm->srcu, idx);
4402 static int init_rmode_identity_map(struct kvm *kvm)
4405 kvm_pfn_t identity_map_pfn;
4411 /* Protect kvm->arch.ept_identity_pagetable_done. */
4412 mutex_lock(&kvm->slots_lock);
4414 if (likely(kvm->arch.ept_identity_pagetable_done))
4417 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4419 r = alloc_identity_pagetable(kvm);
4423 idx = srcu_read_lock(&kvm->srcu);
4424 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4427 /* Set up identity-mapping pagetable for EPT in real mode */
4428 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4429 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4430 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4431 r = kvm_write_guest_page(kvm, identity_map_pfn,
4432 &tmp, i * sizeof(tmp), sizeof(tmp));
4436 kvm->arch.ept_identity_pagetable_done = true;
4439 srcu_read_unlock(&kvm->srcu, idx);
4442 mutex_unlock(&kvm->slots_lock);
4446 static void seg_setup(int seg)
4448 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4451 vmcs_write16(sf->selector, 0);
4452 vmcs_writel(sf->base, 0);
4453 vmcs_write32(sf->limit, 0xffff);
4455 if (seg == VCPU_SREG_CS)
4456 ar |= 0x08; /* code segment */
4458 vmcs_write32(sf->ar_bytes, ar);
4461 static int alloc_apic_access_page(struct kvm *kvm)
4466 mutex_lock(&kvm->slots_lock);
4467 if (kvm->arch.apic_access_page_done)
4469 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4470 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4474 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4475 if (is_error_page(page)) {
4481 * Do not pin the page in memory, so that memory hot-unplug
4482 * is able to migrate it.
4485 kvm->arch.apic_access_page_done = true;
4487 mutex_unlock(&kvm->slots_lock);
4491 static int alloc_identity_pagetable(struct kvm *kvm)
4493 /* Called with kvm->slots_lock held. */
4497 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4499 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4500 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4505 static int allocate_vpid(void)
4511 spin_lock(&vmx_vpid_lock);
4512 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4513 if (vpid < VMX_NR_VPIDS)
4514 __set_bit(vpid, vmx_vpid_bitmap);
4517 spin_unlock(&vmx_vpid_lock);
4521 static void free_vpid(int vpid)
4523 if (!enable_vpid || vpid == 0)
4525 spin_lock(&vmx_vpid_lock);
4526 __clear_bit(vpid, vmx_vpid_bitmap);
4527 spin_unlock(&vmx_vpid_lock);
4530 #define MSR_TYPE_R 1
4531 #define MSR_TYPE_W 2
4532 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4535 int f = sizeof(unsigned long);
4537 if (!cpu_has_vmx_msr_bitmap())
4541 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4542 * have the write-low and read-high bitmap offsets the wrong way round.
4543 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4545 if (msr <= 0x1fff) {
4546 if (type & MSR_TYPE_R)
4548 __clear_bit(msr, msr_bitmap + 0x000 / f);
4550 if (type & MSR_TYPE_W)
4552 __clear_bit(msr, msr_bitmap + 0x800 / f);
4554 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4556 if (type & MSR_TYPE_R)
4558 __clear_bit(msr, msr_bitmap + 0x400 / f);
4560 if (type & MSR_TYPE_W)
4562 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4567 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
4570 int f = sizeof(unsigned long);
4572 if (!cpu_has_vmx_msr_bitmap())
4576 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4577 * have the write-low and read-high bitmap offsets the wrong way round.
4578 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4580 if (msr <= 0x1fff) {
4581 if (type & MSR_TYPE_R)
4583 __set_bit(msr, msr_bitmap + 0x000 / f);
4585 if (type & MSR_TYPE_W)
4587 __set_bit(msr, msr_bitmap + 0x800 / f);
4589 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4591 if (type & MSR_TYPE_R)
4593 __set_bit(msr, msr_bitmap + 0x400 / f);
4595 if (type & MSR_TYPE_W)
4597 __set_bit(msr, msr_bitmap + 0xc00 / f);
4603 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4604 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4606 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4607 unsigned long *msr_bitmap_nested,
4610 int f = sizeof(unsigned long);
4612 if (!cpu_has_vmx_msr_bitmap()) {
4618 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4619 * have the write-low and read-high bitmap offsets the wrong way round.
4620 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4622 if (msr <= 0x1fff) {
4623 if (type & MSR_TYPE_R &&
4624 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4626 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4628 if (type & MSR_TYPE_W &&
4629 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4631 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4633 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4635 if (type & MSR_TYPE_R &&
4636 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4638 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4640 if (type & MSR_TYPE_W &&
4641 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4643 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4648 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4651 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4652 msr, MSR_TYPE_R | MSR_TYPE_W);
4653 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4654 msr, MSR_TYPE_R | MSR_TYPE_W);
4657 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4659 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4661 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4665 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4667 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4669 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4673 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4675 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4677 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4681 static bool vmx_get_enable_apicv(void)
4683 return enable_apicv;
4686 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
4688 struct vcpu_vmx *vmx = to_vmx(vcpu);
4693 if (vmx->nested.pi_desc &&
4694 vmx->nested.pi_pending) {
4695 vmx->nested.pi_pending = false;
4696 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
4699 max_irr = find_last_bit(
4700 (unsigned long *)vmx->nested.pi_desc->pir, 256);
4705 vapic_page = kmap(vmx->nested.virtual_apic_page);
4710 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
4711 kunmap(vmx->nested.virtual_apic_page);
4713 status = vmcs_read16(GUEST_INTR_STATUS);
4714 if ((u8)max_irr > ((u8)status & 0xff)) {
4716 status |= (u8)max_irr;
4717 vmcs_write16(GUEST_INTR_STATUS, status);
4723 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu)
4726 if (vcpu->mode == IN_GUEST_MODE) {
4727 struct vcpu_vmx *vmx = to_vmx(vcpu);
4730 * Currently, we don't support urgent interrupt,
4731 * all interrupts are recognized as non-urgent
4732 * interrupt, so we cannot post interrupts when
4735 * If the vcpu is in guest mode, it means it is
4736 * running instead of being scheduled out and
4737 * waiting in the run queue, and that's the only
4738 * case when 'SN' is set currently, warning if
4741 WARN_ON_ONCE(pi_test_sn(&vmx->pi_desc));
4743 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4744 POSTED_INTR_VECTOR);
4751 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4754 struct vcpu_vmx *vmx = to_vmx(vcpu);
4756 if (is_guest_mode(vcpu) &&
4757 vector == vmx->nested.posted_intr_nv) {
4758 /* the PIR and ON have been set by L1. */
4759 kvm_vcpu_trigger_posted_interrupt(vcpu);
4761 * If a posted intr is not recognized by hardware,
4762 * we will accomplish it in the next vmentry.
4764 vmx->nested.pi_pending = true;
4765 kvm_make_request(KVM_REQ_EVENT, vcpu);
4771 * Send interrupt to vcpu via posted interrupt way.
4772 * 1. If target vcpu is running(non-root mode), send posted interrupt
4773 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4774 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4775 * interrupt from PIR in next vmentry.
4777 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4779 struct vcpu_vmx *vmx = to_vmx(vcpu);
4782 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4786 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4789 r = pi_test_and_set_on(&vmx->pi_desc);
4790 kvm_make_request(KVM_REQ_EVENT, vcpu);
4791 if (r || !kvm_vcpu_trigger_posted_interrupt(vcpu))
4792 kvm_vcpu_kick(vcpu);
4795 static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4797 struct vcpu_vmx *vmx = to_vmx(vcpu);
4799 if (!pi_test_and_clear_on(&vmx->pi_desc))
4802 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4806 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4807 * will not change in the lifetime of the guest.
4808 * Note that host-state that does change is set elsewhere. E.g., host-state
4809 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4811 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4818 vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */
4819 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
4821 /* Save the most likely value for this task's CR4 in the VMCS. */
4822 cr4 = cr4_read_shadow();
4823 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
4824 vmx->host_state.vmcs_host_cr4 = cr4;
4826 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
4827 #ifdef CONFIG_X86_64
4829 * Load null selectors, so we can avoid reloading them in
4830 * __vmx_load_host_state(), in case userspace uses the null selectors
4831 * too (the expected case).
4833 vmcs_write16(HOST_DS_SELECTOR, 0);
4834 vmcs_write16(HOST_ES_SELECTOR, 0);
4836 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4837 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4839 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4840 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
4842 native_store_idt(&dt);
4843 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
4844 vmx->host_idt_base = dt.address;
4846 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4848 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4849 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4850 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4851 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4853 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4854 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4855 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4859 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4861 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4863 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4864 if (is_guest_mode(&vmx->vcpu))
4865 vmx->vcpu.arch.cr4_guest_owned_bits &=
4866 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4867 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4870 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4872 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4874 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4875 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4876 /* Enable the preemption timer dynamically */
4877 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
4878 return pin_based_exec_ctrl;
4881 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
4883 struct vcpu_vmx *vmx = to_vmx(vcpu);
4885 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4886 if (cpu_has_secondary_exec_ctrls()) {
4887 if (kvm_vcpu_apicv_active(vcpu))
4888 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
4889 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4890 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4892 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
4893 SECONDARY_EXEC_APIC_REGISTER_VIRT |
4894 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4897 if (cpu_has_vmx_msr_bitmap())
4898 vmx_set_msr_bitmap(vcpu);
4901 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4903 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4905 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4906 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4908 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4909 exec_control &= ~CPU_BASED_TPR_SHADOW;
4910 #ifdef CONFIG_X86_64
4911 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4912 CPU_BASED_CR8_LOAD_EXITING;
4916 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4917 CPU_BASED_CR3_LOAD_EXITING |
4918 CPU_BASED_INVLPG_EXITING;
4919 return exec_control;
4922 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4924 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4925 if (!cpu_need_virtualize_apic_accesses(&vmx->vcpu))
4926 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4928 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4930 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4931 enable_unrestricted_guest = 0;
4932 /* Enable INVPCID for non-ept guests may cause performance regression. */
4933 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4935 if (!enable_unrestricted_guest)
4936 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4938 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4939 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
4940 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4941 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4942 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4943 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4945 We can NOT enable shadow_vmcs here because we don't have yet
4948 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4951 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4953 /* Currently, we allow L1 guest to directly run pcommit instruction. */
4954 exec_control &= ~SECONDARY_EXEC_PCOMMIT;
4956 return exec_control;
4959 static void ept_set_mmio_spte_mask(void)
4962 * EPT Misconfigurations can be generated if the value of bits 2:0
4963 * of an EPT paging-structure entry is 110b (write/execute).
4964 * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4967 kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4970 #define VMX_XSS_EXIT_BITMAP 0
4972 * Sets up the vmcs for emulated real mode.
4974 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4976 #ifdef CONFIG_X86_64
4982 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4983 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4985 if (enable_shadow_vmcs) {
4986 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4987 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4989 if (cpu_has_vmx_msr_bitmap())
4990 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4992 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4995 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4996 vmx->hv_deadline_tsc = -1;
4998 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
5000 if (cpu_has_secondary_exec_ctrls())
5001 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5002 vmx_secondary_exec_control(vmx));
5004 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
5005 vmcs_write64(EOI_EXIT_BITMAP0, 0);
5006 vmcs_write64(EOI_EXIT_BITMAP1, 0);
5007 vmcs_write64(EOI_EXIT_BITMAP2, 0);
5008 vmcs_write64(EOI_EXIT_BITMAP3, 0);
5010 vmcs_write16(GUEST_INTR_STATUS, 0);
5012 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
5013 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
5017 vmcs_write32(PLE_GAP, ple_gap);
5018 vmx->ple_window = ple_window;
5019 vmx->ple_window_dirty = true;
5022 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
5023 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
5024 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
5026 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
5027 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
5028 vmx_set_constant_host_state(vmx);
5029 #ifdef CONFIG_X86_64
5030 rdmsrl(MSR_FS_BASE, a);
5031 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
5032 rdmsrl(MSR_GS_BASE, a);
5033 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
5035 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
5036 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
5039 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
5040 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
5041 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
5042 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
5043 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
5045 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
5046 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
5048 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
5049 u32 index = vmx_msr_index[i];
5050 u32 data_low, data_high;
5053 if (rdmsr_safe(index, &data_low, &data_high) < 0)
5055 if (wrmsr_safe(index, data_low, data_high) < 0)
5057 vmx->guest_msrs[j].index = i;
5058 vmx->guest_msrs[j].data = 0;
5059 vmx->guest_msrs[j].mask = -1ull;
5064 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
5066 /* 22.2.1, 20.8.1 */
5067 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
5069 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
5070 set_cr4_guest_host_mask(vmx);
5072 if (vmx_xsaves_supported())
5073 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
5078 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
5080 struct vcpu_vmx *vmx = to_vmx(vcpu);
5081 struct msr_data apic_base_msr;
5084 vmx->rmode.vm86_active = 0;
5086 vmx->soft_vnmi_blocked = 0;
5088 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
5089 kvm_set_cr8(vcpu, 0);
5092 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
5093 MSR_IA32_APICBASE_ENABLE;
5094 if (kvm_vcpu_is_reset_bsp(vcpu))
5095 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
5096 apic_base_msr.host_initiated = true;
5097 kvm_set_apic_base(vcpu, &apic_base_msr);
5100 vmx_segment_cache_clear(vmx);
5102 seg_setup(VCPU_SREG_CS);
5103 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
5104 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
5106 seg_setup(VCPU_SREG_DS);
5107 seg_setup(VCPU_SREG_ES);
5108 seg_setup(VCPU_SREG_FS);
5109 seg_setup(VCPU_SREG_GS);
5110 seg_setup(VCPU_SREG_SS);
5112 vmcs_write16(GUEST_TR_SELECTOR, 0);
5113 vmcs_writel(GUEST_TR_BASE, 0);
5114 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
5115 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
5117 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
5118 vmcs_writel(GUEST_LDTR_BASE, 0);
5119 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
5120 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
5123 vmcs_write32(GUEST_SYSENTER_CS, 0);
5124 vmcs_writel(GUEST_SYSENTER_ESP, 0);
5125 vmcs_writel(GUEST_SYSENTER_EIP, 0);
5126 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
5129 vmcs_writel(GUEST_RFLAGS, 0x02);
5130 kvm_rip_write(vcpu, 0xfff0);
5132 vmcs_writel(GUEST_GDTR_BASE, 0);
5133 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
5135 vmcs_writel(GUEST_IDTR_BASE, 0);
5136 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
5138 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
5139 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
5140 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
5144 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
5146 if (cpu_has_vmx_tpr_shadow() && !init_event) {
5147 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
5148 if (cpu_need_tpr_shadow(vcpu))
5149 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
5150 __pa(vcpu->arch.apic->regs));
5151 vmcs_write32(TPR_THRESHOLD, 0);
5154 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5156 if (kvm_vcpu_apicv_active(vcpu))
5157 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
5160 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5162 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5163 vmx->vcpu.arch.cr0 = cr0;
5164 vmx_set_cr0(vcpu, cr0); /* enter rmode */
5165 vmx_set_cr4(vcpu, 0);
5166 vmx_set_efer(vcpu, 0);
5167 vmx_fpu_activate(vcpu);
5168 update_exception_bitmap(vcpu);
5170 vpid_sync_context(vmx->vpid);
5174 * In nested virtualization, check if L1 asked to exit on external interrupts.
5175 * For most existing hypervisors, this will always return true.
5177 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5179 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5180 PIN_BASED_EXT_INTR_MASK;
5184 * In nested virtualization, check if L1 has set
5185 * VM_EXIT_ACK_INTR_ON_EXIT
5187 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5189 return get_vmcs12(vcpu)->vm_exit_controls &
5190 VM_EXIT_ACK_INTR_ON_EXIT;
5193 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5195 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5196 PIN_BASED_NMI_EXITING;
5199 static void enable_irq_window(struct kvm_vcpu *vcpu)
5201 u32 cpu_based_vm_exec_control;
5203 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5204 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
5205 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5208 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5210 u32 cpu_based_vm_exec_control;
5212 if (!cpu_has_virtual_nmis() ||
5213 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5214 enable_irq_window(vcpu);
5218 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5219 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
5220 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5223 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5225 struct vcpu_vmx *vmx = to_vmx(vcpu);
5227 int irq = vcpu->arch.interrupt.nr;
5229 trace_kvm_inj_virq(irq);
5231 ++vcpu->stat.irq_injections;
5232 if (vmx->rmode.vm86_active) {
5234 if (vcpu->arch.interrupt.soft)
5235 inc_eip = vcpu->arch.event_exit_inst_len;
5236 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5237 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5240 intr = irq | INTR_INFO_VALID_MASK;
5241 if (vcpu->arch.interrupt.soft) {
5242 intr |= INTR_TYPE_SOFT_INTR;
5243 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5244 vmx->vcpu.arch.event_exit_inst_len);
5246 intr |= INTR_TYPE_EXT_INTR;
5247 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5250 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5252 struct vcpu_vmx *vmx = to_vmx(vcpu);
5254 if (is_guest_mode(vcpu))
5257 if (!cpu_has_virtual_nmis()) {
5259 * Tracking the NMI-blocked state in software is built upon
5260 * finding the next open IRQ window. This, in turn, depends on
5261 * well-behaving guests: They have to keep IRQs disabled at
5262 * least as long as the NMI handler runs. Otherwise we may
5263 * cause NMI nesting, maybe breaking the guest. But as this is
5264 * highly unlikely, we can live with the residual risk.
5266 vmx->soft_vnmi_blocked = 1;
5267 vmx->vnmi_blocked_time = 0;
5270 ++vcpu->stat.nmi_injections;
5271 vmx->nmi_known_unmasked = false;
5272 if (vmx->rmode.vm86_active) {
5273 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5274 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5277 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5278 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5281 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5283 if (!cpu_has_virtual_nmis())
5284 return to_vmx(vcpu)->soft_vnmi_blocked;
5285 if (to_vmx(vcpu)->nmi_known_unmasked)
5287 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5290 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5292 struct vcpu_vmx *vmx = to_vmx(vcpu);
5294 if (!cpu_has_virtual_nmis()) {
5295 if (vmx->soft_vnmi_blocked != masked) {
5296 vmx->soft_vnmi_blocked = masked;
5297 vmx->vnmi_blocked_time = 0;
5300 vmx->nmi_known_unmasked = !masked;
5302 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5303 GUEST_INTR_STATE_NMI);
5305 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5306 GUEST_INTR_STATE_NMI);
5310 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5312 if (to_vmx(vcpu)->nested.nested_run_pending)
5315 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
5318 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5319 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5320 | GUEST_INTR_STATE_NMI));
5323 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5325 return (!to_vmx(vcpu)->nested.nested_run_pending &&
5326 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5327 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5328 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5331 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5335 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5339 kvm->arch.tss_addr = addr;
5340 return init_rmode_tss(kvm);
5343 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5348 * Update instruction length as we may reinject the exception
5349 * from user space while in guest debugging mode.
5351 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5352 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5353 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5357 if (vcpu->guest_debug &
5358 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5375 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5376 int vec, u32 err_code)
5379 * Instruction with address size override prefix opcode 0x67
5380 * Cause the #SS fault with 0 error code in VM86 mode.
5382 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5383 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5384 if (vcpu->arch.halt_request) {
5385 vcpu->arch.halt_request = 0;
5386 return kvm_vcpu_halt(vcpu);
5394 * Forward all other exceptions that are valid in real mode.
5395 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5396 * the required debugging infrastructure rework.
5398 kvm_queue_exception(vcpu, vec);
5403 * Trigger machine check on the host. We assume all the MSRs are already set up
5404 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5405 * We pass a fake environment to the machine check handler because we want
5406 * the guest to be always treated like user space, no matter what context
5407 * it used internally.
5409 static void kvm_machine_check(void)
5411 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5412 struct pt_regs regs = {
5413 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5414 .flags = X86_EFLAGS_IF,
5417 do_machine_check(®s, 0);
5421 static int handle_machine_check(struct kvm_vcpu *vcpu)
5423 /* already handled by vcpu_run */
5427 static int handle_exception(struct kvm_vcpu *vcpu)
5429 struct vcpu_vmx *vmx = to_vmx(vcpu);
5430 struct kvm_run *kvm_run = vcpu->run;
5431 u32 intr_info, ex_no, error_code;
5432 unsigned long cr2, rip, dr6;
5434 enum emulation_result er;
5436 vect_info = vmx->idt_vectoring_info;
5437 intr_info = vmx->exit_intr_info;
5439 if (is_machine_check(intr_info))
5440 return handle_machine_check(vcpu);
5442 if ((intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR)
5443 return 1; /* already handled by vmx_vcpu_run() */
5445 if (is_no_device(intr_info)) {
5446 vmx_fpu_activate(vcpu);
5450 if (is_invalid_opcode(intr_info)) {
5451 if (is_guest_mode(vcpu)) {
5452 kvm_queue_exception(vcpu, UD_VECTOR);
5455 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5456 if (er != EMULATE_DONE)
5457 kvm_queue_exception(vcpu, UD_VECTOR);
5462 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5463 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5466 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5467 * MMIO, it is better to report an internal error.
5468 * See the comments in vmx_handle_exit.
5470 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5471 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5472 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5473 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5474 vcpu->run->internal.ndata = 3;
5475 vcpu->run->internal.data[0] = vect_info;
5476 vcpu->run->internal.data[1] = intr_info;
5477 vcpu->run->internal.data[2] = error_code;
5481 if (is_page_fault(intr_info)) {
5482 /* EPT won't cause page fault directly */
5484 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5485 trace_kvm_page_fault(cr2, error_code);
5487 if (kvm_event_needs_reinjection(vcpu))
5488 kvm_mmu_unprotect_page_virt(vcpu, cr2);
5489 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
5492 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5494 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5495 return handle_rmode_exception(vcpu, ex_no, error_code);
5499 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5502 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5503 if (!(vcpu->guest_debug &
5504 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5505 vcpu->arch.dr6 &= ~15;
5506 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5507 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5508 skip_emulated_instruction(vcpu);
5510 kvm_queue_exception(vcpu, DB_VECTOR);
5513 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5514 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5518 * Update instruction length as we may reinject #BP from
5519 * user space while in guest debugging mode. Reading it for
5520 * #DB as well causes no harm, it is not used in that case.
5522 vmx->vcpu.arch.event_exit_inst_len =
5523 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5524 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5525 rip = kvm_rip_read(vcpu);
5526 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5527 kvm_run->debug.arch.exception = ex_no;
5530 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5531 kvm_run->ex.exception = ex_no;
5532 kvm_run->ex.error_code = error_code;
5538 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5540 ++vcpu->stat.irq_exits;
5544 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5546 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5550 static int handle_io(struct kvm_vcpu *vcpu)
5552 unsigned long exit_qualification;
5553 int size, in, string;
5556 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5557 string = (exit_qualification & 16) != 0;
5558 in = (exit_qualification & 8) != 0;
5560 ++vcpu->stat.io_exits;
5563 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5565 port = exit_qualification >> 16;
5566 size = (exit_qualification & 7) + 1;
5567 skip_emulated_instruction(vcpu);
5569 return kvm_fast_pio_out(vcpu, size, port);
5573 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5576 * Patch in the VMCALL instruction:
5578 hypercall[0] = 0x0f;
5579 hypercall[1] = 0x01;
5580 hypercall[2] = 0xc1;
5583 static bool nested_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
5585 unsigned long always_on = VMXON_CR0_ALWAYSON;
5586 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5588 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
5589 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
5590 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
5591 always_on &= ~(X86_CR0_PE | X86_CR0_PG);
5592 return (val & always_on) == always_on;
5595 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5596 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5598 if (is_guest_mode(vcpu)) {
5599 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5600 unsigned long orig_val = val;
5603 * We get here when L2 changed cr0 in a way that did not change
5604 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5605 * but did change L0 shadowed bits. So we first calculate the
5606 * effective cr0 value that L1 would like to write into the
5607 * hardware. It consists of the L2-owned bits from the new
5608 * value combined with the L1-owned bits from L1's guest_cr0.
5610 val = (val & ~vmcs12->cr0_guest_host_mask) |
5611 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5613 if (!nested_cr0_valid(vcpu, val))
5616 if (kvm_set_cr0(vcpu, val))
5618 vmcs_writel(CR0_READ_SHADOW, orig_val);
5621 if (to_vmx(vcpu)->nested.vmxon &&
5622 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
5624 return kvm_set_cr0(vcpu, val);
5628 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5630 if (is_guest_mode(vcpu)) {
5631 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5632 unsigned long orig_val = val;
5634 /* analogously to handle_set_cr0 */
5635 val = (val & ~vmcs12->cr4_guest_host_mask) |
5636 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5637 if (kvm_set_cr4(vcpu, val))
5639 vmcs_writel(CR4_READ_SHADOW, orig_val);
5642 return kvm_set_cr4(vcpu, val);
5645 /* called to set cr0 as appropriate for clts instruction exit. */
5646 static void handle_clts(struct kvm_vcpu *vcpu)
5648 if (is_guest_mode(vcpu)) {
5650 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
5651 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
5652 * just pretend it's off (also in arch.cr0 for fpu_activate).
5654 vmcs_writel(CR0_READ_SHADOW,
5655 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
5656 vcpu->arch.cr0 &= ~X86_CR0_TS;
5658 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5661 static int handle_cr(struct kvm_vcpu *vcpu)
5663 unsigned long exit_qualification, val;
5668 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5669 cr = exit_qualification & 15;
5670 reg = (exit_qualification >> 8) & 15;
5671 switch ((exit_qualification >> 4) & 3) {
5672 case 0: /* mov to cr */
5673 val = kvm_register_readl(vcpu, reg);
5674 trace_kvm_cr_write(cr, val);
5677 err = handle_set_cr0(vcpu, val);
5678 kvm_complete_insn_gp(vcpu, err);
5681 err = kvm_set_cr3(vcpu, val);
5682 kvm_complete_insn_gp(vcpu, err);
5685 err = handle_set_cr4(vcpu, val);
5686 kvm_complete_insn_gp(vcpu, err);
5689 u8 cr8_prev = kvm_get_cr8(vcpu);
5691 err = kvm_set_cr8(vcpu, cr8);
5692 kvm_complete_insn_gp(vcpu, err);
5693 if (lapic_in_kernel(vcpu))
5695 if (cr8_prev <= cr8)
5697 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5704 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5705 skip_emulated_instruction(vcpu);
5706 vmx_fpu_activate(vcpu);
5708 case 1: /*mov from cr*/
5711 val = kvm_read_cr3(vcpu);
5712 kvm_register_write(vcpu, reg, val);
5713 trace_kvm_cr_read(cr, val);
5714 skip_emulated_instruction(vcpu);
5717 val = kvm_get_cr8(vcpu);
5718 kvm_register_write(vcpu, reg, val);
5719 trace_kvm_cr_read(cr, val);
5720 skip_emulated_instruction(vcpu);
5725 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5726 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5727 kvm_lmsw(vcpu, val);
5729 skip_emulated_instruction(vcpu);
5734 vcpu->run->exit_reason = 0;
5735 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5736 (int)(exit_qualification >> 4) & 3, cr);
5740 static int handle_dr(struct kvm_vcpu *vcpu)
5742 unsigned long exit_qualification;
5745 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5746 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5748 /* First, if DR does not exist, trigger UD */
5749 if (!kvm_require_dr(vcpu, dr))
5752 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5753 if (!kvm_require_cpl(vcpu, 0))
5755 dr7 = vmcs_readl(GUEST_DR7);
5758 * As the vm-exit takes precedence over the debug trap, we
5759 * need to emulate the latter, either for the host or the
5760 * guest debugging itself.
5762 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5763 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5764 vcpu->run->debug.arch.dr7 = dr7;
5765 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5766 vcpu->run->debug.arch.exception = DB_VECTOR;
5767 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5770 vcpu->arch.dr6 &= ~15;
5771 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5772 kvm_queue_exception(vcpu, DB_VECTOR);
5777 if (vcpu->guest_debug == 0) {
5778 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
5779 CPU_BASED_MOV_DR_EXITING);
5782 * No more DR vmexits; force a reload of the debug registers
5783 * and reenter on this instruction. The next vmexit will
5784 * retrieve the full state of the debug registers.
5786 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5790 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5791 if (exit_qualification & TYPE_MOV_FROM_DR) {
5794 if (kvm_get_dr(vcpu, dr, &val))
5796 kvm_register_write(vcpu, reg, val);
5798 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5801 skip_emulated_instruction(vcpu);
5805 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5807 return vcpu->arch.dr6;
5810 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5814 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5816 get_debugreg(vcpu->arch.db[0], 0);
5817 get_debugreg(vcpu->arch.db[1], 1);
5818 get_debugreg(vcpu->arch.db[2], 2);
5819 get_debugreg(vcpu->arch.db[3], 3);
5820 get_debugreg(vcpu->arch.dr6, 6);
5821 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5823 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5824 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
5827 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5829 vmcs_writel(GUEST_DR7, val);
5832 static int handle_cpuid(struct kvm_vcpu *vcpu)
5834 kvm_emulate_cpuid(vcpu);
5838 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5840 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5841 struct msr_data msr_info;
5843 msr_info.index = ecx;
5844 msr_info.host_initiated = false;
5845 if (vmx_get_msr(vcpu, &msr_info)) {
5846 trace_kvm_msr_read_ex(ecx);
5847 kvm_inject_gp(vcpu, 0);
5851 trace_kvm_msr_read(ecx, msr_info.data);
5853 /* FIXME: handling of bits 32:63 of rax, rdx */
5854 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
5855 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
5856 skip_emulated_instruction(vcpu);
5860 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5862 struct msr_data msr;
5863 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5864 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5865 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5869 msr.host_initiated = false;
5870 if (kvm_set_msr(vcpu, &msr) != 0) {
5871 trace_kvm_msr_write_ex(ecx, data);
5872 kvm_inject_gp(vcpu, 0);
5876 trace_kvm_msr_write(ecx, data);
5877 skip_emulated_instruction(vcpu);
5881 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5883 kvm_make_request(KVM_REQ_EVENT, vcpu);
5887 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5889 u32 cpu_based_vm_exec_control;
5891 /* clear pending irq */
5892 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5893 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5894 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5896 kvm_make_request(KVM_REQ_EVENT, vcpu);
5898 ++vcpu->stat.irq_window_exits;
5902 static int handle_halt(struct kvm_vcpu *vcpu)
5904 return kvm_emulate_halt(vcpu);
5907 static int handle_vmcall(struct kvm_vcpu *vcpu)
5909 return kvm_emulate_hypercall(vcpu);
5912 static int handle_invd(struct kvm_vcpu *vcpu)
5914 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5917 static int handle_invlpg(struct kvm_vcpu *vcpu)
5919 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5921 kvm_mmu_invlpg(vcpu, exit_qualification);
5922 skip_emulated_instruction(vcpu);
5926 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5930 err = kvm_rdpmc(vcpu);
5931 kvm_complete_insn_gp(vcpu, err);
5936 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5938 kvm_emulate_wbinvd(vcpu);
5942 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5944 u64 new_bv = kvm_read_edx_eax(vcpu);
5945 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5947 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5948 skip_emulated_instruction(vcpu);
5952 static int handle_xsaves(struct kvm_vcpu *vcpu)
5954 skip_emulated_instruction(vcpu);
5955 WARN(1, "this should never happen\n");
5959 static int handle_xrstors(struct kvm_vcpu *vcpu)
5961 skip_emulated_instruction(vcpu);
5962 WARN(1, "this should never happen\n");
5966 static int handle_apic_access(struct kvm_vcpu *vcpu)
5968 if (likely(fasteoi)) {
5969 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5970 int access_type, offset;
5972 access_type = exit_qualification & APIC_ACCESS_TYPE;
5973 offset = exit_qualification & APIC_ACCESS_OFFSET;
5975 * Sane guest uses MOV to write EOI, with written value
5976 * not cared. So make a short-circuit here by avoiding
5977 * heavy instruction emulation.
5979 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5980 (offset == APIC_EOI)) {
5981 kvm_lapic_set_eoi(vcpu);
5982 skip_emulated_instruction(vcpu);
5986 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5989 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5991 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5992 int vector = exit_qualification & 0xff;
5994 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5995 kvm_apic_set_eoi_accelerated(vcpu, vector);
5999 static int handle_apic_write(struct kvm_vcpu *vcpu)
6001 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6002 u32 offset = exit_qualification & 0xfff;
6004 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6005 kvm_apic_write_nodecode(vcpu, offset);
6009 static int handle_task_switch(struct kvm_vcpu *vcpu)
6011 struct vcpu_vmx *vmx = to_vmx(vcpu);
6012 unsigned long exit_qualification;
6013 bool has_error_code = false;
6016 int reason, type, idt_v, idt_index;
6018 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
6019 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
6020 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
6022 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6024 reason = (u32)exit_qualification >> 30;
6025 if (reason == TASK_SWITCH_GATE && idt_v) {
6027 case INTR_TYPE_NMI_INTR:
6028 vcpu->arch.nmi_injected = false;
6029 vmx_set_nmi_mask(vcpu, true);
6031 case INTR_TYPE_EXT_INTR:
6032 case INTR_TYPE_SOFT_INTR:
6033 kvm_clear_interrupt_queue(vcpu);
6035 case INTR_TYPE_HARD_EXCEPTION:
6036 if (vmx->idt_vectoring_info &
6037 VECTORING_INFO_DELIVER_CODE_MASK) {
6038 has_error_code = true;
6040 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6043 case INTR_TYPE_SOFT_EXCEPTION:
6044 kvm_clear_exception_queue(vcpu);
6050 tss_selector = exit_qualification;
6052 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
6053 type != INTR_TYPE_EXT_INTR &&
6054 type != INTR_TYPE_NMI_INTR))
6055 skip_emulated_instruction(vcpu);
6057 if (kvm_task_switch(vcpu, tss_selector,
6058 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
6059 has_error_code, error_code) == EMULATE_FAIL) {
6060 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6061 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6062 vcpu->run->internal.ndata = 0;
6067 * TODO: What about debug traps on tss switch?
6068 * Are we supposed to inject them and update dr6?
6074 static int handle_ept_violation(struct kvm_vcpu *vcpu)
6076 unsigned long exit_qualification;
6081 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6083 gla_validity = (exit_qualification >> 7) & 0x3;
6084 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
6085 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
6086 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
6087 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
6088 vmcs_readl(GUEST_LINEAR_ADDRESS));
6089 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
6090 (long unsigned int)exit_qualification);
6091 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6092 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
6097 * EPT violation happened while executing iret from NMI,
6098 * "blocked by NMI" bit has to be set before next VM entry.
6099 * There are errata that may cause this bit to not be set:
6102 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
6103 cpu_has_virtual_nmis() &&
6104 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
6105 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
6107 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6108 trace_kvm_page_fault(gpa, exit_qualification);
6110 /* It is a write fault? */
6111 error_code = exit_qualification & PFERR_WRITE_MASK;
6112 /* It is a fetch fault? */
6113 error_code |= (exit_qualification << 2) & PFERR_FETCH_MASK;
6114 /* ept page table is present? */
6115 error_code |= (exit_qualification >> 3) & PFERR_PRESENT_MASK;
6117 vcpu->arch.exit_qualification = exit_qualification;
6119 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
6122 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
6127 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6128 if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
6129 skip_emulated_instruction(vcpu);
6130 trace_kvm_fast_mmio(gpa);
6134 ret = handle_mmio_page_fault(vcpu, gpa, true);
6135 if (likely(ret == RET_MMIO_PF_EMULATE))
6136 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
6139 if (unlikely(ret == RET_MMIO_PF_INVALID))
6140 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
6142 if (unlikely(ret == RET_MMIO_PF_RETRY))
6145 /* It is the real ept misconfig */
6148 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6149 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6154 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6156 u32 cpu_based_vm_exec_control;
6158 /* clear pending NMI */
6159 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6160 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
6161 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
6162 ++vcpu->stat.nmi_window_exits;
6163 kvm_make_request(KVM_REQ_EVENT, vcpu);
6168 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6170 struct vcpu_vmx *vmx = to_vmx(vcpu);
6171 enum emulation_result err = EMULATE_DONE;
6174 bool intr_window_requested;
6175 unsigned count = 130;
6177 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6178 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6180 while (vmx->emulation_required && count-- != 0) {
6181 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6182 return handle_interrupt_window(&vmx->vcpu);
6184 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
6187 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
6189 if (err == EMULATE_USER_EXIT) {
6190 ++vcpu->stat.mmio_exits;
6195 if (err != EMULATE_DONE) {
6196 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6197 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6198 vcpu->run->internal.ndata = 0;
6202 if (vcpu->arch.halt_request) {
6203 vcpu->arch.halt_request = 0;
6204 ret = kvm_vcpu_halt(vcpu);
6208 if (signal_pending(current))
6218 static int __grow_ple_window(int val)
6220 if (ple_window_grow < 1)
6223 val = min(val, ple_window_actual_max);
6225 if (ple_window_grow < ple_window)
6226 val *= ple_window_grow;
6228 val += ple_window_grow;
6233 static int __shrink_ple_window(int val, int modifier, int minimum)
6238 if (modifier < ple_window)
6243 return max(val, minimum);
6246 static void grow_ple_window(struct kvm_vcpu *vcpu)
6248 struct vcpu_vmx *vmx = to_vmx(vcpu);
6249 int old = vmx->ple_window;
6251 vmx->ple_window = __grow_ple_window(old);
6253 if (vmx->ple_window != old)
6254 vmx->ple_window_dirty = true;
6256 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6259 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6261 struct vcpu_vmx *vmx = to_vmx(vcpu);
6262 int old = vmx->ple_window;
6264 vmx->ple_window = __shrink_ple_window(old,
6265 ple_window_shrink, ple_window);
6267 if (vmx->ple_window != old)
6268 vmx->ple_window_dirty = true;
6270 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6274 * ple_window_actual_max is computed to be one grow_ple_window() below
6275 * ple_window_max. (See __grow_ple_window for the reason.)
6276 * This prevents overflows, because ple_window_max is int.
6277 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6279 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6281 static void update_ple_window_actual_max(void)
6283 ple_window_actual_max =
6284 __shrink_ple_window(max(ple_window_max, ple_window),
6285 ple_window_grow, INT_MIN);
6289 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6291 static void wakeup_handler(void)
6293 struct kvm_vcpu *vcpu;
6294 int cpu = smp_processor_id();
6296 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6297 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6298 blocked_vcpu_list) {
6299 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6301 if (pi_test_on(pi_desc) == 1)
6302 kvm_vcpu_kick(vcpu);
6304 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6307 static __init int hardware_setup(void)
6309 int r = -ENOMEM, i, msr;
6311 rdmsrl_safe(MSR_EFER, &host_efer);
6313 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6314 kvm_define_shared_msr(i, vmx_msr_index[i]);
6316 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
6317 if (!vmx_io_bitmap_a)
6320 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6321 if (!vmx_io_bitmap_b)
6324 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
6325 if (!vmx_msr_bitmap_legacy)
6328 vmx_msr_bitmap_legacy_x2apic =
6329 (unsigned long *)__get_free_page(GFP_KERNEL);
6330 if (!vmx_msr_bitmap_legacy_x2apic)
6333 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
6334 if (!vmx_msr_bitmap_longmode)
6337 vmx_msr_bitmap_longmode_x2apic =
6338 (unsigned long *)__get_free_page(GFP_KERNEL);
6339 if (!vmx_msr_bitmap_longmode_x2apic)
6343 vmx_msr_bitmap_nested =
6344 (unsigned long *)__get_free_page(GFP_KERNEL);
6345 if (!vmx_msr_bitmap_nested)
6349 vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6350 if (!vmx_vmread_bitmap)
6353 vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6354 if (!vmx_vmwrite_bitmap)
6357 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6358 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6361 * Allow direct access to the PC debug port (it is often used for I/O
6362 * delays, but the vmexits simply slow things down).
6364 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6365 clear_bit(0x80, vmx_io_bitmap_a);
6367 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6369 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6370 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6372 memset(vmx_msr_bitmap_nested, 0xff, PAGE_SIZE);
6374 if (setup_vmcs_config(&vmcs_config) < 0) {
6379 if (boot_cpu_has(X86_FEATURE_NX))
6380 kvm_enable_efer_bits(EFER_NX);
6382 if (!cpu_has_vmx_vpid())
6384 if (!cpu_has_vmx_shadow_vmcs())
6385 enable_shadow_vmcs = 0;
6386 if (enable_shadow_vmcs)
6387 init_vmcs_shadow_fields();
6389 if (!cpu_has_vmx_ept() ||
6390 !cpu_has_vmx_ept_4levels()) {
6392 enable_unrestricted_guest = 0;
6393 enable_ept_ad_bits = 0;
6396 if (!cpu_has_vmx_ept_ad_bits())
6397 enable_ept_ad_bits = 0;
6399 if (!cpu_has_vmx_unrestricted_guest())
6400 enable_unrestricted_guest = 0;
6402 if (!cpu_has_vmx_flexpriority())
6403 flexpriority_enabled = 0;
6406 * set_apic_access_page_addr() is used to reload apic access
6407 * page upon invalidation. No need to do anything if not
6408 * using the APIC_ACCESS_ADDR VMCS field.
6410 if (!flexpriority_enabled)
6411 kvm_x86_ops->set_apic_access_page_addr = NULL;
6413 if (!cpu_has_vmx_tpr_shadow())
6414 kvm_x86_ops->update_cr8_intercept = NULL;
6416 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6417 kvm_disable_largepages();
6419 if (!cpu_has_vmx_ple())
6422 if (!cpu_has_vmx_apicv())
6425 if (cpu_has_vmx_tsc_scaling()) {
6426 kvm_has_tsc_control = true;
6427 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6428 kvm_tsc_scaling_ratio_frac_bits = 48;
6431 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6432 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6433 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6434 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6435 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6436 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6437 vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
6439 memcpy(vmx_msr_bitmap_legacy_x2apic,
6440 vmx_msr_bitmap_legacy, PAGE_SIZE);
6441 memcpy(vmx_msr_bitmap_longmode_x2apic,
6442 vmx_msr_bitmap_longmode, PAGE_SIZE);
6444 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6446 for (msr = 0x800; msr <= 0x8ff; msr++)
6447 vmx_disable_intercept_msr_read_x2apic(msr);
6449 /* According SDM, in x2apic mode, the whole id reg is used. But in
6450 * KVM, it only use the highest eight bits. Need to intercept it */
6451 vmx_enable_intercept_msr_read_x2apic(0x802);
6453 vmx_enable_intercept_msr_read_x2apic(0x839);
6455 vmx_disable_intercept_msr_write_x2apic(0x808);
6457 vmx_disable_intercept_msr_write_x2apic(0x80b);
6459 vmx_disable_intercept_msr_write_x2apic(0x83f);
6462 kvm_mmu_set_mask_ptes(0ull,
6463 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
6464 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
6465 0ull, VMX_EPT_EXECUTABLE_MASK);
6466 ept_set_mmio_spte_mask();
6471 update_ple_window_actual_max();
6474 * Only enable PML when hardware supports PML feature, and both EPT
6475 * and EPT A/D bit features are enabled -- PML depends on them to work.
6477 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6481 kvm_x86_ops->slot_enable_log_dirty = NULL;
6482 kvm_x86_ops->slot_disable_log_dirty = NULL;
6483 kvm_x86_ops->flush_log_dirty = NULL;
6484 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6487 if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
6490 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
6491 cpu_preemption_timer_multi =
6492 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
6494 kvm_x86_ops->set_hv_timer = NULL;
6495 kvm_x86_ops->cancel_hv_timer = NULL;
6498 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6500 kvm_mce_cap_supported |= MCG_LMCE_P;
6502 return alloc_kvm_area();
6505 free_page((unsigned long)vmx_vmwrite_bitmap);
6507 free_page((unsigned long)vmx_vmread_bitmap);
6510 free_page((unsigned long)vmx_msr_bitmap_nested);
6512 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6514 free_page((unsigned long)vmx_msr_bitmap_longmode);
6516 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6518 free_page((unsigned long)vmx_msr_bitmap_legacy);
6520 free_page((unsigned long)vmx_io_bitmap_b);
6522 free_page((unsigned long)vmx_io_bitmap_a);
6527 static __exit void hardware_unsetup(void)
6529 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6530 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6531 free_page((unsigned long)vmx_msr_bitmap_legacy);
6532 free_page((unsigned long)vmx_msr_bitmap_longmode);
6533 free_page((unsigned long)vmx_io_bitmap_b);
6534 free_page((unsigned long)vmx_io_bitmap_a);
6535 free_page((unsigned long)vmx_vmwrite_bitmap);
6536 free_page((unsigned long)vmx_vmread_bitmap);
6538 free_page((unsigned long)vmx_msr_bitmap_nested);
6544 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6545 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6547 static int handle_pause(struct kvm_vcpu *vcpu)
6550 grow_ple_window(vcpu);
6552 skip_emulated_instruction(vcpu);
6553 kvm_vcpu_on_spin(vcpu);
6558 static int handle_nop(struct kvm_vcpu *vcpu)
6560 skip_emulated_instruction(vcpu);
6564 static int handle_mwait(struct kvm_vcpu *vcpu)
6566 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6567 return handle_nop(vcpu);
6570 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6575 static int handle_monitor(struct kvm_vcpu *vcpu)
6577 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6578 return handle_nop(vcpu);
6582 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6583 * We could reuse a single VMCS for all the L2 guests, but we also want the
6584 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6585 * allows keeping them loaded on the processor, and in the future will allow
6586 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6587 * every entry if they never change.
6588 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6589 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6591 * The following functions allocate and free a vmcs02 in this pool.
6594 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6595 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6597 struct vmcs02_list *item;
6598 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6599 if (item->vmptr == vmx->nested.current_vmptr) {
6600 list_move(&item->list, &vmx->nested.vmcs02_pool);
6601 return &item->vmcs02;
6604 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6605 /* Recycle the least recently used VMCS. */
6606 item = list_last_entry(&vmx->nested.vmcs02_pool,
6607 struct vmcs02_list, list);
6608 item->vmptr = vmx->nested.current_vmptr;
6609 list_move(&item->list, &vmx->nested.vmcs02_pool);
6610 return &item->vmcs02;
6613 /* Create a new VMCS */
6614 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6617 item->vmcs02.vmcs = alloc_vmcs();
6618 if (!item->vmcs02.vmcs) {
6622 loaded_vmcs_init(&item->vmcs02);
6623 item->vmptr = vmx->nested.current_vmptr;
6624 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6625 vmx->nested.vmcs02_num++;
6626 return &item->vmcs02;
6629 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6630 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6632 struct vmcs02_list *item;
6633 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6634 if (item->vmptr == vmptr) {
6635 free_loaded_vmcs(&item->vmcs02);
6636 list_del(&item->list);
6638 vmx->nested.vmcs02_num--;
6644 * Free all VMCSs saved for this vcpu, except the one pointed by
6645 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6646 * must be &vmx->vmcs01.
6648 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6650 struct vmcs02_list *item, *n;
6652 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6653 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6655 * Something will leak if the above WARN triggers. Better than
6658 if (vmx->loaded_vmcs == &item->vmcs02)
6661 free_loaded_vmcs(&item->vmcs02);
6662 list_del(&item->list);
6664 vmx->nested.vmcs02_num--;
6669 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6670 * set the success or error code of an emulated VMX instruction, as specified
6671 * by Vol 2B, VMX Instruction Reference, "Conventions".
6673 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6675 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6676 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6677 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6680 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6682 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6683 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6684 X86_EFLAGS_SF | X86_EFLAGS_OF))
6688 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6689 u32 vm_instruction_error)
6691 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6693 * failValid writes the error number to the current VMCS, which
6694 * can't be done there isn't a current VMCS.
6696 nested_vmx_failInvalid(vcpu);
6699 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6700 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6701 X86_EFLAGS_SF | X86_EFLAGS_OF))
6703 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6705 * We don't need to force a shadow sync because
6706 * VM_INSTRUCTION_ERROR is not shadowed
6710 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6712 /* TODO: not to reset guest simply here. */
6713 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6714 pr_warn("kvm: nested vmx abort, indicator %d\n", indicator);
6717 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6719 struct vcpu_vmx *vmx =
6720 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6722 vmx->nested.preemption_timer_expired = true;
6723 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6724 kvm_vcpu_kick(&vmx->vcpu);
6726 return HRTIMER_NORESTART;
6730 * Decode the memory-address operand of a vmx instruction, as recorded on an
6731 * exit caused by such an instruction (run by a guest hypervisor).
6732 * On success, returns 0. When the operand is invalid, returns 1 and throws
6735 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6736 unsigned long exit_qualification,
6737 u32 vmx_instruction_info, bool wr, gva_t *ret)
6741 struct kvm_segment s;
6744 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6745 * Execution", on an exit, vmx_instruction_info holds most of the
6746 * addressing components of the operand. Only the displacement part
6747 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6748 * For how an actual address is calculated from all these components,
6749 * refer to Vol. 1, "Operand Addressing".
6751 int scaling = vmx_instruction_info & 3;
6752 int addr_size = (vmx_instruction_info >> 7) & 7;
6753 bool is_reg = vmx_instruction_info & (1u << 10);
6754 int seg_reg = (vmx_instruction_info >> 15) & 7;
6755 int index_reg = (vmx_instruction_info >> 18) & 0xf;
6756 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6757 int base_reg = (vmx_instruction_info >> 23) & 0xf;
6758 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
6761 kvm_queue_exception(vcpu, UD_VECTOR);
6765 /* Addr = segment_base + offset */
6766 /* offset = base + [index * scale] + displacement */
6767 off = exit_qualification; /* holds the displacement */
6769 off += kvm_register_read(vcpu, base_reg);
6771 off += kvm_register_read(vcpu, index_reg)<<scaling;
6772 vmx_get_segment(vcpu, &s, seg_reg);
6773 *ret = s.base + off;
6775 if (addr_size == 1) /* 32 bit */
6778 /* Checks for #GP/#SS exceptions. */
6780 if (is_protmode(vcpu)) {
6781 /* Protected mode: apply checks for segment validity in the
6783 * - segment type check (#GP(0) may be thrown)
6784 * - usability check (#GP(0)/#SS(0))
6785 * - limit check (#GP(0)/#SS(0))
6788 /* #GP(0) if the destination operand is located in a
6789 * read-only data segment or any code segment.
6791 exn = ((s.type & 0xa) == 0 || (s.type & 8));
6793 /* #GP(0) if the source operand is located in an
6794 * execute-only code segment
6796 exn = ((s.type & 0xa) == 8);
6799 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6802 if (is_long_mode(vcpu)) {
6803 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6804 * non-canonical form. This is an only check for long mode.
6806 exn = is_noncanonical_address(*ret);
6807 } else if (is_protmode(vcpu)) {
6808 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
6810 exn = (s.unusable != 0);
6811 /* Protected mode: #GP(0)/#SS(0) if the memory
6812 * operand is outside the segment limit.
6814 exn = exn || (off + sizeof(u64) > s.limit);
6817 kvm_queue_exception_e(vcpu,
6818 seg_reg == VCPU_SREG_SS ?
6819 SS_VECTOR : GP_VECTOR,
6828 * This function performs the various checks including
6829 * - if it's 4KB aligned
6830 * - No bits beyond the physical address width are set
6831 * - Returns 0 on success or else 1
6832 * (Intel SDM Section 30.3)
6834 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6839 struct x86_exception e;
6841 struct vcpu_vmx *vmx = to_vmx(vcpu);
6842 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6844 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6845 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
6848 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6849 sizeof(vmptr), &e)) {
6850 kvm_inject_page_fault(vcpu, &e);
6854 switch (exit_reason) {
6855 case EXIT_REASON_VMON:
6858 * The first 4 bytes of VMXON region contain the supported
6859 * VMCS revision identifier
6861 * Note - IA32_VMX_BASIC[48] will never be 1
6862 * for the nested case;
6863 * which replaces physical address width with 32
6866 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6867 nested_vmx_failInvalid(vcpu);
6868 skip_emulated_instruction(vcpu);
6872 page = nested_get_page(vcpu, vmptr);
6874 *(u32 *)kmap(page) != VMCS12_REVISION) {
6875 nested_vmx_failInvalid(vcpu);
6877 skip_emulated_instruction(vcpu);
6881 vmx->nested.vmxon_ptr = vmptr;
6883 case EXIT_REASON_VMCLEAR:
6884 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6885 nested_vmx_failValid(vcpu,
6886 VMXERR_VMCLEAR_INVALID_ADDRESS);
6887 skip_emulated_instruction(vcpu);
6891 if (vmptr == vmx->nested.vmxon_ptr) {
6892 nested_vmx_failValid(vcpu,
6893 VMXERR_VMCLEAR_VMXON_POINTER);
6894 skip_emulated_instruction(vcpu);
6898 case EXIT_REASON_VMPTRLD:
6899 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6900 nested_vmx_failValid(vcpu,
6901 VMXERR_VMPTRLD_INVALID_ADDRESS);
6902 skip_emulated_instruction(vcpu);
6906 if (vmptr == vmx->nested.vmxon_ptr) {
6907 nested_vmx_failValid(vcpu,
6908 VMXERR_VMCLEAR_VMXON_POINTER);
6909 skip_emulated_instruction(vcpu);
6914 return 1; /* shouldn't happen */
6923 * Emulate the VMXON instruction.
6924 * Currently, we just remember that VMX is active, and do not save or even
6925 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
6926 * do not currently need to store anything in that guest-allocated memory
6927 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
6928 * argument is different from the VMXON pointer (which the spec says they do).
6930 static int handle_vmon(struct kvm_vcpu *vcpu)
6932 struct kvm_segment cs;
6933 struct vcpu_vmx *vmx = to_vmx(vcpu);
6934 struct vmcs *shadow_vmcs;
6935 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
6936 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6938 /* The Intel VMX Instruction Reference lists a bunch of bits that
6939 * are prerequisite to running VMXON, most notably cr4.VMXE must be
6940 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
6941 * Otherwise, we should fail with #UD. We test these now:
6943 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
6944 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
6945 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
6946 kvm_queue_exception(vcpu, UD_VECTOR);
6950 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6951 if (is_long_mode(vcpu) && !cs.l) {
6952 kvm_queue_exception(vcpu, UD_VECTOR);
6956 if (vmx_get_cpl(vcpu)) {
6957 kvm_inject_gp(vcpu, 0);
6961 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
6964 if (vmx->nested.vmxon) {
6965 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
6966 skip_emulated_instruction(vcpu);
6970 if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
6971 != VMXON_NEEDED_FEATURES) {
6972 kvm_inject_gp(vcpu, 0);
6976 if (enable_shadow_vmcs) {
6977 shadow_vmcs = alloc_vmcs();
6980 /* mark vmcs as shadow */
6981 shadow_vmcs->revision_id |= (1u << 31);
6982 /* init shadow vmcs */
6983 vmcs_clear(shadow_vmcs);
6984 vmx->nested.current_shadow_vmcs = shadow_vmcs;
6987 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
6988 vmx->nested.vmcs02_num = 0;
6990 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
6992 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
6994 vmx->nested.vmxon = true;
6996 skip_emulated_instruction(vcpu);
6997 nested_vmx_succeed(vcpu);
7002 * Intel's VMX Instruction Reference specifies a common set of prerequisites
7003 * for running VMX instructions (except VMXON, whose prerequisites are
7004 * slightly different). It also specifies what exception to inject otherwise.
7006 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
7008 struct kvm_segment cs;
7009 struct vcpu_vmx *vmx = to_vmx(vcpu);
7011 if (!vmx->nested.vmxon) {
7012 kvm_queue_exception(vcpu, UD_VECTOR);
7016 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
7017 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
7018 (is_long_mode(vcpu) && !cs.l)) {
7019 kvm_queue_exception(vcpu, UD_VECTOR);
7023 if (vmx_get_cpl(vcpu)) {
7024 kvm_inject_gp(vcpu, 0);
7031 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
7033 if (vmx->nested.current_vmptr == -1ull)
7036 /* current_vmptr and current_vmcs12 are always set/reset together */
7037 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
7040 if (enable_shadow_vmcs) {
7041 /* copy to memory all shadowed fields in case
7042 they were modified */
7043 copy_shadow_to_vmcs12(vmx);
7044 vmx->nested.sync_shadow_vmcs = false;
7045 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
7046 SECONDARY_EXEC_SHADOW_VMCS);
7047 vmcs_write64(VMCS_LINK_POINTER, -1ull);
7049 vmx->nested.posted_intr_nv = -1;
7050 kunmap(vmx->nested.current_vmcs12_page);
7051 nested_release_page(vmx->nested.current_vmcs12_page);
7052 vmx->nested.current_vmptr = -1ull;
7053 vmx->nested.current_vmcs12 = NULL;
7057 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7058 * just stops using VMX.
7060 static void free_nested(struct vcpu_vmx *vmx)
7062 if (!vmx->nested.vmxon)
7065 vmx->nested.vmxon = false;
7066 free_vpid(vmx->nested.vpid02);
7067 nested_release_vmcs12(vmx);
7068 if (enable_shadow_vmcs)
7069 free_vmcs(vmx->nested.current_shadow_vmcs);
7070 /* Unpin physical memory we referred to in current vmcs02 */
7071 if (vmx->nested.apic_access_page) {
7072 nested_release_page(vmx->nested.apic_access_page);
7073 vmx->nested.apic_access_page = NULL;
7075 if (vmx->nested.virtual_apic_page) {
7076 nested_release_page(vmx->nested.virtual_apic_page);
7077 vmx->nested.virtual_apic_page = NULL;
7079 if (vmx->nested.pi_desc_page) {
7080 kunmap(vmx->nested.pi_desc_page);
7081 nested_release_page(vmx->nested.pi_desc_page);
7082 vmx->nested.pi_desc_page = NULL;
7083 vmx->nested.pi_desc = NULL;
7086 nested_free_all_saved_vmcss(vmx);
7089 /* Emulate the VMXOFF instruction */
7090 static int handle_vmoff(struct kvm_vcpu *vcpu)
7092 if (!nested_vmx_check_permission(vcpu))
7094 free_nested(to_vmx(vcpu));
7095 skip_emulated_instruction(vcpu);
7096 nested_vmx_succeed(vcpu);
7100 /* Emulate the VMCLEAR instruction */
7101 static int handle_vmclear(struct kvm_vcpu *vcpu)
7103 struct vcpu_vmx *vmx = to_vmx(vcpu);
7105 struct vmcs12 *vmcs12;
7108 if (!nested_vmx_check_permission(vcpu))
7111 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
7114 if (vmptr == vmx->nested.current_vmptr)
7115 nested_release_vmcs12(vmx);
7117 page = nested_get_page(vcpu, vmptr);
7120 * For accurate processor emulation, VMCLEAR beyond available
7121 * physical memory should do nothing at all. However, it is
7122 * possible that a nested vmx bug, not a guest hypervisor bug,
7123 * resulted in this case, so let's shut down before doing any
7126 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7129 vmcs12 = kmap(page);
7130 vmcs12->launch_state = 0;
7132 nested_release_page(page);
7134 nested_free_vmcs02(vmx, vmptr);
7136 skip_emulated_instruction(vcpu);
7137 nested_vmx_succeed(vcpu);
7141 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
7143 /* Emulate the VMLAUNCH instruction */
7144 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
7146 return nested_vmx_run(vcpu, true);
7149 /* Emulate the VMRESUME instruction */
7150 static int handle_vmresume(struct kvm_vcpu *vcpu)
7153 return nested_vmx_run(vcpu, false);
7156 enum vmcs_field_type {
7157 VMCS_FIELD_TYPE_U16 = 0,
7158 VMCS_FIELD_TYPE_U64 = 1,
7159 VMCS_FIELD_TYPE_U32 = 2,
7160 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
7163 static inline int vmcs_field_type(unsigned long field)
7165 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
7166 return VMCS_FIELD_TYPE_U32;
7167 return (field >> 13) & 0x3 ;
7170 static inline int vmcs_field_readonly(unsigned long field)
7172 return (((field >> 10) & 0x3) == 1);
7176 * Read a vmcs12 field. Since these can have varying lengths and we return
7177 * one type, we chose the biggest type (u64) and zero-extend the return value
7178 * to that size. Note that the caller, handle_vmread, might need to use only
7179 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7180 * 64-bit fields are to be returned).
7182 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7183 unsigned long field, u64 *ret)
7185 short offset = vmcs_field_to_offset(field);
7191 p = ((char *)(get_vmcs12(vcpu))) + offset;
7193 switch (vmcs_field_type(field)) {
7194 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7195 *ret = *((natural_width *)p);
7197 case VMCS_FIELD_TYPE_U16:
7200 case VMCS_FIELD_TYPE_U32:
7203 case VMCS_FIELD_TYPE_U64:
7213 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7214 unsigned long field, u64 field_value){
7215 short offset = vmcs_field_to_offset(field);
7216 char *p = ((char *) get_vmcs12(vcpu)) + offset;
7220 switch (vmcs_field_type(field)) {
7221 case VMCS_FIELD_TYPE_U16:
7222 *(u16 *)p = field_value;
7224 case VMCS_FIELD_TYPE_U32:
7225 *(u32 *)p = field_value;
7227 case VMCS_FIELD_TYPE_U64:
7228 *(u64 *)p = field_value;
7230 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7231 *(natural_width *)p = field_value;
7240 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7243 unsigned long field;
7245 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7246 const unsigned long *fields = shadow_read_write_fields;
7247 const int num_fields = max_shadow_read_write_fields;
7251 vmcs_load(shadow_vmcs);
7253 for (i = 0; i < num_fields; i++) {
7255 switch (vmcs_field_type(field)) {
7256 case VMCS_FIELD_TYPE_U16:
7257 field_value = vmcs_read16(field);
7259 case VMCS_FIELD_TYPE_U32:
7260 field_value = vmcs_read32(field);
7262 case VMCS_FIELD_TYPE_U64:
7263 field_value = vmcs_read64(field);
7265 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7266 field_value = vmcs_readl(field);
7272 vmcs12_write_any(&vmx->vcpu, field, field_value);
7275 vmcs_clear(shadow_vmcs);
7276 vmcs_load(vmx->loaded_vmcs->vmcs);
7281 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7283 const unsigned long *fields[] = {
7284 shadow_read_write_fields,
7285 shadow_read_only_fields
7287 const int max_fields[] = {
7288 max_shadow_read_write_fields,
7289 max_shadow_read_only_fields
7292 unsigned long field;
7293 u64 field_value = 0;
7294 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7296 vmcs_load(shadow_vmcs);
7298 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7299 for (i = 0; i < max_fields[q]; i++) {
7300 field = fields[q][i];
7301 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7303 switch (vmcs_field_type(field)) {
7304 case VMCS_FIELD_TYPE_U16:
7305 vmcs_write16(field, (u16)field_value);
7307 case VMCS_FIELD_TYPE_U32:
7308 vmcs_write32(field, (u32)field_value);
7310 case VMCS_FIELD_TYPE_U64:
7311 vmcs_write64(field, (u64)field_value);
7313 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7314 vmcs_writel(field, (long)field_value);
7323 vmcs_clear(shadow_vmcs);
7324 vmcs_load(vmx->loaded_vmcs->vmcs);
7328 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7329 * used before) all generate the same failure when it is missing.
7331 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7333 struct vcpu_vmx *vmx = to_vmx(vcpu);
7334 if (vmx->nested.current_vmptr == -1ull) {
7335 nested_vmx_failInvalid(vcpu);
7336 skip_emulated_instruction(vcpu);
7342 static int handle_vmread(struct kvm_vcpu *vcpu)
7344 unsigned long field;
7346 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7347 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7350 if (!nested_vmx_check_permission(vcpu) ||
7351 !nested_vmx_check_vmcs12(vcpu))
7354 /* Decode instruction info and find the field to read */
7355 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7356 /* Read the field, zero-extended to a u64 field_value */
7357 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7358 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7359 skip_emulated_instruction(vcpu);
7363 * Now copy part of this value to register or memory, as requested.
7364 * Note that the number of bits actually copied is 32 or 64 depending
7365 * on the guest's mode (32 or 64 bit), not on the given field's length.
7367 if (vmx_instruction_info & (1u << 10)) {
7368 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7371 if (get_vmx_mem_address(vcpu, exit_qualification,
7372 vmx_instruction_info, true, &gva))
7374 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
7375 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7376 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7379 nested_vmx_succeed(vcpu);
7380 skip_emulated_instruction(vcpu);
7385 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7387 unsigned long field;
7389 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7390 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7391 /* The value to write might be 32 or 64 bits, depending on L1's long
7392 * mode, and eventually we need to write that into a field of several
7393 * possible lengths. The code below first zero-extends the value to 64
7394 * bit (field_value), and then copies only the appropriate number of
7395 * bits into the vmcs12 field.
7397 u64 field_value = 0;
7398 struct x86_exception e;
7400 if (!nested_vmx_check_permission(vcpu) ||
7401 !nested_vmx_check_vmcs12(vcpu))
7404 if (vmx_instruction_info & (1u << 10))
7405 field_value = kvm_register_readl(vcpu,
7406 (((vmx_instruction_info) >> 3) & 0xf));
7408 if (get_vmx_mem_address(vcpu, exit_qualification,
7409 vmx_instruction_info, false, &gva))
7411 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7412 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7413 kvm_inject_page_fault(vcpu, &e);
7419 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7420 if (vmcs_field_readonly(field)) {
7421 nested_vmx_failValid(vcpu,
7422 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7423 skip_emulated_instruction(vcpu);
7427 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7428 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7429 skip_emulated_instruction(vcpu);
7433 nested_vmx_succeed(vcpu);
7434 skip_emulated_instruction(vcpu);
7438 /* Emulate the VMPTRLD instruction */
7439 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7441 struct vcpu_vmx *vmx = to_vmx(vcpu);
7444 if (!nested_vmx_check_permission(vcpu))
7447 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
7450 if (vmx->nested.current_vmptr != vmptr) {
7451 struct vmcs12 *new_vmcs12;
7453 page = nested_get_page(vcpu, vmptr);
7455 nested_vmx_failInvalid(vcpu);
7456 skip_emulated_instruction(vcpu);
7459 new_vmcs12 = kmap(page);
7460 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7462 nested_release_page_clean(page);
7463 nested_vmx_failValid(vcpu,
7464 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7465 skip_emulated_instruction(vcpu);
7469 nested_release_vmcs12(vmx);
7470 vmx->nested.current_vmptr = vmptr;
7471 vmx->nested.current_vmcs12 = new_vmcs12;
7472 vmx->nested.current_vmcs12_page = page;
7473 if (enable_shadow_vmcs) {
7474 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7475 SECONDARY_EXEC_SHADOW_VMCS);
7476 vmcs_write64(VMCS_LINK_POINTER,
7477 __pa(vmx->nested.current_shadow_vmcs));
7478 vmx->nested.sync_shadow_vmcs = true;
7482 nested_vmx_succeed(vcpu);
7483 skip_emulated_instruction(vcpu);
7487 /* Emulate the VMPTRST instruction */
7488 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7490 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7491 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7493 struct x86_exception e;
7495 if (!nested_vmx_check_permission(vcpu))
7498 if (get_vmx_mem_address(vcpu, exit_qualification,
7499 vmx_instruction_info, true, &vmcs_gva))
7501 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
7502 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7503 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7505 kvm_inject_page_fault(vcpu, &e);
7508 nested_vmx_succeed(vcpu);
7509 skip_emulated_instruction(vcpu);
7513 /* Emulate the INVEPT instruction */
7514 static int handle_invept(struct kvm_vcpu *vcpu)
7516 struct vcpu_vmx *vmx = to_vmx(vcpu);
7517 u32 vmx_instruction_info, types;
7520 struct x86_exception e;
7525 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7526 SECONDARY_EXEC_ENABLE_EPT) ||
7527 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7528 kvm_queue_exception(vcpu, UD_VECTOR);
7532 if (!nested_vmx_check_permission(vcpu))
7535 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
7536 kvm_queue_exception(vcpu, UD_VECTOR);
7540 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7541 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7543 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7545 if (!(types & (1UL << type))) {
7546 nested_vmx_failValid(vcpu,
7547 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7548 skip_emulated_instruction(vcpu);
7552 /* According to the Intel VMX instruction reference, the memory
7553 * operand is read even if it isn't needed (e.g., for type==global)
7555 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7556 vmx_instruction_info, false, &gva))
7558 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7559 sizeof(operand), &e)) {
7560 kvm_inject_page_fault(vcpu, &e);
7565 case VMX_EPT_EXTENT_GLOBAL:
7566 kvm_mmu_sync_roots(vcpu);
7567 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7568 nested_vmx_succeed(vcpu);
7571 /* Trap single context invalidation invept calls */
7576 skip_emulated_instruction(vcpu);
7580 static int handle_invvpid(struct kvm_vcpu *vcpu)
7582 struct vcpu_vmx *vmx = to_vmx(vcpu);
7583 u32 vmx_instruction_info;
7584 unsigned long type, types;
7586 struct x86_exception e;
7589 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7590 SECONDARY_EXEC_ENABLE_VPID) ||
7591 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7592 kvm_queue_exception(vcpu, UD_VECTOR);
7596 if (!nested_vmx_check_permission(vcpu))
7599 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7600 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7602 types = (vmx->nested.nested_vmx_vpid_caps >> 8) & 0x7;
7604 if (!(types & (1UL << type))) {
7605 nested_vmx_failValid(vcpu,
7606 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7607 skip_emulated_instruction(vcpu);
7611 /* according to the intel vmx instruction reference, the memory
7612 * operand is read even if it isn't needed (e.g., for type==global)
7614 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7615 vmx_instruction_info, false, &gva))
7617 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vpid,
7619 kvm_inject_page_fault(vcpu, &e);
7624 case VMX_VPID_EXTENT_SINGLE_CONTEXT:
7626 * Old versions of KVM use the single-context version so we
7627 * have to support it; just treat it the same as all-context.
7629 case VMX_VPID_EXTENT_ALL_CONTEXT:
7630 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
7631 nested_vmx_succeed(vcpu);
7634 /* Trap individual address invalidation invvpid calls */
7639 skip_emulated_instruction(vcpu);
7643 static int handle_pml_full(struct kvm_vcpu *vcpu)
7645 unsigned long exit_qualification;
7647 trace_kvm_pml_full(vcpu->vcpu_id);
7649 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7652 * PML buffer FULL happened while executing iret from NMI,
7653 * "blocked by NMI" bit has to be set before next VM entry.
7655 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7656 cpu_has_virtual_nmis() &&
7657 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7658 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7659 GUEST_INTR_STATE_NMI);
7662 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7663 * here.., and there's no userspace involvement needed for PML.
7668 static int handle_pcommit(struct kvm_vcpu *vcpu)
7670 /* we never catch pcommit instruct for L1 guest. */
7675 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
7677 kvm_lapic_expired_hv_timer(vcpu);
7682 * The exit handlers return 1 if the exit was handled fully and guest execution
7683 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
7684 * to be done to userspace and return 0.
7686 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7687 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
7688 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
7689 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
7690 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
7691 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
7692 [EXIT_REASON_CR_ACCESS] = handle_cr,
7693 [EXIT_REASON_DR_ACCESS] = handle_dr,
7694 [EXIT_REASON_CPUID] = handle_cpuid,
7695 [EXIT_REASON_MSR_READ] = handle_rdmsr,
7696 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
7697 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
7698 [EXIT_REASON_HLT] = handle_halt,
7699 [EXIT_REASON_INVD] = handle_invd,
7700 [EXIT_REASON_INVLPG] = handle_invlpg,
7701 [EXIT_REASON_RDPMC] = handle_rdpmc,
7702 [EXIT_REASON_VMCALL] = handle_vmcall,
7703 [EXIT_REASON_VMCLEAR] = handle_vmclear,
7704 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
7705 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
7706 [EXIT_REASON_VMPTRST] = handle_vmptrst,
7707 [EXIT_REASON_VMREAD] = handle_vmread,
7708 [EXIT_REASON_VMRESUME] = handle_vmresume,
7709 [EXIT_REASON_VMWRITE] = handle_vmwrite,
7710 [EXIT_REASON_VMOFF] = handle_vmoff,
7711 [EXIT_REASON_VMON] = handle_vmon,
7712 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
7713 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
7714 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
7715 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
7716 [EXIT_REASON_WBINVD] = handle_wbinvd,
7717 [EXIT_REASON_XSETBV] = handle_xsetbv,
7718 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
7719 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
7720 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
7721 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
7722 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
7723 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
7724 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
7725 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
7726 [EXIT_REASON_INVEPT] = handle_invept,
7727 [EXIT_REASON_INVVPID] = handle_invvpid,
7728 [EXIT_REASON_XSAVES] = handle_xsaves,
7729 [EXIT_REASON_XRSTORS] = handle_xrstors,
7730 [EXIT_REASON_PML_FULL] = handle_pml_full,
7731 [EXIT_REASON_PCOMMIT] = handle_pcommit,
7732 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
7735 static const int kvm_vmx_max_exit_handlers =
7736 ARRAY_SIZE(kvm_vmx_exit_handlers);
7738 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7739 struct vmcs12 *vmcs12)
7741 unsigned long exit_qualification;
7742 gpa_t bitmap, last_bitmap;
7747 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7748 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7750 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7752 port = exit_qualification >> 16;
7753 size = (exit_qualification & 7) + 1;
7755 last_bitmap = (gpa_t)-1;
7760 bitmap = vmcs12->io_bitmap_a;
7761 else if (port < 0x10000)
7762 bitmap = vmcs12->io_bitmap_b;
7765 bitmap += (port & 0x7fff) / 8;
7767 if (last_bitmap != bitmap)
7768 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
7770 if (b & (1 << (port & 7)))
7775 last_bitmap = bitmap;
7782 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7783 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7784 * disinterest in the current event (read or write a specific MSR) by using an
7785 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7787 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7788 struct vmcs12 *vmcs12, u32 exit_reason)
7790 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7793 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7797 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7798 * for the four combinations of read/write and low/high MSR numbers.
7799 * First we need to figure out which of the four to use:
7801 bitmap = vmcs12->msr_bitmap;
7802 if (exit_reason == EXIT_REASON_MSR_WRITE)
7804 if (msr_index >= 0xc0000000) {
7805 msr_index -= 0xc0000000;
7809 /* Then read the msr_index'th bit from this bitmap: */
7810 if (msr_index < 1024*8) {
7812 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
7814 return 1 & (b >> (msr_index & 7));
7816 return true; /* let L1 handle the wrong parameter */
7820 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7821 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7822 * intercept (via guest_host_mask etc.) the current event.
7824 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7825 struct vmcs12 *vmcs12)
7827 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7828 int cr = exit_qualification & 15;
7829 int reg = (exit_qualification >> 8) & 15;
7830 unsigned long val = kvm_register_readl(vcpu, reg);
7832 switch ((exit_qualification >> 4) & 3) {
7833 case 0: /* mov to cr */
7836 if (vmcs12->cr0_guest_host_mask &
7837 (val ^ vmcs12->cr0_read_shadow))
7841 if ((vmcs12->cr3_target_count >= 1 &&
7842 vmcs12->cr3_target_value0 == val) ||
7843 (vmcs12->cr3_target_count >= 2 &&
7844 vmcs12->cr3_target_value1 == val) ||
7845 (vmcs12->cr3_target_count >= 3 &&
7846 vmcs12->cr3_target_value2 == val) ||
7847 (vmcs12->cr3_target_count >= 4 &&
7848 vmcs12->cr3_target_value3 == val))
7850 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
7854 if (vmcs12->cr4_guest_host_mask &
7855 (vmcs12->cr4_read_shadow ^ val))
7859 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
7865 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
7866 (vmcs12->cr0_read_shadow & X86_CR0_TS))
7869 case 1: /* mov from cr */
7872 if (vmcs12->cpu_based_vm_exec_control &
7873 CPU_BASED_CR3_STORE_EXITING)
7877 if (vmcs12->cpu_based_vm_exec_control &
7878 CPU_BASED_CR8_STORE_EXITING)
7885 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
7886 * cr0. Other attempted changes are ignored, with no exit.
7888 if (vmcs12->cr0_guest_host_mask & 0xe &
7889 (val ^ vmcs12->cr0_read_shadow))
7891 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
7892 !(vmcs12->cr0_read_shadow & 0x1) &&
7901 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
7902 * should handle it ourselves in L0 (and then continue L2). Only call this
7903 * when in is_guest_mode (L2).
7905 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
7907 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7908 struct vcpu_vmx *vmx = to_vmx(vcpu);
7909 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7910 u32 exit_reason = vmx->exit_reason;
7912 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
7913 vmcs_readl(EXIT_QUALIFICATION),
7914 vmx->idt_vectoring_info,
7916 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
7919 if (vmx->nested.nested_run_pending)
7922 if (unlikely(vmx->fail)) {
7923 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
7924 vmcs_read32(VM_INSTRUCTION_ERROR));
7928 switch (exit_reason) {
7929 case EXIT_REASON_EXCEPTION_NMI:
7930 if (!is_exception(intr_info))
7932 else if (is_page_fault(intr_info))
7934 else if (is_no_device(intr_info) &&
7935 !(vmcs12->guest_cr0 & X86_CR0_TS))
7937 else if (is_debug(intr_info) &&
7939 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
7941 else if (is_breakpoint(intr_info) &&
7942 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
7944 return vmcs12->exception_bitmap &
7945 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
7946 case EXIT_REASON_EXTERNAL_INTERRUPT:
7948 case EXIT_REASON_TRIPLE_FAULT:
7950 case EXIT_REASON_PENDING_INTERRUPT:
7951 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
7952 case EXIT_REASON_NMI_WINDOW:
7953 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
7954 case EXIT_REASON_TASK_SWITCH:
7956 case EXIT_REASON_CPUID:
7957 if (kvm_register_read(vcpu, VCPU_REGS_RAX) == 0xa)
7960 case EXIT_REASON_HLT:
7961 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
7962 case EXIT_REASON_INVD:
7964 case EXIT_REASON_INVLPG:
7965 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
7966 case EXIT_REASON_RDPMC:
7967 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
7968 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
7969 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
7970 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
7971 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
7972 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
7973 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
7974 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
7975 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
7977 * VMX instructions trap unconditionally. This allows L1 to
7978 * emulate them for its L2 guest, i.e., allows 3-level nesting!
7981 case EXIT_REASON_CR_ACCESS:
7982 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
7983 case EXIT_REASON_DR_ACCESS:
7984 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
7985 case EXIT_REASON_IO_INSTRUCTION:
7986 return nested_vmx_exit_handled_io(vcpu, vmcs12);
7987 case EXIT_REASON_MSR_READ:
7988 case EXIT_REASON_MSR_WRITE:
7989 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
7990 case EXIT_REASON_INVALID_STATE:
7992 case EXIT_REASON_MWAIT_INSTRUCTION:
7993 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
7994 case EXIT_REASON_MONITOR_TRAP_FLAG:
7995 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
7996 case EXIT_REASON_MONITOR_INSTRUCTION:
7997 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
7998 case EXIT_REASON_PAUSE_INSTRUCTION:
7999 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
8000 nested_cpu_has2(vmcs12,
8001 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
8002 case EXIT_REASON_MCE_DURING_VMENTRY:
8004 case EXIT_REASON_TPR_BELOW_THRESHOLD:
8005 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
8006 case EXIT_REASON_APIC_ACCESS:
8007 return nested_cpu_has2(vmcs12,
8008 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
8009 case EXIT_REASON_APIC_WRITE:
8010 case EXIT_REASON_EOI_INDUCED:
8011 /* apic_write and eoi_induced should exit unconditionally. */
8013 case EXIT_REASON_EPT_VIOLATION:
8015 * L0 always deals with the EPT violation. If nested EPT is
8016 * used, and the nested mmu code discovers that the address is
8017 * missing in the guest EPT table (EPT12), the EPT violation
8018 * will be injected with nested_ept_inject_page_fault()
8021 case EXIT_REASON_EPT_MISCONFIG:
8023 * L2 never uses directly L1's EPT, but rather L0's own EPT
8024 * table (shadow on EPT) or a merged EPT table that L0 built
8025 * (EPT on EPT). So any problems with the structure of the
8026 * table is L0's fault.
8029 case EXIT_REASON_WBINVD:
8030 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
8031 case EXIT_REASON_XSETBV:
8033 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
8035 * This should never happen, since it is not possible to
8036 * set XSS to a non-zero value---neither in L1 nor in L2.
8037 * If if it were, XSS would have to be checked against
8038 * the XSS exit bitmap in vmcs12.
8040 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
8041 case EXIT_REASON_PCOMMIT:
8042 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_PCOMMIT);
8043 case EXIT_REASON_PREEMPTION_TIMER:
8050 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
8052 *info1 = vmcs_readl(EXIT_QUALIFICATION);
8053 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
8056 static int vmx_create_pml_buffer(struct vcpu_vmx *vmx)
8058 struct page *pml_pg;
8060 pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
8064 vmx->pml_pg = pml_pg;
8066 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
8067 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8072 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
8075 __free_page(vmx->pml_pg);
8080 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
8082 struct vcpu_vmx *vmx = to_vmx(vcpu);
8086 pml_idx = vmcs_read16(GUEST_PML_INDEX);
8088 /* Do nothing if PML buffer is empty */
8089 if (pml_idx == (PML_ENTITY_NUM - 1))
8092 /* PML index always points to next available PML buffer entity */
8093 if (pml_idx >= PML_ENTITY_NUM)
8098 pml_buf = page_address(vmx->pml_pg);
8099 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
8102 gpa = pml_buf[pml_idx];
8103 WARN_ON(gpa & (PAGE_SIZE - 1));
8104 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
8107 /* reset PML index */
8108 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8112 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8113 * Called before reporting dirty_bitmap to userspace.
8115 static void kvm_flush_pml_buffers(struct kvm *kvm)
8118 struct kvm_vcpu *vcpu;
8120 * We only need to kick vcpu out of guest mode here, as PML buffer
8121 * is flushed at beginning of all VMEXITs, and it's obvious that only
8122 * vcpus running in guest are possible to have unflushed GPAs in PML
8125 kvm_for_each_vcpu(i, vcpu, kvm)
8126 kvm_vcpu_kick(vcpu);
8129 static void vmx_dump_sel(char *name, uint32_t sel)
8131 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8132 name, vmcs_read32(sel),
8133 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
8134 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
8135 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
8138 static void vmx_dump_dtsel(char *name, uint32_t limit)
8140 pr_err("%s limit=0x%08x, base=0x%016lx\n",
8141 name, vmcs_read32(limit),
8142 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
8145 static void dump_vmcs(void)
8147 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
8148 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
8149 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
8150 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
8151 u32 secondary_exec_control = 0;
8152 unsigned long cr4 = vmcs_readl(GUEST_CR4);
8153 u64 efer = vmcs_read64(GUEST_IA32_EFER);
8156 if (cpu_has_secondary_exec_ctrls())
8157 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8159 pr_err("*** Guest State ***\n");
8160 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8161 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
8162 vmcs_readl(CR0_GUEST_HOST_MASK));
8163 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8164 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
8165 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
8166 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
8167 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
8169 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
8170 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
8171 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
8172 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
8174 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
8175 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
8176 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
8177 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
8178 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8179 vmcs_readl(GUEST_SYSENTER_ESP),
8180 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8181 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
8182 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
8183 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
8184 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
8185 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
8186 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
8187 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8188 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8189 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8190 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
8191 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8192 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8193 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8194 efer, vmcs_read64(GUEST_IA32_PAT));
8195 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
8196 vmcs_read64(GUEST_IA32_DEBUGCTL),
8197 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8198 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8199 pr_err("PerfGlobCtl = 0x%016llx\n",
8200 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8201 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8202 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8203 pr_err("Interruptibility = %08x ActivityState = %08x\n",
8204 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8205 vmcs_read32(GUEST_ACTIVITY_STATE));
8206 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8207 pr_err("InterruptStatus = %04x\n",
8208 vmcs_read16(GUEST_INTR_STATUS));
8210 pr_err("*** Host State ***\n");
8211 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
8212 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8213 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8214 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8215 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8216 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8217 vmcs_read16(HOST_TR_SELECTOR));
8218 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8219 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8220 vmcs_readl(HOST_TR_BASE));
8221 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8222 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8223 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8224 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8225 vmcs_readl(HOST_CR4));
8226 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8227 vmcs_readl(HOST_IA32_SYSENTER_ESP),
8228 vmcs_read32(HOST_IA32_SYSENTER_CS),
8229 vmcs_readl(HOST_IA32_SYSENTER_EIP));
8230 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8231 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8232 vmcs_read64(HOST_IA32_EFER),
8233 vmcs_read64(HOST_IA32_PAT));
8234 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8235 pr_err("PerfGlobCtl = 0x%016llx\n",
8236 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8238 pr_err("*** Control State ***\n");
8239 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8240 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8241 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8242 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8243 vmcs_read32(EXCEPTION_BITMAP),
8244 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8245 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8246 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8247 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8248 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8249 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8250 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8251 vmcs_read32(VM_EXIT_INTR_INFO),
8252 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8253 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8254 pr_err(" reason=%08x qualification=%016lx\n",
8255 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8256 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8257 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8258 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8259 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8260 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8261 pr_err("TSC Multiplier = 0x%016llx\n",
8262 vmcs_read64(TSC_MULTIPLIER));
8263 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8264 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8265 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8266 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8267 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8268 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8269 n = vmcs_read32(CR3_TARGET_COUNT);
8270 for (i = 0; i + 1 < n; i += 4)
8271 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8272 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8273 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8275 pr_err("CR3 target%u=%016lx\n",
8276 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8277 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8278 pr_err("PLE Gap=%08x Window=%08x\n",
8279 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8280 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8281 pr_err("Virtual processor ID = 0x%04x\n",
8282 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8286 * The guest has exited. See if we can fix it or if we need userspace
8289 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8291 struct vcpu_vmx *vmx = to_vmx(vcpu);
8292 u32 exit_reason = vmx->exit_reason;
8293 u32 vectoring_info = vmx->idt_vectoring_info;
8295 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8298 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8299 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8300 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8301 * mode as if vcpus is in root mode, the PML buffer must has been
8305 vmx_flush_pml_buffer(vcpu);
8307 /* If guest state is invalid, start emulating */
8308 if (vmx->emulation_required)
8309 return handle_invalid_guest_state(vcpu);
8311 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
8312 nested_vmx_vmexit(vcpu, exit_reason,
8313 vmcs_read32(VM_EXIT_INTR_INFO),
8314 vmcs_readl(EXIT_QUALIFICATION));
8318 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8320 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8321 vcpu->run->fail_entry.hardware_entry_failure_reason
8326 if (unlikely(vmx->fail)) {
8327 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8328 vcpu->run->fail_entry.hardware_entry_failure_reason
8329 = vmcs_read32(VM_INSTRUCTION_ERROR);
8335 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8336 * delivery event since it indicates guest is accessing MMIO.
8337 * The vm-exit can be triggered again after return to guest that
8338 * will cause infinite loop.
8340 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8341 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8342 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8343 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8344 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8345 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8346 vcpu->run->internal.ndata = 2;
8347 vcpu->run->internal.data[0] = vectoring_info;
8348 vcpu->run->internal.data[1] = exit_reason;
8352 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
8353 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
8354 get_vmcs12(vcpu))))) {
8355 if (vmx_interrupt_allowed(vcpu)) {
8356 vmx->soft_vnmi_blocked = 0;
8357 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
8358 vcpu->arch.nmi_pending) {
8360 * This CPU don't support us in finding the end of an
8361 * NMI-blocked window if the guest runs with IRQs
8362 * disabled. So we pull the trigger after 1 s of
8363 * futile waiting, but inform the user about this.
8365 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
8366 "state on VCPU %d after 1 s timeout\n",
8367 __func__, vcpu->vcpu_id);
8368 vmx->soft_vnmi_blocked = 0;
8372 if (exit_reason < kvm_vmx_max_exit_handlers
8373 && kvm_vmx_exit_handlers[exit_reason])
8374 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8376 WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_reason);
8377 kvm_queue_exception(vcpu, UD_VECTOR);
8382 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8384 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8386 if (is_guest_mode(vcpu) &&
8387 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8390 if (irr == -1 || tpr < irr) {
8391 vmcs_write32(TPR_THRESHOLD, 0);
8395 vmcs_write32(TPR_THRESHOLD, irr);
8398 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8400 u32 sec_exec_control;
8403 * There is not point to enable virtualize x2apic without enable
8406 if (!cpu_has_vmx_virtualize_x2apic_mode() ||
8407 !kvm_vcpu_apicv_active(vcpu))
8410 if (!cpu_need_tpr_shadow(vcpu))
8413 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8416 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8417 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8419 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8420 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8422 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8424 vmx_set_msr_bitmap(vcpu);
8427 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8429 struct vcpu_vmx *vmx = to_vmx(vcpu);
8432 * Currently we do not handle the nested case where L2 has an
8433 * APIC access page of its own; that page is still pinned.
8434 * Hence, we skip the case where the VCPU is in guest mode _and_
8435 * L1 prepared an APIC access page for L2.
8437 * For the case where L1 and L2 share the same APIC access page
8438 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8439 * in the vmcs12), this function will only update either the vmcs01
8440 * or the vmcs02. If the former, the vmcs02 will be updated by
8441 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8442 * the next L2->L1 exit.
8444 if (!is_guest_mode(vcpu) ||
8445 !nested_cpu_has2(vmx->nested.current_vmcs12,
8446 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
8447 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8450 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
8458 status = vmcs_read16(GUEST_INTR_STATUS);
8460 if (max_isr != old) {
8462 status |= max_isr << 8;
8463 vmcs_write16(GUEST_INTR_STATUS, status);
8467 static void vmx_set_rvi(int vector)
8475 status = vmcs_read16(GUEST_INTR_STATUS);
8476 old = (u8)status & 0xff;
8477 if ((u8)vector != old) {
8479 status |= (u8)vector;
8480 vmcs_write16(GUEST_INTR_STATUS, status);
8484 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8486 if (!is_guest_mode(vcpu)) {
8487 vmx_set_rvi(max_irr);
8495 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8498 if (nested_exit_on_intr(vcpu))
8502 * Else, fall back to pre-APICv interrupt injection since L2
8503 * is run without virtual interrupt delivery.
8505 if (!kvm_event_needs_reinjection(vcpu) &&
8506 vmx_interrupt_allowed(vcpu)) {
8507 kvm_queue_interrupt(vcpu, max_irr, false);
8508 vmx_inject_irq(vcpu);
8512 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
8514 if (!kvm_vcpu_apicv_active(vcpu))
8517 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8518 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8519 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8520 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8523 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8527 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8528 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
8531 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8532 exit_intr_info = vmx->exit_intr_info;
8534 /* Handle machine checks before interrupts are enabled */
8535 if (is_machine_check(exit_intr_info))
8536 kvm_machine_check();
8538 /* We need to handle NMIs before interrupts are enabled */
8539 if ((exit_intr_info & INTR_INFO_INTR_TYPE_MASK) == INTR_TYPE_NMI_INTR &&
8540 (exit_intr_info & INTR_INFO_VALID_MASK)) {
8541 kvm_before_handle_nmi(&vmx->vcpu);
8543 kvm_after_handle_nmi(&vmx->vcpu);
8547 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8549 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8550 register void *__sp asm(_ASM_SP);
8553 * If external interrupt exists, IF bit is set in rflags/eflags on the
8554 * interrupt stack frame, and interrupt will be enabled on a return
8555 * from interrupt handler.
8557 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8558 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8559 unsigned int vector;
8560 unsigned long entry;
8562 struct vcpu_vmx *vmx = to_vmx(vcpu);
8563 #ifdef CONFIG_X86_64
8567 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8568 desc = (gate_desc *)vmx->host_idt_base + vector;
8569 entry = gate_offset(*desc);
8571 #ifdef CONFIG_X86_64
8572 "mov %%" _ASM_SP ", %[sp]\n\t"
8573 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8578 __ASM_SIZE(push) " $%c[cs]\n\t"
8579 "call *%[entry]\n\t"
8581 #ifdef CONFIG_X86_64
8587 [ss]"i"(__KERNEL_DS),
8588 [cs]"i"(__KERNEL_CS)
8593 static bool vmx_has_high_real_mode_segbase(void)
8595 return enable_unrestricted_guest || emulate_invalid_guest_state;
8598 static bool vmx_mpx_supported(void)
8600 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8601 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8604 static bool vmx_xsaves_supported(void)
8606 return vmcs_config.cpu_based_2nd_exec_ctrl &
8607 SECONDARY_EXEC_XSAVES;
8610 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8615 bool idtv_info_valid;
8617 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8619 if (cpu_has_virtual_nmis()) {
8620 if (vmx->nmi_known_unmasked)
8623 * Can't use vmx->exit_intr_info since we're not sure what
8624 * the exit reason is.
8626 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8627 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8628 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8630 * SDM 3: 27.7.1.2 (September 2008)
8631 * Re-set bit "block by NMI" before VM entry if vmexit caused by
8632 * a guest IRET fault.
8633 * SDM 3: 23.2.2 (September 2008)
8634 * Bit 12 is undefined in any of the following cases:
8635 * If the VM exit sets the valid bit in the IDT-vectoring
8636 * information field.
8637 * If the VM exit is due to a double fault.
8639 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8640 vector != DF_VECTOR && !idtv_info_valid)
8641 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8642 GUEST_INTR_STATE_NMI);
8644 vmx->nmi_known_unmasked =
8645 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8646 & GUEST_INTR_STATE_NMI);
8647 } else if (unlikely(vmx->soft_vnmi_blocked))
8648 vmx->vnmi_blocked_time +=
8649 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
8652 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8653 u32 idt_vectoring_info,
8654 int instr_len_field,
8655 int error_code_field)
8659 bool idtv_info_valid;
8661 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8663 vcpu->arch.nmi_injected = false;
8664 kvm_clear_exception_queue(vcpu);
8665 kvm_clear_interrupt_queue(vcpu);
8667 if (!idtv_info_valid)
8670 kvm_make_request(KVM_REQ_EVENT, vcpu);
8672 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8673 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8676 case INTR_TYPE_NMI_INTR:
8677 vcpu->arch.nmi_injected = true;
8679 * SDM 3: 27.7.1.2 (September 2008)
8680 * Clear bit "block by NMI" before VM entry if a NMI
8683 vmx_set_nmi_mask(vcpu, false);
8685 case INTR_TYPE_SOFT_EXCEPTION:
8686 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8688 case INTR_TYPE_HARD_EXCEPTION:
8689 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8690 u32 err = vmcs_read32(error_code_field);
8691 kvm_requeue_exception_e(vcpu, vector, err);
8693 kvm_requeue_exception(vcpu, vector);
8695 case INTR_TYPE_SOFT_INTR:
8696 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8698 case INTR_TYPE_EXT_INTR:
8699 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8706 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8708 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8709 VM_EXIT_INSTRUCTION_LEN,
8710 IDT_VECTORING_ERROR_CODE);
8713 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8715 __vmx_complete_interrupts(vcpu,
8716 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8717 VM_ENTRY_INSTRUCTION_LEN,
8718 VM_ENTRY_EXCEPTION_ERROR_CODE);
8720 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8723 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8726 struct perf_guest_switch_msr *msrs;
8728 msrs = perf_guest_get_msrs(&nr_msrs);
8733 for (i = 0; i < nr_msrs; i++)
8734 if (msrs[i].host == msrs[i].guest)
8735 clear_atomic_switch_msr(vmx, msrs[i].msr);
8737 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8741 void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
8743 struct vcpu_vmx *vmx = to_vmx(vcpu);
8747 if (vmx->hv_deadline_tsc == -1)
8751 if (vmx->hv_deadline_tsc > tscl)
8752 /* sure to be 32 bit only because checked on set_hv_timer */
8753 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
8754 cpu_preemption_timer_multi);
8758 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
8761 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8763 struct vcpu_vmx *vmx = to_vmx(vcpu);
8764 unsigned long debugctlmsr, cr4;
8766 /* Record the guest's net vcpu time for enforced NMI injections. */
8767 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
8768 vmx->entry_time = ktime_get();
8770 /* Don't enter VMX if guest state is invalid, let the exit handler
8771 start emulation until we arrive back to a valid state */
8772 if (vmx->emulation_required)
8775 if (vmx->ple_window_dirty) {
8776 vmx->ple_window_dirty = false;
8777 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8780 if (vmx->nested.sync_shadow_vmcs) {
8781 copy_vmcs12_to_shadow(vmx);
8782 vmx->nested.sync_shadow_vmcs = false;
8785 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8786 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
8787 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
8788 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
8790 cr4 = cr4_read_shadow();
8791 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
8792 vmcs_writel(HOST_CR4, cr4);
8793 vmx->host_state.vmcs_host_cr4 = cr4;
8796 /* When single-stepping over STI and MOV SS, we must clear the
8797 * corresponding interruptibility bits in the guest state. Otherwise
8798 * vmentry fails as it then expects bit 14 (BS) in pending debug
8799 * exceptions being set, but that's not correct for the guest debugging
8801 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8802 vmx_set_interrupt_shadow(vcpu, 0);
8804 if (vmx->guest_pkru_valid)
8805 __write_pkru(vmx->guest_pkru);
8807 atomic_switch_perf_msrs(vmx);
8808 debugctlmsr = get_debugctlmsr();
8810 vmx_arm_hv_timer(vcpu);
8812 vmx->__launched = vmx->loaded_vmcs->launched;
8814 /* Store host registers */
8815 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
8816 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
8817 "push %%" _ASM_CX " \n\t"
8818 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8820 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8821 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
8823 /* Reload cr2 if changed */
8824 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
8825 "mov %%cr2, %%" _ASM_DX " \n\t"
8826 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
8828 "mov %%" _ASM_AX", %%cr2 \n\t"
8830 /* Check if vmlaunch of vmresume is needed */
8831 "cmpl $0, %c[launched](%0) \n\t"
8832 /* Load guest registers. Don't clobber flags. */
8833 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
8834 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
8835 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
8836 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
8837 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
8838 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
8839 #ifdef CONFIG_X86_64
8840 "mov %c[r8](%0), %%r8 \n\t"
8841 "mov %c[r9](%0), %%r9 \n\t"
8842 "mov %c[r10](%0), %%r10 \n\t"
8843 "mov %c[r11](%0), %%r11 \n\t"
8844 "mov %c[r12](%0), %%r12 \n\t"
8845 "mov %c[r13](%0), %%r13 \n\t"
8846 "mov %c[r14](%0), %%r14 \n\t"
8847 "mov %c[r15](%0), %%r15 \n\t"
8849 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
8851 /* Enter guest mode */
8853 __ex(ASM_VMX_VMLAUNCH) "\n\t"
8855 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
8857 /* Save guest registers, load host registers, keep flags */
8858 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
8860 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
8861 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
8862 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
8863 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
8864 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
8865 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
8866 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
8867 #ifdef CONFIG_X86_64
8868 "mov %%r8, %c[r8](%0) \n\t"
8869 "mov %%r9, %c[r9](%0) \n\t"
8870 "mov %%r10, %c[r10](%0) \n\t"
8871 "mov %%r11, %c[r11](%0) \n\t"
8872 "mov %%r12, %c[r12](%0) \n\t"
8873 "mov %%r13, %c[r13](%0) \n\t"
8874 "mov %%r14, %c[r14](%0) \n\t"
8875 "mov %%r15, %c[r15](%0) \n\t"
8877 "mov %%cr2, %%" _ASM_AX " \n\t"
8878 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
8880 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
8881 "setbe %c[fail](%0) \n\t"
8882 ".pushsection .rodata \n\t"
8883 ".global vmx_return \n\t"
8884 "vmx_return: " _ASM_PTR " 2b \n\t"
8886 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
8887 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
8888 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
8889 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
8890 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
8891 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
8892 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
8893 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
8894 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
8895 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
8896 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
8897 #ifdef CONFIG_X86_64
8898 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
8899 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
8900 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
8901 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
8902 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
8903 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
8904 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
8905 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
8907 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
8908 [wordsize]"i"(sizeof(ulong))
8910 #ifdef CONFIG_X86_64
8911 , "rax", "rbx", "rdi", "rsi"
8912 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
8914 , "eax", "ebx", "edi", "esi"
8918 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
8920 update_debugctlmsr(debugctlmsr);
8922 #ifndef CONFIG_X86_64
8924 * The sysexit path does not restore ds/es, so we must set them to
8925 * a reasonable value ourselves.
8927 * We can't defer this to vmx_load_host_state() since that function
8928 * may be executed in interrupt context, which saves and restore segments
8929 * around it, nullifying its effect.
8931 loadsegment(ds, __USER_DS);
8932 loadsegment(es, __USER_DS);
8935 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
8936 | (1 << VCPU_EXREG_RFLAGS)
8937 | (1 << VCPU_EXREG_PDPTR)
8938 | (1 << VCPU_EXREG_SEGMENTS)
8939 | (1 << VCPU_EXREG_CR3));
8940 vcpu->arch.regs_dirty = 0;
8942 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
8944 vmx->loaded_vmcs->launched = 1;
8946 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
8949 * eager fpu is enabled if PKEY is supported and CR4 is switched
8950 * back on host, so it is safe to read guest PKRU from current
8953 if (boot_cpu_has(X86_FEATURE_OSPKE)) {
8954 vmx->guest_pkru = __read_pkru();
8955 if (vmx->guest_pkru != vmx->host_pkru) {
8956 vmx->guest_pkru_valid = true;
8957 __write_pkru(vmx->host_pkru);
8959 vmx->guest_pkru_valid = false;
8963 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
8964 * we did not inject a still-pending event to L1 now because of
8965 * nested_run_pending, we need to re-enable this bit.
8967 if (vmx->nested.nested_run_pending)
8968 kvm_make_request(KVM_REQ_EVENT, vcpu);
8970 vmx->nested.nested_run_pending = 0;
8972 vmx_complete_atomic_exit(vmx);
8973 vmx_recover_nmi_blocking(vmx);
8974 vmx_complete_interrupts(vmx);
8977 static void vmx_load_vmcs01(struct kvm_vcpu *vcpu)
8979 struct vcpu_vmx *vmx = to_vmx(vcpu);
8982 if (vmx->loaded_vmcs == &vmx->vmcs01)
8986 vmx->loaded_vmcs = &vmx->vmcs01;
8988 vmx_vcpu_load(vcpu, cpu);
8993 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
8995 struct vcpu_vmx *vmx = to_vmx(vcpu);
8998 vmx_destroy_pml_buffer(vmx);
8999 free_vpid(vmx->vpid);
9000 leave_guest_mode(vcpu);
9001 vmx_load_vmcs01(vcpu);
9003 free_loaded_vmcs(vmx->loaded_vmcs);
9004 kfree(vmx->guest_msrs);
9005 kvm_vcpu_uninit(vcpu);
9006 kmem_cache_free(kvm_vcpu_cache, vmx);
9009 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
9012 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
9016 return ERR_PTR(-ENOMEM);
9018 vmx->vpid = allocate_vpid();
9020 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
9024 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
9025 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
9029 if (!vmx->guest_msrs) {
9033 vmx->loaded_vmcs = &vmx->vmcs01;
9034 vmx->loaded_vmcs->vmcs = alloc_vmcs();
9035 if (!vmx->loaded_vmcs->vmcs)
9038 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
9039 loaded_vmcs_init(vmx->loaded_vmcs);
9044 vmx_vcpu_load(&vmx->vcpu, cpu);
9045 vmx->vcpu.cpu = cpu;
9046 err = vmx_vcpu_setup(vmx);
9047 vmx_vcpu_put(&vmx->vcpu);
9051 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9052 err = alloc_apic_access_page(kvm);
9058 if (!kvm->arch.ept_identity_map_addr)
9059 kvm->arch.ept_identity_map_addr =
9060 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
9061 err = init_rmode_identity_map(kvm);
9067 nested_vmx_setup_ctls_msrs(vmx);
9068 vmx->nested.vpid02 = allocate_vpid();
9071 vmx->nested.posted_intr_nv = -1;
9072 vmx->nested.current_vmptr = -1ull;
9073 vmx->nested.current_vmcs12 = NULL;
9076 * If PML is turned on, failure on enabling PML just results in failure
9077 * of creating the vcpu, therefore we can simplify PML logic (by
9078 * avoiding dealing with cases, such as enabling PML partially on vcpus
9079 * for the guest, etc.
9082 err = vmx_create_pml_buffer(vmx);
9087 vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
9092 free_vpid(vmx->nested.vpid02);
9093 free_loaded_vmcs(vmx->loaded_vmcs);
9095 kfree(vmx->guest_msrs);
9097 kvm_vcpu_uninit(&vmx->vcpu);
9099 free_vpid(vmx->vpid);
9100 kmem_cache_free(kvm_vcpu_cache, vmx);
9101 return ERR_PTR(err);
9104 static void __init vmx_check_processor_compat(void *rtn)
9106 struct vmcs_config vmcs_conf;
9109 if (setup_vmcs_config(&vmcs_conf) < 0)
9111 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
9112 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
9113 smp_processor_id());
9118 static int get_ept_level(void)
9120 return VMX_EPT_DEFAULT_GAW + 1;
9123 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
9128 /* For VT-d and EPT combination
9129 * 1. MMIO: always map as UC
9131 * a. VT-d without snooping control feature: can't guarantee the
9132 * result, try to trust guest.
9133 * b. VT-d with snooping control feature: snooping control feature of
9134 * VT-d engine can guarantee the cache correctness. Just set it
9135 * to WB to keep consistent with host. So the same as item 3.
9136 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9137 * consistent with host MTRR
9140 cache = MTRR_TYPE_UNCACHABLE;
9144 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
9145 ipat = VMX_EPT_IPAT_BIT;
9146 cache = MTRR_TYPE_WRBACK;
9150 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
9151 ipat = VMX_EPT_IPAT_BIT;
9152 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
9153 cache = MTRR_TYPE_WRBACK;
9155 cache = MTRR_TYPE_UNCACHABLE;
9159 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
9162 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
9165 static int vmx_get_lpage_level(void)
9167 if (enable_ept && !cpu_has_vmx_ept_1g_page())
9168 return PT_DIRECTORY_LEVEL;
9170 /* For shadow and EPT supported 1GB page */
9171 return PT_PDPE_LEVEL;
9174 static void vmcs_set_secondary_exec_control(u32 new_ctl)
9177 * These bits in the secondary execution controls field
9178 * are dynamic, the others are mostly based on the hypervisor
9179 * architecture and the guest's CPUID. Do not touch the
9183 SECONDARY_EXEC_SHADOW_VMCS |
9184 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
9185 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9187 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9189 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
9190 (new_ctl & ~mask) | (cur_ctl & mask));
9193 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
9195 struct kvm_cpuid_entry2 *best;
9196 struct vcpu_vmx *vmx = to_vmx(vcpu);
9197 u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx);
9199 if (vmx_rdtscp_supported()) {
9200 bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu);
9201 if (!rdtscp_enabled)
9202 secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP;
9206 vmx->nested.nested_vmx_secondary_ctls_high |=
9207 SECONDARY_EXEC_RDTSCP;
9209 vmx->nested.nested_vmx_secondary_ctls_high &=
9210 ~SECONDARY_EXEC_RDTSCP;
9214 /* Exposing INVPCID only when PCID is exposed */
9215 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9216 if (vmx_invpcid_supported() &&
9217 (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) ||
9218 !guest_cpuid_has_pcid(vcpu))) {
9219 secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID;
9222 best->ebx &= ~bit(X86_FEATURE_INVPCID);
9225 if (cpu_has_secondary_exec_ctrls())
9226 vmcs_set_secondary_exec_control(secondary_exec_ctl);
9228 if (static_cpu_has(X86_FEATURE_PCOMMIT) && nested) {
9229 if (guest_cpuid_has_pcommit(vcpu))
9230 vmx->nested.nested_vmx_secondary_ctls_high |=
9231 SECONDARY_EXEC_PCOMMIT;
9233 vmx->nested.nested_vmx_secondary_ctls_high &=
9234 ~SECONDARY_EXEC_PCOMMIT;
9237 if (nested_vmx_allowed(vcpu))
9238 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
9239 FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9241 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
9242 ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9245 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9247 if (func == 1 && nested)
9248 entry->ecx |= bit(X86_FEATURE_VMX);
9251 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9252 struct x86_exception *fault)
9254 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9257 if (fault->error_code & PFERR_RSVD_MASK)
9258 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9260 exit_reason = EXIT_REASON_EPT_VIOLATION;
9261 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
9262 vmcs12->guest_physical_address = fault->address;
9265 /* Callbacks for nested_ept_init_mmu_context: */
9267 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9269 /* return the page table to be shadowed - in our case, EPT12 */
9270 return get_vmcs12(vcpu)->ept_pointer;
9273 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9275 WARN_ON(mmu_is_nested(vcpu));
9276 kvm_init_shadow_ept_mmu(vcpu,
9277 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
9278 VMX_EPT_EXECUTE_ONLY_BIT);
9279 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
9280 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
9281 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
9283 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
9286 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
9288 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
9291 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9294 bool inequality, bit;
9296 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9298 (error_code & vmcs12->page_fault_error_code_mask) !=
9299 vmcs12->page_fault_error_code_match;
9300 return inequality ^ bit;
9303 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9304 struct x86_exception *fault)
9306 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9308 WARN_ON(!is_guest_mode(vcpu));
9310 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code))
9311 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
9312 vmcs_read32(VM_EXIT_INTR_INFO),
9313 vmcs_readl(EXIT_QUALIFICATION));
9315 kvm_inject_page_fault(vcpu, fault);
9318 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9319 struct vmcs12 *vmcs12)
9321 struct vcpu_vmx *vmx = to_vmx(vcpu);
9322 int maxphyaddr = cpuid_maxphyaddr(vcpu);
9324 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9325 if (!PAGE_ALIGNED(vmcs12->apic_access_addr) ||
9326 vmcs12->apic_access_addr >> maxphyaddr)
9330 * Translate L1 physical address to host physical
9331 * address for vmcs02. Keep the page pinned, so this
9332 * physical address remains valid. We keep a reference
9333 * to it so we can release it later.
9335 if (vmx->nested.apic_access_page) /* shouldn't happen */
9336 nested_release_page(vmx->nested.apic_access_page);
9337 vmx->nested.apic_access_page =
9338 nested_get_page(vcpu, vmcs12->apic_access_addr);
9341 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9342 if (!PAGE_ALIGNED(vmcs12->virtual_apic_page_addr) ||
9343 vmcs12->virtual_apic_page_addr >> maxphyaddr)
9346 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
9347 nested_release_page(vmx->nested.virtual_apic_page);
9348 vmx->nested.virtual_apic_page =
9349 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
9352 * Failing the vm entry is _not_ what the processor does
9353 * but it's basically the only possibility we have.
9354 * We could still enter the guest if CR8 load exits are
9355 * enabled, CR8 store exits are enabled, and virtualize APIC
9356 * access is disabled; in this case the processor would never
9357 * use the TPR shadow and we could simply clear the bit from
9358 * the execution control. But such a configuration is useless,
9359 * so let's keep the code simple.
9361 if (!vmx->nested.virtual_apic_page)
9365 if (nested_cpu_has_posted_intr(vmcs12)) {
9366 if (!IS_ALIGNED(vmcs12->posted_intr_desc_addr, 64) ||
9367 vmcs12->posted_intr_desc_addr >> maxphyaddr)
9370 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9371 kunmap(vmx->nested.pi_desc_page);
9372 nested_release_page(vmx->nested.pi_desc_page);
9374 vmx->nested.pi_desc_page =
9375 nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
9376 if (!vmx->nested.pi_desc_page)
9379 vmx->nested.pi_desc =
9380 (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
9381 if (!vmx->nested.pi_desc) {
9382 nested_release_page_clean(vmx->nested.pi_desc_page);
9385 vmx->nested.pi_desc =
9386 (struct pi_desc *)((void *)vmx->nested.pi_desc +
9387 (unsigned long)(vmcs12->posted_intr_desc_addr &
9394 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9396 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9397 struct vcpu_vmx *vmx = to_vmx(vcpu);
9399 if (vcpu->arch.virtual_tsc_khz == 0)
9402 /* Make sure short timeouts reliably trigger an immediate vmexit.
9403 * hrtimer_start does not guarantee this. */
9404 if (preemption_timeout <= 1) {
9405 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9409 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9410 preemption_timeout *= 1000000;
9411 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9412 hrtimer_start(&vmx->nested.preemption_timer,
9413 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9416 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9417 struct vmcs12 *vmcs12)
9422 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9425 if (vmcs12_read_any(vcpu, MSR_BITMAP, &addr)) {
9429 maxphyaddr = cpuid_maxphyaddr(vcpu);
9431 if (!PAGE_ALIGNED(vmcs12->msr_bitmap) ||
9432 ((addr + PAGE_SIZE) >> maxphyaddr))
9439 * Merge L0's and L1's MSR bitmap, return false to indicate that
9440 * we do not use the hardware.
9442 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9443 struct vmcs12 *vmcs12)
9447 unsigned long *msr_bitmap;
9449 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
9452 page = nested_get_page(vcpu, vmcs12->msr_bitmap);
9457 msr_bitmap = (unsigned long *)kmap(page);
9459 nested_release_page_clean(page);
9464 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
9465 if (nested_cpu_has_apic_reg_virt(vmcs12))
9466 for (msr = 0x800; msr <= 0x8ff; msr++)
9467 nested_vmx_disable_intercept_for_msr(
9469 vmx_msr_bitmap_nested,
9471 /* TPR is allowed */
9472 nested_vmx_disable_intercept_for_msr(msr_bitmap,
9473 vmx_msr_bitmap_nested,
9474 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9475 MSR_TYPE_R | MSR_TYPE_W);
9476 if (nested_cpu_has_vid(vmcs12)) {
9477 /* EOI and self-IPI are allowed */
9478 nested_vmx_disable_intercept_for_msr(
9480 vmx_msr_bitmap_nested,
9481 APIC_BASE_MSR + (APIC_EOI >> 4),
9483 nested_vmx_disable_intercept_for_msr(
9485 vmx_msr_bitmap_nested,
9486 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9491 * Enable reading intercept of all the x2apic
9492 * MSRs. We should not rely on vmcs12 to do any
9493 * optimizations here, it may have been modified
9496 for (msr = 0x800; msr <= 0x8ff; msr++)
9497 __vmx_enable_intercept_for_msr(
9498 vmx_msr_bitmap_nested,
9502 __vmx_enable_intercept_for_msr(
9503 vmx_msr_bitmap_nested,
9504 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9506 __vmx_enable_intercept_for_msr(
9507 vmx_msr_bitmap_nested,
9508 APIC_BASE_MSR + (APIC_EOI >> 4),
9510 __vmx_enable_intercept_for_msr(
9511 vmx_msr_bitmap_nested,
9512 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9516 nested_release_page_clean(page);
9521 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
9522 struct vmcs12 *vmcs12)
9524 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9525 !nested_cpu_has_apic_reg_virt(vmcs12) &&
9526 !nested_cpu_has_vid(vmcs12) &&
9527 !nested_cpu_has_posted_intr(vmcs12))
9531 * If virtualize x2apic mode is enabled,
9532 * virtualize apic access must be disabled.
9534 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9535 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9539 * If virtual interrupt delivery is enabled,
9540 * we must exit on external interrupts.
9542 if (nested_cpu_has_vid(vmcs12) &&
9543 !nested_exit_on_intr(vcpu))
9547 * bits 15:8 should be zero in posted_intr_nv,
9548 * the descriptor address has been already checked
9549 * in nested_get_vmcs12_pages.
9551 if (nested_cpu_has_posted_intr(vmcs12) &&
9552 (!nested_cpu_has_vid(vmcs12) ||
9553 !nested_exit_intr_ack_set(vcpu) ||
9554 vmcs12->posted_intr_nv & 0xff00))
9557 /* tpr shadow is needed by all apicv features. */
9558 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9564 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9565 unsigned long count_field,
9566 unsigned long addr_field)
9571 if (vmcs12_read_any(vcpu, count_field, &count) ||
9572 vmcs12_read_any(vcpu, addr_field, &addr)) {
9578 maxphyaddr = cpuid_maxphyaddr(vcpu);
9579 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9580 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9581 pr_warn_ratelimited(
9582 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9583 addr_field, maxphyaddr, count, addr);
9589 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9590 struct vmcs12 *vmcs12)
9592 if (vmcs12->vm_exit_msr_load_count == 0 &&
9593 vmcs12->vm_exit_msr_store_count == 0 &&
9594 vmcs12->vm_entry_msr_load_count == 0)
9595 return 0; /* Fast path */
9596 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9597 VM_EXIT_MSR_LOAD_ADDR) ||
9598 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9599 VM_EXIT_MSR_STORE_ADDR) ||
9600 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9601 VM_ENTRY_MSR_LOAD_ADDR))
9606 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9607 struct vmx_msr_entry *e)
9609 /* x2APIC MSR accesses are not allowed */
9610 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9612 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9613 e->index == MSR_IA32_UCODE_REV)
9615 if (e->reserved != 0)
9620 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9621 struct vmx_msr_entry *e)
9623 if (e->index == MSR_FS_BASE ||
9624 e->index == MSR_GS_BASE ||
9625 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9626 nested_vmx_msr_check_common(vcpu, e))
9631 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9632 struct vmx_msr_entry *e)
9634 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9635 nested_vmx_msr_check_common(vcpu, e))
9641 * Load guest's/host's msr at nested entry/exit.
9642 * return 0 for success, entry index for failure.
9644 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9647 struct vmx_msr_entry e;
9648 struct msr_data msr;
9650 msr.host_initiated = false;
9651 for (i = 0; i < count; i++) {
9652 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
9654 pr_warn_ratelimited(
9655 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9656 __func__, i, gpa + i * sizeof(e));
9659 if (nested_vmx_load_msr_check(vcpu, &e)) {
9660 pr_warn_ratelimited(
9661 "%s check failed (%u, 0x%x, 0x%x)\n",
9662 __func__, i, e.index, e.reserved);
9665 msr.index = e.index;
9667 if (kvm_set_msr(vcpu, &msr)) {
9668 pr_warn_ratelimited(
9669 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9670 __func__, i, e.index, e.value);
9679 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9682 struct vmx_msr_entry e;
9684 for (i = 0; i < count; i++) {
9685 struct msr_data msr_info;
9686 if (kvm_vcpu_read_guest(vcpu,
9687 gpa + i * sizeof(e),
9688 &e, 2 * sizeof(u32))) {
9689 pr_warn_ratelimited(
9690 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9691 __func__, i, gpa + i * sizeof(e));
9694 if (nested_vmx_store_msr_check(vcpu, &e)) {
9695 pr_warn_ratelimited(
9696 "%s check failed (%u, 0x%x, 0x%x)\n",
9697 __func__, i, e.index, e.reserved);
9700 msr_info.host_initiated = false;
9701 msr_info.index = e.index;
9702 if (kvm_get_msr(vcpu, &msr_info)) {
9703 pr_warn_ratelimited(
9704 "%s cannot read MSR (%u, 0x%x)\n",
9705 __func__, i, e.index);
9708 if (kvm_vcpu_write_guest(vcpu,
9709 gpa + i * sizeof(e) +
9710 offsetof(struct vmx_msr_entry, value),
9711 &msr_info.data, sizeof(msr_info.data))) {
9712 pr_warn_ratelimited(
9713 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9714 __func__, i, e.index, msr_info.data);
9722 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
9723 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
9724 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
9725 * guest in a way that will both be appropriate to L1's requests, and our
9726 * needs. In addition to modifying the active vmcs (which is vmcs02), this
9727 * function also has additional necessary side-effects, like setting various
9728 * vcpu->arch fields.
9730 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9732 struct vcpu_vmx *vmx = to_vmx(vcpu);
9735 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
9736 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
9737 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
9738 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
9739 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
9740 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
9741 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
9742 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
9743 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
9744 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
9745 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
9746 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
9747 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
9748 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
9749 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
9750 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
9751 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
9752 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
9753 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
9754 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
9755 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
9756 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
9757 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
9758 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
9759 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
9760 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
9761 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
9762 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
9763 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
9764 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
9765 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
9766 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
9767 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
9768 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
9769 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
9770 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
9772 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
9773 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
9774 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
9776 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
9777 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
9779 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
9780 vmcs12->vm_entry_intr_info_field);
9781 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
9782 vmcs12->vm_entry_exception_error_code);
9783 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
9784 vmcs12->vm_entry_instruction_len);
9785 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
9786 vmcs12->guest_interruptibility_info);
9787 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
9788 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
9789 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
9790 vmcs12->guest_pending_dbg_exceptions);
9791 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
9792 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
9794 if (nested_cpu_has_xsaves(vmcs12))
9795 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
9796 vmcs_write64(VMCS_LINK_POINTER, -1ull);
9798 exec_control = vmcs12->pin_based_vm_exec_control;
9799 exec_control |= vmcs_config.pin_based_exec_ctrl;
9800 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
9802 if (nested_cpu_has_posted_intr(vmcs12)) {
9804 * Note that we use L0's vector here and in
9805 * vmx_deliver_nested_posted_interrupt.
9807 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
9808 vmx->nested.pi_pending = false;
9809 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
9810 vmcs_write64(POSTED_INTR_DESC_ADDR,
9811 page_to_phys(vmx->nested.pi_desc_page) +
9812 (unsigned long)(vmcs12->posted_intr_desc_addr &
9815 exec_control &= ~PIN_BASED_POSTED_INTR;
9817 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
9819 vmx->nested.preemption_timer_expired = false;
9820 if (nested_cpu_has_preemption_timer(vmcs12))
9821 vmx_start_preemption_timer(vcpu);
9824 * Whether page-faults are trapped is determined by a combination of
9825 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
9826 * If enable_ept, L0 doesn't care about page faults and we should
9827 * set all of these to L1's desires. However, if !enable_ept, L0 does
9828 * care about (at least some) page faults, and because it is not easy
9829 * (if at all possible?) to merge L0 and L1's desires, we simply ask
9830 * to exit on each and every L2 page fault. This is done by setting
9831 * MASK=MATCH=0 and (see below) EB.PF=1.
9832 * Note that below we don't need special code to set EB.PF beyond the
9833 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
9834 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
9835 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
9837 * A problem with this approach (when !enable_ept) is that L1 may be
9838 * injected with more page faults than it asked for. This could have
9839 * caused problems, but in practice existing hypervisors don't care.
9840 * To fix this, we will need to emulate the PFEC checking (on the L1
9841 * page tables), using walk_addr(), when injecting PFs to L1.
9843 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
9844 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
9845 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
9846 enable_ept ? vmcs12->page_fault_error_code_match : 0);
9848 if (cpu_has_secondary_exec_ctrls()) {
9849 exec_control = vmx_secondary_exec_control(vmx);
9851 /* Take the following fields only from vmcs12 */
9852 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
9853 SECONDARY_EXEC_RDTSCP |
9854 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
9855 SECONDARY_EXEC_APIC_REGISTER_VIRT |
9856 SECONDARY_EXEC_PCOMMIT);
9857 if (nested_cpu_has(vmcs12,
9858 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
9859 exec_control |= vmcs12->secondary_vm_exec_control;
9861 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
9863 * If translation failed, no matter: This feature asks
9864 * to exit when accessing the given address, and if it
9865 * can never be accessed, this feature won't do
9868 if (!vmx->nested.apic_access_page)
9870 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9872 vmcs_write64(APIC_ACCESS_ADDR,
9873 page_to_phys(vmx->nested.apic_access_page));
9874 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9875 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9877 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9878 kvm_vcpu_reload_apic_access_page(vcpu);
9881 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
9882 vmcs_write64(EOI_EXIT_BITMAP0,
9883 vmcs12->eoi_exit_bitmap0);
9884 vmcs_write64(EOI_EXIT_BITMAP1,
9885 vmcs12->eoi_exit_bitmap1);
9886 vmcs_write64(EOI_EXIT_BITMAP2,
9887 vmcs12->eoi_exit_bitmap2);
9888 vmcs_write64(EOI_EXIT_BITMAP3,
9889 vmcs12->eoi_exit_bitmap3);
9890 vmcs_write16(GUEST_INTR_STATUS,
9891 vmcs12->guest_intr_status);
9894 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
9899 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
9900 * Some constant fields are set here by vmx_set_constant_host_state().
9901 * Other fields are different per CPU, and will be set later when
9902 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
9904 vmx_set_constant_host_state(vmx);
9907 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
9908 * entry, but only if the current (host) sp changed from the value
9909 * we wrote last (vmx->host_rsp). This cache is no longer relevant
9910 * if we switch vmcs, and rather than hold a separate cache per vmcs,
9911 * here we just force the write to happen on entry.
9915 exec_control = vmx_exec_control(vmx); /* L0's desires */
9916 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
9917 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
9918 exec_control &= ~CPU_BASED_TPR_SHADOW;
9919 exec_control |= vmcs12->cpu_based_vm_exec_control;
9921 if (exec_control & CPU_BASED_TPR_SHADOW) {
9922 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
9923 page_to_phys(vmx->nested.virtual_apic_page));
9924 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
9927 if (cpu_has_vmx_msr_bitmap() &&
9928 exec_control & CPU_BASED_USE_MSR_BITMAPS) {
9929 nested_vmx_merge_msr_bitmap(vcpu, vmcs12);
9930 /* MSR_BITMAP will be set by following vmx_set_efer. */
9932 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
9935 * Merging of IO bitmap not currently supported.
9936 * Rather, exit every time.
9938 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
9939 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
9941 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
9943 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
9944 * bitwise-or of what L1 wants to trap for L2, and what we want to
9945 * trap. Note that CR0.TS also needs updating - we do this later.
9947 update_exception_bitmap(vcpu);
9948 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
9949 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
9951 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
9952 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
9953 * bits are further modified by vmx_set_efer() below.
9955 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
9957 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
9958 * emulated by vmx_set_efer(), below.
9960 vm_entry_controls_init(vmx,
9961 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
9962 ~VM_ENTRY_IA32E_MODE) |
9963 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
9965 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) {
9966 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
9967 vcpu->arch.pat = vmcs12->guest_ia32_pat;
9968 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
9969 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
9972 set_cr4_guest_host_mask(vmx);
9974 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
9975 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
9977 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
9978 vmcs_write64(TSC_OFFSET,
9979 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
9981 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
9985 * There is no direct mapping between vpid02 and vpid12, the
9986 * vpid02 is per-vCPU for L0 and reused while the value of
9987 * vpid12 is changed w/ one invvpid during nested vmentry.
9988 * The vpid12 is allocated by L1 for L2, so it will not
9989 * influence global bitmap(for vpid01 and vpid02 allocation)
9990 * even if spawn a lot of nested vCPUs.
9992 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
9993 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
9994 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
9995 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
9996 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
9999 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
10000 vmx_flush_tlb(vcpu);
10005 if (nested_cpu_has_ept(vmcs12)) {
10006 kvm_mmu_unload(vcpu);
10007 nested_ept_init_mmu_context(vcpu);
10010 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
10011 vcpu->arch.efer = vmcs12->guest_ia32_efer;
10012 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
10013 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10015 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10016 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10017 vmx_set_efer(vcpu, vcpu->arch.efer);
10020 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
10021 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
10022 * The CR0_READ_SHADOW is what L2 should have expected to read given
10023 * the specifications by L1; It's not enough to take
10024 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10025 * have more bits than L1 expected.
10027 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
10028 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
10030 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
10031 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
10033 /* shadow page tables on either EPT or shadow page tables */
10034 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
10035 kvm_mmu_reset_context(vcpu);
10038 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
10041 * L1 may access the L2's PDPTR, so save them to construct vmcs12
10044 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
10045 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
10046 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
10047 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
10050 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
10051 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
10055 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
10056 * for running an L2 nested guest.
10058 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
10060 struct vmcs12 *vmcs12;
10061 struct vcpu_vmx *vmx = to_vmx(vcpu);
10063 struct loaded_vmcs *vmcs02;
10067 if (!nested_vmx_check_permission(vcpu) ||
10068 !nested_vmx_check_vmcs12(vcpu))
10071 skip_emulated_instruction(vcpu);
10072 vmcs12 = get_vmcs12(vcpu);
10074 if (enable_shadow_vmcs)
10075 copy_shadow_to_vmcs12(vmx);
10078 * The nested entry process starts with enforcing various prerequisites
10079 * on vmcs12 as required by the Intel SDM, and act appropriately when
10080 * they fail: As the SDM explains, some conditions should cause the
10081 * instruction to fail, while others will cause the instruction to seem
10082 * to succeed, but return an EXIT_REASON_INVALID_STATE.
10083 * To speed up the normal (success) code path, we should avoid checking
10084 * for misconfigurations which will anyway be caught by the processor
10085 * when using the merged vmcs02.
10087 if (vmcs12->launch_state == launch) {
10088 nested_vmx_failValid(vcpu,
10089 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
10090 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
10094 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
10095 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT) {
10096 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10100 if (!nested_get_vmcs12_pages(vcpu, vmcs12)) {
10101 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10105 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12)) {
10106 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10110 if (nested_vmx_check_apicv_controls(vcpu, vmcs12)) {
10111 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10115 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12)) {
10116 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10120 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
10121 vmx->nested.nested_vmx_true_procbased_ctls_low,
10122 vmx->nested.nested_vmx_procbased_ctls_high) ||
10123 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
10124 vmx->nested.nested_vmx_secondary_ctls_low,
10125 vmx->nested.nested_vmx_secondary_ctls_high) ||
10126 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
10127 vmx->nested.nested_vmx_pinbased_ctls_low,
10128 vmx->nested.nested_vmx_pinbased_ctls_high) ||
10129 !vmx_control_verify(vmcs12->vm_exit_controls,
10130 vmx->nested.nested_vmx_true_exit_ctls_low,
10131 vmx->nested.nested_vmx_exit_ctls_high) ||
10132 !vmx_control_verify(vmcs12->vm_entry_controls,
10133 vmx->nested.nested_vmx_true_entry_ctls_low,
10134 vmx->nested.nested_vmx_entry_ctls_high))
10136 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
10140 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
10141 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
10142 nested_vmx_failValid(vcpu,
10143 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
10147 if (!nested_cr0_valid(vcpu, vmcs12->guest_cr0) ||
10148 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
10149 nested_vmx_entry_failure(vcpu, vmcs12,
10150 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
10153 if (vmcs12->vmcs_link_pointer != -1ull) {
10154 nested_vmx_entry_failure(vcpu, vmcs12,
10155 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
10160 * If the load IA32_EFER VM-entry control is 1, the following checks
10161 * are performed on the field for the IA32_EFER MSR:
10162 * - Bits reserved in the IA32_EFER MSR must be 0.
10163 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
10164 * the IA-32e mode guest VM-exit control. It must also be identical
10165 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
10168 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
10169 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
10170 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
10171 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
10172 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
10173 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
10174 nested_vmx_entry_failure(vcpu, vmcs12,
10175 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
10181 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
10182 * IA32_EFER MSR must be 0 in the field for that register. In addition,
10183 * the values of the LMA and LME bits in the field must each be that of
10184 * the host address-space size VM-exit control.
10186 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
10187 ia32e = (vmcs12->vm_exit_controls &
10188 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
10189 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
10190 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
10191 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
10192 nested_vmx_entry_failure(vcpu, vmcs12,
10193 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
10199 * We're finally done with prerequisite checking, and can start with
10200 * the nested entry.
10203 vmcs02 = nested_get_current_vmcs02(vmx);
10207 enter_guest_mode(vcpu);
10209 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
10211 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
10212 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10215 vmx->loaded_vmcs = vmcs02;
10216 vmx_vcpu_put(vcpu);
10217 vmx_vcpu_load(vcpu, cpu);
10221 vmx_segment_cache_clear(vmx);
10223 prepare_vmcs02(vcpu, vmcs12);
10225 msr_entry_idx = nested_vmx_load_msr(vcpu,
10226 vmcs12->vm_entry_msr_load_addr,
10227 vmcs12->vm_entry_msr_load_count);
10228 if (msr_entry_idx) {
10229 leave_guest_mode(vcpu);
10230 vmx_load_vmcs01(vcpu);
10231 nested_vmx_entry_failure(vcpu, vmcs12,
10232 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
10236 vmcs12->launch_state = 1;
10238 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
10239 return kvm_vcpu_halt(vcpu);
10241 vmx->nested.nested_run_pending = 1;
10244 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10245 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10246 * returned as far as L1 is concerned. It will only return (and set
10247 * the success flag) when L2 exits (see nested_vmx_vmexit()).
10253 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
10254 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
10255 * This function returns the new value we should put in vmcs12.guest_cr0.
10256 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
10257 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
10258 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
10259 * didn't trap the bit, because if L1 did, so would L0).
10260 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
10261 * been modified by L2, and L1 knows it. So just leave the old value of
10262 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
10263 * isn't relevant, because if L0 traps this bit it can set it to anything.
10264 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
10265 * changed these bits, and therefore they need to be updated, but L0
10266 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
10267 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
10269 static inline unsigned long
10270 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10273 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
10274 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
10275 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
10276 vcpu->arch.cr0_guest_owned_bits));
10279 static inline unsigned long
10280 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10283 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
10284 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
10285 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
10286 vcpu->arch.cr4_guest_owned_bits));
10289 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
10290 struct vmcs12 *vmcs12)
10295 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
10296 nr = vcpu->arch.exception.nr;
10297 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10299 if (kvm_exception_is_soft(nr)) {
10300 vmcs12->vm_exit_instruction_len =
10301 vcpu->arch.event_exit_inst_len;
10302 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
10304 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
10306 if (vcpu->arch.exception.has_error_code) {
10307 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
10308 vmcs12->idt_vectoring_error_code =
10309 vcpu->arch.exception.error_code;
10312 vmcs12->idt_vectoring_info_field = idt_vectoring;
10313 } else if (vcpu->arch.nmi_injected) {
10314 vmcs12->idt_vectoring_info_field =
10315 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
10316 } else if (vcpu->arch.interrupt.pending) {
10317 nr = vcpu->arch.interrupt.nr;
10318 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10320 if (vcpu->arch.interrupt.soft) {
10321 idt_vectoring |= INTR_TYPE_SOFT_INTR;
10322 vmcs12->vm_entry_instruction_len =
10323 vcpu->arch.event_exit_inst_len;
10325 idt_vectoring |= INTR_TYPE_EXT_INTR;
10327 vmcs12->idt_vectoring_info_field = idt_vectoring;
10331 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
10333 struct vcpu_vmx *vmx = to_vmx(vcpu);
10335 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
10336 vmx->nested.preemption_timer_expired) {
10337 if (vmx->nested.nested_run_pending)
10339 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
10343 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
10344 if (vmx->nested.nested_run_pending ||
10345 vcpu->arch.interrupt.pending)
10347 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10348 NMI_VECTOR | INTR_TYPE_NMI_INTR |
10349 INTR_INFO_VALID_MASK, 0);
10351 * The NMI-triggered VM exit counts as injection:
10352 * clear this one and block further NMIs.
10354 vcpu->arch.nmi_pending = 0;
10355 vmx_set_nmi_mask(vcpu, true);
10359 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
10360 nested_exit_on_intr(vcpu)) {
10361 if (vmx->nested.nested_run_pending)
10363 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
10367 return vmx_complete_nested_posted_interrupt(vcpu);
10370 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
10372 ktime_t remaining =
10373 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
10376 if (ktime_to_ns(remaining) <= 0)
10379 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
10380 do_div(value, 1000000);
10381 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10385 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10386 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10387 * and this function updates it to reflect the changes to the guest state while
10388 * L2 was running (and perhaps made some exits which were handled directly by L0
10389 * without going back to L1), and to reflect the exit reason.
10390 * Note that we do not have to copy here all VMCS fields, just those that
10391 * could have changed by the L2 guest or the exit - i.e., the guest-state and
10392 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10393 * which already writes to vmcs12 directly.
10395 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10396 u32 exit_reason, u32 exit_intr_info,
10397 unsigned long exit_qualification)
10399 /* update guest state fields: */
10400 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
10401 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
10403 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
10404 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
10405 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
10407 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
10408 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
10409 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
10410 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
10411 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
10412 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
10413 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
10414 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
10415 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
10416 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
10417 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
10418 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
10419 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
10420 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
10421 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
10422 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
10423 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
10424 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
10425 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
10426 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
10427 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
10428 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
10429 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
10430 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
10431 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
10432 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
10433 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
10434 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
10435 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
10436 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
10437 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
10438 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
10439 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
10440 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
10441 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
10442 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
10444 vmcs12->guest_interruptibility_info =
10445 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
10446 vmcs12->guest_pending_dbg_exceptions =
10447 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
10448 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10449 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
10451 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
10453 if (nested_cpu_has_preemption_timer(vmcs12)) {
10454 if (vmcs12->vm_exit_controls &
10455 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
10456 vmcs12->vmx_preemption_timer_value =
10457 vmx_get_preemption_timer_value(vcpu);
10458 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
10462 * In some cases (usually, nested EPT), L2 is allowed to change its
10463 * own CR3 without exiting. If it has changed it, we must keep it.
10464 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10465 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10467 * Additionally, restore L2's PDPTR to vmcs12.
10470 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
10471 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
10472 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
10473 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
10474 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
10477 if (nested_cpu_has_vid(vmcs12))
10478 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
10480 vmcs12->vm_entry_controls =
10481 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
10482 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
10484 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
10485 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
10486 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10489 /* TODO: These cannot have changed unless we have MSR bitmaps and
10490 * the relevant bit asks not to trap the change */
10491 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
10492 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
10493 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
10494 vmcs12->guest_ia32_efer = vcpu->arch.efer;
10495 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
10496 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
10497 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
10498 if (kvm_mpx_supported())
10499 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
10500 if (nested_cpu_has_xsaves(vmcs12))
10501 vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP);
10503 /* update exit information fields: */
10505 vmcs12->vm_exit_reason = exit_reason;
10506 vmcs12->exit_qualification = exit_qualification;
10508 vmcs12->vm_exit_intr_info = exit_intr_info;
10509 if ((vmcs12->vm_exit_intr_info &
10510 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
10511 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
10512 vmcs12->vm_exit_intr_error_code =
10513 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
10514 vmcs12->idt_vectoring_info_field = 0;
10515 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
10516 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
10518 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
10519 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10520 * instead of reading the real value. */
10521 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
10524 * Transfer the event that L0 or L1 may wanted to inject into
10525 * L2 to IDT_VECTORING_INFO_FIELD.
10527 vmcs12_save_pending_event(vcpu, vmcs12);
10531 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10532 * preserved above and would only end up incorrectly in L1.
10534 vcpu->arch.nmi_injected = false;
10535 kvm_clear_exception_queue(vcpu);
10536 kvm_clear_interrupt_queue(vcpu);
10540 * A part of what we need to when the nested L2 guest exits and we want to
10541 * run its L1 parent, is to reset L1's guest state to the host state specified
10543 * This function is to be called not only on normal nested exit, but also on
10544 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10545 * Failures During or After Loading Guest State").
10546 * This function should be called when the active VMCS is L1's (vmcs01).
10548 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10549 struct vmcs12 *vmcs12)
10551 struct kvm_segment seg;
10553 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10554 vcpu->arch.efer = vmcs12->host_ia32_efer;
10555 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10556 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10558 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10559 vmx_set_efer(vcpu, vcpu->arch.efer);
10561 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10562 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10563 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10565 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10566 * actually changed, because it depends on the current state of
10567 * fpu_active (which may have changed).
10568 * Note that vmx_set_cr0 refers to efer set above.
10570 vmx_set_cr0(vcpu, vmcs12->host_cr0);
10572 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
10573 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
10574 * but we also need to update cr0_guest_host_mask and exception_bitmap.
10576 update_exception_bitmap(vcpu);
10577 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
10578 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10581 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
10582 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
10584 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
10585 kvm_set_cr4(vcpu, vmcs12->host_cr4);
10587 nested_ept_uninit_mmu_context(vcpu);
10589 kvm_set_cr3(vcpu, vmcs12->host_cr3);
10590 kvm_mmu_reset_context(vcpu);
10593 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
10597 * Trivially support vpid by letting L2s share their parent
10598 * L1's vpid. TODO: move to a more elaborate solution, giving
10599 * each L2 its own vpid and exposing the vpid feature to L1.
10601 vmx_flush_tlb(vcpu);
10605 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
10606 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
10607 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
10608 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
10609 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
10611 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
10612 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
10613 vmcs_write64(GUEST_BNDCFGS, 0);
10615 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
10616 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
10617 vcpu->arch.pat = vmcs12->host_ia32_pat;
10619 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
10620 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
10621 vmcs12->host_ia32_perf_global_ctrl);
10623 /* Set L1 segment info according to Intel SDM
10624 27.5.2 Loading Host Segment and Descriptor-Table Registers */
10625 seg = (struct kvm_segment) {
10627 .limit = 0xFFFFFFFF,
10628 .selector = vmcs12->host_cs_selector,
10634 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10638 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
10639 seg = (struct kvm_segment) {
10641 .limit = 0xFFFFFFFF,
10648 seg.selector = vmcs12->host_ds_selector;
10649 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
10650 seg.selector = vmcs12->host_es_selector;
10651 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
10652 seg.selector = vmcs12->host_ss_selector;
10653 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
10654 seg.selector = vmcs12->host_fs_selector;
10655 seg.base = vmcs12->host_fs_base;
10656 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
10657 seg.selector = vmcs12->host_gs_selector;
10658 seg.base = vmcs12->host_gs_base;
10659 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
10660 seg = (struct kvm_segment) {
10661 .base = vmcs12->host_tr_base,
10663 .selector = vmcs12->host_tr_selector,
10667 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
10669 kvm_set_dr(vcpu, 7, 0x400);
10670 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
10672 if (cpu_has_vmx_msr_bitmap())
10673 vmx_set_msr_bitmap(vcpu);
10675 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
10676 vmcs12->vm_exit_msr_load_count))
10677 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
10681 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
10682 * and modify vmcs12 to make it see what it would expect to see there if
10683 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
10685 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
10686 u32 exit_intr_info,
10687 unsigned long exit_qualification)
10689 struct vcpu_vmx *vmx = to_vmx(vcpu);
10690 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10692 /* trying to cancel vmlaunch/vmresume is a bug */
10693 WARN_ON_ONCE(vmx->nested.nested_run_pending);
10695 leave_guest_mode(vcpu);
10696 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
10697 exit_qualification);
10699 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
10700 vmcs12->vm_exit_msr_store_count))
10701 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
10703 vmx_load_vmcs01(vcpu);
10705 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
10706 && nested_exit_intr_ack_set(vcpu)) {
10707 int irq = kvm_cpu_get_interrupt(vcpu);
10709 vmcs12->vm_exit_intr_info = irq |
10710 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
10713 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
10714 vmcs12->exit_qualification,
10715 vmcs12->idt_vectoring_info_field,
10716 vmcs12->vm_exit_intr_info,
10717 vmcs12->vm_exit_intr_error_code,
10720 vm_entry_controls_init(vmx, vmcs_read32(VM_ENTRY_CONTROLS));
10721 vm_exit_controls_init(vmx, vmcs_read32(VM_EXIT_CONTROLS));
10722 vmx_segment_cache_clear(vmx);
10724 /* if no vmcs02 cache requested, remove the one we used */
10725 if (VMCS02_POOL_SIZE == 0)
10726 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
10728 load_vmcs12_host_state(vcpu, vmcs12);
10730 /* Update TSC_OFFSET if TSC was changed while L2 ran */
10731 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
10733 /* This is needed for same reason as it was needed in prepare_vmcs02 */
10736 /* Unpin physical memory we referred to in vmcs02 */
10737 if (vmx->nested.apic_access_page) {
10738 nested_release_page(vmx->nested.apic_access_page);
10739 vmx->nested.apic_access_page = NULL;
10741 if (vmx->nested.virtual_apic_page) {
10742 nested_release_page(vmx->nested.virtual_apic_page);
10743 vmx->nested.virtual_apic_page = NULL;
10745 if (vmx->nested.pi_desc_page) {
10746 kunmap(vmx->nested.pi_desc_page);
10747 nested_release_page(vmx->nested.pi_desc_page);
10748 vmx->nested.pi_desc_page = NULL;
10749 vmx->nested.pi_desc = NULL;
10753 * We are now running in L2, mmu_notifier will force to reload the
10754 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
10756 kvm_vcpu_reload_apic_access_page(vcpu);
10759 * Exiting from L2 to L1, we're now back to L1 which thinks it just
10760 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
10761 * success or failure flag accordingly.
10763 if (unlikely(vmx->fail)) {
10765 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
10767 nested_vmx_succeed(vcpu);
10768 if (enable_shadow_vmcs)
10769 vmx->nested.sync_shadow_vmcs = true;
10771 /* in case we halted in L2 */
10772 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10776 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
10778 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
10780 if (is_guest_mode(vcpu))
10781 nested_vmx_vmexit(vcpu, -1, 0, 0);
10782 free_nested(to_vmx(vcpu));
10786 * L1's failure to enter L2 is a subset of a normal exit, as explained in
10787 * 23.7 "VM-entry failures during or after loading guest state" (this also
10788 * lists the acceptable exit-reason and exit-qualification parameters).
10789 * It should only be called before L2 actually succeeded to run, and when
10790 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
10792 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
10793 struct vmcs12 *vmcs12,
10794 u32 reason, unsigned long qualification)
10796 load_vmcs12_host_state(vcpu, vmcs12);
10797 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
10798 vmcs12->exit_qualification = qualification;
10799 nested_vmx_succeed(vcpu);
10800 if (enable_shadow_vmcs)
10801 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
10804 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
10805 struct x86_instruction_info *info,
10806 enum x86_intercept_stage stage)
10808 return X86EMUL_CONTINUE;
10811 #ifdef CONFIG_X86_64
10812 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
10813 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
10814 u64 divisor, u64 *result)
10816 u64 low = a << shift, high = a >> (64 - shift);
10818 /* To avoid the overflow on divq */
10819 if (high >= divisor)
10822 /* Low hold the result, high hold rem which is discarded */
10823 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
10824 "rm" (divisor), "0" (low), "1" (high));
10830 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
10832 struct vcpu_vmx *vmx = to_vmx(vcpu);
10833 u64 tscl = rdtsc();
10834 u64 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
10835 u64 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
10837 /* Convert to host delta tsc if tsc scaling is enabled */
10838 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
10839 u64_shl_div_u64(delta_tsc,
10840 kvm_tsc_scaling_ratio_frac_bits,
10841 vcpu->arch.tsc_scaling_ratio,
10846 * If the delta tsc can't fit in the 32 bit after the multi shift,
10847 * we can't use the preemption timer.
10848 * It's possible that it fits on later vmentries, but checking
10849 * on every vmentry is costly so we just use an hrtimer.
10851 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
10854 vmx->hv_deadline_tsc = tscl + delta_tsc;
10855 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
10856 PIN_BASED_VMX_PREEMPTION_TIMER);
10860 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
10862 struct vcpu_vmx *vmx = to_vmx(vcpu);
10863 vmx->hv_deadline_tsc = -1;
10864 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
10865 PIN_BASED_VMX_PREEMPTION_TIMER);
10869 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
10872 shrink_ple_window(vcpu);
10875 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
10876 struct kvm_memory_slot *slot)
10878 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
10879 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
10882 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
10883 struct kvm_memory_slot *slot)
10885 kvm_mmu_slot_set_dirty(kvm, slot);
10888 static void vmx_flush_log_dirty(struct kvm *kvm)
10890 kvm_flush_pml_buffers(kvm);
10893 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
10894 struct kvm_memory_slot *memslot,
10895 gfn_t offset, unsigned long mask)
10897 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
10901 * This routine does the following things for vCPU which is going
10902 * to be blocked if VT-d PI is enabled.
10903 * - Store the vCPU to the wakeup list, so when interrupts happen
10904 * we can find the right vCPU to wake up.
10905 * - Change the Posted-interrupt descriptor as below:
10906 * 'NDST' <-- vcpu->pre_pcpu
10907 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
10908 * - If 'ON' is set during this process, which means at least one
10909 * interrupt is posted for this vCPU, we cannot block it, in
10910 * this case, return 1, otherwise, return 0.
10913 static int pi_pre_block(struct kvm_vcpu *vcpu)
10915 unsigned long flags;
10917 struct pi_desc old, new;
10918 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10920 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10921 !irq_remapping_cap(IRQ_POSTING_CAP))
10924 vcpu->pre_pcpu = vcpu->cpu;
10925 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10926 vcpu->pre_pcpu), flags);
10927 list_add_tail(&vcpu->blocked_vcpu_list,
10928 &per_cpu(blocked_vcpu_on_cpu,
10930 spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
10931 vcpu->pre_pcpu), flags);
10934 old.control = new.control = pi_desc->control;
10937 * We should not block the vCPU if
10938 * an interrupt is posted for it.
10940 if (pi_test_on(pi_desc) == 1) {
10941 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10942 vcpu->pre_pcpu), flags);
10943 list_del(&vcpu->blocked_vcpu_list);
10944 spin_unlock_irqrestore(
10945 &per_cpu(blocked_vcpu_on_cpu_lock,
10946 vcpu->pre_pcpu), flags);
10947 vcpu->pre_pcpu = -1;
10952 WARN((pi_desc->sn == 1),
10953 "Warning: SN field of posted-interrupts "
10954 "is set before blocking\n");
10957 * Since vCPU can be preempted during this process,
10958 * vcpu->cpu could be different with pre_pcpu, we
10959 * need to set pre_pcpu as the destination of wakeup
10960 * notification event, then we can find the right vCPU
10961 * to wakeup in wakeup handler if interrupts happen
10962 * when the vCPU is in blocked state.
10964 dest = cpu_physical_id(vcpu->pre_pcpu);
10966 if (x2apic_enabled())
10969 new.ndst = (dest << 8) & 0xFF00;
10971 /* set 'NV' to 'wakeup vector' */
10972 new.nv = POSTED_INTR_WAKEUP_VECTOR;
10973 } while (cmpxchg(&pi_desc->control, old.control,
10974 new.control) != old.control);
10979 static int vmx_pre_block(struct kvm_vcpu *vcpu)
10981 if (pi_pre_block(vcpu))
10984 if (kvm_lapic_hv_timer_in_use(vcpu))
10985 kvm_lapic_switch_to_sw_timer(vcpu);
10990 static void pi_post_block(struct kvm_vcpu *vcpu)
10992 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10993 struct pi_desc old, new;
10995 unsigned long flags;
10997 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10998 !irq_remapping_cap(IRQ_POSTING_CAP))
11002 old.control = new.control = pi_desc->control;
11004 dest = cpu_physical_id(vcpu->cpu);
11006 if (x2apic_enabled())
11009 new.ndst = (dest << 8) & 0xFF00;
11011 /* Allow posting non-urgent interrupts */
11014 /* set 'NV' to 'notification vector' */
11015 new.nv = POSTED_INTR_VECTOR;
11016 } while (cmpxchg(&pi_desc->control, old.control,
11017 new.control) != old.control);
11019 if(vcpu->pre_pcpu != -1) {
11021 &per_cpu(blocked_vcpu_on_cpu_lock,
11022 vcpu->pre_pcpu), flags);
11023 list_del(&vcpu->blocked_vcpu_list);
11024 spin_unlock_irqrestore(
11025 &per_cpu(blocked_vcpu_on_cpu_lock,
11026 vcpu->pre_pcpu), flags);
11027 vcpu->pre_pcpu = -1;
11031 static void vmx_post_block(struct kvm_vcpu *vcpu)
11033 if (kvm_x86_ops->set_hv_timer)
11034 kvm_lapic_switch_to_hv_timer(vcpu);
11036 pi_post_block(vcpu);
11040 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
11043 * @host_irq: host irq of the interrupt
11044 * @guest_irq: gsi of the interrupt
11045 * @set: set or unset PI
11046 * returns 0 on success, < 0 on failure
11048 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
11049 uint32_t guest_irq, bool set)
11051 struct kvm_kernel_irq_routing_entry *e;
11052 struct kvm_irq_routing_table *irq_rt;
11053 struct kvm_lapic_irq irq;
11054 struct kvm_vcpu *vcpu;
11055 struct vcpu_data vcpu_info;
11056 int idx, ret = -EINVAL;
11058 if (!kvm_arch_has_assigned_device(kvm) ||
11059 !irq_remapping_cap(IRQ_POSTING_CAP))
11062 idx = srcu_read_lock(&kvm->irq_srcu);
11063 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
11064 BUG_ON(guest_irq >= irq_rt->nr_rt_entries);
11066 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
11067 if (e->type != KVM_IRQ_ROUTING_MSI)
11070 * VT-d PI cannot support posting multicast/broadcast
11071 * interrupts to a vCPU, we still use interrupt remapping
11072 * for these kind of interrupts.
11074 * For lowest-priority interrupts, we only support
11075 * those with single CPU as the destination, e.g. user
11076 * configures the interrupts via /proc/irq or uses
11077 * irqbalance to make the interrupts single-CPU.
11079 * We will support full lowest-priority interrupt later.
11082 kvm_set_msi_irq(e, &irq);
11083 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
11085 * Make sure the IRTE is in remapped mode if
11086 * we don't handle it in posted mode.
11088 ret = irq_set_vcpu_affinity(host_irq, NULL);
11091 "failed to back to remapped mode, irq: %u\n",
11099 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
11100 vcpu_info.vector = irq.vector;
11102 trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
11103 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
11106 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
11108 /* suppress notification event before unposting */
11109 pi_set_sn(vcpu_to_pi_desc(vcpu));
11110 ret = irq_set_vcpu_affinity(host_irq, NULL);
11111 pi_clear_sn(vcpu_to_pi_desc(vcpu));
11115 printk(KERN_INFO "%s: failed to update PI IRTE\n",
11123 srcu_read_unlock(&kvm->irq_srcu, idx);
11127 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
11129 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
11130 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
11131 FEATURE_CONTROL_LMCE;
11133 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
11134 ~FEATURE_CONTROL_LMCE;
11137 static struct kvm_x86_ops vmx_x86_ops = {
11138 .cpu_has_kvm_support = cpu_has_kvm_support,
11139 .disabled_by_bios = vmx_disabled_by_bios,
11140 .hardware_setup = hardware_setup,
11141 .hardware_unsetup = hardware_unsetup,
11142 .check_processor_compatibility = vmx_check_processor_compat,
11143 .hardware_enable = hardware_enable,
11144 .hardware_disable = hardware_disable,
11145 .cpu_has_accelerated_tpr = report_flexpriority,
11146 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
11148 .vcpu_create = vmx_create_vcpu,
11149 .vcpu_free = vmx_free_vcpu,
11150 .vcpu_reset = vmx_vcpu_reset,
11152 .prepare_guest_switch = vmx_save_host_state,
11153 .vcpu_load = vmx_vcpu_load,
11154 .vcpu_put = vmx_vcpu_put,
11156 .update_bp_intercept = update_exception_bitmap,
11157 .get_msr = vmx_get_msr,
11158 .set_msr = vmx_set_msr,
11159 .get_segment_base = vmx_get_segment_base,
11160 .get_segment = vmx_get_segment,
11161 .set_segment = vmx_set_segment,
11162 .get_cpl = vmx_get_cpl,
11163 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
11164 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
11165 .decache_cr3 = vmx_decache_cr3,
11166 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
11167 .set_cr0 = vmx_set_cr0,
11168 .set_cr3 = vmx_set_cr3,
11169 .set_cr4 = vmx_set_cr4,
11170 .set_efer = vmx_set_efer,
11171 .get_idt = vmx_get_idt,
11172 .set_idt = vmx_set_idt,
11173 .get_gdt = vmx_get_gdt,
11174 .set_gdt = vmx_set_gdt,
11175 .get_dr6 = vmx_get_dr6,
11176 .set_dr6 = vmx_set_dr6,
11177 .set_dr7 = vmx_set_dr7,
11178 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
11179 .cache_reg = vmx_cache_reg,
11180 .get_rflags = vmx_get_rflags,
11181 .set_rflags = vmx_set_rflags,
11183 .get_pkru = vmx_get_pkru,
11185 .fpu_activate = vmx_fpu_activate,
11186 .fpu_deactivate = vmx_fpu_deactivate,
11188 .tlb_flush = vmx_flush_tlb,
11190 .run = vmx_vcpu_run,
11191 .handle_exit = vmx_handle_exit,
11192 .skip_emulated_instruction = skip_emulated_instruction,
11193 .set_interrupt_shadow = vmx_set_interrupt_shadow,
11194 .get_interrupt_shadow = vmx_get_interrupt_shadow,
11195 .patch_hypercall = vmx_patch_hypercall,
11196 .set_irq = vmx_inject_irq,
11197 .set_nmi = vmx_inject_nmi,
11198 .queue_exception = vmx_queue_exception,
11199 .cancel_injection = vmx_cancel_injection,
11200 .interrupt_allowed = vmx_interrupt_allowed,
11201 .nmi_allowed = vmx_nmi_allowed,
11202 .get_nmi_mask = vmx_get_nmi_mask,
11203 .set_nmi_mask = vmx_set_nmi_mask,
11204 .enable_nmi_window = enable_nmi_window,
11205 .enable_irq_window = enable_irq_window,
11206 .update_cr8_intercept = update_cr8_intercept,
11207 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
11208 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
11209 .get_enable_apicv = vmx_get_enable_apicv,
11210 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
11211 .load_eoi_exitmap = vmx_load_eoi_exitmap,
11212 .hwapic_irr_update = vmx_hwapic_irr_update,
11213 .hwapic_isr_update = vmx_hwapic_isr_update,
11214 .sync_pir_to_irr = vmx_sync_pir_to_irr,
11215 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
11217 .set_tss_addr = vmx_set_tss_addr,
11218 .get_tdp_level = get_ept_level,
11219 .get_mt_mask = vmx_get_mt_mask,
11221 .get_exit_info = vmx_get_exit_info,
11223 .get_lpage_level = vmx_get_lpage_level,
11225 .cpuid_update = vmx_cpuid_update,
11227 .rdtscp_supported = vmx_rdtscp_supported,
11228 .invpcid_supported = vmx_invpcid_supported,
11230 .set_supported_cpuid = vmx_set_supported_cpuid,
11232 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
11234 .read_tsc_offset = vmx_read_tsc_offset,
11235 .write_tsc_offset = vmx_write_tsc_offset,
11236 .adjust_tsc_offset_guest = vmx_adjust_tsc_offset_guest,
11237 .read_l1_tsc = vmx_read_l1_tsc,
11239 .set_tdp_cr3 = vmx_set_cr3,
11241 .check_intercept = vmx_check_intercept,
11242 .handle_external_intr = vmx_handle_external_intr,
11243 .mpx_supported = vmx_mpx_supported,
11244 .xsaves_supported = vmx_xsaves_supported,
11246 .check_nested_events = vmx_check_nested_events,
11248 .sched_in = vmx_sched_in,
11250 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
11251 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
11252 .flush_log_dirty = vmx_flush_log_dirty,
11253 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
11255 .pre_block = vmx_pre_block,
11256 .post_block = vmx_post_block,
11258 .pmu_ops = &intel_pmu_ops,
11260 .update_pi_irte = vmx_update_pi_irte,
11262 #ifdef CONFIG_X86_64
11263 .set_hv_timer = vmx_set_hv_timer,
11264 .cancel_hv_timer = vmx_cancel_hv_timer,
11267 .setup_mce = vmx_setup_mce,
11270 static int __init vmx_init(void)
11272 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
11273 __alignof__(struct vcpu_vmx), THIS_MODULE);
11277 #ifdef CONFIG_KEXEC_CORE
11278 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
11279 crash_vmclear_local_loaded_vmcss);
11285 static void __exit vmx_exit(void)
11287 #ifdef CONFIG_KEXEC_CORE
11288 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
11295 module_init(vmx_init)
11296 module_exit(vmx_exit)