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 <linux/frame.h>
37 #include "kvm_cache_regs.h"
44 #include <asm/virtext.h>
46 #include <asm/fpu/internal.h>
47 #include <asm/perf_event.h>
48 #include <asm/debugreg.h>
49 #include <asm/kexec.h>
51 #include <asm/irq_remapping.h>
52 #include <asm/mmu_context.h>
57 #define __ex(x) __kvm_handle_fault_on_reboot(x)
58 #define __ex_clear(x, reg) \
59 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
61 MODULE_AUTHOR("Qumranet");
62 MODULE_LICENSE("GPL");
64 static const struct x86_cpu_id vmx_cpu_id[] = {
65 X86_FEATURE_MATCH(X86_FEATURE_VMX),
68 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
70 static bool __read_mostly enable_vpid = 1;
71 module_param_named(vpid, enable_vpid, bool, 0444);
73 static bool __read_mostly flexpriority_enabled = 1;
74 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
76 static bool __read_mostly enable_ept = 1;
77 module_param_named(ept, enable_ept, bool, S_IRUGO);
79 static bool __read_mostly enable_unrestricted_guest = 1;
80 module_param_named(unrestricted_guest,
81 enable_unrestricted_guest, bool, S_IRUGO);
83 static bool __read_mostly enable_ept_ad_bits = 1;
84 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
86 static bool __read_mostly emulate_invalid_guest_state = true;
87 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
89 static bool __read_mostly fasteoi = 1;
90 module_param(fasteoi, bool, S_IRUGO);
92 static bool __read_mostly enable_apicv = 1;
93 module_param(enable_apicv, bool, S_IRUGO);
95 static bool __read_mostly enable_shadow_vmcs = 1;
96 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
98 * If nested=1, nested virtualization is supported, i.e., guests may use
99 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
100 * use VMX instructions.
102 static bool __read_mostly nested = 0;
103 module_param(nested, bool, S_IRUGO);
105 static u64 __read_mostly host_xss;
107 static bool __read_mostly enable_pml = 1;
108 module_param_named(pml, enable_pml, bool, S_IRUGO);
110 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
112 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
113 static int __read_mostly cpu_preemption_timer_multi;
114 static bool __read_mostly enable_preemption_timer = 1;
116 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
119 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
120 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
121 #define KVM_VM_CR0_ALWAYS_ON \
122 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
123 #define KVM_CR4_GUEST_OWNED_BITS \
124 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
125 | X86_CR4_OSXMMEXCPT | X86_CR4_TSD)
127 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
128 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
130 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
132 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
135 * Hyper-V requires all of these, so mark them as supported even though
136 * they are just treated the same as all-context.
138 #define VMX_VPID_EXTENT_SUPPORTED_MASK \
139 (VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT | \
140 VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT | \
141 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT | \
142 VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
145 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
146 * ple_gap: upper bound on the amount of time between two successive
147 * executions of PAUSE in a loop. Also indicate if ple enabled.
148 * According to test, this time is usually smaller than 128 cycles.
149 * ple_window: upper bound on the amount of time a guest is allowed to execute
150 * in a PAUSE loop. Tests indicate that most spinlocks are held for
151 * less than 2^12 cycles
152 * Time is measured based on a counter that runs at the same rate as the TSC,
153 * refer SDM volume 3b section 21.6.13 & 22.1.3.
155 #define KVM_VMX_DEFAULT_PLE_GAP 128
156 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
157 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
158 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
159 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
160 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
162 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
163 module_param(ple_gap, int, S_IRUGO);
165 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
166 module_param(ple_window, int, S_IRUGO);
168 /* Default doubles per-vcpu window every exit. */
169 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
170 module_param(ple_window_grow, int, S_IRUGO);
172 /* Default resets per-vcpu window every exit to ple_window. */
173 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
174 module_param(ple_window_shrink, int, S_IRUGO);
176 /* Default is to compute the maximum so we can never overflow. */
177 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
178 static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
179 module_param(ple_window_max, int, S_IRUGO);
181 extern const ulong vmx_return;
183 #define NR_AUTOLOAD_MSRS 8
184 #define VMCS02_POOL_SIZE 1
193 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
194 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
195 * loaded on this CPU (so we can clear them if the CPU goes down).
199 struct vmcs *shadow_vmcs;
202 bool nmi_known_unmasked;
203 struct list_head loaded_vmcss_on_cpu_link;
206 struct shared_msr_entry {
213 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
214 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
215 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
216 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
217 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
218 * More than one of these structures may exist, if L1 runs multiple L2 guests.
219 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
220 * underlying hardware which will be used to run L2.
221 * This structure is packed to ensure that its layout is identical across
222 * machines (necessary for live migration).
223 * If there are changes in this struct, VMCS12_REVISION must be changed.
225 typedef u64 natural_width;
226 struct __packed vmcs12 {
227 /* According to the Intel spec, a VMCS region must start with the
228 * following two fields. Then follow implementation-specific data.
233 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
234 u32 padding[7]; /* room for future expansion */
239 u64 vm_exit_msr_store_addr;
240 u64 vm_exit_msr_load_addr;
241 u64 vm_entry_msr_load_addr;
243 u64 virtual_apic_page_addr;
244 u64 apic_access_addr;
245 u64 posted_intr_desc_addr;
247 u64 eoi_exit_bitmap0;
248 u64 eoi_exit_bitmap1;
249 u64 eoi_exit_bitmap2;
250 u64 eoi_exit_bitmap3;
252 u64 guest_physical_address;
253 u64 vmcs_link_pointer;
255 u64 guest_ia32_debugctl;
258 u64 guest_ia32_perf_global_ctrl;
266 u64 host_ia32_perf_global_ctrl;
267 u64 padding64[8]; /* room for future expansion */
269 * To allow migration of L1 (complete with its L2 guests) between
270 * machines of different natural widths (32 or 64 bit), we cannot have
271 * unsigned long fields with no explict size. We use u64 (aliased
272 * natural_width) instead. Luckily, x86 is little-endian.
274 natural_width cr0_guest_host_mask;
275 natural_width cr4_guest_host_mask;
276 natural_width cr0_read_shadow;
277 natural_width cr4_read_shadow;
278 natural_width cr3_target_value0;
279 natural_width cr3_target_value1;
280 natural_width cr3_target_value2;
281 natural_width cr3_target_value3;
282 natural_width exit_qualification;
283 natural_width guest_linear_address;
284 natural_width guest_cr0;
285 natural_width guest_cr3;
286 natural_width guest_cr4;
287 natural_width guest_es_base;
288 natural_width guest_cs_base;
289 natural_width guest_ss_base;
290 natural_width guest_ds_base;
291 natural_width guest_fs_base;
292 natural_width guest_gs_base;
293 natural_width guest_ldtr_base;
294 natural_width guest_tr_base;
295 natural_width guest_gdtr_base;
296 natural_width guest_idtr_base;
297 natural_width guest_dr7;
298 natural_width guest_rsp;
299 natural_width guest_rip;
300 natural_width guest_rflags;
301 natural_width guest_pending_dbg_exceptions;
302 natural_width guest_sysenter_esp;
303 natural_width guest_sysenter_eip;
304 natural_width host_cr0;
305 natural_width host_cr3;
306 natural_width host_cr4;
307 natural_width host_fs_base;
308 natural_width host_gs_base;
309 natural_width host_tr_base;
310 natural_width host_gdtr_base;
311 natural_width host_idtr_base;
312 natural_width host_ia32_sysenter_esp;
313 natural_width host_ia32_sysenter_eip;
314 natural_width host_rsp;
315 natural_width host_rip;
316 natural_width paddingl[8]; /* room for future expansion */
317 u32 pin_based_vm_exec_control;
318 u32 cpu_based_vm_exec_control;
319 u32 exception_bitmap;
320 u32 page_fault_error_code_mask;
321 u32 page_fault_error_code_match;
322 u32 cr3_target_count;
323 u32 vm_exit_controls;
324 u32 vm_exit_msr_store_count;
325 u32 vm_exit_msr_load_count;
326 u32 vm_entry_controls;
327 u32 vm_entry_msr_load_count;
328 u32 vm_entry_intr_info_field;
329 u32 vm_entry_exception_error_code;
330 u32 vm_entry_instruction_len;
332 u32 secondary_vm_exec_control;
333 u32 vm_instruction_error;
335 u32 vm_exit_intr_info;
336 u32 vm_exit_intr_error_code;
337 u32 idt_vectoring_info_field;
338 u32 idt_vectoring_error_code;
339 u32 vm_exit_instruction_len;
340 u32 vmx_instruction_info;
347 u32 guest_ldtr_limit;
349 u32 guest_gdtr_limit;
350 u32 guest_idtr_limit;
351 u32 guest_es_ar_bytes;
352 u32 guest_cs_ar_bytes;
353 u32 guest_ss_ar_bytes;
354 u32 guest_ds_ar_bytes;
355 u32 guest_fs_ar_bytes;
356 u32 guest_gs_ar_bytes;
357 u32 guest_ldtr_ar_bytes;
358 u32 guest_tr_ar_bytes;
359 u32 guest_interruptibility_info;
360 u32 guest_activity_state;
361 u32 guest_sysenter_cs;
362 u32 host_ia32_sysenter_cs;
363 u32 vmx_preemption_timer_value;
364 u32 padding32[7]; /* room for future expansion */
365 u16 virtual_processor_id;
367 u16 guest_es_selector;
368 u16 guest_cs_selector;
369 u16 guest_ss_selector;
370 u16 guest_ds_selector;
371 u16 guest_fs_selector;
372 u16 guest_gs_selector;
373 u16 guest_ldtr_selector;
374 u16 guest_tr_selector;
375 u16 guest_intr_status;
377 u16 host_es_selector;
378 u16 host_cs_selector;
379 u16 host_ss_selector;
380 u16 host_ds_selector;
381 u16 host_fs_selector;
382 u16 host_gs_selector;
383 u16 host_tr_selector;
387 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
388 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
389 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
391 #define VMCS12_REVISION 0x11e57ed0
394 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
395 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
396 * current implementation, 4K are reserved to avoid future complications.
398 #define VMCS12_SIZE 0x1000
400 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
402 struct list_head list;
404 struct loaded_vmcs vmcs02;
408 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
409 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
412 /* Has the level1 guest done vmxon? */
417 /* The guest-physical address of the current VMCS L1 keeps for L2 */
420 * Cache of the guest's VMCS, existing outside of guest memory.
421 * Loaded from guest memory during VMPTRLD. Flushed to guest
422 * memory during VMCLEAR and VMPTRLD.
424 struct vmcs12 *cached_vmcs12;
426 * Indicates if the shadow vmcs must be updated with the
427 * data hold by vmcs12
429 bool sync_shadow_vmcs;
431 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
432 struct list_head vmcs02_pool;
434 bool change_vmcs01_virtual_x2apic_mode;
435 /* L2 must run next, and mustn't decide to exit to L1. */
436 bool nested_run_pending;
438 * Guest pages referred to in vmcs02 with host-physical pointers, so
439 * we must keep them pinned while L2 runs.
441 struct page *apic_access_page;
442 struct page *virtual_apic_page;
443 struct page *pi_desc_page;
444 struct pi_desc *pi_desc;
448 unsigned long *msr_bitmap;
450 struct hrtimer preemption_timer;
451 bool preemption_timer_expired;
453 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
460 * We only store the "true" versions of the VMX capability MSRs. We
461 * generate the "non-true" versions by setting the must-be-1 bits
462 * according to the SDM.
464 u32 nested_vmx_procbased_ctls_low;
465 u32 nested_vmx_procbased_ctls_high;
466 u32 nested_vmx_secondary_ctls_low;
467 u32 nested_vmx_secondary_ctls_high;
468 u32 nested_vmx_pinbased_ctls_low;
469 u32 nested_vmx_pinbased_ctls_high;
470 u32 nested_vmx_exit_ctls_low;
471 u32 nested_vmx_exit_ctls_high;
472 u32 nested_vmx_entry_ctls_low;
473 u32 nested_vmx_entry_ctls_high;
474 u32 nested_vmx_misc_low;
475 u32 nested_vmx_misc_high;
476 u32 nested_vmx_ept_caps;
477 u32 nested_vmx_vpid_caps;
478 u64 nested_vmx_basic;
479 u64 nested_vmx_cr0_fixed0;
480 u64 nested_vmx_cr0_fixed1;
481 u64 nested_vmx_cr4_fixed0;
482 u64 nested_vmx_cr4_fixed1;
483 u64 nested_vmx_vmcs_enum;
486 #define POSTED_INTR_ON 0
487 #define POSTED_INTR_SN 1
489 /* Posted-Interrupt Descriptor */
491 u32 pir[8]; /* Posted interrupt requested */
494 /* bit 256 - Outstanding Notification */
496 /* bit 257 - Suppress Notification */
498 /* bit 271:258 - Reserved */
500 /* bit 279:272 - Notification Vector */
502 /* bit 287:280 - Reserved */
504 /* bit 319:288 - Notification Destination */
512 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
514 return test_and_set_bit(POSTED_INTR_ON,
515 (unsigned long *)&pi_desc->control);
518 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
520 return test_and_clear_bit(POSTED_INTR_ON,
521 (unsigned long *)&pi_desc->control);
524 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
526 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
529 static inline void pi_clear_sn(struct pi_desc *pi_desc)
531 return clear_bit(POSTED_INTR_SN,
532 (unsigned long *)&pi_desc->control);
535 static inline void pi_set_sn(struct pi_desc *pi_desc)
537 return set_bit(POSTED_INTR_SN,
538 (unsigned long *)&pi_desc->control);
541 static inline void pi_clear_on(struct pi_desc *pi_desc)
543 clear_bit(POSTED_INTR_ON,
544 (unsigned long *)&pi_desc->control);
547 static inline int pi_test_on(struct pi_desc *pi_desc)
549 return test_bit(POSTED_INTR_ON,
550 (unsigned long *)&pi_desc->control);
553 static inline int pi_test_sn(struct pi_desc *pi_desc)
555 return test_bit(POSTED_INTR_SN,
556 (unsigned long *)&pi_desc->control);
560 struct kvm_vcpu vcpu;
561 unsigned long host_rsp;
564 u32 idt_vectoring_info;
566 struct shared_msr_entry *guest_msrs;
569 unsigned long host_idt_base;
571 u64 msr_host_kernel_gs_base;
572 u64 msr_guest_kernel_gs_base;
574 u32 vm_entry_controls_shadow;
575 u32 vm_exit_controls_shadow;
577 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
578 * non-nested (L1) guest, it always points to vmcs01. For a nested
579 * guest (L2), it points to a different VMCS.
581 struct loaded_vmcs vmcs01;
582 struct loaded_vmcs *loaded_vmcs;
583 bool __launched; /* temporary, used in vmx_vcpu_run */
584 struct msr_autoload {
586 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
587 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
591 u16 fs_sel, gs_sel, ldt_sel;
595 int gs_ldt_reload_needed;
596 int fs_reload_needed;
597 u64 msr_host_bndcfgs;
598 unsigned long vmcs_host_cr3; /* May not match real cr3 */
599 unsigned long vmcs_host_cr4; /* May not match real cr4 */
604 struct kvm_segment segs[8];
607 u32 bitmask; /* 4 bits per segment (1 bit per field) */
608 struct kvm_save_segment {
616 bool emulation_required;
620 /* Posted interrupt descriptor */
621 struct pi_desc pi_desc;
623 /* Support for a guest hypervisor (nested VMX) */
624 struct nested_vmx nested;
626 /* Dynamic PLE window. */
628 bool ple_window_dirty;
630 /* Support for PML */
631 #define PML_ENTITY_NUM 512
634 /* apic deadline value in host tsc */
637 u64 current_tsc_ratio;
639 bool guest_pkru_valid;
644 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
645 * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
646 * in msr_ia32_feature_control_valid_bits.
648 u64 msr_ia32_feature_control;
649 u64 msr_ia32_feature_control_valid_bits;
652 enum segment_cache_field {
661 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
663 return container_of(vcpu, struct vcpu_vmx, vcpu);
666 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
668 return &(to_vmx(vcpu)->pi_desc);
671 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
672 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
673 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
674 [number##_HIGH] = VMCS12_OFFSET(name)+4
677 static unsigned long shadow_read_only_fields[] = {
679 * We do NOT shadow fields that are modified when L0
680 * traps and emulates any vmx instruction (e.g. VMPTRLD,
681 * VMXON...) executed by L1.
682 * For example, VM_INSTRUCTION_ERROR is read
683 * by L1 if a vmx instruction fails (part of the error path).
684 * Note the code assumes this logic. If for some reason
685 * we start shadowing these fields then we need to
686 * force a shadow sync when L0 emulates vmx instructions
687 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
688 * by nested_vmx_failValid)
692 VM_EXIT_INSTRUCTION_LEN,
693 IDT_VECTORING_INFO_FIELD,
694 IDT_VECTORING_ERROR_CODE,
695 VM_EXIT_INTR_ERROR_CODE,
697 GUEST_LINEAR_ADDRESS,
698 GUEST_PHYSICAL_ADDRESS
700 static int max_shadow_read_only_fields =
701 ARRAY_SIZE(shadow_read_only_fields);
703 static unsigned long shadow_read_write_fields[] = {
710 GUEST_INTERRUPTIBILITY_INFO,
723 CPU_BASED_VM_EXEC_CONTROL,
724 VM_ENTRY_EXCEPTION_ERROR_CODE,
725 VM_ENTRY_INTR_INFO_FIELD,
726 VM_ENTRY_INSTRUCTION_LEN,
727 VM_ENTRY_EXCEPTION_ERROR_CODE,
733 static int max_shadow_read_write_fields =
734 ARRAY_SIZE(shadow_read_write_fields);
736 static const unsigned short vmcs_field_to_offset_table[] = {
737 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
738 FIELD(POSTED_INTR_NV, posted_intr_nv),
739 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
740 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
741 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
742 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
743 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
744 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
745 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
746 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
747 FIELD(GUEST_INTR_STATUS, guest_intr_status),
748 FIELD(GUEST_PML_INDEX, guest_pml_index),
749 FIELD(HOST_ES_SELECTOR, host_es_selector),
750 FIELD(HOST_CS_SELECTOR, host_cs_selector),
751 FIELD(HOST_SS_SELECTOR, host_ss_selector),
752 FIELD(HOST_DS_SELECTOR, host_ds_selector),
753 FIELD(HOST_FS_SELECTOR, host_fs_selector),
754 FIELD(HOST_GS_SELECTOR, host_gs_selector),
755 FIELD(HOST_TR_SELECTOR, host_tr_selector),
756 FIELD64(IO_BITMAP_A, io_bitmap_a),
757 FIELD64(IO_BITMAP_B, io_bitmap_b),
758 FIELD64(MSR_BITMAP, msr_bitmap),
759 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
760 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
761 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
762 FIELD64(TSC_OFFSET, tsc_offset),
763 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
764 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
765 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
766 FIELD64(EPT_POINTER, ept_pointer),
767 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
768 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
769 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
770 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
771 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
772 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
773 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
774 FIELD64(PML_ADDRESS, pml_address),
775 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
776 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
777 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
778 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
779 FIELD64(GUEST_PDPTR0, guest_pdptr0),
780 FIELD64(GUEST_PDPTR1, guest_pdptr1),
781 FIELD64(GUEST_PDPTR2, guest_pdptr2),
782 FIELD64(GUEST_PDPTR3, guest_pdptr3),
783 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
784 FIELD64(HOST_IA32_PAT, host_ia32_pat),
785 FIELD64(HOST_IA32_EFER, host_ia32_efer),
786 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
787 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
788 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
789 FIELD(EXCEPTION_BITMAP, exception_bitmap),
790 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
791 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
792 FIELD(CR3_TARGET_COUNT, cr3_target_count),
793 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
794 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
795 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
796 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
797 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
798 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
799 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
800 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
801 FIELD(TPR_THRESHOLD, tpr_threshold),
802 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
803 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
804 FIELD(VM_EXIT_REASON, vm_exit_reason),
805 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
806 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
807 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
808 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
809 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
810 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
811 FIELD(GUEST_ES_LIMIT, guest_es_limit),
812 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
813 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
814 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
815 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
816 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
817 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
818 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
819 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
820 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
821 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
822 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
823 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
824 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
825 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
826 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
827 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
828 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
829 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
830 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
831 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
832 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
833 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
834 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
835 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
836 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
837 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
838 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
839 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
840 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
841 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
842 FIELD(EXIT_QUALIFICATION, exit_qualification),
843 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
844 FIELD(GUEST_CR0, guest_cr0),
845 FIELD(GUEST_CR3, guest_cr3),
846 FIELD(GUEST_CR4, guest_cr4),
847 FIELD(GUEST_ES_BASE, guest_es_base),
848 FIELD(GUEST_CS_BASE, guest_cs_base),
849 FIELD(GUEST_SS_BASE, guest_ss_base),
850 FIELD(GUEST_DS_BASE, guest_ds_base),
851 FIELD(GUEST_FS_BASE, guest_fs_base),
852 FIELD(GUEST_GS_BASE, guest_gs_base),
853 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
854 FIELD(GUEST_TR_BASE, guest_tr_base),
855 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
856 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
857 FIELD(GUEST_DR7, guest_dr7),
858 FIELD(GUEST_RSP, guest_rsp),
859 FIELD(GUEST_RIP, guest_rip),
860 FIELD(GUEST_RFLAGS, guest_rflags),
861 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
862 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
863 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
864 FIELD(HOST_CR0, host_cr0),
865 FIELD(HOST_CR3, host_cr3),
866 FIELD(HOST_CR4, host_cr4),
867 FIELD(HOST_FS_BASE, host_fs_base),
868 FIELD(HOST_GS_BASE, host_gs_base),
869 FIELD(HOST_TR_BASE, host_tr_base),
870 FIELD(HOST_GDTR_BASE, host_gdtr_base),
871 FIELD(HOST_IDTR_BASE, host_idtr_base),
872 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
873 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
874 FIELD(HOST_RSP, host_rsp),
875 FIELD(HOST_RIP, host_rip),
878 static inline short vmcs_field_to_offset(unsigned long field)
880 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
882 if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
883 vmcs_field_to_offset_table[field] == 0)
886 return vmcs_field_to_offset_table[field];
889 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
891 return to_vmx(vcpu)->nested.cached_vmcs12;
894 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
896 struct page *page = kvm_vcpu_gfn_to_page(vcpu, addr >> PAGE_SHIFT);
897 if (is_error_page(page))
903 static void nested_release_page(struct page *page)
905 kvm_release_page_dirty(page);
908 static void nested_release_page_clean(struct page *page)
910 kvm_release_page_clean(page);
913 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu);
914 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
915 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
916 static bool vmx_xsaves_supported(void);
917 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
918 static void vmx_set_segment(struct kvm_vcpu *vcpu,
919 struct kvm_segment *var, int seg);
920 static void vmx_get_segment(struct kvm_vcpu *vcpu,
921 struct kvm_segment *var, int seg);
922 static bool guest_state_valid(struct kvm_vcpu *vcpu);
923 static u32 vmx_segment_access_rights(struct kvm_segment *var);
924 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
925 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
926 static int alloc_identity_pagetable(struct kvm *kvm);
927 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
928 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
929 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
932 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
933 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
935 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
936 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
938 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
941 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
942 * can find which vCPU should be waken up.
944 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
945 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
950 VMX_MSR_BITMAP_LEGACY,
951 VMX_MSR_BITMAP_LONGMODE,
952 VMX_MSR_BITMAP_LEGACY_X2APIC_APICV,
953 VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV,
954 VMX_MSR_BITMAP_LEGACY_X2APIC,
955 VMX_MSR_BITMAP_LONGMODE_X2APIC,
961 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
963 #define vmx_io_bitmap_a (vmx_bitmap[VMX_IO_BITMAP_A])
964 #define vmx_io_bitmap_b (vmx_bitmap[VMX_IO_BITMAP_B])
965 #define vmx_msr_bitmap_legacy (vmx_bitmap[VMX_MSR_BITMAP_LEGACY])
966 #define vmx_msr_bitmap_longmode (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE])
967 #define vmx_msr_bitmap_legacy_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC_APICV])
968 #define vmx_msr_bitmap_longmode_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV])
969 #define vmx_msr_bitmap_legacy_x2apic (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC])
970 #define vmx_msr_bitmap_longmode_x2apic (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC])
971 #define vmx_vmread_bitmap (vmx_bitmap[VMX_VMREAD_BITMAP])
972 #define vmx_vmwrite_bitmap (vmx_bitmap[VMX_VMWRITE_BITMAP])
974 static bool cpu_has_load_ia32_efer;
975 static bool cpu_has_load_perf_global_ctrl;
977 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
978 static DEFINE_SPINLOCK(vmx_vpid_lock);
980 static struct vmcs_config {
985 u32 pin_based_exec_ctrl;
986 u32 cpu_based_exec_ctrl;
987 u32 cpu_based_2nd_exec_ctrl;
992 static struct vmx_capability {
997 #define VMX_SEGMENT_FIELD(seg) \
998 [VCPU_SREG_##seg] = { \
999 .selector = GUEST_##seg##_SELECTOR, \
1000 .base = GUEST_##seg##_BASE, \
1001 .limit = GUEST_##seg##_LIMIT, \
1002 .ar_bytes = GUEST_##seg##_AR_BYTES, \
1005 static const struct kvm_vmx_segment_field {
1010 } kvm_vmx_segment_fields[] = {
1011 VMX_SEGMENT_FIELD(CS),
1012 VMX_SEGMENT_FIELD(DS),
1013 VMX_SEGMENT_FIELD(ES),
1014 VMX_SEGMENT_FIELD(FS),
1015 VMX_SEGMENT_FIELD(GS),
1016 VMX_SEGMENT_FIELD(SS),
1017 VMX_SEGMENT_FIELD(TR),
1018 VMX_SEGMENT_FIELD(LDTR),
1021 static u64 host_efer;
1023 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
1026 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
1027 * away by decrementing the array size.
1029 static const u32 vmx_msr_index[] = {
1030 #ifdef CONFIG_X86_64
1031 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
1033 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
1036 static inline bool is_exception_n(u32 intr_info, u8 vector)
1038 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1039 INTR_INFO_VALID_MASK)) ==
1040 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
1043 static inline bool is_debug(u32 intr_info)
1045 return is_exception_n(intr_info, DB_VECTOR);
1048 static inline bool is_breakpoint(u32 intr_info)
1050 return is_exception_n(intr_info, BP_VECTOR);
1053 static inline bool is_page_fault(u32 intr_info)
1055 return is_exception_n(intr_info, PF_VECTOR);
1058 static inline bool is_no_device(u32 intr_info)
1060 return is_exception_n(intr_info, NM_VECTOR);
1063 static inline bool is_invalid_opcode(u32 intr_info)
1065 return is_exception_n(intr_info, UD_VECTOR);
1068 static inline bool is_external_interrupt(u32 intr_info)
1070 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1071 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1074 static inline bool is_machine_check(u32 intr_info)
1076 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1077 INTR_INFO_VALID_MASK)) ==
1078 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1081 static inline bool cpu_has_vmx_msr_bitmap(void)
1083 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1086 static inline bool cpu_has_vmx_tpr_shadow(void)
1088 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1091 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1093 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1096 static inline bool cpu_has_secondary_exec_ctrls(void)
1098 return vmcs_config.cpu_based_exec_ctrl &
1099 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1102 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1104 return vmcs_config.cpu_based_2nd_exec_ctrl &
1105 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1108 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1110 return vmcs_config.cpu_based_2nd_exec_ctrl &
1111 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1114 static inline bool cpu_has_vmx_apic_register_virt(void)
1116 return vmcs_config.cpu_based_2nd_exec_ctrl &
1117 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1120 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1122 return vmcs_config.cpu_based_2nd_exec_ctrl &
1123 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1127 * Comment's format: document - errata name - stepping - processor name.
1129 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1131 static u32 vmx_preemption_cpu_tfms[] = {
1132 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
1134 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
1135 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1136 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1138 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1140 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
1141 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
1143 * 320767.pdf - AAP86 - B1 -
1144 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1147 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1149 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1151 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1153 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1154 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1155 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1159 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1161 u32 eax = cpuid_eax(0x00000001), i;
1163 /* Clear the reserved bits */
1164 eax &= ~(0x3U << 14 | 0xfU << 28);
1165 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
1166 if (eax == vmx_preemption_cpu_tfms[i])
1172 static inline bool cpu_has_vmx_preemption_timer(void)
1174 return vmcs_config.pin_based_exec_ctrl &
1175 PIN_BASED_VMX_PREEMPTION_TIMER;
1178 static inline bool cpu_has_vmx_posted_intr(void)
1180 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1181 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1184 static inline bool cpu_has_vmx_apicv(void)
1186 return cpu_has_vmx_apic_register_virt() &&
1187 cpu_has_vmx_virtual_intr_delivery() &&
1188 cpu_has_vmx_posted_intr();
1191 static inline bool cpu_has_vmx_flexpriority(void)
1193 return cpu_has_vmx_tpr_shadow() &&
1194 cpu_has_vmx_virtualize_apic_accesses();
1197 static inline bool cpu_has_vmx_ept_execute_only(void)
1199 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1202 static inline bool cpu_has_vmx_ept_2m_page(void)
1204 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1207 static inline bool cpu_has_vmx_ept_1g_page(void)
1209 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1212 static inline bool cpu_has_vmx_ept_4levels(void)
1214 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1217 static inline bool cpu_has_vmx_ept_ad_bits(void)
1219 return vmx_capability.ept & VMX_EPT_AD_BIT;
1222 static inline bool cpu_has_vmx_invept_context(void)
1224 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1227 static inline bool cpu_has_vmx_invept_global(void)
1229 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1232 static inline bool cpu_has_vmx_invvpid_single(void)
1234 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1237 static inline bool cpu_has_vmx_invvpid_global(void)
1239 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1242 static inline bool cpu_has_vmx_invvpid(void)
1244 return vmx_capability.vpid & VMX_VPID_INVVPID_BIT;
1247 static inline bool cpu_has_vmx_ept(void)
1249 return vmcs_config.cpu_based_2nd_exec_ctrl &
1250 SECONDARY_EXEC_ENABLE_EPT;
1253 static inline bool cpu_has_vmx_unrestricted_guest(void)
1255 return vmcs_config.cpu_based_2nd_exec_ctrl &
1256 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1259 static inline bool cpu_has_vmx_ple(void)
1261 return vmcs_config.cpu_based_2nd_exec_ctrl &
1262 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1265 static inline bool cpu_has_vmx_basic_inout(void)
1267 return (((u64)vmcs_config.basic_cap << 32) & VMX_BASIC_INOUT);
1270 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1272 return flexpriority_enabled && lapic_in_kernel(vcpu);
1275 static inline bool cpu_has_vmx_vpid(void)
1277 return vmcs_config.cpu_based_2nd_exec_ctrl &
1278 SECONDARY_EXEC_ENABLE_VPID;
1281 static inline bool cpu_has_vmx_rdtscp(void)
1283 return vmcs_config.cpu_based_2nd_exec_ctrl &
1284 SECONDARY_EXEC_RDTSCP;
1287 static inline bool cpu_has_vmx_invpcid(void)
1289 return vmcs_config.cpu_based_2nd_exec_ctrl &
1290 SECONDARY_EXEC_ENABLE_INVPCID;
1293 static inline bool cpu_has_vmx_wbinvd_exit(void)
1295 return vmcs_config.cpu_based_2nd_exec_ctrl &
1296 SECONDARY_EXEC_WBINVD_EXITING;
1299 static inline bool cpu_has_vmx_shadow_vmcs(void)
1302 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1303 /* check if the cpu supports writing r/o exit information fields */
1304 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1307 return vmcs_config.cpu_based_2nd_exec_ctrl &
1308 SECONDARY_EXEC_SHADOW_VMCS;
1311 static inline bool cpu_has_vmx_pml(void)
1313 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1316 static inline bool cpu_has_vmx_tsc_scaling(void)
1318 return vmcs_config.cpu_based_2nd_exec_ctrl &
1319 SECONDARY_EXEC_TSC_SCALING;
1322 static inline bool report_flexpriority(void)
1324 return flexpriority_enabled;
1327 static inline unsigned nested_cpu_vmx_misc_cr3_count(struct kvm_vcpu *vcpu)
1329 return vmx_misc_cr3_count(to_vmx(vcpu)->nested.nested_vmx_misc_low);
1332 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1334 return vmcs12->cpu_based_vm_exec_control & bit;
1337 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1339 return (vmcs12->cpu_based_vm_exec_control &
1340 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1341 (vmcs12->secondary_vm_exec_control & bit);
1344 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1346 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1349 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1351 return vmcs12->pin_based_vm_exec_control &
1352 PIN_BASED_VMX_PREEMPTION_TIMER;
1355 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1357 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1360 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1362 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) &&
1363 vmx_xsaves_supported();
1366 static inline bool nested_cpu_has_pml(struct vmcs12 *vmcs12)
1368 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML);
1371 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1373 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1376 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1378 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1381 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1383 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1386 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1388 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1391 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1393 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1396 static inline bool is_nmi(u32 intr_info)
1398 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1399 == (INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK);
1402 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1404 unsigned long exit_qualification);
1405 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1406 struct vmcs12 *vmcs12,
1407 u32 reason, unsigned long qualification);
1409 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1413 for (i = 0; i < vmx->nmsrs; ++i)
1414 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1419 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1425 } operand = { vpid, 0, gva };
1427 asm volatile (__ex(ASM_VMX_INVVPID)
1428 /* CF==1 or ZF==1 --> rc = -1 */
1429 "; ja 1f ; ud2 ; 1:"
1430 : : "a"(&operand), "c"(ext) : "cc", "memory");
1433 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1437 } operand = {eptp, gpa};
1439 asm volatile (__ex(ASM_VMX_INVEPT)
1440 /* CF==1 or ZF==1 --> rc = -1 */
1441 "; ja 1f ; ud2 ; 1:\n"
1442 : : "a" (&operand), "c" (ext) : "cc", "memory");
1445 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1449 i = __find_msr_index(vmx, msr);
1451 return &vmx->guest_msrs[i];
1455 static void vmcs_clear(struct vmcs *vmcs)
1457 u64 phys_addr = __pa(vmcs);
1460 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1461 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1464 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1468 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1470 vmcs_clear(loaded_vmcs->vmcs);
1471 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
1472 vmcs_clear(loaded_vmcs->shadow_vmcs);
1473 loaded_vmcs->cpu = -1;
1474 loaded_vmcs->launched = 0;
1477 static void vmcs_load(struct vmcs *vmcs)
1479 u64 phys_addr = __pa(vmcs);
1482 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1483 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1486 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1490 #ifdef CONFIG_KEXEC_CORE
1492 * This bitmap is used to indicate whether the vmclear
1493 * operation is enabled on all cpus. All disabled by
1496 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1498 static inline void crash_enable_local_vmclear(int cpu)
1500 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1503 static inline void crash_disable_local_vmclear(int cpu)
1505 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1508 static inline int crash_local_vmclear_enabled(int cpu)
1510 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1513 static void crash_vmclear_local_loaded_vmcss(void)
1515 int cpu = raw_smp_processor_id();
1516 struct loaded_vmcs *v;
1518 if (!crash_local_vmclear_enabled(cpu))
1521 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1522 loaded_vmcss_on_cpu_link)
1523 vmcs_clear(v->vmcs);
1526 static inline void crash_enable_local_vmclear(int cpu) { }
1527 static inline void crash_disable_local_vmclear(int cpu) { }
1528 #endif /* CONFIG_KEXEC_CORE */
1530 static void __loaded_vmcs_clear(void *arg)
1532 struct loaded_vmcs *loaded_vmcs = arg;
1533 int cpu = raw_smp_processor_id();
1535 if (loaded_vmcs->cpu != cpu)
1536 return; /* vcpu migration can race with cpu offline */
1537 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1538 per_cpu(current_vmcs, cpu) = NULL;
1539 crash_disable_local_vmclear(cpu);
1540 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1543 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1544 * is before setting loaded_vmcs->vcpu to -1 which is done in
1545 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1546 * then adds the vmcs into percpu list before it is deleted.
1550 loaded_vmcs_init(loaded_vmcs);
1551 crash_enable_local_vmclear(cpu);
1554 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1556 int cpu = loaded_vmcs->cpu;
1559 smp_call_function_single(cpu,
1560 __loaded_vmcs_clear, loaded_vmcs, 1);
1563 static inline void vpid_sync_vcpu_single(int vpid)
1568 if (cpu_has_vmx_invvpid_single())
1569 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1572 static inline void vpid_sync_vcpu_global(void)
1574 if (cpu_has_vmx_invvpid_global())
1575 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1578 static inline void vpid_sync_context(int vpid)
1580 if (cpu_has_vmx_invvpid_single())
1581 vpid_sync_vcpu_single(vpid);
1583 vpid_sync_vcpu_global();
1586 static inline void ept_sync_global(void)
1588 if (cpu_has_vmx_invept_global())
1589 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1592 static inline void ept_sync_context(u64 eptp)
1595 if (cpu_has_vmx_invept_context())
1596 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1602 static __always_inline void vmcs_check16(unsigned long field)
1604 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1605 "16-bit accessor invalid for 64-bit field");
1606 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1607 "16-bit accessor invalid for 64-bit high field");
1608 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1609 "16-bit accessor invalid for 32-bit high field");
1610 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1611 "16-bit accessor invalid for natural width field");
1614 static __always_inline void vmcs_check32(unsigned long field)
1616 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1617 "32-bit accessor invalid for 16-bit field");
1618 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1619 "32-bit accessor invalid for natural width field");
1622 static __always_inline void vmcs_check64(unsigned long field)
1624 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1625 "64-bit accessor invalid for 16-bit field");
1626 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1627 "64-bit accessor invalid for 64-bit high field");
1628 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1629 "64-bit accessor invalid for 32-bit field");
1630 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1631 "64-bit accessor invalid for natural width field");
1634 static __always_inline void vmcs_checkl(unsigned long field)
1636 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1637 "Natural width accessor invalid for 16-bit field");
1638 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1639 "Natural width accessor invalid for 64-bit field");
1640 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1641 "Natural width accessor invalid for 64-bit high field");
1642 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1643 "Natural width accessor invalid for 32-bit field");
1646 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1648 unsigned long value;
1650 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1651 : "=a"(value) : "d"(field) : "cc");
1655 static __always_inline u16 vmcs_read16(unsigned long field)
1657 vmcs_check16(field);
1658 return __vmcs_readl(field);
1661 static __always_inline u32 vmcs_read32(unsigned long field)
1663 vmcs_check32(field);
1664 return __vmcs_readl(field);
1667 static __always_inline u64 vmcs_read64(unsigned long field)
1669 vmcs_check64(field);
1670 #ifdef CONFIG_X86_64
1671 return __vmcs_readl(field);
1673 return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1677 static __always_inline unsigned long vmcs_readl(unsigned long field)
1680 return __vmcs_readl(field);
1683 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1685 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1686 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1690 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1694 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1695 : "=q"(error) : "a"(value), "d"(field) : "cc");
1696 if (unlikely(error))
1697 vmwrite_error(field, value);
1700 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1702 vmcs_check16(field);
1703 __vmcs_writel(field, value);
1706 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1708 vmcs_check32(field);
1709 __vmcs_writel(field, value);
1712 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1714 vmcs_check64(field);
1715 __vmcs_writel(field, value);
1716 #ifndef CONFIG_X86_64
1718 __vmcs_writel(field+1, value >> 32);
1722 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1725 __vmcs_writel(field, value);
1728 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1730 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1731 "vmcs_clear_bits does not support 64-bit fields");
1732 __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1735 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1737 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1738 "vmcs_set_bits does not support 64-bit fields");
1739 __vmcs_writel(field, __vmcs_readl(field) | mask);
1742 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
1744 vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
1747 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1749 vmcs_write32(VM_ENTRY_CONTROLS, val);
1750 vmx->vm_entry_controls_shadow = val;
1753 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1755 if (vmx->vm_entry_controls_shadow != val)
1756 vm_entry_controls_init(vmx, val);
1759 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1761 return vmx->vm_entry_controls_shadow;
1765 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1767 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1770 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1772 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1775 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
1777 vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
1780 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1782 vmcs_write32(VM_EXIT_CONTROLS, val);
1783 vmx->vm_exit_controls_shadow = val;
1786 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1788 if (vmx->vm_exit_controls_shadow != val)
1789 vm_exit_controls_init(vmx, val);
1792 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1794 return vmx->vm_exit_controls_shadow;
1798 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1800 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1803 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1805 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1808 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1810 vmx->segment_cache.bitmask = 0;
1813 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1817 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1819 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1820 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1821 vmx->segment_cache.bitmask = 0;
1823 ret = vmx->segment_cache.bitmask & mask;
1824 vmx->segment_cache.bitmask |= mask;
1828 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1830 u16 *p = &vmx->segment_cache.seg[seg].selector;
1832 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1833 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1837 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1839 ulong *p = &vmx->segment_cache.seg[seg].base;
1841 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1842 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1846 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1848 u32 *p = &vmx->segment_cache.seg[seg].limit;
1850 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1851 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1855 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1857 u32 *p = &vmx->segment_cache.seg[seg].ar;
1859 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1860 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1864 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1868 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1869 (1u << DB_VECTOR) | (1u << AC_VECTOR);
1870 if ((vcpu->guest_debug &
1871 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1872 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1873 eb |= 1u << BP_VECTOR;
1874 if (to_vmx(vcpu)->rmode.vm86_active)
1877 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1879 /* When we are running a nested L2 guest and L1 specified for it a
1880 * certain exception bitmap, we must trap the same exceptions and pass
1881 * them to L1. When running L2, we will only handle the exceptions
1882 * specified above if L1 did not want them.
1884 if (is_guest_mode(vcpu))
1885 eb |= get_vmcs12(vcpu)->exception_bitmap;
1887 vmcs_write32(EXCEPTION_BITMAP, eb);
1890 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1891 unsigned long entry, unsigned long exit)
1893 vm_entry_controls_clearbit(vmx, entry);
1894 vm_exit_controls_clearbit(vmx, exit);
1897 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1900 struct msr_autoload *m = &vmx->msr_autoload;
1904 if (cpu_has_load_ia32_efer) {
1905 clear_atomic_switch_msr_special(vmx,
1906 VM_ENTRY_LOAD_IA32_EFER,
1907 VM_EXIT_LOAD_IA32_EFER);
1911 case MSR_CORE_PERF_GLOBAL_CTRL:
1912 if (cpu_has_load_perf_global_ctrl) {
1913 clear_atomic_switch_msr_special(vmx,
1914 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1915 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1921 for (i = 0; i < m->nr; ++i)
1922 if (m->guest[i].index == msr)
1928 m->guest[i] = m->guest[m->nr];
1929 m->host[i] = m->host[m->nr];
1930 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1931 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1934 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1935 unsigned long entry, unsigned long exit,
1936 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1937 u64 guest_val, u64 host_val)
1939 vmcs_write64(guest_val_vmcs, guest_val);
1940 vmcs_write64(host_val_vmcs, host_val);
1941 vm_entry_controls_setbit(vmx, entry);
1942 vm_exit_controls_setbit(vmx, exit);
1945 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1946 u64 guest_val, u64 host_val)
1949 struct msr_autoload *m = &vmx->msr_autoload;
1953 if (cpu_has_load_ia32_efer) {
1954 add_atomic_switch_msr_special(vmx,
1955 VM_ENTRY_LOAD_IA32_EFER,
1956 VM_EXIT_LOAD_IA32_EFER,
1959 guest_val, host_val);
1963 case MSR_CORE_PERF_GLOBAL_CTRL:
1964 if (cpu_has_load_perf_global_ctrl) {
1965 add_atomic_switch_msr_special(vmx,
1966 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1967 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1968 GUEST_IA32_PERF_GLOBAL_CTRL,
1969 HOST_IA32_PERF_GLOBAL_CTRL,
1970 guest_val, host_val);
1974 case MSR_IA32_PEBS_ENABLE:
1975 /* PEBS needs a quiescent period after being disabled (to write
1976 * a record). Disabling PEBS through VMX MSR swapping doesn't
1977 * provide that period, so a CPU could write host's record into
1980 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1983 for (i = 0; i < m->nr; ++i)
1984 if (m->guest[i].index == msr)
1987 if (i == NR_AUTOLOAD_MSRS) {
1988 printk_once(KERN_WARNING "Not enough msr switch entries. "
1989 "Can't add msr %x\n", msr);
1991 } else if (i == m->nr) {
1993 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1994 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1997 m->guest[i].index = msr;
1998 m->guest[i].value = guest_val;
1999 m->host[i].index = msr;
2000 m->host[i].value = host_val;
2003 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
2005 u64 guest_efer = vmx->vcpu.arch.efer;
2006 u64 ignore_bits = 0;
2010 * NX is needed to handle CR0.WP=1, CR4.SMEP=1. Testing
2011 * host CPUID is more efficient than testing guest CPUID
2012 * or CR4. Host SMEP is anyway a requirement for guest SMEP.
2014 if (boot_cpu_has(X86_FEATURE_SMEP))
2015 guest_efer |= EFER_NX;
2016 else if (!(guest_efer & EFER_NX))
2017 ignore_bits |= EFER_NX;
2021 * LMA and LME handled by hardware; SCE meaningless outside long mode.
2023 ignore_bits |= EFER_SCE;
2024 #ifdef CONFIG_X86_64
2025 ignore_bits |= EFER_LMA | EFER_LME;
2026 /* SCE is meaningful only in long mode on Intel */
2027 if (guest_efer & EFER_LMA)
2028 ignore_bits &= ~(u64)EFER_SCE;
2031 clear_atomic_switch_msr(vmx, MSR_EFER);
2034 * On EPT, we can't emulate NX, so we must switch EFER atomically.
2035 * On CPUs that support "load IA32_EFER", always switch EFER
2036 * atomically, since it's faster than switching it manually.
2038 if (cpu_has_load_ia32_efer ||
2039 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
2040 if (!(guest_efer & EFER_LMA))
2041 guest_efer &= ~EFER_LME;
2042 if (guest_efer != host_efer)
2043 add_atomic_switch_msr(vmx, MSR_EFER,
2044 guest_efer, host_efer);
2047 guest_efer &= ~ignore_bits;
2048 guest_efer |= host_efer & ignore_bits;
2050 vmx->guest_msrs[efer_offset].data = guest_efer;
2051 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
2057 #ifdef CONFIG_X86_32
2059 * On 32-bit kernels, VM exits still load the FS and GS bases from the
2060 * VMCS rather than the segment table. KVM uses this helper to figure
2061 * out the current bases to poke them into the VMCS before entry.
2063 static unsigned long segment_base(u16 selector)
2065 struct desc_struct *table;
2068 if (!(selector & ~SEGMENT_RPL_MASK))
2071 table = get_current_gdt_ro();
2073 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
2074 u16 ldt_selector = kvm_read_ldt();
2076 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
2079 table = (struct desc_struct *)segment_base(ldt_selector);
2081 v = get_desc_base(&table[selector >> 3]);
2086 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
2088 struct vcpu_vmx *vmx = to_vmx(vcpu);
2091 if (vmx->host_state.loaded)
2094 vmx->host_state.loaded = 1;
2096 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
2097 * allow segment selectors with cpl > 0 or ti == 1.
2099 vmx->host_state.ldt_sel = kvm_read_ldt();
2100 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
2101 savesegment(fs, vmx->host_state.fs_sel);
2102 if (!(vmx->host_state.fs_sel & 7)) {
2103 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
2104 vmx->host_state.fs_reload_needed = 0;
2106 vmcs_write16(HOST_FS_SELECTOR, 0);
2107 vmx->host_state.fs_reload_needed = 1;
2109 savesegment(gs, vmx->host_state.gs_sel);
2110 if (!(vmx->host_state.gs_sel & 7))
2111 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
2113 vmcs_write16(HOST_GS_SELECTOR, 0);
2114 vmx->host_state.gs_ldt_reload_needed = 1;
2117 #ifdef CONFIG_X86_64
2118 savesegment(ds, vmx->host_state.ds_sel);
2119 savesegment(es, vmx->host_state.es_sel);
2122 #ifdef CONFIG_X86_64
2123 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
2124 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
2126 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
2127 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
2130 #ifdef CONFIG_X86_64
2131 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2132 if (is_long_mode(&vmx->vcpu))
2133 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2135 if (boot_cpu_has(X86_FEATURE_MPX))
2136 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2137 for (i = 0; i < vmx->save_nmsrs; ++i)
2138 kvm_set_shared_msr(vmx->guest_msrs[i].index,
2139 vmx->guest_msrs[i].data,
2140 vmx->guest_msrs[i].mask);
2143 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
2145 if (!vmx->host_state.loaded)
2148 ++vmx->vcpu.stat.host_state_reload;
2149 vmx->host_state.loaded = 0;
2150 #ifdef CONFIG_X86_64
2151 if (is_long_mode(&vmx->vcpu))
2152 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2154 if (vmx->host_state.gs_ldt_reload_needed) {
2155 kvm_load_ldt(vmx->host_state.ldt_sel);
2156 #ifdef CONFIG_X86_64
2157 load_gs_index(vmx->host_state.gs_sel);
2159 loadsegment(gs, vmx->host_state.gs_sel);
2162 if (vmx->host_state.fs_reload_needed)
2163 loadsegment(fs, vmx->host_state.fs_sel);
2164 #ifdef CONFIG_X86_64
2165 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2166 loadsegment(ds, vmx->host_state.ds_sel);
2167 loadsegment(es, vmx->host_state.es_sel);
2170 invalidate_tss_limit();
2171 #ifdef CONFIG_X86_64
2172 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2174 if (vmx->host_state.msr_host_bndcfgs)
2175 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2176 load_fixmap_gdt(raw_smp_processor_id());
2179 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2182 __vmx_load_host_state(vmx);
2186 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2188 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2189 struct pi_desc old, new;
2192 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2193 !irq_remapping_cap(IRQ_POSTING_CAP) ||
2194 !kvm_vcpu_apicv_active(vcpu))
2198 old.control = new.control = pi_desc->control;
2201 * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
2202 * are two possible cases:
2203 * 1. After running 'pre_block', context switch
2204 * happened. For this case, 'sn' was set in
2205 * vmx_vcpu_put(), so we need to clear it here.
2206 * 2. After running 'pre_block', we were blocked,
2207 * and woken up by some other guy. For this case,
2208 * we don't need to do anything, 'pi_post_block'
2209 * will do everything for us. However, we cannot
2210 * check whether it is case #1 or case #2 here
2211 * (maybe, not needed), so we also clear sn here,
2212 * I think it is not a big deal.
2214 if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) {
2215 if (vcpu->cpu != cpu) {
2216 dest = cpu_physical_id(cpu);
2218 if (x2apic_enabled())
2221 new.ndst = (dest << 8) & 0xFF00;
2224 /* set 'NV' to 'notification vector' */
2225 new.nv = POSTED_INTR_VECTOR;
2228 /* Allow posting non-urgent interrupts */
2230 } while (cmpxchg(&pi_desc->control, old.control,
2231 new.control) != old.control);
2234 static void decache_tsc_multiplier(struct vcpu_vmx *vmx)
2236 vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
2237 vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2241 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2242 * vcpu mutex is already taken.
2244 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2246 struct vcpu_vmx *vmx = to_vmx(vcpu);
2247 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
2249 if (!already_loaded) {
2250 loaded_vmcs_clear(vmx->loaded_vmcs);
2251 local_irq_disable();
2252 crash_disable_local_vmclear(cpu);
2255 * Read loaded_vmcs->cpu should be before fetching
2256 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2257 * See the comments in __loaded_vmcs_clear().
2261 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2262 &per_cpu(loaded_vmcss_on_cpu, cpu));
2263 crash_enable_local_vmclear(cpu);
2267 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2268 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2269 vmcs_load(vmx->loaded_vmcs->vmcs);
2272 if (!already_loaded) {
2273 void *gdt = get_current_gdt_ro();
2274 unsigned long sysenter_esp;
2276 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2279 * Linux uses per-cpu TSS and GDT, so set these when switching
2280 * processors. See 22.2.4.
2282 vmcs_writel(HOST_TR_BASE,
2283 (unsigned long)this_cpu_ptr(&cpu_tss));
2284 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
2287 * VM exits change the host TR limit to 0x67 after a VM
2288 * exit. This is okay, since 0x67 covers everything except
2289 * the IO bitmap and have have code to handle the IO bitmap
2290 * being lost after a VM exit.
2292 BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
2294 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2295 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2297 vmx->loaded_vmcs->cpu = cpu;
2300 /* Setup TSC multiplier */
2301 if (kvm_has_tsc_control &&
2302 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
2303 decache_tsc_multiplier(vmx);
2305 vmx_vcpu_pi_load(vcpu, cpu);
2306 vmx->host_pkru = read_pkru();
2309 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2311 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2313 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2314 !irq_remapping_cap(IRQ_POSTING_CAP) ||
2315 !kvm_vcpu_apicv_active(vcpu))
2318 /* Set SN when the vCPU is preempted */
2319 if (vcpu->preempted)
2323 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2325 vmx_vcpu_pi_put(vcpu);
2327 __vmx_load_host_state(to_vmx(vcpu));
2330 static bool emulation_required(struct kvm_vcpu *vcpu)
2332 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
2335 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2338 * Return the cr0 value that a nested guest would read. This is a combination
2339 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2340 * its hypervisor (cr0_read_shadow).
2342 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2344 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2345 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2347 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2349 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2350 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2353 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2355 unsigned long rflags, save_rflags;
2357 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2358 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2359 rflags = vmcs_readl(GUEST_RFLAGS);
2360 if (to_vmx(vcpu)->rmode.vm86_active) {
2361 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2362 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2363 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2365 to_vmx(vcpu)->rflags = rflags;
2367 return to_vmx(vcpu)->rflags;
2370 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2372 unsigned long old_rflags = vmx_get_rflags(vcpu);
2374 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2375 to_vmx(vcpu)->rflags = rflags;
2376 if (to_vmx(vcpu)->rmode.vm86_active) {
2377 to_vmx(vcpu)->rmode.save_rflags = rflags;
2378 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2380 vmcs_writel(GUEST_RFLAGS, rflags);
2382 if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
2383 to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
2386 static u32 vmx_get_pkru(struct kvm_vcpu *vcpu)
2388 return to_vmx(vcpu)->guest_pkru;
2391 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2393 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2396 if (interruptibility & GUEST_INTR_STATE_STI)
2397 ret |= KVM_X86_SHADOW_INT_STI;
2398 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2399 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2404 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2406 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2407 u32 interruptibility = interruptibility_old;
2409 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2411 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2412 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2413 else if (mask & KVM_X86_SHADOW_INT_STI)
2414 interruptibility |= GUEST_INTR_STATE_STI;
2416 if ((interruptibility != interruptibility_old))
2417 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2420 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2424 rip = kvm_rip_read(vcpu);
2425 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2426 kvm_rip_write(vcpu, rip);
2428 /* skipping an emulated instruction also counts */
2429 vmx_set_interrupt_shadow(vcpu, 0);
2432 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
2433 unsigned long exit_qual)
2435 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2436 unsigned int nr = vcpu->arch.exception.nr;
2437 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2439 if (vcpu->arch.exception.has_error_code) {
2440 vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
2441 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2444 if (kvm_exception_is_soft(nr))
2445 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2447 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2449 if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
2450 vmx_get_nmi_mask(vcpu))
2451 intr_info |= INTR_INFO_UNBLOCK_NMI;
2453 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
2457 * KVM wants to inject page-faults which it got to the guest. This function
2458 * checks whether in a nested guest, we need to inject them to L1 or L2.
2460 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu)
2462 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2463 unsigned int nr = vcpu->arch.exception.nr;
2465 if (nr == PF_VECTOR) {
2466 if (vcpu->arch.exception.nested_apf) {
2467 nested_vmx_inject_exception_vmexit(vcpu,
2468 vcpu->arch.apf.nested_apf_token);
2472 * FIXME: we must not write CR2 when L1 intercepts an L2 #PF exception.
2473 * The fix is to add the ancillary datum (CR2 or DR6) to structs
2474 * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6
2475 * can be written only when inject_pending_event runs. This should be
2476 * conditional on a new capability---if the capability is disabled,
2477 * kvm_multiple_exception would write the ancillary information to
2478 * CR2 or DR6, for backwards ABI-compatibility.
2480 if (nested_vmx_is_page_fault_vmexit(vmcs12,
2481 vcpu->arch.exception.error_code)) {
2482 nested_vmx_inject_exception_vmexit(vcpu, vcpu->arch.cr2);
2486 unsigned long exit_qual = 0;
2487 if (nr == DB_VECTOR)
2488 exit_qual = vcpu->arch.dr6;
2490 if (vmcs12->exception_bitmap & (1u << nr)) {
2491 nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
2499 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
2501 struct vcpu_vmx *vmx = to_vmx(vcpu);
2502 unsigned nr = vcpu->arch.exception.nr;
2503 bool has_error_code = vcpu->arch.exception.has_error_code;
2504 bool reinject = vcpu->arch.exception.reinject;
2505 u32 error_code = vcpu->arch.exception.error_code;
2506 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2508 if (!reinject && is_guest_mode(vcpu) &&
2509 nested_vmx_check_exception(vcpu))
2512 if (has_error_code) {
2513 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2514 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2517 if (vmx->rmode.vm86_active) {
2519 if (kvm_exception_is_soft(nr))
2520 inc_eip = vcpu->arch.event_exit_inst_len;
2521 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2522 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2526 if (kvm_exception_is_soft(nr)) {
2527 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2528 vmx->vcpu.arch.event_exit_inst_len);
2529 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2531 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2533 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2536 static bool vmx_rdtscp_supported(void)
2538 return cpu_has_vmx_rdtscp();
2541 static bool vmx_invpcid_supported(void)
2543 return cpu_has_vmx_invpcid() && enable_ept;
2547 * Swap MSR entry in host/guest MSR entry array.
2549 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2551 struct shared_msr_entry tmp;
2553 tmp = vmx->guest_msrs[to];
2554 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2555 vmx->guest_msrs[from] = tmp;
2558 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2560 unsigned long *msr_bitmap;
2562 if (is_guest_mode(vcpu))
2563 msr_bitmap = to_vmx(vcpu)->nested.msr_bitmap;
2564 else if (cpu_has_secondary_exec_ctrls() &&
2565 (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
2566 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
2567 if (enable_apicv && kvm_vcpu_apicv_active(vcpu)) {
2568 if (is_long_mode(vcpu))
2569 msr_bitmap = vmx_msr_bitmap_longmode_x2apic_apicv;
2571 msr_bitmap = vmx_msr_bitmap_legacy_x2apic_apicv;
2573 if (is_long_mode(vcpu))
2574 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2576 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2579 if (is_long_mode(vcpu))
2580 msr_bitmap = vmx_msr_bitmap_longmode;
2582 msr_bitmap = vmx_msr_bitmap_legacy;
2585 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2589 * Set up the vmcs to automatically save and restore system
2590 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2591 * mode, as fiddling with msrs is very expensive.
2593 static void setup_msrs(struct vcpu_vmx *vmx)
2595 int save_nmsrs, index;
2598 #ifdef CONFIG_X86_64
2599 if (is_long_mode(&vmx->vcpu)) {
2600 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2602 move_msr_up(vmx, index, save_nmsrs++);
2603 index = __find_msr_index(vmx, MSR_LSTAR);
2605 move_msr_up(vmx, index, save_nmsrs++);
2606 index = __find_msr_index(vmx, MSR_CSTAR);
2608 move_msr_up(vmx, index, save_nmsrs++);
2609 index = __find_msr_index(vmx, MSR_TSC_AUX);
2610 if (index >= 0 && guest_cpuid_has_rdtscp(&vmx->vcpu))
2611 move_msr_up(vmx, index, save_nmsrs++);
2613 * MSR_STAR is only needed on long mode guests, and only
2614 * if efer.sce is enabled.
2616 index = __find_msr_index(vmx, MSR_STAR);
2617 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2618 move_msr_up(vmx, index, save_nmsrs++);
2621 index = __find_msr_index(vmx, MSR_EFER);
2622 if (index >= 0 && update_transition_efer(vmx, index))
2623 move_msr_up(vmx, index, save_nmsrs++);
2625 vmx->save_nmsrs = save_nmsrs;
2627 if (cpu_has_vmx_msr_bitmap())
2628 vmx_set_msr_bitmap(&vmx->vcpu);
2632 * reads and returns guest's timestamp counter "register"
2633 * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2634 * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2636 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2638 u64 host_tsc, tsc_offset;
2641 tsc_offset = vmcs_read64(TSC_OFFSET);
2642 return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2646 * writes 'offset' into guest's timestamp counter offset register
2648 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2650 if (is_guest_mode(vcpu)) {
2652 * We're here if L1 chose not to trap WRMSR to TSC. According
2653 * to the spec, this should set L1's TSC; The offset that L1
2654 * set for L2 remains unchanged, and still needs to be added
2655 * to the newly set TSC to get L2's TSC.
2657 struct vmcs12 *vmcs12;
2658 /* recalculate vmcs02.TSC_OFFSET: */
2659 vmcs12 = get_vmcs12(vcpu);
2660 vmcs_write64(TSC_OFFSET, offset +
2661 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2662 vmcs12->tsc_offset : 0));
2664 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2665 vmcs_read64(TSC_OFFSET), offset);
2666 vmcs_write64(TSC_OFFSET, offset);
2670 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2672 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2673 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2677 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2678 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2679 * all guests if the "nested" module option is off, and can also be disabled
2680 * for a single guest by disabling its VMX cpuid bit.
2682 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2684 return nested && guest_cpuid_has_vmx(vcpu);
2688 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2689 * returned for the various VMX controls MSRs when nested VMX is enabled.
2690 * The same values should also be used to verify that vmcs12 control fields are
2691 * valid during nested entry from L1 to L2.
2692 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2693 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2694 * bit in the high half is on if the corresponding bit in the control field
2695 * may be on. See also vmx_control_verify().
2697 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2700 * Note that as a general rule, the high half of the MSRs (bits in
2701 * the control fields which may be 1) should be initialized by the
2702 * intersection of the underlying hardware's MSR (i.e., features which
2703 * can be supported) and the list of features we want to expose -
2704 * because they are known to be properly supported in our code.
2705 * Also, usually, the low half of the MSRs (bits which must be 1) can
2706 * be set to 0, meaning that L1 may turn off any of these bits. The
2707 * reason is that if one of these bits is necessary, it will appear
2708 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2709 * fields of vmcs01 and vmcs02, will turn these bits off - and
2710 * nested_vmx_exit_reflected() will not pass related exits to L1.
2711 * These rules have exceptions below.
2714 /* pin-based controls */
2715 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2716 vmx->nested.nested_vmx_pinbased_ctls_low,
2717 vmx->nested.nested_vmx_pinbased_ctls_high);
2718 vmx->nested.nested_vmx_pinbased_ctls_low |=
2719 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2720 vmx->nested.nested_vmx_pinbased_ctls_high &=
2721 PIN_BASED_EXT_INTR_MASK |
2722 PIN_BASED_NMI_EXITING |
2723 PIN_BASED_VIRTUAL_NMIS;
2724 vmx->nested.nested_vmx_pinbased_ctls_high |=
2725 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2726 PIN_BASED_VMX_PREEMPTION_TIMER;
2727 if (kvm_vcpu_apicv_active(&vmx->vcpu))
2728 vmx->nested.nested_vmx_pinbased_ctls_high |=
2729 PIN_BASED_POSTED_INTR;
2732 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2733 vmx->nested.nested_vmx_exit_ctls_low,
2734 vmx->nested.nested_vmx_exit_ctls_high);
2735 vmx->nested.nested_vmx_exit_ctls_low =
2736 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2738 vmx->nested.nested_vmx_exit_ctls_high &=
2739 #ifdef CONFIG_X86_64
2740 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2742 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2743 vmx->nested.nested_vmx_exit_ctls_high |=
2744 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2745 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2746 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2748 if (kvm_mpx_supported())
2749 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2751 /* We support free control of debug control saving. */
2752 vmx->nested.nested_vmx_exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2754 /* entry controls */
2755 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2756 vmx->nested.nested_vmx_entry_ctls_low,
2757 vmx->nested.nested_vmx_entry_ctls_high);
2758 vmx->nested.nested_vmx_entry_ctls_low =
2759 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2760 vmx->nested.nested_vmx_entry_ctls_high &=
2761 #ifdef CONFIG_X86_64
2762 VM_ENTRY_IA32E_MODE |
2764 VM_ENTRY_LOAD_IA32_PAT;
2765 vmx->nested.nested_vmx_entry_ctls_high |=
2766 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2767 if (kvm_mpx_supported())
2768 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2770 /* We support free control of debug control loading. */
2771 vmx->nested.nested_vmx_entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2773 /* cpu-based controls */
2774 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2775 vmx->nested.nested_vmx_procbased_ctls_low,
2776 vmx->nested.nested_vmx_procbased_ctls_high);
2777 vmx->nested.nested_vmx_procbased_ctls_low =
2778 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2779 vmx->nested.nested_vmx_procbased_ctls_high &=
2780 CPU_BASED_VIRTUAL_INTR_PENDING |
2781 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2782 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2783 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2784 CPU_BASED_CR3_STORE_EXITING |
2785 #ifdef CONFIG_X86_64
2786 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2788 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2789 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2790 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2791 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2792 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2794 * We can allow some features even when not supported by the
2795 * hardware. For example, L1 can specify an MSR bitmap - and we
2796 * can use it to avoid exits to L1 - even when L0 runs L2
2797 * without MSR bitmaps.
2799 vmx->nested.nested_vmx_procbased_ctls_high |=
2800 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2801 CPU_BASED_USE_MSR_BITMAPS;
2803 /* We support free control of CR3 access interception. */
2804 vmx->nested.nested_vmx_procbased_ctls_low &=
2805 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2807 /* secondary cpu-based controls */
2808 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2809 vmx->nested.nested_vmx_secondary_ctls_low,
2810 vmx->nested.nested_vmx_secondary_ctls_high);
2811 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2812 vmx->nested.nested_vmx_secondary_ctls_high &=
2813 SECONDARY_EXEC_RDRAND | SECONDARY_EXEC_RDSEED |
2814 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2815 SECONDARY_EXEC_RDTSCP |
2816 SECONDARY_EXEC_DESC |
2817 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2818 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2819 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2820 SECONDARY_EXEC_WBINVD_EXITING |
2821 SECONDARY_EXEC_XSAVES;
2824 /* nested EPT: emulate EPT also to L1 */
2825 vmx->nested.nested_vmx_secondary_ctls_high |=
2826 SECONDARY_EXEC_ENABLE_EPT;
2827 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2828 VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
2829 if (cpu_has_vmx_ept_execute_only())
2830 vmx->nested.nested_vmx_ept_caps |=
2831 VMX_EPT_EXECUTE_ONLY_BIT;
2832 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2833 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
2834 VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
2835 VMX_EPT_1GB_PAGE_BIT;
2836 if (enable_ept_ad_bits) {
2837 vmx->nested.nested_vmx_secondary_ctls_high |=
2838 SECONDARY_EXEC_ENABLE_PML;
2839 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_AD_BIT;
2842 vmx->nested.nested_vmx_ept_caps = 0;
2845 * Old versions of KVM use the single-context version without
2846 * checking for support, so declare that it is supported even
2847 * though it is treated as global context. The alternative is
2848 * not failing the single-context invvpid, and it is worse.
2851 vmx->nested.nested_vmx_secondary_ctls_high |=
2852 SECONDARY_EXEC_ENABLE_VPID;
2853 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2854 VMX_VPID_EXTENT_SUPPORTED_MASK;
2856 vmx->nested.nested_vmx_vpid_caps = 0;
2858 if (enable_unrestricted_guest)
2859 vmx->nested.nested_vmx_secondary_ctls_high |=
2860 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2862 /* miscellaneous data */
2863 rdmsr(MSR_IA32_VMX_MISC,
2864 vmx->nested.nested_vmx_misc_low,
2865 vmx->nested.nested_vmx_misc_high);
2866 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2867 vmx->nested.nested_vmx_misc_low |=
2868 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2869 VMX_MISC_ACTIVITY_HLT;
2870 vmx->nested.nested_vmx_misc_high = 0;
2873 * This MSR reports some information about VMX support. We
2874 * should return information about the VMX we emulate for the
2875 * guest, and the VMCS structure we give it - not about the
2876 * VMX support of the underlying hardware.
2878 vmx->nested.nested_vmx_basic =
2880 VMX_BASIC_TRUE_CTLS |
2881 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2882 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2884 if (cpu_has_vmx_basic_inout())
2885 vmx->nested.nested_vmx_basic |= VMX_BASIC_INOUT;
2888 * These MSRs specify bits which the guest must keep fixed on
2889 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2890 * We picked the standard core2 setting.
2892 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2893 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2894 vmx->nested.nested_vmx_cr0_fixed0 = VMXON_CR0_ALWAYSON;
2895 vmx->nested.nested_vmx_cr4_fixed0 = VMXON_CR4_ALWAYSON;
2897 /* These MSRs specify bits which the guest must keep fixed off. */
2898 rdmsrl(MSR_IA32_VMX_CR0_FIXED1, vmx->nested.nested_vmx_cr0_fixed1);
2899 rdmsrl(MSR_IA32_VMX_CR4_FIXED1, vmx->nested.nested_vmx_cr4_fixed1);
2901 /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2902 vmx->nested.nested_vmx_vmcs_enum = 0x2e;
2906 * if fixed0[i] == 1: val[i] must be 1
2907 * if fixed1[i] == 0: val[i] must be 0
2909 static inline bool fixed_bits_valid(u64 val, u64 fixed0, u64 fixed1)
2911 return ((val & fixed1) | fixed0) == val;
2914 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2916 return fixed_bits_valid(control, low, high);
2919 static inline u64 vmx_control_msr(u32 low, u32 high)
2921 return low | ((u64)high << 32);
2924 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
2929 return (superset | subset) == superset;
2932 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
2934 const u64 feature_and_reserved =
2935 /* feature (except bit 48; see below) */
2936 BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
2938 BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
2939 u64 vmx_basic = vmx->nested.nested_vmx_basic;
2941 if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
2945 * KVM does not emulate a version of VMX that constrains physical
2946 * addresses of VMX structures (e.g. VMCS) to 32-bits.
2948 if (data & BIT_ULL(48))
2951 if (vmx_basic_vmcs_revision_id(vmx_basic) !=
2952 vmx_basic_vmcs_revision_id(data))
2955 if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
2958 vmx->nested.nested_vmx_basic = data;
2963 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
2968 switch (msr_index) {
2969 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2970 lowp = &vmx->nested.nested_vmx_pinbased_ctls_low;
2971 highp = &vmx->nested.nested_vmx_pinbased_ctls_high;
2973 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2974 lowp = &vmx->nested.nested_vmx_procbased_ctls_low;
2975 highp = &vmx->nested.nested_vmx_procbased_ctls_high;
2977 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2978 lowp = &vmx->nested.nested_vmx_exit_ctls_low;
2979 highp = &vmx->nested.nested_vmx_exit_ctls_high;
2981 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2982 lowp = &vmx->nested.nested_vmx_entry_ctls_low;
2983 highp = &vmx->nested.nested_vmx_entry_ctls_high;
2985 case MSR_IA32_VMX_PROCBASED_CTLS2:
2986 lowp = &vmx->nested.nested_vmx_secondary_ctls_low;
2987 highp = &vmx->nested.nested_vmx_secondary_ctls_high;
2993 supported = vmx_control_msr(*lowp, *highp);
2995 /* Check must-be-1 bits are still 1. */
2996 if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
2999 /* Check must-be-0 bits are still 0. */
3000 if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
3004 *highp = data >> 32;
3008 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
3010 const u64 feature_and_reserved_bits =
3012 BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
3013 BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
3015 GENMASK_ULL(13, 9) | BIT_ULL(31);
3018 vmx_misc = vmx_control_msr(vmx->nested.nested_vmx_misc_low,
3019 vmx->nested.nested_vmx_misc_high);
3021 if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
3024 if ((vmx->nested.nested_vmx_pinbased_ctls_high &
3025 PIN_BASED_VMX_PREEMPTION_TIMER) &&
3026 vmx_misc_preemption_timer_rate(data) !=
3027 vmx_misc_preemption_timer_rate(vmx_misc))
3030 if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
3033 if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
3036 if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
3039 vmx->nested.nested_vmx_misc_low = data;
3040 vmx->nested.nested_vmx_misc_high = data >> 32;
3044 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
3046 u64 vmx_ept_vpid_cap;
3048 vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.nested_vmx_ept_caps,
3049 vmx->nested.nested_vmx_vpid_caps);
3051 /* Every bit is either reserved or a feature bit. */
3052 if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
3055 vmx->nested.nested_vmx_ept_caps = data;
3056 vmx->nested.nested_vmx_vpid_caps = data >> 32;
3060 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
3064 switch (msr_index) {
3065 case MSR_IA32_VMX_CR0_FIXED0:
3066 msr = &vmx->nested.nested_vmx_cr0_fixed0;
3068 case MSR_IA32_VMX_CR4_FIXED0:
3069 msr = &vmx->nested.nested_vmx_cr4_fixed0;
3076 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
3077 * must be 1 in the restored value.
3079 if (!is_bitwise_subset(data, *msr, -1ULL))
3087 * Called when userspace is restoring VMX MSRs.
3089 * Returns 0 on success, non-0 otherwise.
3091 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
3093 struct vcpu_vmx *vmx = to_vmx(vcpu);
3095 switch (msr_index) {
3096 case MSR_IA32_VMX_BASIC:
3097 return vmx_restore_vmx_basic(vmx, data);
3098 case MSR_IA32_VMX_PINBASED_CTLS:
3099 case MSR_IA32_VMX_PROCBASED_CTLS:
3100 case MSR_IA32_VMX_EXIT_CTLS:
3101 case MSR_IA32_VMX_ENTRY_CTLS:
3103 * The "non-true" VMX capability MSRs are generated from the
3104 * "true" MSRs, so we do not support restoring them directly.
3106 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
3107 * should restore the "true" MSRs with the must-be-1 bits
3108 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
3109 * DEFAULT SETTINGS".
3112 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3113 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3114 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3115 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3116 case MSR_IA32_VMX_PROCBASED_CTLS2:
3117 return vmx_restore_control_msr(vmx, msr_index, data);
3118 case MSR_IA32_VMX_MISC:
3119 return vmx_restore_vmx_misc(vmx, data);
3120 case MSR_IA32_VMX_CR0_FIXED0:
3121 case MSR_IA32_VMX_CR4_FIXED0:
3122 return vmx_restore_fixed0_msr(vmx, msr_index, data);
3123 case MSR_IA32_VMX_CR0_FIXED1:
3124 case MSR_IA32_VMX_CR4_FIXED1:
3126 * These MSRs are generated based on the vCPU's CPUID, so we
3127 * do not support restoring them directly.
3130 case MSR_IA32_VMX_EPT_VPID_CAP:
3131 return vmx_restore_vmx_ept_vpid_cap(vmx, data);
3132 case MSR_IA32_VMX_VMCS_ENUM:
3133 vmx->nested.nested_vmx_vmcs_enum = data;
3137 * The rest of the VMX capability MSRs do not support restore.
3143 /* Returns 0 on success, non-0 otherwise. */
3144 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
3146 struct vcpu_vmx *vmx = to_vmx(vcpu);
3148 switch (msr_index) {
3149 case MSR_IA32_VMX_BASIC:
3150 *pdata = vmx->nested.nested_vmx_basic;
3152 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3153 case MSR_IA32_VMX_PINBASED_CTLS:
3154 *pdata = vmx_control_msr(
3155 vmx->nested.nested_vmx_pinbased_ctls_low,
3156 vmx->nested.nested_vmx_pinbased_ctls_high);
3157 if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
3158 *pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3160 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3161 case MSR_IA32_VMX_PROCBASED_CTLS:
3162 *pdata = vmx_control_msr(
3163 vmx->nested.nested_vmx_procbased_ctls_low,
3164 vmx->nested.nested_vmx_procbased_ctls_high);
3165 if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
3166 *pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3168 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3169 case MSR_IA32_VMX_EXIT_CTLS:
3170 *pdata = vmx_control_msr(
3171 vmx->nested.nested_vmx_exit_ctls_low,
3172 vmx->nested.nested_vmx_exit_ctls_high);
3173 if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
3174 *pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
3176 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3177 case MSR_IA32_VMX_ENTRY_CTLS:
3178 *pdata = vmx_control_msr(
3179 vmx->nested.nested_vmx_entry_ctls_low,
3180 vmx->nested.nested_vmx_entry_ctls_high);
3181 if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
3182 *pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
3184 case MSR_IA32_VMX_MISC:
3185 *pdata = vmx_control_msr(
3186 vmx->nested.nested_vmx_misc_low,
3187 vmx->nested.nested_vmx_misc_high);
3189 case MSR_IA32_VMX_CR0_FIXED0:
3190 *pdata = vmx->nested.nested_vmx_cr0_fixed0;
3192 case MSR_IA32_VMX_CR0_FIXED1:
3193 *pdata = vmx->nested.nested_vmx_cr0_fixed1;
3195 case MSR_IA32_VMX_CR4_FIXED0:
3196 *pdata = vmx->nested.nested_vmx_cr4_fixed0;
3198 case MSR_IA32_VMX_CR4_FIXED1:
3199 *pdata = vmx->nested.nested_vmx_cr4_fixed1;
3201 case MSR_IA32_VMX_VMCS_ENUM:
3202 *pdata = vmx->nested.nested_vmx_vmcs_enum;
3204 case MSR_IA32_VMX_PROCBASED_CTLS2:
3205 *pdata = vmx_control_msr(
3206 vmx->nested.nested_vmx_secondary_ctls_low,
3207 vmx->nested.nested_vmx_secondary_ctls_high);
3209 case MSR_IA32_VMX_EPT_VPID_CAP:
3210 *pdata = vmx->nested.nested_vmx_ept_caps |
3211 ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
3220 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
3223 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
3225 return !(val & ~valid_bits);
3229 * Reads an msr value (of 'msr_index') into 'pdata'.
3230 * Returns 0 on success, non-0 otherwise.
3231 * Assumes vcpu_load() was already called.
3233 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3235 struct shared_msr_entry *msr;
3237 switch (msr_info->index) {
3238 #ifdef CONFIG_X86_64
3240 msr_info->data = vmcs_readl(GUEST_FS_BASE);
3243 msr_info->data = vmcs_readl(GUEST_GS_BASE);
3245 case MSR_KERNEL_GS_BASE:
3246 vmx_load_host_state(to_vmx(vcpu));
3247 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
3251 return kvm_get_msr_common(vcpu, msr_info);
3253 msr_info->data = guest_read_tsc(vcpu);
3255 case MSR_IA32_SYSENTER_CS:
3256 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
3258 case MSR_IA32_SYSENTER_EIP:
3259 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
3261 case MSR_IA32_SYSENTER_ESP:
3262 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
3264 case MSR_IA32_BNDCFGS:
3265 if (!kvm_mpx_supported() ||
3266 (!msr_info->host_initiated && !guest_cpuid_has_mpx(vcpu)))
3268 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
3270 case MSR_IA32_MCG_EXT_CTL:
3271 if (!msr_info->host_initiated &&
3272 !(to_vmx(vcpu)->msr_ia32_feature_control &
3273 FEATURE_CONTROL_LMCE))
3275 msr_info->data = vcpu->arch.mcg_ext_ctl;
3277 case MSR_IA32_FEATURE_CONTROL:
3278 msr_info->data = to_vmx(vcpu)->msr_ia32_feature_control;
3280 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3281 if (!nested_vmx_allowed(vcpu))
3283 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
3285 if (!vmx_xsaves_supported())
3287 msr_info->data = vcpu->arch.ia32_xss;
3290 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3292 /* Otherwise falls through */
3294 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
3296 msr_info->data = msr->data;
3299 return kvm_get_msr_common(vcpu, msr_info);
3305 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3308 * Writes msr value into into the appropriate "register".
3309 * Returns 0 on success, non-0 otherwise.
3310 * Assumes vcpu_load() was already called.
3312 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3314 struct vcpu_vmx *vmx = to_vmx(vcpu);
3315 struct shared_msr_entry *msr;
3317 u32 msr_index = msr_info->index;
3318 u64 data = msr_info->data;
3320 switch (msr_index) {
3322 ret = kvm_set_msr_common(vcpu, msr_info);
3324 #ifdef CONFIG_X86_64
3326 vmx_segment_cache_clear(vmx);
3327 vmcs_writel(GUEST_FS_BASE, data);
3330 vmx_segment_cache_clear(vmx);
3331 vmcs_writel(GUEST_GS_BASE, data);
3333 case MSR_KERNEL_GS_BASE:
3334 vmx_load_host_state(vmx);
3335 vmx->msr_guest_kernel_gs_base = data;
3338 case MSR_IA32_SYSENTER_CS:
3339 vmcs_write32(GUEST_SYSENTER_CS, data);
3341 case MSR_IA32_SYSENTER_EIP:
3342 vmcs_writel(GUEST_SYSENTER_EIP, data);
3344 case MSR_IA32_SYSENTER_ESP:
3345 vmcs_writel(GUEST_SYSENTER_ESP, data);
3347 case MSR_IA32_BNDCFGS:
3348 if (!kvm_mpx_supported() ||
3349 (!msr_info->host_initiated && !guest_cpuid_has_mpx(vcpu)))
3351 if (is_noncanonical_address(data & PAGE_MASK) ||
3352 (data & MSR_IA32_BNDCFGS_RSVD))
3354 vmcs_write64(GUEST_BNDCFGS, data);
3357 kvm_write_tsc(vcpu, msr_info);
3359 case MSR_IA32_CR_PAT:
3360 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3361 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3363 vmcs_write64(GUEST_IA32_PAT, data);
3364 vcpu->arch.pat = data;
3367 ret = kvm_set_msr_common(vcpu, msr_info);
3369 case MSR_IA32_TSC_ADJUST:
3370 ret = kvm_set_msr_common(vcpu, msr_info);
3372 case MSR_IA32_MCG_EXT_CTL:
3373 if ((!msr_info->host_initiated &&
3374 !(to_vmx(vcpu)->msr_ia32_feature_control &
3375 FEATURE_CONTROL_LMCE)) ||
3376 (data & ~MCG_EXT_CTL_LMCE_EN))
3378 vcpu->arch.mcg_ext_ctl = data;
3380 case MSR_IA32_FEATURE_CONTROL:
3381 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3382 (to_vmx(vcpu)->msr_ia32_feature_control &
3383 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3385 vmx->msr_ia32_feature_control = data;
3386 if (msr_info->host_initiated && data == 0)
3387 vmx_leave_nested(vcpu);
3389 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3390 if (!msr_info->host_initiated)
3391 return 1; /* they are read-only */
3392 if (!nested_vmx_allowed(vcpu))
3394 return vmx_set_vmx_msr(vcpu, msr_index, data);
3396 if (!vmx_xsaves_supported())
3399 * The only supported bit as of Skylake is bit 8, but
3400 * it is not supported on KVM.
3404 vcpu->arch.ia32_xss = data;
3405 if (vcpu->arch.ia32_xss != host_xss)
3406 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3407 vcpu->arch.ia32_xss, host_xss);
3409 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3412 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3414 /* Check reserved bit, higher 32 bits should be zero */
3415 if ((data >> 32) != 0)
3417 /* Otherwise falls through */
3419 msr = find_msr_entry(vmx, msr_index);
3421 u64 old_msr_data = msr->data;
3423 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3425 ret = kvm_set_shared_msr(msr->index, msr->data,
3429 msr->data = old_msr_data;
3433 ret = kvm_set_msr_common(vcpu, msr_info);
3439 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3441 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3444 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3447 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3449 case VCPU_EXREG_PDPTR:
3451 ept_save_pdptrs(vcpu);
3458 static __init int cpu_has_kvm_support(void)
3460 return cpu_has_vmx();
3463 static __init int vmx_disabled_by_bios(void)
3467 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3468 if (msr & FEATURE_CONTROL_LOCKED) {
3469 /* launched w/ TXT and VMX disabled */
3470 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3473 /* launched w/o TXT and VMX only enabled w/ TXT */
3474 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3475 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3476 && !tboot_enabled()) {
3477 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3478 "activate TXT before enabling KVM\n");
3481 /* launched w/o TXT and VMX disabled */
3482 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3483 && !tboot_enabled())
3490 static void kvm_cpu_vmxon(u64 addr)
3492 cr4_set_bits(X86_CR4_VMXE);
3493 intel_pt_handle_vmx(1);
3495 asm volatile (ASM_VMX_VMXON_RAX
3496 : : "a"(&addr), "m"(addr)
3500 static int hardware_enable(void)
3502 int cpu = raw_smp_processor_id();
3503 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3506 if (cr4_read_shadow() & X86_CR4_VMXE)
3509 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3510 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3511 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3514 * Now we can enable the vmclear operation in kdump
3515 * since the loaded_vmcss_on_cpu list on this cpu
3516 * has been initialized.
3518 * Though the cpu is not in VMX operation now, there
3519 * is no problem to enable the vmclear operation
3520 * for the loaded_vmcss_on_cpu list is empty!
3522 crash_enable_local_vmclear(cpu);
3524 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3526 test_bits = FEATURE_CONTROL_LOCKED;
3527 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3528 if (tboot_enabled())
3529 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3531 if ((old & test_bits) != test_bits) {
3532 /* enable and lock */
3533 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3535 kvm_cpu_vmxon(phys_addr);
3541 static void vmclear_local_loaded_vmcss(void)
3543 int cpu = raw_smp_processor_id();
3544 struct loaded_vmcs *v, *n;
3546 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3547 loaded_vmcss_on_cpu_link)
3548 __loaded_vmcs_clear(v);
3552 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3555 static void kvm_cpu_vmxoff(void)
3557 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3559 intel_pt_handle_vmx(0);
3560 cr4_clear_bits(X86_CR4_VMXE);
3563 static void hardware_disable(void)
3565 vmclear_local_loaded_vmcss();
3569 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3570 u32 msr, u32 *result)
3572 u32 vmx_msr_low, vmx_msr_high;
3573 u32 ctl = ctl_min | ctl_opt;
3575 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3577 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3578 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3580 /* Ensure minimum (required) set of control bits are supported. */
3588 static __init bool allow_1_setting(u32 msr, u32 ctl)
3590 u32 vmx_msr_low, vmx_msr_high;
3592 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3593 return vmx_msr_high & ctl;
3596 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3598 u32 vmx_msr_low, vmx_msr_high;
3599 u32 min, opt, min2, opt2;
3600 u32 _pin_based_exec_control = 0;
3601 u32 _cpu_based_exec_control = 0;
3602 u32 _cpu_based_2nd_exec_control = 0;
3603 u32 _vmexit_control = 0;
3604 u32 _vmentry_control = 0;
3606 min = CPU_BASED_HLT_EXITING |
3607 #ifdef CONFIG_X86_64
3608 CPU_BASED_CR8_LOAD_EXITING |
3609 CPU_BASED_CR8_STORE_EXITING |
3611 CPU_BASED_CR3_LOAD_EXITING |
3612 CPU_BASED_CR3_STORE_EXITING |
3613 CPU_BASED_USE_IO_BITMAPS |
3614 CPU_BASED_MOV_DR_EXITING |
3615 CPU_BASED_USE_TSC_OFFSETING |
3616 CPU_BASED_INVLPG_EXITING |
3617 CPU_BASED_RDPMC_EXITING;
3619 if (!kvm_mwait_in_guest())
3620 min |= CPU_BASED_MWAIT_EXITING |
3621 CPU_BASED_MONITOR_EXITING;
3623 opt = CPU_BASED_TPR_SHADOW |
3624 CPU_BASED_USE_MSR_BITMAPS |
3625 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3626 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3627 &_cpu_based_exec_control) < 0)
3629 #ifdef CONFIG_X86_64
3630 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3631 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3632 ~CPU_BASED_CR8_STORE_EXITING;
3634 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3636 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3637 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3638 SECONDARY_EXEC_WBINVD_EXITING |
3639 SECONDARY_EXEC_ENABLE_VPID |
3640 SECONDARY_EXEC_ENABLE_EPT |
3641 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3642 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3643 SECONDARY_EXEC_RDTSCP |
3644 SECONDARY_EXEC_ENABLE_INVPCID |
3645 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3646 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3647 SECONDARY_EXEC_SHADOW_VMCS |
3648 SECONDARY_EXEC_XSAVES |
3649 SECONDARY_EXEC_ENABLE_PML |
3650 SECONDARY_EXEC_TSC_SCALING;
3651 if (adjust_vmx_controls(min2, opt2,
3652 MSR_IA32_VMX_PROCBASED_CTLS2,
3653 &_cpu_based_2nd_exec_control) < 0)
3656 #ifndef CONFIG_X86_64
3657 if (!(_cpu_based_2nd_exec_control &
3658 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3659 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3662 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3663 _cpu_based_2nd_exec_control &= ~(
3664 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3665 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3666 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3668 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3669 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3671 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3672 CPU_BASED_CR3_STORE_EXITING |
3673 CPU_BASED_INVLPG_EXITING);
3674 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3675 vmx_capability.ept, vmx_capability.vpid);
3678 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
3679 #ifdef CONFIG_X86_64
3680 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3682 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3683 VM_EXIT_CLEAR_BNDCFGS;
3684 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3685 &_vmexit_control) < 0)
3688 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
3689 PIN_BASED_VIRTUAL_NMIS;
3690 opt = PIN_BASED_POSTED_INTR | PIN_BASED_VMX_PREEMPTION_TIMER;
3691 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3692 &_pin_based_exec_control) < 0)
3695 if (cpu_has_broken_vmx_preemption_timer())
3696 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3697 if (!(_cpu_based_2nd_exec_control &
3698 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
3699 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3701 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3702 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3703 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3704 &_vmentry_control) < 0)
3707 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3709 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3710 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3713 #ifdef CONFIG_X86_64
3714 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3715 if (vmx_msr_high & (1u<<16))
3719 /* Require Write-Back (WB) memory type for VMCS accesses. */
3720 if (((vmx_msr_high >> 18) & 15) != 6)
3723 vmcs_conf->size = vmx_msr_high & 0x1fff;
3724 vmcs_conf->order = get_order(vmcs_conf->size);
3725 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
3726 vmcs_conf->revision_id = vmx_msr_low;
3728 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3729 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3730 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3731 vmcs_conf->vmexit_ctrl = _vmexit_control;
3732 vmcs_conf->vmentry_ctrl = _vmentry_control;
3734 cpu_has_load_ia32_efer =
3735 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3736 VM_ENTRY_LOAD_IA32_EFER)
3737 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3738 VM_EXIT_LOAD_IA32_EFER);
3740 cpu_has_load_perf_global_ctrl =
3741 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3742 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3743 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3744 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3747 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3748 * but due to errata below it can't be used. Workaround is to use
3749 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3751 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3756 * BC86,AAY89,BD102 (model 44)
3760 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3761 switch (boot_cpu_data.x86_model) {
3767 cpu_has_load_perf_global_ctrl = false;
3768 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3769 "does not work properly. Using workaround\n");
3776 if (boot_cpu_has(X86_FEATURE_XSAVES))
3777 rdmsrl(MSR_IA32_XSS, host_xss);
3782 static struct vmcs *alloc_vmcs_cpu(int cpu)
3784 int node = cpu_to_node(cpu);
3788 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3791 vmcs = page_address(pages);
3792 memset(vmcs, 0, vmcs_config.size);
3793 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3797 static struct vmcs *alloc_vmcs(void)
3799 return alloc_vmcs_cpu(raw_smp_processor_id());
3802 static void free_vmcs(struct vmcs *vmcs)
3804 free_pages((unsigned long)vmcs, vmcs_config.order);
3808 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3810 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3812 if (!loaded_vmcs->vmcs)
3814 loaded_vmcs_clear(loaded_vmcs);
3815 free_vmcs(loaded_vmcs->vmcs);
3816 loaded_vmcs->vmcs = NULL;
3817 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
3820 static void free_kvm_area(void)
3824 for_each_possible_cpu(cpu) {
3825 free_vmcs(per_cpu(vmxarea, cpu));
3826 per_cpu(vmxarea, cpu) = NULL;
3830 enum vmcs_field_type {
3831 VMCS_FIELD_TYPE_U16 = 0,
3832 VMCS_FIELD_TYPE_U64 = 1,
3833 VMCS_FIELD_TYPE_U32 = 2,
3834 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
3837 static inline int vmcs_field_type(unsigned long field)
3839 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
3840 return VMCS_FIELD_TYPE_U32;
3841 return (field >> 13) & 0x3 ;
3844 static inline int vmcs_field_readonly(unsigned long field)
3846 return (((field >> 10) & 0x3) == 1);
3849 static void init_vmcs_shadow_fields(void)
3853 /* No checks for read only fields yet */
3855 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3856 switch (shadow_read_write_fields[i]) {
3858 if (!kvm_mpx_supported())
3866 shadow_read_write_fields[j] =
3867 shadow_read_write_fields[i];
3870 max_shadow_read_write_fields = j;
3872 /* shadowed fields guest access without vmexit */
3873 for (i = 0; i < max_shadow_read_write_fields; i++) {
3874 unsigned long field = shadow_read_write_fields[i];
3876 clear_bit(field, vmx_vmwrite_bitmap);
3877 clear_bit(field, vmx_vmread_bitmap);
3878 if (vmcs_field_type(field) == VMCS_FIELD_TYPE_U64) {
3879 clear_bit(field + 1, vmx_vmwrite_bitmap);
3880 clear_bit(field + 1, vmx_vmread_bitmap);
3883 for (i = 0; i < max_shadow_read_only_fields; i++) {
3884 unsigned long field = shadow_read_only_fields[i];
3886 clear_bit(field, vmx_vmread_bitmap);
3887 if (vmcs_field_type(field) == VMCS_FIELD_TYPE_U64)
3888 clear_bit(field + 1, vmx_vmread_bitmap);
3892 static __init int alloc_kvm_area(void)
3896 for_each_possible_cpu(cpu) {
3899 vmcs = alloc_vmcs_cpu(cpu);
3905 per_cpu(vmxarea, cpu) = vmcs;
3910 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3911 struct kvm_segment *save)
3913 if (!emulate_invalid_guest_state) {
3915 * CS and SS RPL should be equal during guest entry according
3916 * to VMX spec, but in reality it is not always so. Since vcpu
3917 * is in the middle of the transition from real mode to
3918 * protected mode it is safe to assume that RPL 0 is a good
3921 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3922 save->selector &= ~SEGMENT_RPL_MASK;
3923 save->dpl = save->selector & SEGMENT_RPL_MASK;
3926 vmx_set_segment(vcpu, save, seg);
3929 static void enter_pmode(struct kvm_vcpu *vcpu)
3931 unsigned long flags;
3932 struct vcpu_vmx *vmx = to_vmx(vcpu);
3935 * Update real mode segment cache. It may be not up-to-date if sement
3936 * register was written while vcpu was in a guest mode.
3938 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3939 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3940 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3941 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3942 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3943 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3945 vmx->rmode.vm86_active = 0;
3947 vmx_segment_cache_clear(vmx);
3949 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3951 flags = vmcs_readl(GUEST_RFLAGS);
3952 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3953 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3954 vmcs_writel(GUEST_RFLAGS, flags);
3956 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3957 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3959 update_exception_bitmap(vcpu);
3961 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3962 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3963 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3964 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3965 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3966 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3969 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3971 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3972 struct kvm_segment var = *save;
3975 if (seg == VCPU_SREG_CS)
3978 if (!emulate_invalid_guest_state) {
3979 var.selector = var.base >> 4;
3980 var.base = var.base & 0xffff0;
3990 if (save->base & 0xf)
3991 printk_once(KERN_WARNING "kvm: segment base is not "
3992 "paragraph aligned when entering "
3993 "protected mode (seg=%d)", seg);
3996 vmcs_write16(sf->selector, var.selector);
3997 vmcs_writel(sf->base, var.base);
3998 vmcs_write32(sf->limit, var.limit);
3999 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
4002 static void enter_rmode(struct kvm_vcpu *vcpu)
4004 unsigned long flags;
4005 struct vcpu_vmx *vmx = to_vmx(vcpu);
4007 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
4008 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
4009 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
4010 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
4011 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
4012 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
4013 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
4015 vmx->rmode.vm86_active = 1;
4018 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
4019 * vcpu. Warn the user that an update is overdue.
4021 if (!vcpu->kvm->arch.tss_addr)
4022 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
4023 "called before entering vcpu\n");
4025 vmx_segment_cache_clear(vmx);
4027 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
4028 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
4029 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4031 flags = vmcs_readl(GUEST_RFLAGS);
4032 vmx->rmode.save_rflags = flags;
4034 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
4036 vmcs_writel(GUEST_RFLAGS, flags);
4037 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
4038 update_exception_bitmap(vcpu);
4040 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
4041 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
4042 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
4043 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
4044 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
4045 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
4047 kvm_mmu_reset_context(vcpu);
4050 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
4052 struct vcpu_vmx *vmx = to_vmx(vcpu);
4053 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
4059 * Force kernel_gs_base reloading before EFER changes, as control
4060 * of this msr depends on is_long_mode().
4062 vmx_load_host_state(to_vmx(vcpu));
4063 vcpu->arch.efer = efer;
4064 if (efer & EFER_LMA) {
4065 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4068 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4070 msr->data = efer & ~EFER_LME;
4075 #ifdef CONFIG_X86_64
4077 static void enter_lmode(struct kvm_vcpu *vcpu)
4081 vmx_segment_cache_clear(to_vmx(vcpu));
4083 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
4084 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
4085 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
4087 vmcs_write32(GUEST_TR_AR_BYTES,
4088 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
4089 | VMX_AR_TYPE_BUSY_64_TSS);
4091 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
4094 static void exit_lmode(struct kvm_vcpu *vcpu)
4096 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4097 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
4102 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
4105 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
4107 ept_sync_context(construct_eptp(vcpu, vcpu->arch.mmu.root_hpa));
4109 vpid_sync_context(vpid);
4113 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
4115 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
4118 static void vmx_flush_tlb_ept_only(struct kvm_vcpu *vcpu)
4121 vmx_flush_tlb(vcpu);
4124 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
4126 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
4128 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
4129 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
4132 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
4134 if (enable_ept && is_paging(vcpu))
4135 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4136 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
4139 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
4141 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
4143 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
4144 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
4147 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
4149 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4151 if (!test_bit(VCPU_EXREG_PDPTR,
4152 (unsigned long *)&vcpu->arch.regs_dirty))
4155 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4156 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
4157 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
4158 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
4159 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
4163 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
4165 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4167 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4168 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
4169 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
4170 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
4171 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
4174 __set_bit(VCPU_EXREG_PDPTR,
4175 (unsigned long *)&vcpu->arch.regs_avail);
4176 __set_bit(VCPU_EXREG_PDPTR,
4177 (unsigned long *)&vcpu->arch.regs_dirty);
4180 static bool nested_guest_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4182 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4183 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4184 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4186 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
4187 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4188 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4189 fixed0 &= ~(X86_CR0_PE | X86_CR0_PG);
4191 return fixed_bits_valid(val, fixed0, fixed1);
4194 static bool nested_host_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4196 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4197 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4199 return fixed_bits_valid(val, fixed0, fixed1);
4202 static bool nested_cr4_valid(struct kvm_vcpu *vcpu, unsigned long val)
4204 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed0;
4205 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed1;
4207 return fixed_bits_valid(val, fixed0, fixed1);
4210 /* No difference in the restrictions on guest and host CR4 in VMX operation. */
4211 #define nested_guest_cr4_valid nested_cr4_valid
4212 #define nested_host_cr4_valid nested_cr4_valid
4214 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
4216 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
4218 struct kvm_vcpu *vcpu)
4220 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
4221 vmx_decache_cr3(vcpu);
4222 if (!(cr0 & X86_CR0_PG)) {
4223 /* From paging/starting to nonpaging */
4224 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4225 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
4226 (CPU_BASED_CR3_LOAD_EXITING |
4227 CPU_BASED_CR3_STORE_EXITING));
4228 vcpu->arch.cr0 = cr0;
4229 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4230 } else if (!is_paging(vcpu)) {
4231 /* From nonpaging to paging */
4232 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4233 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
4234 ~(CPU_BASED_CR3_LOAD_EXITING |
4235 CPU_BASED_CR3_STORE_EXITING));
4236 vcpu->arch.cr0 = cr0;
4237 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4240 if (!(cr0 & X86_CR0_WP))
4241 *hw_cr0 &= ~X86_CR0_WP;
4244 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
4246 struct vcpu_vmx *vmx = to_vmx(vcpu);
4247 unsigned long hw_cr0;
4249 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
4250 if (enable_unrestricted_guest)
4251 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
4253 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
4255 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
4258 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
4262 #ifdef CONFIG_X86_64
4263 if (vcpu->arch.efer & EFER_LME) {
4264 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
4266 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
4272 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
4274 vmcs_writel(CR0_READ_SHADOW, cr0);
4275 vmcs_writel(GUEST_CR0, hw_cr0);
4276 vcpu->arch.cr0 = cr0;
4278 /* depends on vcpu->arch.cr0 to be set to a new value */
4279 vmx->emulation_required = emulation_required(vcpu);
4282 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
4286 /* TODO write the value reading from MSR */
4287 eptp = VMX_EPT_DEFAULT_MT |
4288 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
4289 if (enable_ept_ad_bits &&
4290 (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
4291 eptp |= VMX_EPT_AD_ENABLE_BIT;
4292 eptp |= (root_hpa & PAGE_MASK);
4297 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
4299 unsigned long guest_cr3;
4304 eptp = construct_eptp(vcpu, cr3);
4305 vmcs_write64(EPT_POINTER, eptp);
4306 if (is_paging(vcpu) || is_guest_mode(vcpu))
4307 guest_cr3 = kvm_read_cr3(vcpu);
4309 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
4310 ept_load_pdptrs(vcpu);
4313 vmx_flush_tlb(vcpu);
4314 vmcs_writel(GUEST_CR3, guest_cr3);
4317 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
4320 * Pass through host's Machine Check Enable value to hw_cr4, which
4321 * is in force while we are in guest mode. Do not let guests control
4322 * this bit, even if host CR4.MCE == 0.
4324 unsigned long hw_cr4 =
4325 (cr4_read_shadow() & X86_CR4_MCE) |
4326 (cr4 & ~X86_CR4_MCE) |
4327 (to_vmx(vcpu)->rmode.vm86_active ?
4328 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
4330 if (cr4 & X86_CR4_VMXE) {
4332 * To use VMXON (and later other VMX instructions), a guest
4333 * must first be able to turn on cr4.VMXE (see handle_vmon()).
4334 * So basically the check on whether to allow nested VMX
4337 if (!nested_vmx_allowed(vcpu))
4341 if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
4344 vcpu->arch.cr4 = cr4;
4346 if (!is_paging(vcpu)) {
4347 hw_cr4 &= ~X86_CR4_PAE;
4348 hw_cr4 |= X86_CR4_PSE;
4349 } else if (!(cr4 & X86_CR4_PAE)) {
4350 hw_cr4 &= ~X86_CR4_PAE;
4354 if (!enable_unrestricted_guest && !is_paging(vcpu))
4356 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4357 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
4358 * to be manually disabled when guest switches to non-paging
4361 * If !enable_unrestricted_guest, the CPU is always running
4362 * with CR0.PG=1 and CR4 needs to be modified.
4363 * If enable_unrestricted_guest, the CPU automatically
4364 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4366 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
4368 vmcs_writel(CR4_READ_SHADOW, cr4);
4369 vmcs_writel(GUEST_CR4, hw_cr4);
4373 static void vmx_get_segment(struct kvm_vcpu *vcpu,
4374 struct kvm_segment *var, int seg)
4376 struct vcpu_vmx *vmx = to_vmx(vcpu);
4379 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4380 *var = vmx->rmode.segs[seg];
4381 if (seg == VCPU_SREG_TR
4382 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
4384 var->base = vmx_read_guest_seg_base(vmx, seg);
4385 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4388 var->base = vmx_read_guest_seg_base(vmx, seg);
4389 var->limit = vmx_read_guest_seg_limit(vmx, seg);
4390 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4391 ar = vmx_read_guest_seg_ar(vmx, seg);
4392 var->unusable = (ar >> 16) & 1;
4393 var->type = ar & 15;
4394 var->s = (ar >> 4) & 1;
4395 var->dpl = (ar >> 5) & 3;
4397 * Some userspaces do not preserve unusable property. Since usable
4398 * segment has to be present according to VMX spec we can use present
4399 * property to amend userspace bug by making unusable segment always
4400 * nonpresent. vmx_segment_access_rights() already marks nonpresent
4401 * segment as unusable.
4403 var->present = !var->unusable;
4404 var->avl = (ar >> 12) & 1;
4405 var->l = (ar >> 13) & 1;
4406 var->db = (ar >> 14) & 1;
4407 var->g = (ar >> 15) & 1;
4410 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
4412 struct kvm_segment s;
4414 if (to_vmx(vcpu)->rmode.vm86_active) {
4415 vmx_get_segment(vcpu, &s, seg);
4418 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
4421 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
4423 struct vcpu_vmx *vmx = to_vmx(vcpu);
4425 if (unlikely(vmx->rmode.vm86_active))
4428 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
4429 return VMX_AR_DPL(ar);
4433 static u32 vmx_segment_access_rights(struct kvm_segment *var)
4437 if (var->unusable || !var->present)
4440 ar = var->type & 15;
4441 ar |= (var->s & 1) << 4;
4442 ar |= (var->dpl & 3) << 5;
4443 ar |= (var->present & 1) << 7;
4444 ar |= (var->avl & 1) << 12;
4445 ar |= (var->l & 1) << 13;
4446 ar |= (var->db & 1) << 14;
4447 ar |= (var->g & 1) << 15;
4453 static void vmx_set_segment(struct kvm_vcpu *vcpu,
4454 struct kvm_segment *var, int seg)
4456 struct vcpu_vmx *vmx = to_vmx(vcpu);
4457 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4459 vmx_segment_cache_clear(vmx);
4461 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4462 vmx->rmode.segs[seg] = *var;
4463 if (seg == VCPU_SREG_TR)
4464 vmcs_write16(sf->selector, var->selector);
4466 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
4470 vmcs_writel(sf->base, var->base);
4471 vmcs_write32(sf->limit, var->limit);
4472 vmcs_write16(sf->selector, var->selector);
4475 * Fix the "Accessed" bit in AR field of segment registers for older
4477 * IA32 arch specifies that at the time of processor reset the
4478 * "Accessed" bit in the AR field of segment registers is 1. And qemu
4479 * is setting it to 0 in the userland code. This causes invalid guest
4480 * state vmexit when "unrestricted guest" mode is turned on.
4481 * Fix for this setup issue in cpu_reset is being pushed in the qemu
4482 * tree. Newer qemu binaries with that qemu fix would not need this
4485 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
4486 var->type |= 0x1; /* Accessed */
4488 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4491 vmx->emulation_required = emulation_required(vcpu);
4494 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4496 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4498 *db = (ar >> 14) & 1;
4499 *l = (ar >> 13) & 1;
4502 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4504 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4505 dt->address = vmcs_readl(GUEST_IDTR_BASE);
4508 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4510 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4511 vmcs_writel(GUEST_IDTR_BASE, dt->address);
4514 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4516 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4517 dt->address = vmcs_readl(GUEST_GDTR_BASE);
4520 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4522 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4523 vmcs_writel(GUEST_GDTR_BASE, dt->address);
4526 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4528 struct kvm_segment var;
4531 vmx_get_segment(vcpu, &var, seg);
4533 if (seg == VCPU_SREG_CS)
4535 ar = vmx_segment_access_rights(&var);
4537 if (var.base != (var.selector << 4))
4539 if (var.limit != 0xffff)
4547 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4549 struct kvm_segment cs;
4550 unsigned int cs_rpl;
4552 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4553 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4557 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4561 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4562 if (cs.dpl > cs_rpl)
4565 if (cs.dpl != cs_rpl)
4571 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4575 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4577 struct kvm_segment ss;
4578 unsigned int ss_rpl;
4580 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4581 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4585 if (ss.type != 3 && ss.type != 7)
4589 if (ss.dpl != ss_rpl) /* DPL != RPL */
4597 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4599 struct kvm_segment var;
4602 vmx_get_segment(vcpu, &var, seg);
4603 rpl = var.selector & SEGMENT_RPL_MASK;
4611 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4612 if (var.dpl < rpl) /* DPL < RPL */
4616 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4622 static bool tr_valid(struct kvm_vcpu *vcpu)
4624 struct kvm_segment tr;
4626 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4630 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4632 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4640 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4642 struct kvm_segment ldtr;
4644 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4648 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4658 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4660 struct kvm_segment cs, ss;
4662 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4663 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4665 return ((cs.selector & SEGMENT_RPL_MASK) ==
4666 (ss.selector & SEGMENT_RPL_MASK));
4670 * Check if guest state is valid. Returns true if valid, false if
4672 * We assume that registers are always usable
4674 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4676 if (enable_unrestricted_guest)
4679 /* real mode guest state checks */
4680 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4681 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4683 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4685 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4687 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4689 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4691 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4694 /* protected mode guest state checks */
4695 if (!cs_ss_rpl_check(vcpu))
4697 if (!code_segment_valid(vcpu))
4699 if (!stack_segment_valid(vcpu))
4701 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4703 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4705 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4707 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4709 if (!tr_valid(vcpu))
4711 if (!ldtr_valid(vcpu))
4715 * - Add checks on RIP
4716 * - Add checks on RFLAGS
4722 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
4724 return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
4727 static int init_rmode_tss(struct kvm *kvm)
4733 idx = srcu_read_lock(&kvm->srcu);
4734 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4735 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4738 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4739 r = kvm_write_guest_page(kvm, fn++, &data,
4740 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4743 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4746 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4750 r = kvm_write_guest_page(kvm, fn, &data,
4751 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4754 srcu_read_unlock(&kvm->srcu, idx);
4758 static int init_rmode_identity_map(struct kvm *kvm)
4761 kvm_pfn_t identity_map_pfn;
4767 /* Protect kvm->arch.ept_identity_pagetable_done. */
4768 mutex_lock(&kvm->slots_lock);
4770 if (likely(kvm->arch.ept_identity_pagetable_done))
4773 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4775 r = alloc_identity_pagetable(kvm);
4779 idx = srcu_read_lock(&kvm->srcu);
4780 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4783 /* Set up identity-mapping pagetable for EPT in real mode */
4784 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4785 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4786 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4787 r = kvm_write_guest_page(kvm, identity_map_pfn,
4788 &tmp, i * sizeof(tmp), sizeof(tmp));
4792 kvm->arch.ept_identity_pagetable_done = true;
4795 srcu_read_unlock(&kvm->srcu, idx);
4798 mutex_unlock(&kvm->slots_lock);
4802 static void seg_setup(int seg)
4804 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4807 vmcs_write16(sf->selector, 0);
4808 vmcs_writel(sf->base, 0);
4809 vmcs_write32(sf->limit, 0xffff);
4811 if (seg == VCPU_SREG_CS)
4812 ar |= 0x08; /* code segment */
4814 vmcs_write32(sf->ar_bytes, ar);
4817 static int alloc_apic_access_page(struct kvm *kvm)
4822 mutex_lock(&kvm->slots_lock);
4823 if (kvm->arch.apic_access_page_done)
4825 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4826 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4830 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4831 if (is_error_page(page)) {
4837 * Do not pin the page in memory, so that memory hot-unplug
4838 * is able to migrate it.
4841 kvm->arch.apic_access_page_done = true;
4843 mutex_unlock(&kvm->slots_lock);
4847 static int alloc_identity_pagetable(struct kvm *kvm)
4849 /* Called with kvm->slots_lock held. */
4853 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4855 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4856 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4861 static int allocate_vpid(void)
4867 spin_lock(&vmx_vpid_lock);
4868 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4869 if (vpid < VMX_NR_VPIDS)
4870 __set_bit(vpid, vmx_vpid_bitmap);
4873 spin_unlock(&vmx_vpid_lock);
4877 static void free_vpid(int vpid)
4879 if (!enable_vpid || vpid == 0)
4881 spin_lock(&vmx_vpid_lock);
4882 __clear_bit(vpid, vmx_vpid_bitmap);
4883 spin_unlock(&vmx_vpid_lock);
4886 #define MSR_TYPE_R 1
4887 #define MSR_TYPE_W 2
4888 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4891 int f = sizeof(unsigned long);
4893 if (!cpu_has_vmx_msr_bitmap())
4897 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4898 * have the write-low and read-high bitmap offsets the wrong way round.
4899 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4901 if (msr <= 0x1fff) {
4902 if (type & MSR_TYPE_R)
4904 __clear_bit(msr, msr_bitmap + 0x000 / f);
4906 if (type & MSR_TYPE_W)
4908 __clear_bit(msr, msr_bitmap + 0x800 / f);
4910 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4912 if (type & MSR_TYPE_R)
4914 __clear_bit(msr, msr_bitmap + 0x400 / f);
4916 if (type & MSR_TYPE_W)
4918 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4924 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4925 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4927 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4928 unsigned long *msr_bitmap_nested,
4931 int f = sizeof(unsigned long);
4933 if (!cpu_has_vmx_msr_bitmap()) {
4939 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4940 * have the write-low and read-high bitmap offsets the wrong way round.
4941 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4943 if (msr <= 0x1fff) {
4944 if (type & MSR_TYPE_R &&
4945 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4947 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4949 if (type & MSR_TYPE_W &&
4950 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4952 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4954 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4956 if (type & MSR_TYPE_R &&
4957 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4959 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4961 if (type & MSR_TYPE_W &&
4962 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4964 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4969 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4972 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4973 msr, MSR_TYPE_R | MSR_TYPE_W);
4974 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4975 msr, MSR_TYPE_R | MSR_TYPE_W);
4978 static void vmx_disable_intercept_msr_x2apic(u32 msr, int type, bool apicv_active)
4981 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic_apicv,
4983 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic_apicv,
4986 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4988 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4993 static bool vmx_get_enable_apicv(void)
4995 return enable_apicv;
4998 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
5000 struct vcpu_vmx *vmx = to_vmx(vcpu);
5005 if (vmx->nested.pi_desc &&
5006 vmx->nested.pi_pending) {
5007 vmx->nested.pi_pending = false;
5008 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
5011 max_irr = find_last_bit(
5012 (unsigned long *)vmx->nested.pi_desc->pir, 256);
5017 vapic_page = kmap(vmx->nested.virtual_apic_page);
5018 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
5019 kunmap(vmx->nested.virtual_apic_page);
5021 status = vmcs_read16(GUEST_INTR_STATUS);
5022 if ((u8)max_irr > ((u8)status & 0xff)) {
5024 status |= (u8)max_irr;
5025 vmcs_write16(GUEST_INTR_STATUS, status);
5030 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
5034 int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
5036 if (vcpu->mode == IN_GUEST_MODE) {
5037 struct vcpu_vmx *vmx = to_vmx(vcpu);
5040 * Currently, we don't support urgent interrupt,
5041 * all interrupts are recognized as non-urgent
5042 * interrupt, so we cannot post interrupts when
5045 * If the vcpu is in guest mode, it means it is
5046 * running instead of being scheduled out and
5047 * waiting in the run queue, and that's the only
5048 * case when 'SN' is set currently, warning if
5051 WARN_ON_ONCE(pi_test_sn(&vmx->pi_desc));
5053 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
5060 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
5063 struct vcpu_vmx *vmx = to_vmx(vcpu);
5065 if (is_guest_mode(vcpu) &&
5066 vector == vmx->nested.posted_intr_nv) {
5067 /* the PIR and ON have been set by L1. */
5068 kvm_vcpu_trigger_posted_interrupt(vcpu, true);
5070 * If a posted intr is not recognized by hardware,
5071 * we will accomplish it in the next vmentry.
5073 vmx->nested.pi_pending = true;
5074 kvm_make_request(KVM_REQ_EVENT, vcpu);
5080 * Send interrupt to vcpu via posted interrupt way.
5081 * 1. If target vcpu is running(non-root mode), send posted interrupt
5082 * notification to vcpu and hardware will sync PIR to vIRR atomically.
5083 * 2. If target vcpu isn't running(root mode), kick it to pick up the
5084 * interrupt from PIR in next vmentry.
5086 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
5088 struct vcpu_vmx *vmx = to_vmx(vcpu);
5091 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
5095 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
5098 /* If a previous notification has sent the IPI, nothing to do. */
5099 if (pi_test_and_set_on(&vmx->pi_desc))
5102 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
5103 kvm_vcpu_kick(vcpu);
5107 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
5108 * will not change in the lifetime of the guest.
5109 * Note that host-state that does change is set elsewhere. E.g., host-state
5110 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
5112 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
5117 unsigned long cr0, cr3, cr4;
5120 WARN_ON(cr0 & X86_CR0_TS);
5121 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
5124 * Save the most likely value for this task's CR3 in the VMCS.
5125 * We can't use __get_current_cr3_fast() because we're not atomic.
5128 vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */
5129 vmx->host_state.vmcs_host_cr3 = cr3;
5131 /* Save the most likely value for this task's CR4 in the VMCS. */
5132 cr4 = cr4_read_shadow();
5133 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
5134 vmx->host_state.vmcs_host_cr4 = cr4;
5136 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
5137 #ifdef CONFIG_X86_64
5139 * Load null selectors, so we can avoid reloading them in
5140 * __vmx_load_host_state(), in case userspace uses the null selectors
5141 * too (the expected case).
5143 vmcs_write16(HOST_DS_SELECTOR, 0);
5144 vmcs_write16(HOST_ES_SELECTOR, 0);
5146 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5147 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5149 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5150 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
5152 native_store_idt(&dt);
5153 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
5154 vmx->host_idt_base = dt.address;
5156 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
5158 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
5159 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
5160 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
5161 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
5163 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
5164 rdmsr(MSR_IA32_CR_PAT, low32, high32);
5165 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
5169 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
5171 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
5173 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
5174 if (is_guest_mode(&vmx->vcpu))
5175 vmx->vcpu.arch.cr4_guest_owned_bits &=
5176 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
5177 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
5180 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
5182 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
5184 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5185 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
5186 /* Enable the preemption timer dynamically */
5187 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
5188 return pin_based_exec_ctrl;
5191 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5193 struct vcpu_vmx *vmx = to_vmx(vcpu);
5195 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5196 if (cpu_has_secondary_exec_ctrls()) {
5197 if (kvm_vcpu_apicv_active(vcpu))
5198 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
5199 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5200 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5202 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
5203 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5204 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5207 if (cpu_has_vmx_msr_bitmap())
5208 vmx_set_msr_bitmap(vcpu);
5211 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
5213 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
5215 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
5216 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5218 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
5219 exec_control &= ~CPU_BASED_TPR_SHADOW;
5220 #ifdef CONFIG_X86_64
5221 exec_control |= CPU_BASED_CR8_STORE_EXITING |
5222 CPU_BASED_CR8_LOAD_EXITING;
5226 exec_control |= CPU_BASED_CR3_STORE_EXITING |
5227 CPU_BASED_CR3_LOAD_EXITING |
5228 CPU_BASED_INVLPG_EXITING;
5229 return exec_control;
5232 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
5234 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
5235 if (!cpu_need_virtualize_apic_accesses(&vmx->vcpu))
5236 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5238 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
5240 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
5241 enable_unrestricted_guest = 0;
5242 /* Enable INVPCID for non-ept guests may cause performance regression. */
5243 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5245 if (!enable_unrestricted_guest)
5246 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
5248 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
5249 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5250 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
5251 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5252 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
5253 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
5255 We can NOT enable shadow_vmcs here because we don't have yet
5258 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
5261 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
5263 return exec_control;
5266 static void ept_set_mmio_spte_mask(void)
5269 * EPT Misconfigurations can be generated if the value of bits 2:0
5270 * of an EPT paging-structure entry is 110b (write/execute).
5272 kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
5273 VMX_EPT_MISCONFIG_WX_VALUE);
5276 #define VMX_XSS_EXIT_BITMAP 0
5278 * Sets up the vmcs for emulated real mode.
5280 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
5282 #ifdef CONFIG_X86_64
5288 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
5289 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
5291 if (enable_shadow_vmcs) {
5292 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5293 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5295 if (cpu_has_vmx_msr_bitmap())
5296 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
5298 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
5301 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5302 vmx->hv_deadline_tsc = -1;
5304 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
5306 if (cpu_has_secondary_exec_ctrls()) {
5307 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5308 vmx_secondary_exec_control(vmx));
5311 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
5312 vmcs_write64(EOI_EXIT_BITMAP0, 0);
5313 vmcs_write64(EOI_EXIT_BITMAP1, 0);
5314 vmcs_write64(EOI_EXIT_BITMAP2, 0);
5315 vmcs_write64(EOI_EXIT_BITMAP3, 0);
5317 vmcs_write16(GUEST_INTR_STATUS, 0);
5319 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
5320 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
5324 vmcs_write32(PLE_GAP, ple_gap);
5325 vmx->ple_window = ple_window;
5326 vmx->ple_window_dirty = true;
5329 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
5330 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
5331 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
5333 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
5334 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
5335 vmx_set_constant_host_state(vmx);
5336 #ifdef CONFIG_X86_64
5337 rdmsrl(MSR_FS_BASE, a);
5338 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
5339 rdmsrl(MSR_GS_BASE, a);
5340 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
5342 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
5343 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
5346 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
5347 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
5348 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
5349 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
5350 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
5352 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
5353 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
5355 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
5356 u32 index = vmx_msr_index[i];
5357 u32 data_low, data_high;
5360 if (rdmsr_safe(index, &data_low, &data_high) < 0)
5362 if (wrmsr_safe(index, data_low, data_high) < 0)
5364 vmx->guest_msrs[j].index = i;
5365 vmx->guest_msrs[j].data = 0;
5366 vmx->guest_msrs[j].mask = -1ull;
5371 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
5373 /* 22.2.1, 20.8.1 */
5374 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
5376 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
5377 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
5379 set_cr4_guest_host_mask(vmx);
5381 if (vmx_xsaves_supported())
5382 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
5385 ASSERT(vmx->pml_pg);
5386 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
5387 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5393 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
5395 struct vcpu_vmx *vmx = to_vmx(vcpu);
5396 struct msr_data apic_base_msr;
5399 vmx->rmode.vm86_active = 0;
5401 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
5402 kvm_set_cr8(vcpu, 0);
5405 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
5406 MSR_IA32_APICBASE_ENABLE;
5407 if (kvm_vcpu_is_reset_bsp(vcpu))
5408 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
5409 apic_base_msr.host_initiated = true;
5410 kvm_set_apic_base(vcpu, &apic_base_msr);
5413 vmx_segment_cache_clear(vmx);
5415 seg_setup(VCPU_SREG_CS);
5416 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
5417 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
5419 seg_setup(VCPU_SREG_DS);
5420 seg_setup(VCPU_SREG_ES);
5421 seg_setup(VCPU_SREG_FS);
5422 seg_setup(VCPU_SREG_GS);
5423 seg_setup(VCPU_SREG_SS);
5425 vmcs_write16(GUEST_TR_SELECTOR, 0);
5426 vmcs_writel(GUEST_TR_BASE, 0);
5427 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
5428 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
5430 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
5431 vmcs_writel(GUEST_LDTR_BASE, 0);
5432 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
5433 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
5436 vmcs_write32(GUEST_SYSENTER_CS, 0);
5437 vmcs_writel(GUEST_SYSENTER_ESP, 0);
5438 vmcs_writel(GUEST_SYSENTER_EIP, 0);
5439 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
5442 vmcs_writel(GUEST_RFLAGS, 0x02);
5443 kvm_rip_write(vcpu, 0xfff0);
5445 vmcs_writel(GUEST_GDTR_BASE, 0);
5446 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
5448 vmcs_writel(GUEST_IDTR_BASE, 0);
5449 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
5451 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
5452 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
5453 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
5457 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
5459 if (cpu_has_vmx_tpr_shadow() && !init_event) {
5460 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
5461 if (cpu_need_tpr_shadow(vcpu))
5462 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
5463 __pa(vcpu->arch.apic->regs));
5464 vmcs_write32(TPR_THRESHOLD, 0);
5467 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5469 if (kvm_vcpu_apicv_active(vcpu))
5470 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
5473 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5475 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5476 vmx->vcpu.arch.cr0 = cr0;
5477 vmx_set_cr0(vcpu, cr0); /* enter rmode */
5478 vmx_set_cr4(vcpu, 0);
5479 vmx_set_efer(vcpu, 0);
5481 update_exception_bitmap(vcpu);
5483 vpid_sync_context(vmx->vpid);
5487 * In nested virtualization, check if L1 asked to exit on external interrupts.
5488 * For most existing hypervisors, this will always return true.
5490 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5492 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5493 PIN_BASED_EXT_INTR_MASK;
5497 * In nested virtualization, check if L1 has set
5498 * VM_EXIT_ACK_INTR_ON_EXIT
5500 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5502 return get_vmcs12(vcpu)->vm_exit_controls &
5503 VM_EXIT_ACK_INTR_ON_EXIT;
5506 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5508 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5509 PIN_BASED_NMI_EXITING;
5512 static void enable_irq_window(struct kvm_vcpu *vcpu)
5514 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5515 CPU_BASED_VIRTUAL_INTR_PENDING);
5518 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5520 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5521 enable_irq_window(vcpu);
5525 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5526 CPU_BASED_VIRTUAL_NMI_PENDING);
5529 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5531 struct vcpu_vmx *vmx = to_vmx(vcpu);
5533 int irq = vcpu->arch.interrupt.nr;
5535 trace_kvm_inj_virq(irq);
5537 ++vcpu->stat.irq_injections;
5538 if (vmx->rmode.vm86_active) {
5540 if (vcpu->arch.interrupt.soft)
5541 inc_eip = vcpu->arch.event_exit_inst_len;
5542 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5543 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5546 intr = irq | INTR_INFO_VALID_MASK;
5547 if (vcpu->arch.interrupt.soft) {
5548 intr |= INTR_TYPE_SOFT_INTR;
5549 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5550 vmx->vcpu.arch.event_exit_inst_len);
5552 intr |= INTR_TYPE_EXT_INTR;
5553 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5556 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5558 struct vcpu_vmx *vmx = to_vmx(vcpu);
5560 ++vcpu->stat.nmi_injections;
5561 vmx->loaded_vmcs->nmi_known_unmasked = false;
5563 if (vmx->rmode.vm86_active) {
5564 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5565 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5569 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5570 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5573 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5575 struct vcpu_vmx *vmx = to_vmx(vcpu);
5578 if (vmx->loaded_vmcs->nmi_known_unmasked)
5580 masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5581 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5585 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5587 struct vcpu_vmx *vmx = to_vmx(vcpu);
5589 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5591 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5592 GUEST_INTR_STATE_NMI);
5594 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5595 GUEST_INTR_STATE_NMI);
5598 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5600 if (to_vmx(vcpu)->nested.nested_run_pending)
5603 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5604 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5605 | GUEST_INTR_STATE_NMI));
5608 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5610 return (!to_vmx(vcpu)->nested.nested_run_pending &&
5611 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5612 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5613 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5616 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5620 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5624 kvm->arch.tss_addr = addr;
5625 return init_rmode_tss(kvm);
5628 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5633 * Update instruction length as we may reinject the exception
5634 * from user space while in guest debugging mode.
5636 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5637 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5638 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5642 if (vcpu->guest_debug &
5643 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5660 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5661 int vec, u32 err_code)
5664 * Instruction with address size override prefix opcode 0x67
5665 * Cause the #SS fault with 0 error code in VM86 mode.
5667 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5668 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5669 if (vcpu->arch.halt_request) {
5670 vcpu->arch.halt_request = 0;
5671 return kvm_vcpu_halt(vcpu);
5679 * Forward all other exceptions that are valid in real mode.
5680 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5681 * the required debugging infrastructure rework.
5683 kvm_queue_exception(vcpu, vec);
5688 * Trigger machine check on the host. We assume all the MSRs are already set up
5689 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5690 * We pass a fake environment to the machine check handler because we want
5691 * the guest to be always treated like user space, no matter what context
5692 * it used internally.
5694 static void kvm_machine_check(void)
5696 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5697 struct pt_regs regs = {
5698 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5699 .flags = X86_EFLAGS_IF,
5702 do_machine_check(®s, 0);
5706 static int handle_machine_check(struct kvm_vcpu *vcpu)
5708 /* already handled by vcpu_run */
5712 static int handle_exception(struct kvm_vcpu *vcpu)
5714 struct vcpu_vmx *vmx = to_vmx(vcpu);
5715 struct kvm_run *kvm_run = vcpu->run;
5716 u32 intr_info, ex_no, error_code;
5717 unsigned long cr2, rip, dr6;
5719 enum emulation_result er;
5721 vect_info = vmx->idt_vectoring_info;
5722 intr_info = vmx->exit_intr_info;
5724 if (is_machine_check(intr_info))
5725 return handle_machine_check(vcpu);
5727 if (is_nmi(intr_info))
5728 return 1; /* already handled by vmx_vcpu_run() */
5730 if (is_invalid_opcode(intr_info)) {
5731 if (is_guest_mode(vcpu)) {
5732 kvm_queue_exception(vcpu, UD_VECTOR);
5735 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5736 if (er != EMULATE_DONE)
5737 kvm_queue_exception(vcpu, UD_VECTOR);
5742 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5743 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5746 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5747 * MMIO, it is better to report an internal error.
5748 * See the comments in vmx_handle_exit.
5750 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5751 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5752 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5753 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5754 vcpu->run->internal.ndata = 3;
5755 vcpu->run->internal.data[0] = vect_info;
5756 vcpu->run->internal.data[1] = intr_info;
5757 vcpu->run->internal.data[2] = error_code;
5761 if (is_page_fault(intr_info)) {
5762 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5763 /* EPT won't cause page fault directly */
5764 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
5765 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0,
5769 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5771 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5772 return handle_rmode_exception(vcpu, ex_no, error_code);
5776 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5779 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5780 if (!(vcpu->guest_debug &
5781 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5782 vcpu->arch.dr6 &= ~15;
5783 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5784 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5785 skip_emulated_instruction(vcpu);
5787 kvm_queue_exception(vcpu, DB_VECTOR);
5790 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5791 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5795 * Update instruction length as we may reinject #BP from
5796 * user space while in guest debugging mode. Reading it for
5797 * #DB as well causes no harm, it is not used in that case.
5799 vmx->vcpu.arch.event_exit_inst_len =
5800 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5801 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5802 rip = kvm_rip_read(vcpu);
5803 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5804 kvm_run->debug.arch.exception = ex_no;
5807 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5808 kvm_run->ex.exception = ex_no;
5809 kvm_run->ex.error_code = error_code;
5815 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5817 ++vcpu->stat.irq_exits;
5821 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5823 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5827 static int handle_io(struct kvm_vcpu *vcpu)
5829 unsigned long exit_qualification;
5830 int size, in, string, ret;
5833 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5834 string = (exit_qualification & 16) != 0;
5835 in = (exit_qualification & 8) != 0;
5837 ++vcpu->stat.io_exits;
5840 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5842 port = exit_qualification >> 16;
5843 size = (exit_qualification & 7) + 1;
5845 ret = kvm_skip_emulated_instruction(vcpu);
5848 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
5849 * KVM_EXIT_DEBUG here.
5851 return kvm_fast_pio_out(vcpu, size, port) && ret;
5855 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5858 * Patch in the VMCALL instruction:
5860 hypercall[0] = 0x0f;
5861 hypercall[1] = 0x01;
5862 hypercall[2] = 0xc1;
5865 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5866 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5868 if (is_guest_mode(vcpu)) {
5869 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5870 unsigned long orig_val = val;
5873 * We get here when L2 changed cr0 in a way that did not change
5874 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5875 * but did change L0 shadowed bits. So we first calculate the
5876 * effective cr0 value that L1 would like to write into the
5877 * hardware. It consists of the L2-owned bits from the new
5878 * value combined with the L1-owned bits from L1's guest_cr0.
5880 val = (val & ~vmcs12->cr0_guest_host_mask) |
5881 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5883 if (!nested_guest_cr0_valid(vcpu, val))
5886 if (kvm_set_cr0(vcpu, val))
5888 vmcs_writel(CR0_READ_SHADOW, orig_val);
5891 if (to_vmx(vcpu)->nested.vmxon &&
5892 !nested_host_cr0_valid(vcpu, val))
5895 return kvm_set_cr0(vcpu, val);
5899 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5901 if (is_guest_mode(vcpu)) {
5902 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5903 unsigned long orig_val = val;
5905 /* analogously to handle_set_cr0 */
5906 val = (val & ~vmcs12->cr4_guest_host_mask) |
5907 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5908 if (kvm_set_cr4(vcpu, val))
5910 vmcs_writel(CR4_READ_SHADOW, orig_val);
5913 return kvm_set_cr4(vcpu, val);
5916 static int handle_cr(struct kvm_vcpu *vcpu)
5918 unsigned long exit_qualification, val;
5924 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5925 cr = exit_qualification & 15;
5926 reg = (exit_qualification >> 8) & 15;
5927 switch ((exit_qualification >> 4) & 3) {
5928 case 0: /* mov to cr */
5929 val = kvm_register_readl(vcpu, reg);
5930 trace_kvm_cr_write(cr, val);
5933 err = handle_set_cr0(vcpu, val);
5934 return kvm_complete_insn_gp(vcpu, err);
5936 err = kvm_set_cr3(vcpu, val);
5937 return kvm_complete_insn_gp(vcpu, err);
5939 err = handle_set_cr4(vcpu, val);
5940 return kvm_complete_insn_gp(vcpu, err);
5942 u8 cr8_prev = kvm_get_cr8(vcpu);
5944 err = kvm_set_cr8(vcpu, cr8);
5945 ret = kvm_complete_insn_gp(vcpu, err);
5946 if (lapic_in_kernel(vcpu))
5948 if (cr8_prev <= cr8)
5951 * TODO: we might be squashing a
5952 * KVM_GUESTDBG_SINGLESTEP-triggered
5953 * KVM_EXIT_DEBUG here.
5955 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5961 WARN_ONCE(1, "Guest should always own CR0.TS");
5962 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5963 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5964 return kvm_skip_emulated_instruction(vcpu);
5965 case 1: /*mov from cr*/
5968 val = kvm_read_cr3(vcpu);
5969 kvm_register_write(vcpu, reg, val);
5970 trace_kvm_cr_read(cr, val);
5971 return kvm_skip_emulated_instruction(vcpu);
5973 val = kvm_get_cr8(vcpu);
5974 kvm_register_write(vcpu, reg, val);
5975 trace_kvm_cr_read(cr, val);
5976 return kvm_skip_emulated_instruction(vcpu);
5980 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5981 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5982 kvm_lmsw(vcpu, val);
5984 return kvm_skip_emulated_instruction(vcpu);
5988 vcpu->run->exit_reason = 0;
5989 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5990 (int)(exit_qualification >> 4) & 3, cr);
5994 static int handle_dr(struct kvm_vcpu *vcpu)
5996 unsigned long exit_qualification;
5999 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6000 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
6002 /* First, if DR does not exist, trigger UD */
6003 if (!kvm_require_dr(vcpu, dr))
6006 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
6007 if (!kvm_require_cpl(vcpu, 0))
6009 dr7 = vmcs_readl(GUEST_DR7);
6012 * As the vm-exit takes precedence over the debug trap, we
6013 * need to emulate the latter, either for the host or the
6014 * guest debugging itself.
6016 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6017 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
6018 vcpu->run->debug.arch.dr7 = dr7;
6019 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
6020 vcpu->run->debug.arch.exception = DB_VECTOR;
6021 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
6024 vcpu->arch.dr6 &= ~15;
6025 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
6026 kvm_queue_exception(vcpu, DB_VECTOR);
6031 if (vcpu->guest_debug == 0) {
6032 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6033 CPU_BASED_MOV_DR_EXITING);
6036 * No more DR vmexits; force a reload of the debug registers
6037 * and reenter on this instruction. The next vmexit will
6038 * retrieve the full state of the debug registers.
6040 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
6044 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
6045 if (exit_qualification & TYPE_MOV_FROM_DR) {
6048 if (kvm_get_dr(vcpu, dr, &val))
6050 kvm_register_write(vcpu, reg, val);
6052 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
6055 return kvm_skip_emulated_instruction(vcpu);
6058 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
6060 return vcpu->arch.dr6;
6063 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
6067 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
6069 get_debugreg(vcpu->arch.db[0], 0);
6070 get_debugreg(vcpu->arch.db[1], 1);
6071 get_debugreg(vcpu->arch.db[2], 2);
6072 get_debugreg(vcpu->arch.db[3], 3);
6073 get_debugreg(vcpu->arch.dr6, 6);
6074 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
6076 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
6077 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
6080 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
6082 vmcs_writel(GUEST_DR7, val);
6085 static int handle_cpuid(struct kvm_vcpu *vcpu)
6087 return kvm_emulate_cpuid(vcpu);
6090 static int handle_rdmsr(struct kvm_vcpu *vcpu)
6092 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6093 struct msr_data msr_info;
6095 msr_info.index = ecx;
6096 msr_info.host_initiated = false;
6097 if (vmx_get_msr(vcpu, &msr_info)) {
6098 trace_kvm_msr_read_ex(ecx);
6099 kvm_inject_gp(vcpu, 0);
6103 trace_kvm_msr_read(ecx, msr_info.data);
6105 /* FIXME: handling of bits 32:63 of rax, rdx */
6106 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
6107 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
6108 return kvm_skip_emulated_instruction(vcpu);
6111 static int handle_wrmsr(struct kvm_vcpu *vcpu)
6113 struct msr_data msr;
6114 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6115 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
6116 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6120 msr.host_initiated = false;
6121 if (kvm_set_msr(vcpu, &msr) != 0) {
6122 trace_kvm_msr_write_ex(ecx, data);
6123 kvm_inject_gp(vcpu, 0);
6127 trace_kvm_msr_write(ecx, data);
6128 return kvm_skip_emulated_instruction(vcpu);
6131 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
6133 kvm_apic_update_ppr(vcpu);
6137 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
6139 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6140 CPU_BASED_VIRTUAL_INTR_PENDING);
6142 kvm_make_request(KVM_REQ_EVENT, vcpu);
6144 ++vcpu->stat.irq_window_exits;
6148 static int handle_halt(struct kvm_vcpu *vcpu)
6150 return kvm_emulate_halt(vcpu);
6153 static int handle_vmcall(struct kvm_vcpu *vcpu)
6155 return kvm_emulate_hypercall(vcpu);
6158 static int handle_invd(struct kvm_vcpu *vcpu)
6160 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6163 static int handle_invlpg(struct kvm_vcpu *vcpu)
6165 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6167 kvm_mmu_invlpg(vcpu, exit_qualification);
6168 return kvm_skip_emulated_instruction(vcpu);
6171 static int handle_rdpmc(struct kvm_vcpu *vcpu)
6175 err = kvm_rdpmc(vcpu);
6176 return kvm_complete_insn_gp(vcpu, err);
6179 static int handle_wbinvd(struct kvm_vcpu *vcpu)
6181 return kvm_emulate_wbinvd(vcpu);
6184 static int handle_xsetbv(struct kvm_vcpu *vcpu)
6186 u64 new_bv = kvm_read_edx_eax(vcpu);
6187 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
6189 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
6190 return kvm_skip_emulated_instruction(vcpu);
6194 static int handle_xsaves(struct kvm_vcpu *vcpu)
6196 kvm_skip_emulated_instruction(vcpu);
6197 WARN(1, "this should never happen\n");
6201 static int handle_xrstors(struct kvm_vcpu *vcpu)
6203 kvm_skip_emulated_instruction(vcpu);
6204 WARN(1, "this should never happen\n");
6208 static int handle_apic_access(struct kvm_vcpu *vcpu)
6210 if (likely(fasteoi)) {
6211 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6212 int access_type, offset;
6214 access_type = exit_qualification & APIC_ACCESS_TYPE;
6215 offset = exit_qualification & APIC_ACCESS_OFFSET;
6217 * Sane guest uses MOV to write EOI, with written value
6218 * not cared. So make a short-circuit here by avoiding
6219 * heavy instruction emulation.
6221 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
6222 (offset == APIC_EOI)) {
6223 kvm_lapic_set_eoi(vcpu);
6224 return kvm_skip_emulated_instruction(vcpu);
6227 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6230 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
6232 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6233 int vector = exit_qualification & 0xff;
6235 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
6236 kvm_apic_set_eoi_accelerated(vcpu, vector);
6240 static int handle_apic_write(struct kvm_vcpu *vcpu)
6242 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6243 u32 offset = exit_qualification & 0xfff;
6245 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6246 kvm_apic_write_nodecode(vcpu, offset);
6250 static int handle_task_switch(struct kvm_vcpu *vcpu)
6252 struct vcpu_vmx *vmx = to_vmx(vcpu);
6253 unsigned long exit_qualification;
6254 bool has_error_code = false;
6257 int reason, type, idt_v, idt_index;
6259 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
6260 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
6261 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
6263 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6265 reason = (u32)exit_qualification >> 30;
6266 if (reason == TASK_SWITCH_GATE && idt_v) {
6268 case INTR_TYPE_NMI_INTR:
6269 vcpu->arch.nmi_injected = false;
6270 vmx_set_nmi_mask(vcpu, true);
6272 case INTR_TYPE_EXT_INTR:
6273 case INTR_TYPE_SOFT_INTR:
6274 kvm_clear_interrupt_queue(vcpu);
6276 case INTR_TYPE_HARD_EXCEPTION:
6277 if (vmx->idt_vectoring_info &
6278 VECTORING_INFO_DELIVER_CODE_MASK) {
6279 has_error_code = true;
6281 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6284 case INTR_TYPE_SOFT_EXCEPTION:
6285 kvm_clear_exception_queue(vcpu);
6291 tss_selector = exit_qualification;
6293 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
6294 type != INTR_TYPE_EXT_INTR &&
6295 type != INTR_TYPE_NMI_INTR))
6296 skip_emulated_instruction(vcpu);
6298 if (kvm_task_switch(vcpu, tss_selector,
6299 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
6300 has_error_code, error_code) == EMULATE_FAIL) {
6301 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6302 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6303 vcpu->run->internal.ndata = 0;
6308 * TODO: What about debug traps on tss switch?
6309 * Are we supposed to inject them and update dr6?
6315 static int handle_ept_violation(struct kvm_vcpu *vcpu)
6317 unsigned long exit_qualification;
6321 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6324 * EPT violation happened while executing iret from NMI,
6325 * "blocked by NMI" bit has to be set before next VM entry.
6326 * There are errata that may cause this bit to not be set:
6329 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
6330 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
6331 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
6333 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6334 trace_kvm_page_fault(gpa, exit_qualification);
6336 /* Is it a read fault? */
6337 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
6338 ? PFERR_USER_MASK : 0;
6339 /* Is it a write fault? */
6340 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
6341 ? PFERR_WRITE_MASK : 0;
6342 /* Is it a fetch fault? */
6343 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
6344 ? PFERR_FETCH_MASK : 0;
6345 /* ept page table entry is present? */
6346 error_code |= (exit_qualification &
6347 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
6348 EPT_VIOLATION_EXECUTABLE))
6349 ? PFERR_PRESENT_MASK : 0;
6351 vcpu->arch.gpa_available = true;
6352 vcpu->arch.exit_qualification = exit_qualification;
6354 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
6357 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
6362 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6363 if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
6364 trace_kvm_fast_mmio(gpa);
6365 return kvm_skip_emulated_instruction(vcpu);
6368 ret = handle_mmio_page_fault(vcpu, gpa, true);
6369 vcpu->arch.gpa_available = true;
6370 if (likely(ret == RET_MMIO_PF_EMULATE))
6371 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
6374 if (unlikely(ret == RET_MMIO_PF_INVALID))
6375 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
6377 if (unlikely(ret == RET_MMIO_PF_RETRY))
6380 /* It is the real ept misconfig */
6383 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6384 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6389 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6391 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6392 CPU_BASED_VIRTUAL_NMI_PENDING);
6393 ++vcpu->stat.nmi_window_exits;
6394 kvm_make_request(KVM_REQ_EVENT, vcpu);
6399 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6401 struct vcpu_vmx *vmx = to_vmx(vcpu);
6402 enum emulation_result err = EMULATE_DONE;
6405 bool intr_window_requested;
6406 unsigned count = 130;
6408 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6409 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6411 while (vmx->emulation_required && count-- != 0) {
6412 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6413 return handle_interrupt_window(&vmx->vcpu);
6415 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
6418 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
6420 if (err == EMULATE_USER_EXIT) {
6421 ++vcpu->stat.mmio_exits;
6426 if (err != EMULATE_DONE) {
6427 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6428 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6429 vcpu->run->internal.ndata = 0;
6433 if (vcpu->arch.halt_request) {
6434 vcpu->arch.halt_request = 0;
6435 ret = kvm_vcpu_halt(vcpu);
6439 if (signal_pending(current))
6449 static int __grow_ple_window(int val)
6451 if (ple_window_grow < 1)
6454 val = min(val, ple_window_actual_max);
6456 if (ple_window_grow < ple_window)
6457 val *= ple_window_grow;
6459 val += ple_window_grow;
6464 static int __shrink_ple_window(int val, int modifier, int minimum)
6469 if (modifier < ple_window)
6474 return max(val, minimum);
6477 static void grow_ple_window(struct kvm_vcpu *vcpu)
6479 struct vcpu_vmx *vmx = to_vmx(vcpu);
6480 int old = vmx->ple_window;
6482 vmx->ple_window = __grow_ple_window(old);
6484 if (vmx->ple_window != old)
6485 vmx->ple_window_dirty = true;
6487 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6490 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6492 struct vcpu_vmx *vmx = to_vmx(vcpu);
6493 int old = vmx->ple_window;
6495 vmx->ple_window = __shrink_ple_window(old,
6496 ple_window_shrink, ple_window);
6498 if (vmx->ple_window != old)
6499 vmx->ple_window_dirty = true;
6501 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6505 * ple_window_actual_max is computed to be one grow_ple_window() below
6506 * ple_window_max. (See __grow_ple_window for the reason.)
6507 * This prevents overflows, because ple_window_max is int.
6508 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6510 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6512 static void update_ple_window_actual_max(void)
6514 ple_window_actual_max =
6515 __shrink_ple_window(max(ple_window_max, ple_window),
6516 ple_window_grow, INT_MIN);
6520 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6522 static void wakeup_handler(void)
6524 struct kvm_vcpu *vcpu;
6525 int cpu = smp_processor_id();
6527 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6528 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6529 blocked_vcpu_list) {
6530 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6532 if (pi_test_on(pi_desc) == 1)
6533 kvm_vcpu_kick(vcpu);
6535 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6538 void vmx_enable_tdp(void)
6540 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
6541 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
6542 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
6543 0ull, VMX_EPT_EXECUTABLE_MASK,
6544 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
6547 ept_set_mmio_spte_mask();
6551 static __init int hardware_setup(void)
6553 int r = -ENOMEM, i, msr;
6555 rdmsrl_safe(MSR_EFER, &host_efer);
6557 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6558 kvm_define_shared_msr(i, vmx_msr_index[i]);
6560 for (i = 0; i < VMX_BITMAP_NR; i++) {
6561 vmx_bitmap[i] = (unsigned long *)__get_free_page(GFP_KERNEL);
6566 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6567 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6568 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6571 * Allow direct access to the PC debug port (it is often used for I/O
6572 * delays, but the vmexits simply slow things down).
6574 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6575 clear_bit(0x80, vmx_io_bitmap_a);
6577 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6579 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6580 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6582 if (setup_vmcs_config(&vmcs_config) < 0) {
6587 if (boot_cpu_has(X86_FEATURE_NX))
6588 kvm_enable_efer_bits(EFER_NX);
6590 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
6591 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
6594 if (!cpu_has_vmx_shadow_vmcs())
6595 enable_shadow_vmcs = 0;
6596 if (enable_shadow_vmcs)
6597 init_vmcs_shadow_fields();
6599 if (!cpu_has_vmx_ept() ||
6600 !cpu_has_vmx_ept_4levels()) {
6602 enable_unrestricted_guest = 0;
6603 enable_ept_ad_bits = 0;
6606 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
6607 enable_ept_ad_bits = 0;
6609 if (!cpu_has_vmx_unrestricted_guest())
6610 enable_unrestricted_guest = 0;
6612 if (!cpu_has_vmx_flexpriority())
6613 flexpriority_enabled = 0;
6616 * set_apic_access_page_addr() is used to reload apic access
6617 * page upon invalidation. No need to do anything if not
6618 * using the APIC_ACCESS_ADDR VMCS field.
6620 if (!flexpriority_enabled)
6621 kvm_x86_ops->set_apic_access_page_addr = NULL;
6623 if (!cpu_has_vmx_tpr_shadow())
6624 kvm_x86_ops->update_cr8_intercept = NULL;
6626 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6627 kvm_disable_largepages();
6629 if (!cpu_has_vmx_ple())
6632 if (!cpu_has_vmx_apicv()) {
6634 kvm_x86_ops->sync_pir_to_irr = NULL;
6637 if (cpu_has_vmx_tsc_scaling()) {
6638 kvm_has_tsc_control = true;
6639 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6640 kvm_tsc_scaling_ratio_frac_bits = 48;
6643 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6644 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6645 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6646 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6647 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6648 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6650 memcpy(vmx_msr_bitmap_legacy_x2apic_apicv,
6651 vmx_msr_bitmap_legacy, PAGE_SIZE);
6652 memcpy(vmx_msr_bitmap_longmode_x2apic_apicv,
6653 vmx_msr_bitmap_longmode, PAGE_SIZE);
6654 memcpy(vmx_msr_bitmap_legacy_x2apic,
6655 vmx_msr_bitmap_legacy, PAGE_SIZE);
6656 memcpy(vmx_msr_bitmap_longmode_x2apic,
6657 vmx_msr_bitmap_longmode, PAGE_SIZE);
6659 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6661 for (msr = 0x800; msr <= 0x8ff; msr++) {
6662 if (msr == 0x839 /* TMCCT */)
6664 vmx_disable_intercept_msr_x2apic(msr, MSR_TYPE_R, true);
6668 * TPR reads and writes can be virtualized even if virtual interrupt
6669 * delivery is not in use.
6671 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_W, true);
6672 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_R | MSR_TYPE_W, false);
6675 vmx_disable_intercept_msr_x2apic(0x80b, MSR_TYPE_W, true);
6677 vmx_disable_intercept_msr_x2apic(0x83f, MSR_TYPE_W, true);
6684 update_ple_window_actual_max();
6687 * Only enable PML when hardware supports PML feature, and both EPT
6688 * and EPT A/D bit features are enabled -- PML depends on them to work.
6690 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6694 kvm_x86_ops->slot_enable_log_dirty = NULL;
6695 kvm_x86_ops->slot_disable_log_dirty = NULL;
6696 kvm_x86_ops->flush_log_dirty = NULL;
6697 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6700 if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
6703 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
6704 cpu_preemption_timer_multi =
6705 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
6707 kvm_x86_ops->set_hv_timer = NULL;
6708 kvm_x86_ops->cancel_hv_timer = NULL;
6711 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6713 kvm_mce_cap_supported |= MCG_LMCE_P;
6715 return alloc_kvm_area();
6718 for (i = 0; i < VMX_BITMAP_NR; i++)
6719 free_page((unsigned long)vmx_bitmap[i]);
6724 static __exit void hardware_unsetup(void)
6728 for (i = 0; i < VMX_BITMAP_NR; i++)
6729 free_page((unsigned long)vmx_bitmap[i]);
6735 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6736 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6738 static int handle_pause(struct kvm_vcpu *vcpu)
6741 grow_ple_window(vcpu);
6743 kvm_vcpu_on_spin(vcpu);
6744 return kvm_skip_emulated_instruction(vcpu);
6747 static int handle_nop(struct kvm_vcpu *vcpu)
6749 return kvm_skip_emulated_instruction(vcpu);
6752 static int handle_mwait(struct kvm_vcpu *vcpu)
6754 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6755 return handle_nop(vcpu);
6758 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6763 static int handle_monitor(struct kvm_vcpu *vcpu)
6765 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6766 return handle_nop(vcpu);
6770 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6771 * We could reuse a single VMCS for all the L2 guests, but we also want the
6772 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6773 * allows keeping them loaded on the processor, and in the future will allow
6774 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6775 * every entry if they never change.
6776 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6777 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6779 * The following functions allocate and free a vmcs02 in this pool.
6782 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6783 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6785 struct vmcs02_list *item;
6786 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6787 if (item->vmptr == vmx->nested.current_vmptr) {
6788 list_move(&item->list, &vmx->nested.vmcs02_pool);
6789 return &item->vmcs02;
6792 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6793 /* Recycle the least recently used VMCS. */
6794 item = list_last_entry(&vmx->nested.vmcs02_pool,
6795 struct vmcs02_list, list);
6796 item->vmptr = vmx->nested.current_vmptr;
6797 list_move(&item->list, &vmx->nested.vmcs02_pool);
6798 return &item->vmcs02;
6801 /* Create a new VMCS */
6802 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6805 item->vmcs02.vmcs = alloc_vmcs();
6806 item->vmcs02.shadow_vmcs = NULL;
6807 if (!item->vmcs02.vmcs) {
6811 loaded_vmcs_init(&item->vmcs02);
6812 item->vmptr = vmx->nested.current_vmptr;
6813 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6814 vmx->nested.vmcs02_num++;
6815 return &item->vmcs02;
6818 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6819 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6821 struct vmcs02_list *item;
6822 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6823 if (item->vmptr == vmptr) {
6824 free_loaded_vmcs(&item->vmcs02);
6825 list_del(&item->list);
6827 vmx->nested.vmcs02_num--;
6833 * Free all VMCSs saved for this vcpu, except the one pointed by
6834 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6835 * must be &vmx->vmcs01.
6837 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6839 struct vmcs02_list *item, *n;
6841 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6842 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6844 * Something will leak if the above WARN triggers. Better than
6847 if (vmx->loaded_vmcs == &item->vmcs02)
6850 free_loaded_vmcs(&item->vmcs02);
6851 list_del(&item->list);
6853 vmx->nested.vmcs02_num--;
6858 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6859 * set the success or error code of an emulated VMX instruction, as specified
6860 * by Vol 2B, VMX Instruction Reference, "Conventions".
6862 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6864 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6865 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6866 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6869 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6871 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6872 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6873 X86_EFLAGS_SF | X86_EFLAGS_OF))
6877 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6878 u32 vm_instruction_error)
6880 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6882 * failValid writes the error number to the current VMCS, which
6883 * can't be done there isn't a current VMCS.
6885 nested_vmx_failInvalid(vcpu);
6888 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6889 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6890 X86_EFLAGS_SF | X86_EFLAGS_OF))
6892 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6894 * We don't need to force a shadow sync because
6895 * VM_INSTRUCTION_ERROR is not shadowed
6899 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6901 /* TODO: not to reset guest simply here. */
6902 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6903 pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
6906 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6908 struct vcpu_vmx *vmx =
6909 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6911 vmx->nested.preemption_timer_expired = true;
6912 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6913 kvm_vcpu_kick(&vmx->vcpu);
6915 return HRTIMER_NORESTART;
6919 * Decode the memory-address operand of a vmx instruction, as recorded on an
6920 * exit caused by such an instruction (run by a guest hypervisor).
6921 * On success, returns 0. When the operand is invalid, returns 1 and throws
6924 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6925 unsigned long exit_qualification,
6926 u32 vmx_instruction_info, bool wr, gva_t *ret)
6930 struct kvm_segment s;
6933 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6934 * Execution", on an exit, vmx_instruction_info holds most of the
6935 * addressing components of the operand. Only the displacement part
6936 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6937 * For how an actual address is calculated from all these components,
6938 * refer to Vol. 1, "Operand Addressing".
6940 int scaling = vmx_instruction_info & 3;
6941 int addr_size = (vmx_instruction_info >> 7) & 7;
6942 bool is_reg = vmx_instruction_info & (1u << 10);
6943 int seg_reg = (vmx_instruction_info >> 15) & 7;
6944 int index_reg = (vmx_instruction_info >> 18) & 0xf;
6945 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6946 int base_reg = (vmx_instruction_info >> 23) & 0xf;
6947 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
6950 kvm_queue_exception(vcpu, UD_VECTOR);
6954 /* Addr = segment_base + offset */
6955 /* offset = base + [index * scale] + displacement */
6956 off = exit_qualification; /* holds the displacement */
6958 off += kvm_register_read(vcpu, base_reg);
6960 off += kvm_register_read(vcpu, index_reg)<<scaling;
6961 vmx_get_segment(vcpu, &s, seg_reg);
6962 *ret = s.base + off;
6964 if (addr_size == 1) /* 32 bit */
6967 /* Checks for #GP/#SS exceptions. */
6969 if (is_long_mode(vcpu)) {
6970 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6971 * non-canonical form. This is the only check on the memory
6972 * destination for long mode!
6974 exn = is_noncanonical_address(*ret);
6975 } else if (is_protmode(vcpu)) {
6976 /* Protected mode: apply checks for segment validity in the
6978 * - segment type check (#GP(0) may be thrown)
6979 * - usability check (#GP(0)/#SS(0))
6980 * - limit check (#GP(0)/#SS(0))
6983 /* #GP(0) if the destination operand is located in a
6984 * read-only data segment or any code segment.
6986 exn = ((s.type & 0xa) == 0 || (s.type & 8));
6988 /* #GP(0) if the source operand is located in an
6989 * execute-only code segment
6991 exn = ((s.type & 0xa) == 8);
6993 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6996 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
6998 exn = (s.unusable != 0);
6999 /* Protected mode: #GP(0)/#SS(0) if the memory
7000 * operand is outside the segment limit.
7002 exn = exn || (off + sizeof(u64) > s.limit);
7005 kvm_queue_exception_e(vcpu,
7006 seg_reg == VCPU_SREG_SS ?
7007 SS_VECTOR : GP_VECTOR,
7015 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
7018 struct x86_exception e;
7020 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7021 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
7024 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, vmpointer,
7025 sizeof(*vmpointer), &e)) {
7026 kvm_inject_page_fault(vcpu, &e);
7033 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
7035 struct vcpu_vmx *vmx = to_vmx(vcpu);
7036 struct vmcs *shadow_vmcs;
7038 if (cpu_has_vmx_msr_bitmap()) {
7039 vmx->nested.msr_bitmap =
7040 (unsigned long *)__get_free_page(GFP_KERNEL);
7041 if (!vmx->nested.msr_bitmap)
7042 goto out_msr_bitmap;
7045 vmx->nested.cached_vmcs12 = kmalloc(VMCS12_SIZE, GFP_KERNEL);
7046 if (!vmx->nested.cached_vmcs12)
7047 goto out_cached_vmcs12;
7049 if (enable_shadow_vmcs) {
7050 shadow_vmcs = alloc_vmcs();
7052 goto out_shadow_vmcs;
7053 /* mark vmcs as shadow */
7054 shadow_vmcs->revision_id |= (1u << 31);
7055 /* init shadow vmcs */
7056 vmcs_clear(shadow_vmcs);
7057 vmx->vmcs01.shadow_vmcs = shadow_vmcs;
7060 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
7061 vmx->nested.vmcs02_num = 0;
7063 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
7064 HRTIMER_MODE_REL_PINNED);
7065 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
7067 vmx->nested.vmxon = true;
7071 kfree(vmx->nested.cached_vmcs12);
7074 free_page((unsigned long)vmx->nested.msr_bitmap);
7081 * Emulate the VMXON instruction.
7082 * Currently, we just remember that VMX is active, and do not save or even
7083 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
7084 * do not currently need to store anything in that guest-allocated memory
7085 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
7086 * argument is different from the VMXON pointer (which the spec says they do).
7088 static int handle_vmon(struct kvm_vcpu *vcpu)
7093 struct vcpu_vmx *vmx = to_vmx(vcpu);
7094 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
7095 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7098 * The Intel VMX Instruction Reference lists a bunch of bits that are
7099 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
7100 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
7101 * Otherwise, we should fail with #UD. But most faulting conditions
7102 * have already been checked by hardware, prior to the VM-exit for
7103 * VMXON. We do test guest cr4.VMXE because processor CR4 always has
7104 * that bit set to 1 in non-root mode.
7106 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
7107 kvm_queue_exception(vcpu, UD_VECTOR);
7111 if (vmx->nested.vmxon) {
7112 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
7113 return kvm_skip_emulated_instruction(vcpu);
7116 if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
7117 != VMXON_NEEDED_FEATURES) {
7118 kvm_inject_gp(vcpu, 0);
7122 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7127 * The first 4 bytes of VMXON region contain the supported
7128 * VMCS revision identifier
7130 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
7131 * which replaces physical address width with 32
7133 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7134 nested_vmx_failInvalid(vcpu);
7135 return kvm_skip_emulated_instruction(vcpu);
7138 page = nested_get_page(vcpu, vmptr);
7140 nested_vmx_failInvalid(vcpu);
7141 return kvm_skip_emulated_instruction(vcpu);
7143 if (*(u32 *)kmap(page) != VMCS12_REVISION) {
7145 nested_release_page_clean(page);
7146 nested_vmx_failInvalid(vcpu);
7147 return kvm_skip_emulated_instruction(vcpu);
7150 nested_release_page_clean(page);
7152 vmx->nested.vmxon_ptr = vmptr;
7153 ret = enter_vmx_operation(vcpu);
7157 nested_vmx_succeed(vcpu);
7158 return kvm_skip_emulated_instruction(vcpu);
7162 * Intel's VMX Instruction Reference specifies a common set of prerequisites
7163 * for running VMX instructions (except VMXON, whose prerequisites are
7164 * slightly different). It also specifies what exception to inject otherwise.
7165 * Note that many of these exceptions have priority over VM exits, so they
7166 * don't have to be checked again here.
7168 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
7170 if (!to_vmx(vcpu)->nested.vmxon) {
7171 kvm_queue_exception(vcpu, UD_VECTOR);
7177 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
7179 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL, SECONDARY_EXEC_SHADOW_VMCS);
7180 vmcs_write64(VMCS_LINK_POINTER, -1ull);
7183 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
7185 if (vmx->nested.current_vmptr == -1ull)
7188 if (enable_shadow_vmcs) {
7189 /* copy to memory all shadowed fields in case
7190 they were modified */
7191 copy_shadow_to_vmcs12(vmx);
7192 vmx->nested.sync_shadow_vmcs = false;
7193 vmx_disable_shadow_vmcs(vmx);
7195 vmx->nested.posted_intr_nv = -1;
7197 /* Flush VMCS12 to guest memory */
7198 kvm_vcpu_write_guest_page(&vmx->vcpu,
7199 vmx->nested.current_vmptr >> PAGE_SHIFT,
7200 vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
7202 vmx->nested.current_vmptr = -1ull;
7206 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7207 * just stops using VMX.
7209 static void free_nested(struct vcpu_vmx *vmx)
7211 if (!vmx->nested.vmxon)
7214 vmx->nested.vmxon = false;
7215 free_vpid(vmx->nested.vpid02);
7216 vmx->nested.posted_intr_nv = -1;
7217 vmx->nested.current_vmptr = -1ull;
7218 if (vmx->nested.msr_bitmap) {
7219 free_page((unsigned long)vmx->nested.msr_bitmap);
7220 vmx->nested.msr_bitmap = NULL;
7222 if (enable_shadow_vmcs) {
7223 vmx_disable_shadow_vmcs(vmx);
7224 vmcs_clear(vmx->vmcs01.shadow_vmcs);
7225 free_vmcs(vmx->vmcs01.shadow_vmcs);
7226 vmx->vmcs01.shadow_vmcs = NULL;
7228 kfree(vmx->nested.cached_vmcs12);
7229 /* Unpin physical memory we referred to in current vmcs02 */
7230 if (vmx->nested.apic_access_page) {
7231 nested_release_page(vmx->nested.apic_access_page);
7232 vmx->nested.apic_access_page = NULL;
7234 if (vmx->nested.virtual_apic_page) {
7235 nested_release_page(vmx->nested.virtual_apic_page);
7236 vmx->nested.virtual_apic_page = NULL;
7238 if (vmx->nested.pi_desc_page) {
7239 kunmap(vmx->nested.pi_desc_page);
7240 nested_release_page(vmx->nested.pi_desc_page);
7241 vmx->nested.pi_desc_page = NULL;
7242 vmx->nested.pi_desc = NULL;
7245 nested_free_all_saved_vmcss(vmx);
7248 /* Emulate the VMXOFF instruction */
7249 static int handle_vmoff(struct kvm_vcpu *vcpu)
7251 if (!nested_vmx_check_permission(vcpu))
7253 free_nested(to_vmx(vcpu));
7254 nested_vmx_succeed(vcpu);
7255 return kvm_skip_emulated_instruction(vcpu);
7258 /* Emulate the VMCLEAR instruction */
7259 static int handle_vmclear(struct kvm_vcpu *vcpu)
7261 struct vcpu_vmx *vmx = to_vmx(vcpu);
7265 if (!nested_vmx_check_permission(vcpu))
7268 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7271 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7272 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
7273 return kvm_skip_emulated_instruction(vcpu);
7276 if (vmptr == vmx->nested.vmxon_ptr) {
7277 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
7278 return kvm_skip_emulated_instruction(vcpu);
7281 if (vmptr == vmx->nested.current_vmptr)
7282 nested_release_vmcs12(vmx);
7284 kvm_vcpu_write_guest(vcpu,
7285 vmptr + offsetof(struct vmcs12, launch_state),
7286 &zero, sizeof(zero));
7288 nested_free_vmcs02(vmx, vmptr);
7290 nested_vmx_succeed(vcpu);
7291 return kvm_skip_emulated_instruction(vcpu);
7294 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
7296 /* Emulate the VMLAUNCH instruction */
7297 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
7299 return nested_vmx_run(vcpu, true);
7302 /* Emulate the VMRESUME instruction */
7303 static int handle_vmresume(struct kvm_vcpu *vcpu)
7306 return nested_vmx_run(vcpu, false);
7310 * Read a vmcs12 field. Since these can have varying lengths and we return
7311 * one type, we chose the biggest type (u64) and zero-extend the return value
7312 * to that size. Note that the caller, handle_vmread, might need to use only
7313 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7314 * 64-bit fields are to be returned).
7316 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7317 unsigned long field, u64 *ret)
7319 short offset = vmcs_field_to_offset(field);
7325 p = ((char *)(get_vmcs12(vcpu))) + offset;
7327 switch (vmcs_field_type(field)) {
7328 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7329 *ret = *((natural_width *)p);
7331 case VMCS_FIELD_TYPE_U16:
7334 case VMCS_FIELD_TYPE_U32:
7337 case VMCS_FIELD_TYPE_U64:
7347 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7348 unsigned long field, u64 field_value){
7349 short offset = vmcs_field_to_offset(field);
7350 char *p = ((char *) get_vmcs12(vcpu)) + offset;
7354 switch (vmcs_field_type(field)) {
7355 case VMCS_FIELD_TYPE_U16:
7356 *(u16 *)p = field_value;
7358 case VMCS_FIELD_TYPE_U32:
7359 *(u32 *)p = field_value;
7361 case VMCS_FIELD_TYPE_U64:
7362 *(u64 *)p = field_value;
7364 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7365 *(natural_width *)p = field_value;
7374 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7377 unsigned long field;
7379 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7380 const unsigned long *fields = shadow_read_write_fields;
7381 const int num_fields = max_shadow_read_write_fields;
7385 vmcs_load(shadow_vmcs);
7387 for (i = 0; i < num_fields; i++) {
7389 switch (vmcs_field_type(field)) {
7390 case VMCS_FIELD_TYPE_U16:
7391 field_value = vmcs_read16(field);
7393 case VMCS_FIELD_TYPE_U32:
7394 field_value = vmcs_read32(field);
7396 case VMCS_FIELD_TYPE_U64:
7397 field_value = vmcs_read64(field);
7399 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7400 field_value = vmcs_readl(field);
7406 vmcs12_write_any(&vmx->vcpu, field, field_value);
7409 vmcs_clear(shadow_vmcs);
7410 vmcs_load(vmx->loaded_vmcs->vmcs);
7415 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7417 const unsigned long *fields[] = {
7418 shadow_read_write_fields,
7419 shadow_read_only_fields
7421 const int max_fields[] = {
7422 max_shadow_read_write_fields,
7423 max_shadow_read_only_fields
7426 unsigned long field;
7427 u64 field_value = 0;
7428 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7430 vmcs_load(shadow_vmcs);
7432 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7433 for (i = 0; i < max_fields[q]; i++) {
7434 field = fields[q][i];
7435 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7437 switch (vmcs_field_type(field)) {
7438 case VMCS_FIELD_TYPE_U16:
7439 vmcs_write16(field, (u16)field_value);
7441 case VMCS_FIELD_TYPE_U32:
7442 vmcs_write32(field, (u32)field_value);
7444 case VMCS_FIELD_TYPE_U64:
7445 vmcs_write64(field, (u64)field_value);
7447 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7448 vmcs_writel(field, (long)field_value);
7457 vmcs_clear(shadow_vmcs);
7458 vmcs_load(vmx->loaded_vmcs->vmcs);
7462 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7463 * used before) all generate the same failure when it is missing.
7465 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7467 struct vcpu_vmx *vmx = to_vmx(vcpu);
7468 if (vmx->nested.current_vmptr == -1ull) {
7469 nested_vmx_failInvalid(vcpu);
7475 static int handle_vmread(struct kvm_vcpu *vcpu)
7477 unsigned long field;
7479 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7480 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7483 if (!nested_vmx_check_permission(vcpu))
7486 if (!nested_vmx_check_vmcs12(vcpu))
7487 return kvm_skip_emulated_instruction(vcpu);
7489 /* Decode instruction info and find the field to read */
7490 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7491 /* Read the field, zero-extended to a u64 field_value */
7492 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7493 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7494 return kvm_skip_emulated_instruction(vcpu);
7497 * Now copy part of this value to register or memory, as requested.
7498 * Note that the number of bits actually copied is 32 or 64 depending
7499 * on the guest's mode (32 or 64 bit), not on the given field's length.
7501 if (vmx_instruction_info & (1u << 10)) {
7502 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7505 if (get_vmx_mem_address(vcpu, exit_qualification,
7506 vmx_instruction_info, true, &gva))
7508 /* _system ok, as hardware has verified cpl=0 */
7509 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7510 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7513 nested_vmx_succeed(vcpu);
7514 return kvm_skip_emulated_instruction(vcpu);
7518 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7520 unsigned long field;
7522 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7523 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7524 /* The value to write might be 32 or 64 bits, depending on L1's long
7525 * mode, and eventually we need to write that into a field of several
7526 * possible lengths. The code below first zero-extends the value to 64
7527 * bit (field_value), and then copies only the appropriate number of
7528 * bits into the vmcs12 field.
7530 u64 field_value = 0;
7531 struct x86_exception e;
7533 if (!nested_vmx_check_permission(vcpu))
7536 if (!nested_vmx_check_vmcs12(vcpu))
7537 return kvm_skip_emulated_instruction(vcpu);
7539 if (vmx_instruction_info & (1u << 10))
7540 field_value = kvm_register_readl(vcpu,
7541 (((vmx_instruction_info) >> 3) & 0xf));
7543 if (get_vmx_mem_address(vcpu, exit_qualification,
7544 vmx_instruction_info, false, &gva))
7546 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7547 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7548 kvm_inject_page_fault(vcpu, &e);
7554 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7555 if (vmcs_field_readonly(field)) {
7556 nested_vmx_failValid(vcpu,
7557 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7558 return kvm_skip_emulated_instruction(vcpu);
7561 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7562 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7563 return kvm_skip_emulated_instruction(vcpu);
7566 nested_vmx_succeed(vcpu);
7567 return kvm_skip_emulated_instruction(vcpu);
7570 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
7572 vmx->nested.current_vmptr = vmptr;
7573 if (enable_shadow_vmcs) {
7574 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7575 SECONDARY_EXEC_SHADOW_VMCS);
7576 vmcs_write64(VMCS_LINK_POINTER,
7577 __pa(vmx->vmcs01.shadow_vmcs));
7578 vmx->nested.sync_shadow_vmcs = true;
7582 /* Emulate the VMPTRLD instruction */
7583 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7585 struct vcpu_vmx *vmx = to_vmx(vcpu);
7588 if (!nested_vmx_check_permission(vcpu))
7591 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7594 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7595 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
7596 return kvm_skip_emulated_instruction(vcpu);
7599 if (vmptr == vmx->nested.vmxon_ptr) {
7600 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
7601 return kvm_skip_emulated_instruction(vcpu);
7604 if (vmx->nested.current_vmptr != vmptr) {
7605 struct vmcs12 *new_vmcs12;
7607 page = nested_get_page(vcpu, vmptr);
7609 nested_vmx_failInvalid(vcpu);
7610 return kvm_skip_emulated_instruction(vcpu);
7612 new_vmcs12 = kmap(page);
7613 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7615 nested_release_page_clean(page);
7616 nested_vmx_failValid(vcpu,
7617 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7618 return kvm_skip_emulated_instruction(vcpu);
7621 nested_release_vmcs12(vmx);
7623 * Load VMCS12 from guest memory since it is not already
7626 memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
7628 nested_release_page_clean(page);
7630 set_current_vmptr(vmx, vmptr);
7633 nested_vmx_succeed(vcpu);
7634 return kvm_skip_emulated_instruction(vcpu);
7637 /* Emulate the VMPTRST instruction */
7638 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7640 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7641 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7643 struct x86_exception e;
7645 if (!nested_vmx_check_permission(vcpu))
7648 if (get_vmx_mem_address(vcpu, exit_qualification,
7649 vmx_instruction_info, true, &vmcs_gva))
7651 /* ok to use *_system, as hardware has verified cpl=0 */
7652 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7653 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7655 kvm_inject_page_fault(vcpu, &e);
7658 nested_vmx_succeed(vcpu);
7659 return kvm_skip_emulated_instruction(vcpu);
7662 /* Emulate the INVEPT instruction */
7663 static int handle_invept(struct kvm_vcpu *vcpu)
7665 struct vcpu_vmx *vmx = to_vmx(vcpu);
7666 u32 vmx_instruction_info, types;
7669 struct x86_exception e;
7674 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7675 SECONDARY_EXEC_ENABLE_EPT) ||
7676 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7677 kvm_queue_exception(vcpu, UD_VECTOR);
7681 if (!nested_vmx_check_permission(vcpu))
7684 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7685 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7687 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7689 if (type >= 32 || !(types & (1 << type))) {
7690 nested_vmx_failValid(vcpu,
7691 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7692 return kvm_skip_emulated_instruction(vcpu);
7695 /* According to the Intel VMX instruction reference, the memory
7696 * operand is read even if it isn't needed (e.g., for type==global)
7698 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7699 vmx_instruction_info, false, &gva))
7701 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7702 sizeof(operand), &e)) {
7703 kvm_inject_page_fault(vcpu, &e);
7708 case VMX_EPT_EXTENT_GLOBAL:
7710 * TODO: track mappings and invalidate
7711 * single context requests appropriately
7713 case VMX_EPT_EXTENT_CONTEXT:
7714 kvm_mmu_sync_roots(vcpu);
7715 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7716 nested_vmx_succeed(vcpu);
7723 return kvm_skip_emulated_instruction(vcpu);
7726 static int handle_invvpid(struct kvm_vcpu *vcpu)
7728 struct vcpu_vmx *vmx = to_vmx(vcpu);
7729 u32 vmx_instruction_info;
7730 unsigned long type, types;
7732 struct x86_exception e;
7738 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7739 SECONDARY_EXEC_ENABLE_VPID) ||
7740 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7741 kvm_queue_exception(vcpu, UD_VECTOR);
7745 if (!nested_vmx_check_permission(vcpu))
7748 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7749 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7751 types = (vmx->nested.nested_vmx_vpid_caps &
7752 VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
7754 if (type >= 32 || !(types & (1 << type))) {
7755 nested_vmx_failValid(vcpu,
7756 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7757 return kvm_skip_emulated_instruction(vcpu);
7760 /* according to the intel vmx instruction reference, the memory
7761 * operand is read even if it isn't needed (e.g., for type==global)
7763 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7764 vmx_instruction_info, false, &gva))
7766 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7767 sizeof(operand), &e)) {
7768 kvm_inject_page_fault(vcpu, &e);
7771 if (operand.vpid >> 16) {
7772 nested_vmx_failValid(vcpu,
7773 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7774 return kvm_skip_emulated_instruction(vcpu);
7778 case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
7779 if (is_noncanonical_address(operand.gla)) {
7780 nested_vmx_failValid(vcpu,
7781 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7782 return kvm_skip_emulated_instruction(vcpu);
7785 case VMX_VPID_EXTENT_SINGLE_CONTEXT:
7786 case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
7787 if (!operand.vpid) {
7788 nested_vmx_failValid(vcpu,
7789 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7790 return kvm_skip_emulated_instruction(vcpu);
7793 case VMX_VPID_EXTENT_ALL_CONTEXT:
7797 return kvm_skip_emulated_instruction(vcpu);
7800 __vmx_flush_tlb(vcpu, vmx->nested.vpid02);
7801 nested_vmx_succeed(vcpu);
7803 return kvm_skip_emulated_instruction(vcpu);
7806 static int handle_pml_full(struct kvm_vcpu *vcpu)
7808 unsigned long exit_qualification;
7810 trace_kvm_pml_full(vcpu->vcpu_id);
7812 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7815 * PML buffer FULL happened while executing iret from NMI,
7816 * "blocked by NMI" bit has to be set before next VM entry.
7818 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7819 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7820 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7821 GUEST_INTR_STATE_NMI);
7824 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7825 * here.., and there's no userspace involvement needed for PML.
7830 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
7832 kvm_lapic_expired_hv_timer(vcpu);
7837 * The exit handlers return 1 if the exit was handled fully and guest execution
7838 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
7839 * to be done to userspace and return 0.
7841 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7842 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
7843 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
7844 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
7845 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
7846 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
7847 [EXIT_REASON_CR_ACCESS] = handle_cr,
7848 [EXIT_REASON_DR_ACCESS] = handle_dr,
7849 [EXIT_REASON_CPUID] = handle_cpuid,
7850 [EXIT_REASON_MSR_READ] = handle_rdmsr,
7851 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
7852 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
7853 [EXIT_REASON_HLT] = handle_halt,
7854 [EXIT_REASON_INVD] = handle_invd,
7855 [EXIT_REASON_INVLPG] = handle_invlpg,
7856 [EXIT_REASON_RDPMC] = handle_rdpmc,
7857 [EXIT_REASON_VMCALL] = handle_vmcall,
7858 [EXIT_REASON_VMCLEAR] = handle_vmclear,
7859 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
7860 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
7861 [EXIT_REASON_VMPTRST] = handle_vmptrst,
7862 [EXIT_REASON_VMREAD] = handle_vmread,
7863 [EXIT_REASON_VMRESUME] = handle_vmresume,
7864 [EXIT_REASON_VMWRITE] = handle_vmwrite,
7865 [EXIT_REASON_VMOFF] = handle_vmoff,
7866 [EXIT_REASON_VMON] = handle_vmon,
7867 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
7868 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
7869 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
7870 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
7871 [EXIT_REASON_WBINVD] = handle_wbinvd,
7872 [EXIT_REASON_XSETBV] = handle_xsetbv,
7873 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
7874 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
7875 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
7876 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
7877 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
7878 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
7879 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
7880 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
7881 [EXIT_REASON_INVEPT] = handle_invept,
7882 [EXIT_REASON_INVVPID] = handle_invvpid,
7883 [EXIT_REASON_XSAVES] = handle_xsaves,
7884 [EXIT_REASON_XRSTORS] = handle_xrstors,
7885 [EXIT_REASON_PML_FULL] = handle_pml_full,
7886 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
7889 static const int kvm_vmx_max_exit_handlers =
7890 ARRAY_SIZE(kvm_vmx_exit_handlers);
7892 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7893 struct vmcs12 *vmcs12)
7895 unsigned long exit_qualification;
7896 gpa_t bitmap, last_bitmap;
7901 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7902 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7904 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7906 port = exit_qualification >> 16;
7907 size = (exit_qualification & 7) + 1;
7909 last_bitmap = (gpa_t)-1;
7914 bitmap = vmcs12->io_bitmap_a;
7915 else if (port < 0x10000)
7916 bitmap = vmcs12->io_bitmap_b;
7919 bitmap += (port & 0x7fff) / 8;
7921 if (last_bitmap != bitmap)
7922 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
7924 if (b & (1 << (port & 7)))
7929 last_bitmap = bitmap;
7936 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7937 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7938 * disinterest in the current event (read or write a specific MSR) by using an
7939 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7941 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7942 struct vmcs12 *vmcs12, u32 exit_reason)
7944 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7947 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7951 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7952 * for the four combinations of read/write and low/high MSR numbers.
7953 * First we need to figure out which of the four to use:
7955 bitmap = vmcs12->msr_bitmap;
7956 if (exit_reason == EXIT_REASON_MSR_WRITE)
7958 if (msr_index >= 0xc0000000) {
7959 msr_index -= 0xc0000000;
7963 /* Then read the msr_index'th bit from this bitmap: */
7964 if (msr_index < 1024*8) {
7966 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
7968 return 1 & (b >> (msr_index & 7));
7970 return true; /* let L1 handle the wrong parameter */
7974 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7975 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7976 * intercept (via guest_host_mask etc.) the current event.
7978 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7979 struct vmcs12 *vmcs12)
7981 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7982 int cr = exit_qualification & 15;
7986 switch ((exit_qualification >> 4) & 3) {
7987 case 0: /* mov to cr */
7988 reg = (exit_qualification >> 8) & 15;
7989 val = kvm_register_readl(vcpu, reg);
7992 if (vmcs12->cr0_guest_host_mask &
7993 (val ^ vmcs12->cr0_read_shadow))
7997 if ((vmcs12->cr3_target_count >= 1 &&
7998 vmcs12->cr3_target_value0 == val) ||
7999 (vmcs12->cr3_target_count >= 2 &&
8000 vmcs12->cr3_target_value1 == val) ||
8001 (vmcs12->cr3_target_count >= 3 &&
8002 vmcs12->cr3_target_value2 == val) ||
8003 (vmcs12->cr3_target_count >= 4 &&
8004 vmcs12->cr3_target_value3 == val))
8006 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
8010 if (vmcs12->cr4_guest_host_mask &
8011 (vmcs12->cr4_read_shadow ^ val))
8015 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
8021 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
8022 (vmcs12->cr0_read_shadow & X86_CR0_TS))
8025 case 1: /* mov from cr */
8028 if (vmcs12->cpu_based_vm_exec_control &
8029 CPU_BASED_CR3_STORE_EXITING)
8033 if (vmcs12->cpu_based_vm_exec_control &
8034 CPU_BASED_CR8_STORE_EXITING)
8041 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
8042 * cr0. Other attempted changes are ignored, with no exit.
8044 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
8045 if (vmcs12->cr0_guest_host_mask & 0xe &
8046 (val ^ vmcs12->cr0_read_shadow))
8048 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
8049 !(vmcs12->cr0_read_shadow & 0x1) &&
8058 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
8059 * should handle it ourselves in L0 (and then continue L2). Only call this
8060 * when in is_guest_mode (L2).
8062 static bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
8064 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8065 struct vcpu_vmx *vmx = to_vmx(vcpu);
8066 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8068 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
8069 vmcs_readl(EXIT_QUALIFICATION),
8070 vmx->idt_vectoring_info,
8072 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8075 if (vmx->nested.nested_run_pending)
8078 if (unlikely(vmx->fail)) {
8079 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
8080 vmcs_read32(VM_INSTRUCTION_ERROR));
8084 switch (exit_reason) {
8085 case EXIT_REASON_EXCEPTION_NMI:
8086 if (is_nmi(intr_info))
8088 else if (is_page_fault(intr_info))
8089 return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
8090 else if (is_no_device(intr_info) &&
8091 !(vmcs12->guest_cr0 & X86_CR0_TS))
8093 else if (is_debug(intr_info) &&
8095 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
8097 else if (is_breakpoint(intr_info) &&
8098 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
8100 return vmcs12->exception_bitmap &
8101 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
8102 case EXIT_REASON_EXTERNAL_INTERRUPT:
8104 case EXIT_REASON_TRIPLE_FAULT:
8106 case EXIT_REASON_PENDING_INTERRUPT:
8107 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
8108 case EXIT_REASON_NMI_WINDOW:
8109 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
8110 case EXIT_REASON_TASK_SWITCH:
8112 case EXIT_REASON_CPUID:
8114 case EXIT_REASON_HLT:
8115 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
8116 case EXIT_REASON_INVD:
8118 case EXIT_REASON_INVLPG:
8119 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8120 case EXIT_REASON_RDPMC:
8121 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
8122 case EXIT_REASON_RDRAND:
8123 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND);
8124 case EXIT_REASON_RDSEED:
8125 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED);
8126 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
8127 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
8128 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
8129 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
8130 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
8131 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
8132 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
8133 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
8135 * VMX instructions trap unconditionally. This allows L1 to
8136 * emulate them for its L2 guest, i.e., allows 3-level nesting!
8139 case EXIT_REASON_CR_ACCESS:
8140 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
8141 case EXIT_REASON_DR_ACCESS:
8142 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
8143 case EXIT_REASON_IO_INSTRUCTION:
8144 return nested_vmx_exit_handled_io(vcpu, vmcs12);
8145 case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
8146 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
8147 case EXIT_REASON_MSR_READ:
8148 case EXIT_REASON_MSR_WRITE:
8149 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
8150 case EXIT_REASON_INVALID_STATE:
8152 case EXIT_REASON_MWAIT_INSTRUCTION:
8153 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
8154 case EXIT_REASON_MONITOR_TRAP_FLAG:
8155 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
8156 case EXIT_REASON_MONITOR_INSTRUCTION:
8157 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
8158 case EXIT_REASON_PAUSE_INSTRUCTION:
8159 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
8160 nested_cpu_has2(vmcs12,
8161 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
8162 case EXIT_REASON_MCE_DURING_VMENTRY:
8164 case EXIT_REASON_TPR_BELOW_THRESHOLD:
8165 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
8166 case EXIT_REASON_APIC_ACCESS:
8167 return nested_cpu_has2(vmcs12,
8168 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
8169 case EXIT_REASON_APIC_WRITE:
8170 case EXIT_REASON_EOI_INDUCED:
8171 /* apic_write and eoi_induced should exit unconditionally. */
8173 case EXIT_REASON_EPT_VIOLATION:
8175 * L0 always deals with the EPT violation. If nested EPT is
8176 * used, and the nested mmu code discovers that the address is
8177 * missing in the guest EPT table (EPT12), the EPT violation
8178 * will be injected with nested_ept_inject_page_fault()
8181 case EXIT_REASON_EPT_MISCONFIG:
8183 * L2 never uses directly L1's EPT, but rather L0's own EPT
8184 * table (shadow on EPT) or a merged EPT table that L0 built
8185 * (EPT on EPT). So any problems with the structure of the
8186 * table is L0's fault.
8189 case EXIT_REASON_WBINVD:
8190 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
8191 case EXIT_REASON_XSETBV:
8193 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
8195 * This should never happen, since it is not possible to
8196 * set XSS to a non-zero value---neither in L1 nor in L2.
8197 * If if it were, XSS would have to be checked against
8198 * the XSS exit bitmap in vmcs12.
8200 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
8201 case EXIT_REASON_PREEMPTION_TIMER:
8203 case EXIT_REASON_PML_FULL:
8204 /* We emulate PML support to L1. */
8211 static int nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason)
8213 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8216 * At this point, the exit interruption info in exit_intr_info
8217 * is only valid for EXCEPTION_NMI exits. For EXTERNAL_INTERRUPT
8218 * we need to query the in-kernel LAPIC.
8220 WARN_ON(exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT);
8221 if ((exit_intr_info &
8222 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
8223 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) {
8224 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8225 vmcs12->vm_exit_intr_error_code =
8226 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
8229 nested_vmx_vmexit(vcpu, exit_reason, exit_intr_info,
8230 vmcs_readl(EXIT_QUALIFICATION));
8234 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
8236 *info1 = vmcs_readl(EXIT_QUALIFICATION);
8237 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
8240 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
8243 __free_page(vmx->pml_pg);
8248 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
8250 struct vcpu_vmx *vmx = to_vmx(vcpu);
8254 pml_idx = vmcs_read16(GUEST_PML_INDEX);
8256 /* Do nothing if PML buffer is empty */
8257 if (pml_idx == (PML_ENTITY_NUM - 1))
8260 /* PML index always points to next available PML buffer entity */
8261 if (pml_idx >= PML_ENTITY_NUM)
8266 pml_buf = page_address(vmx->pml_pg);
8267 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
8270 gpa = pml_buf[pml_idx];
8271 WARN_ON(gpa & (PAGE_SIZE - 1));
8272 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
8275 /* reset PML index */
8276 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8280 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8281 * Called before reporting dirty_bitmap to userspace.
8283 static void kvm_flush_pml_buffers(struct kvm *kvm)
8286 struct kvm_vcpu *vcpu;
8288 * We only need to kick vcpu out of guest mode here, as PML buffer
8289 * is flushed at beginning of all VMEXITs, and it's obvious that only
8290 * vcpus running in guest are possible to have unflushed GPAs in PML
8293 kvm_for_each_vcpu(i, vcpu, kvm)
8294 kvm_vcpu_kick(vcpu);
8297 static void vmx_dump_sel(char *name, uint32_t sel)
8299 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8300 name, vmcs_read16(sel),
8301 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
8302 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
8303 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
8306 static void vmx_dump_dtsel(char *name, uint32_t limit)
8308 pr_err("%s limit=0x%08x, base=0x%016lx\n",
8309 name, vmcs_read32(limit),
8310 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
8313 static void dump_vmcs(void)
8315 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
8316 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
8317 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
8318 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
8319 u32 secondary_exec_control = 0;
8320 unsigned long cr4 = vmcs_readl(GUEST_CR4);
8321 u64 efer = vmcs_read64(GUEST_IA32_EFER);
8324 if (cpu_has_secondary_exec_ctrls())
8325 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8327 pr_err("*** Guest State ***\n");
8328 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8329 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
8330 vmcs_readl(CR0_GUEST_HOST_MASK));
8331 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8332 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
8333 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
8334 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
8335 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
8337 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
8338 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
8339 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
8340 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
8342 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
8343 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
8344 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
8345 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
8346 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8347 vmcs_readl(GUEST_SYSENTER_ESP),
8348 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8349 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
8350 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
8351 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
8352 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
8353 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
8354 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
8355 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8356 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8357 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8358 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
8359 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8360 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8361 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8362 efer, vmcs_read64(GUEST_IA32_PAT));
8363 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
8364 vmcs_read64(GUEST_IA32_DEBUGCTL),
8365 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8366 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8367 pr_err("PerfGlobCtl = 0x%016llx\n",
8368 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8369 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8370 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8371 pr_err("Interruptibility = %08x ActivityState = %08x\n",
8372 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8373 vmcs_read32(GUEST_ACTIVITY_STATE));
8374 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8375 pr_err("InterruptStatus = %04x\n",
8376 vmcs_read16(GUEST_INTR_STATUS));
8378 pr_err("*** Host State ***\n");
8379 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
8380 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8381 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8382 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8383 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8384 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8385 vmcs_read16(HOST_TR_SELECTOR));
8386 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8387 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8388 vmcs_readl(HOST_TR_BASE));
8389 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8390 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8391 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8392 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8393 vmcs_readl(HOST_CR4));
8394 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8395 vmcs_readl(HOST_IA32_SYSENTER_ESP),
8396 vmcs_read32(HOST_IA32_SYSENTER_CS),
8397 vmcs_readl(HOST_IA32_SYSENTER_EIP));
8398 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8399 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8400 vmcs_read64(HOST_IA32_EFER),
8401 vmcs_read64(HOST_IA32_PAT));
8402 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8403 pr_err("PerfGlobCtl = 0x%016llx\n",
8404 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8406 pr_err("*** Control State ***\n");
8407 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8408 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8409 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8410 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8411 vmcs_read32(EXCEPTION_BITMAP),
8412 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8413 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8414 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8415 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8416 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8417 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8418 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8419 vmcs_read32(VM_EXIT_INTR_INFO),
8420 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8421 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8422 pr_err(" reason=%08x qualification=%016lx\n",
8423 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8424 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8425 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8426 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8427 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8428 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8429 pr_err("TSC Multiplier = 0x%016llx\n",
8430 vmcs_read64(TSC_MULTIPLIER));
8431 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8432 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8433 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8434 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8435 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8436 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8437 n = vmcs_read32(CR3_TARGET_COUNT);
8438 for (i = 0; i + 1 < n; i += 4)
8439 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8440 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8441 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8443 pr_err("CR3 target%u=%016lx\n",
8444 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8445 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8446 pr_err("PLE Gap=%08x Window=%08x\n",
8447 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8448 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8449 pr_err("Virtual processor ID = 0x%04x\n",
8450 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8454 * The guest has exited. See if we can fix it or if we need userspace
8457 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8459 struct vcpu_vmx *vmx = to_vmx(vcpu);
8460 u32 exit_reason = vmx->exit_reason;
8461 u32 vectoring_info = vmx->idt_vectoring_info;
8463 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8464 vcpu->arch.gpa_available = false;
8467 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8468 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8469 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8470 * mode as if vcpus is in root mode, the PML buffer must has been
8474 vmx_flush_pml_buffer(vcpu);
8476 /* If guest state is invalid, start emulating */
8477 if (vmx->emulation_required)
8478 return handle_invalid_guest_state(vcpu);
8480 if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
8481 return nested_vmx_reflect_vmexit(vcpu, exit_reason);
8483 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8485 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8486 vcpu->run->fail_entry.hardware_entry_failure_reason
8491 if (unlikely(vmx->fail)) {
8492 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8493 vcpu->run->fail_entry.hardware_entry_failure_reason
8494 = vmcs_read32(VM_INSTRUCTION_ERROR);
8500 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8501 * delivery event since it indicates guest is accessing MMIO.
8502 * The vm-exit can be triggered again after return to guest that
8503 * will cause infinite loop.
8505 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8506 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8507 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8508 exit_reason != EXIT_REASON_PML_FULL &&
8509 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8510 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8511 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8512 vcpu->run->internal.ndata = 3;
8513 vcpu->run->internal.data[0] = vectoring_info;
8514 vcpu->run->internal.data[1] = exit_reason;
8515 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
8516 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
8517 vcpu->run->internal.ndata++;
8518 vcpu->run->internal.data[3] =
8519 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
8524 if (exit_reason < kvm_vmx_max_exit_handlers
8525 && kvm_vmx_exit_handlers[exit_reason])
8526 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8528 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
8530 kvm_queue_exception(vcpu, UD_VECTOR);
8535 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8537 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8539 if (is_guest_mode(vcpu) &&
8540 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8543 if (irr == -1 || tpr < irr) {
8544 vmcs_write32(TPR_THRESHOLD, 0);
8548 vmcs_write32(TPR_THRESHOLD, irr);
8551 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8553 u32 sec_exec_control;
8555 /* Postpone execution until vmcs01 is the current VMCS. */
8556 if (is_guest_mode(vcpu)) {
8557 to_vmx(vcpu)->nested.change_vmcs01_virtual_x2apic_mode = true;
8561 if (!cpu_has_vmx_virtualize_x2apic_mode())
8564 if (!cpu_need_tpr_shadow(vcpu))
8567 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8570 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8571 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8573 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8574 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8575 vmx_flush_tlb_ept_only(vcpu);
8577 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8579 vmx_set_msr_bitmap(vcpu);
8582 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8584 struct vcpu_vmx *vmx = to_vmx(vcpu);
8587 * Currently we do not handle the nested case where L2 has an
8588 * APIC access page of its own; that page is still pinned.
8589 * Hence, we skip the case where the VCPU is in guest mode _and_
8590 * L1 prepared an APIC access page for L2.
8592 * For the case where L1 and L2 share the same APIC access page
8593 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8594 * in the vmcs12), this function will only update either the vmcs01
8595 * or the vmcs02. If the former, the vmcs02 will be updated by
8596 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8597 * the next L2->L1 exit.
8599 if (!is_guest_mode(vcpu) ||
8600 !nested_cpu_has2(get_vmcs12(&vmx->vcpu),
8601 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
8602 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8603 vmx_flush_tlb_ept_only(vcpu);
8607 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
8615 status = vmcs_read16(GUEST_INTR_STATUS);
8617 if (max_isr != old) {
8619 status |= max_isr << 8;
8620 vmcs_write16(GUEST_INTR_STATUS, status);
8624 static void vmx_set_rvi(int vector)
8632 status = vmcs_read16(GUEST_INTR_STATUS);
8633 old = (u8)status & 0xff;
8634 if ((u8)vector != old) {
8636 status |= (u8)vector;
8637 vmcs_write16(GUEST_INTR_STATUS, status);
8641 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8643 if (!is_guest_mode(vcpu)) {
8644 vmx_set_rvi(max_irr);
8652 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8655 if (nested_exit_on_intr(vcpu))
8659 * Else, fall back to pre-APICv interrupt injection since L2
8660 * is run without virtual interrupt delivery.
8662 if (!kvm_event_needs_reinjection(vcpu) &&
8663 vmx_interrupt_allowed(vcpu)) {
8664 kvm_queue_interrupt(vcpu, max_irr, false);
8665 vmx_inject_irq(vcpu);
8669 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
8671 struct vcpu_vmx *vmx = to_vmx(vcpu);
8674 WARN_ON(!vcpu->arch.apicv_active);
8675 if (pi_test_on(&vmx->pi_desc)) {
8676 pi_clear_on(&vmx->pi_desc);
8678 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
8679 * But on x86 this is just a compiler barrier anyway.
8681 smp_mb__after_atomic();
8682 max_irr = kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
8684 max_irr = kvm_lapic_find_highest_irr(vcpu);
8686 vmx_hwapic_irr_update(vcpu, max_irr);
8690 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
8692 if (!kvm_vcpu_apicv_active(vcpu))
8695 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8696 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8697 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8698 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8701 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
8703 struct vcpu_vmx *vmx = to_vmx(vcpu);
8705 pi_clear_on(&vmx->pi_desc);
8706 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
8709 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8711 u32 exit_intr_info = 0;
8712 u16 basic_exit_reason = (u16)vmx->exit_reason;
8714 if (!(basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8715 || basic_exit_reason == EXIT_REASON_EXCEPTION_NMI))
8718 if (!(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
8719 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8720 vmx->exit_intr_info = exit_intr_info;
8722 /* if exit due to PF check for async PF */
8723 if (is_page_fault(exit_intr_info))
8724 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
8726 /* Handle machine checks before interrupts are enabled */
8727 if (basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY ||
8728 is_machine_check(exit_intr_info))
8729 kvm_machine_check();
8731 /* We need to handle NMIs before interrupts are enabled */
8732 if (is_nmi(exit_intr_info)) {
8733 kvm_before_handle_nmi(&vmx->vcpu);
8735 kvm_after_handle_nmi(&vmx->vcpu);
8739 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8741 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8742 register void *__sp asm(_ASM_SP);
8744 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8745 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8746 unsigned int vector;
8747 unsigned long entry;
8749 struct vcpu_vmx *vmx = to_vmx(vcpu);
8750 #ifdef CONFIG_X86_64
8754 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8755 desc = (gate_desc *)vmx->host_idt_base + vector;
8756 entry = gate_offset(*desc);
8758 #ifdef CONFIG_X86_64
8759 "mov %%" _ASM_SP ", %[sp]\n\t"
8760 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8765 __ASM_SIZE(push) " $%c[cs]\n\t"
8766 "call *%[entry]\n\t"
8768 #ifdef CONFIG_X86_64
8774 [ss]"i"(__KERNEL_DS),
8775 [cs]"i"(__KERNEL_CS)
8779 STACK_FRAME_NON_STANDARD(vmx_handle_external_intr);
8781 static bool vmx_has_high_real_mode_segbase(void)
8783 return enable_unrestricted_guest || emulate_invalid_guest_state;
8786 static bool vmx_mpx_supported(void)
8788 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8789 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8792 static bool vmx_xsaves_supported(void)
8794 return vmcs_config.cpu_based_2nd_exec_ctrl &
8795 SECONDARY_EXEC_XSAVES;
8798 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8803 bool idtv_info_valid;
8805 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8807 if (vmx->loaded_vmcs->nmi_known_unmasked)
8810 * Can't use vmx->exit_intr_info since we're not sure what
8811 * the exit reason is.
8813 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8814 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8815 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8817 * SDM 3: 27.7.1.2 (September 2008)
8818 * Re-set bit "block by NMI" before VM entry if vmexit caused by
8819 * a guest IRET fault.
8820 * SDM 3: 23.2.2 (September 2008)
8821 * Bit 12 is undefined in any of the following cases:
8822 * If the VM exit sets the valid bit in the IDT-vectoring
8823 * information field.
8824 * If the VM exit is due to a double fault.
8826 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8827 vector != DF_VECTOR && !idtv_info_valid)
8828 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8829 GUEST_INTR_STATE_NMI);
8831 vmx->loaded_vmcs->nmi_known_unmasked =
8832 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8833 & GUEST_INTR_STATE_NMI);
8836 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8837 u32 idt_vectoring_info,
8838 int instr_len_field,
8839 int error_code_field)
8843 bool idtv_info_valid;
8845 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8847 vcpu->arch.nmi_injected = false;
8848 kvm_clear_exception_queue(vcpu);
8849 kvm_clear_interrupt_queue(vcpu);
8851 if (!idtv_info_valid)
8854 kvm_make_request(KVM_REQ_EVENT, vcpu);
8856 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8857 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8860 case INTR_TYPE_NMI_INTR:
8861 vcpu->arch.nmi_injected = true;
8863 * SDM 3: 27.7.1.2 (September 2008)
8864 * Clear bit "block by NMI" before VM entry if a NMI
8867 vmx_set_nmi_mask(vcpu, false);
8869 case INTR_TYPE_SOFT_EXCEPTION:
8870 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8872 case INTR_TYPE_HARD_EXCEPTION:
8873 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8874 u32 err = vmcs_read32(error_code_field);
8875 kvm_requeue_exception_e(vcpu, vector, err);
8877 kvm_requeue_exception(vcpu, vector);
8879 case INTR_TYPE_SOFT_INTR:
8880 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8882 case INTR_TYPE_EXT_INTR:
8883 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8890 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8892 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8893 VM_EXIT_INSTRUCTION_LEN,
8894 IDT_VECTORING_ERROR_CODE);
8897 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8899 __vmx_complete_interrupts(vcpu,
8900 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8901 VM_ENTRY_INSTRUCTION_LEN,
8902 VM_ENTRY_EXCEPTION_ERROR_CODE);
8904 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8907 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8910 struct perf_guest_switch_msr *msrs;
8912 msrs = perf_guest_get_msrs(&nr_msrs);
8917 for (i = 0; i < nr_msrs; i++)
8918 if (msrs[i].host == msrs[i].guest)
8919 clear_atomic_switch_msr(vmx, msrs[i].msr);
8921 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8925 static void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
8927 struct vcpu_vmx *vmx = to_vmx(vcpu);
8931 if (vmx->hv_deadline_tsc == -1)
8935 if (vmx->hv_deadline_tsc > tscl)
8936 /* sure to be 32 bit only because checked on set_hv_timer */
8937 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
8938 cpu_preemption_timer_multi);
8942 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
8945 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8947 struct vcpu_vmx *vmx = to_vmx(vcpu);
8948 unsigned long debugctlmsr, cr3, cr4;
8950 /* Don't enter VMX if guest state is invalid, let the exit handler
8951 start emulation until we arrive back to a valid state */
8952 if (vmx->emulation_required)
8955 if (vmx->ple_window_dirty) {
8956 vmx->ple_window_dirty = false;
8957 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8960 if (vmx->nested.sync_shadow_vmcs) {
8961 copy_vmcs12_to_shadow(vmx);
8962 vmx->nested.sync_shadow_vmcs = false;
8965 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8966 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
8967 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
8968 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
8970 cr3 = __get_current_cr3_fast();
8971 if (unlikely(cr3 != vmx->host_state.vmcs_host_cr3)) {
8972 vmcs_writel(HOST_CR3, cr3);
8973 vmx->host_state.vmcs_host_cr3 = cr3;
8976 cr4 = cr4_read_shadow();
8977 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
8978 vmcs_writel(HOST_CR4, cr4);
8979 vmx->host_state.vmcs_host_cr4 = cr4;
8982 /* When single-stepping over STI and MOV SS, we must clear the
8983 * corresponding interruptibility bits in the guest state. Otherwise
8984 * vmentry fails as it then expects bit 14 (BS) in pending debug
8985 * exceptions being set, but that's not correct for the guest debugging
8987 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8988 vmx_set_interrupt_shadow(vcpu, 0);
8990 if (vmx->guest_pkru_valid)
8991 __write_pkru(vmx->guest_pkru);
8993 atomic_switch_perf_msrs(vmx);
8994 debugctlmsr = get_debugctlmsr();
8996 vmx_arm_hv_timer(vcpu);
8998 vmx->__launched = vmx->loaded_vmcs->launched;
9000 /* Store host registers */
9001 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
9002 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
9003 "push %%" _ASM_CX " \n\t"
9004 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9006 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9007 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
9009 /* Reload cr2 if changed */
9010 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
9011 "mov %%cr2, %%" _ASM_DX " \n\t"
9012 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
9014 "mov %%" _ASM_AX", %%cr2 \n\t"
9016 /* Check if vmlaunch of vmresume is needed */
9017 "cmpl $0, %c[launched](%0) \n\t"
9018 /* Load guest registers. Don't clobber flags. */
9019 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
9020 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
9021 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
9022 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
9023 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
9024 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
9025 #ifdef CONFIG_X86_64
9026 "mov %c[r8](%0), %%r8 \n\t"
9027 "mov %c[r9](%0), %%r9 \n\t"
9028 "mov %c[r10](%0), %%r10 \n\t"
9029 "mov %c[r11](%0), %%r11 \n\t"
9030 "mov %c[r12](%0), %%r12 \n\t"
9031 "mov %c[r13](%0), %%r13 \n\t"
9032 "mov %c[r14](%0), %%r14 \n\t"
9033 "mov %c[r15](%0), %%r15 \n\t"
9035 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
9037 /* Enter guest mode */
9039 __ex(ASM_VMX_VMLAUNCH) "\n\t"
9041 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
9043 /* Save guest registers, load host registers, keep flags */
9044 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
9046 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
9047 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
9048 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
9049 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
9050 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
9051 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
9052 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
9053 #ifdef CONFIG_X86_64
9054 "mov %%r8, %c[r8](%0) \n\t"
9055 "mov %%r9, %c[r9](%0) \n\t"
9056 "mov %%r10, %c[r10](%0) \n\t"
9057 "mov %%r11, %c[r11](%0) \n\t"
9058 "mov %%r12, %c[r12](%0) \n\t"
9059 "mov %%r13, %c[r13](%0) \n\t"
9060 "mov %%r14, %c[r14](%0) \n\t"
9061 "mov %%r15, %c[r15](%0) \n\t"
9063 "mov %%cr2, %%" _ASM_AX " \n\t"
9064 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
9066 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
9067 "setbe %c[fail](%0) \n\t"
9068 ".pushsection .rodata \n\t"
9069 ".global vmx_return \n\t"
9070 "vmx_return: " _ASM_PTR " 2b \n\t"
9072 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
9073 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
9074 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
9075 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
9076 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
9077 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
9078 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
9079 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
9080 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
9081 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
9082 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
9083 #ifdef CONFIG_X86_64
9084 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
9085 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
9086 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
9087 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
9088 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
9089 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
9090 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
9091 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
9093 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
9094 [wordsize]"i"(sizeof(ulong))
9096 #ifdef CONFIG_X86_64
9097 , "rax", "rbx", "rdi", "rsi"
9098 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
9100 , "eax", "ebx", "edi", "esi"
9104 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
9106 update_debugctlmsr(debugctlmsr);
9108 #ifndef CONFIG_X86_64
9110 * The sysexit path does not restore ds/es, so we must set them to
9111 * a reasonable value ourselves.
9113 * We can't defer this to vmx_load_host_state() since that function
9114 * may be executed in interrupt context, which saves and restore segments
9115 * around it, nullifying its effect.
9117 loadsegment(ds, __USER_DS);
9118 loadsegment(es, __USER_DS);
9121 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
9122 | (1 << VCPU_EXREG_RFLAGS)
9123 | (1 << VCPU_EXREG_PDPTR)
9124 | (1 << VCPU_EXREG_SEGMENTS)
9125 | (1 << VCPU_EXREG_CR3));
9126 vcpu->arch.regs_dirty = 0;
9128 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9130 vmx->loaded_vmcs->launched = 1;
9132 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
9135 * eager fpu is enabled if PKEY is supported and CR4 is switched
9136 * back on host, so it is safe to read guest PKRU from current
9139 if (boot_cpu_has(X86_FEATURE_OSPKE)) {
9140 vmx->guest_pkru = __read_pkru();
9141 if (vmx->guest_pkru != vmx->host_pkru) {
9142 vmx->guest_pkru_valid = true;
9143 __write_pkru(vmx->host_pkru);
9145 vmx->guest_pkru_valid = false;
9149 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
9150 * we did not inject a still-pending event to L1 now because of
9151 * nested_run_pending, we need to re-enable this bit.
9153 if (vmx->nested.nested_run_pending)
9154 kvm_make_request(KVM_REQ_EVENT, vcpu);
9156 vmx->nested.nested_run_pending = 0;
9158 vmx_complete_atomic_exit(vmx);
9159 vmx_recover_nmi_blocking(vmx);
9160 vmx_complete_interrupts(vmx);
9162 STACK_FRAME_NON_STANDARD(vmx_vcpu_run);
9164 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
9166 struct vcpu_vmx *vmx = to_vmx(vcpu);
9169 if (vmx->loaded_vmcs == vmcs)
9173 vmx->loaded_vmcs = vmcs;
9175 vmx_vcpu_load(vcpu, cpu);
9181 * Ensure that the current vmcs of the logical processor is the
9182 * vmcs01 of the vcpu before calling free_nested().
9184 static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
9186 struct vcpu_vmx *vmx = to_vmx(vcpu);
9189 r = vcpu_load(vcpu);
9191 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
9196 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
9198 struct vcpu_vmx *vmx = to_vmx(vcpu);
9201 vmx_destroy_pml_buffer(vmx);
9202 free_vpid(vmx->vpid);
9203 leave_guest_mode(vcpu);
9204 vmx_free_vcpu_nested(vcpu);
9205 free_loaded_vmcs(vmx->loaded_vmcs);
9206 kfree(vmx->guest_msrs);
9207 kvm_vcpu_uninit(vcpu);
9208 kmem_cache_free(kvm_vcpu_cache, vmx);
9211 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
9214 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
9218 return ERR_PTR(-ENOMEM);
9220 vmx->vpid = allocate_vpid();
9222 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
9229 * If PML is turned on, failure on enabling PML just results in failure
9230 * of creating the vcpu, therefore we can simplify PML logic (by
9231 * avoiding dealing with cases, such as enabling PML partially on vcpus
9232 * for the guest, etc.
9235 vmx->pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
9240 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
9241 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
9244 if (!vmx->guest_msrs)
9247 vmx->loaded_vmcs = &vmx->vmcs01;
9248 vmx->loaded_vmcs->vmcs = alloc_vmcs();
9249 vmx->loaded_vmcs->shadow_vmcs = NULL;
9250 if (!vmx->loaded_vmcs->vmcs)
9252 loaded_vmcs_init(vmx->loaded_vmcs);
9255 vmx_vcpu_load(&vmx->vcpu, cpu);
9256 vmx->vcpu.cpu = cpu;
9257 err = vmx_vcpu_setup(vmx);
9258 vmx_vcpu_put(&vmx->vcpu);
9262 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9263 err = alloc_apic_access_page(kvm);
9269 if (!kvm->arch.ept_identity_map_addr)
9270 kvm->arch.ept_identity_map_addr =
9271 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
9272 err = init_rmode_identity_map(kvm);
9278 nested_vmx_setup_ctls_msrs(vmx);
9279 vmx->nested.vpid02 = allocate_vpid();
9282 vmx->nested.posted_intr_nv = -1;
9283 vmx->nested.current_vmptr = -1ull;
9285 vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
9290 free_vpid(vmx->nested.vpid02);
9291 free_loaded_vmcs(vmx->loaded_vmcs);
9293 kfree(vmx->guest_msrs);
9295 vmx_destroy_pml_buffer(vmx);
9297 kvm_vcpu_uninit(&vmx->vcpu);
9299 free_vpid(vmx->vpid);
9300 kmem_cache_free(kvm_vcpu_cache, vmx);
9301 return ERR_PTR(err);
9304 static void __init vmx_check_processor_compat(void *rtn)
9306 struct vmcs_config vmcs_conf;
9309 if (setup_vmcs_config(&vmcs_conf) < 0)
9311 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
9312 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
9313 smp_processor_id());
9318 static int get_ept_level(void)
9320 return VMX_EPT_DEFAULT_GAW + 1;
9323 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
9328 /* For VT-d and EPT combination
9329 * 1. MMIO: always map as UC
9331 * a. VT-d without snooping control feature: can't guarantee the
9332 * result, try to trust guest.
9333 * b. VT-d with snooping control feature: snooping control feature of
9334 * VT-d engine can guarantee the cache correctness. Just set it
9335 * to WB to keep consistent with host. So the same as item 3.
9336 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9337 * consistent with host MTRR
9340 cache = MTRR_TYPE_UNCACHABLE;
9344 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
9345 ipat = VMX_EPT_IPAT_BIT;
9346 cache = MTRR_TYPE_WRBACK;
9350 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
9351 ipat = VMX_EPT_IPAT_BIT;
9352 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
9353 cache = MTRR_TYPE_WRBACK;
9355 cache = MTRR_TYPE_UNCACHABLE;
9359 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
9362 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
9365 static int vmx_get_lpage_level(void)
9367 if (enable_ept && !cpu_has_vmx_ept_1g_page())
9368 return PT_DIRECTORY_LEVEL;
9370 /* For shadow and EPT supported 1GB page */
9371 return PT_PDPE_LEVEL;
9374 static void vmcs_set_secondary_exec_control(u32 new_ctl)
9377 * These bits in the secondary execution controls field
9378 * are dynamic, the others are mostly based on the hypervisor
9379 * architecture and the guest's CPUID. Do not touch the
9383 SECONDARY_EXEC_SHADOW_VMCS |
9384 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
9385 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9387 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9389 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
9390 (new_ctl & ~mask) | (cur_ctl & mask));
9394 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
9395 * (indicating "allowed-1") if they are supported in the guest's CPUID.
9397 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
9399 struct vcpu_vmx *vmx = to_vmx(vcpu);
9400 struct kvm_cpuid_entry2 *entry;
9402 vmx->nested.nested_vmx_cr0_fixed1 = 0xffffffff;
9403 vmx->nested.nested_vmx_cr4_fixed1 = X86_CR4_PCE;
9405 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
9406 if (entry && (entry->_reg & (_cpuid_mask))) \
9407 vmx->nested.nested_vmx_cr4_fixed1 |= (_cr4_mask); \
9410 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
9411 cr4_fixed1_update(X86_CR4_VME, edx, bit(X86_FEATURE_VME));
9412 cr4_fixed1_update(X86_CR4_PVI, edx, bit(X86_FEATURE_VME));
9413 cr4_fixed1_update(X86_CR4_TSD, edx, bit(X86_FEATURE_TSC));
9414 cr4_fixed1_update(X86_CR4_DE, edx, bit(X86_FEATURE_DE));
9415 cr4_fixed1_update(X86_CR4_PSE, edx, bit(X86_FEATURE_PSE));
9416 cr4_fixed1_update(X86_CR4_PAE, edx, bit(X86_FEATURE_PAE));
9417 cr4_fixed1_update(X86_CR4_MCE, edx, bit(X86_FEATURE_MCE));
9418 cr4_fixed1_update(X86_CR4_PGE, edx, bit(X86_FEATURE_PGE));
9419 cr4_fixed1_update(X86_CR4_OSFXSR, edx, bit(X86_FEATURE_FXSR));
9420 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
9421 cr4_fixed1_update(X86_CR4_VMXE, ecx, bit(X86_FEATURE_VMX));
9422 cr4_fixed1_update(X86_CR4_SMXE, ecx, bit(X86_FEATURE_SMX));
9423 cr4_fixed1_update(X86_CR4_PCIDE, ecx, bit(X86_FEATURE_PCID));
9424 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, bit(X86_FEATURE_XSAVE));
9426 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9427 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, bit(X86_FEATURE_FSGSBASE));
9428 cr4_fixed1_update(X86_CR4_SMEP, ebx, bit(X86_FEATURE_SMEP));
9429 cr4_fixed1_update(X86_CR4_SMAP, ebx, bit(X86_FEATURE_SMAP));
9430 cr4_fixed1_update(X86_CR4_PKE, ecx, bit(X86_FEATURE_PKU));
9431 /* TODO: Use X86_CR4_UMIP and X86_FEATURE_UMIP macros */
9432 cr4_fixed1_update(bit(11), ecx, bit(2));
9434 #undef cr4_fixed1_update
9437 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
9439 struct kvm_cpuid_entry2 *best;
9440 struct vcpu_vmx *vmx = to_vmx(vcpu);
9441 u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx);
9443 if (vmx_rdtscp_supported()) {
9444 bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu);
9445 if (!rdtscp_enabled)
9446 secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP;
9450 vmx->nested.nested_vmx_secondary_ctls_high |=
9451 SECONDARY_EXEC_RDTSCP;
9453 vmx->nested.nested_vmx_secondary_ctls_high &=
9454 ~SECONDARY_EXEC_RDTSCP;
9458 /* Exposing INVPCID only when PCID is exposed */
9459 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9460 if (vmx_invpcid_supported() &&
9461 (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) ||
9462 !guest_cpuid_has_pcid(vcpu))) {
9463 secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID;
9466 best->ebx &= ~bit(X86_FEATURE_INVPCID);
9469 if (cpu_has_secondary_exec_ctrls())
9470 vmcs_set_secondary_exec_control(secondary_exec_ctl);
9472 if (nested_vmx_allowed(vcpu))
9473 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
9474 FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9476 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
9477 ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9479 if (nested_vmx_allowed(vcpu))
9480 nested_vmx_cr_fixed1_bits_update(vcpu);
9483 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9485 if (func == 1 && nested)
9486 entry->ecx |= bit(X86_FEATURE_VMX);
9489 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9490 struct x86_exception *fault)
9492 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9493 struct vcpu_vmx *vmx = to_vmx(vcpu);
9495 unsigned long exit_qualification = vcpu->arch.exit_qualification;
9497 if (vmx->nested.pml_full) {
9498 exit_reason = EXIT_REASON_PML_FULL;
9499 vmx->nested.pml_full = false;
9500 exit_qualification &= INTR_INFO_UNBLOCK_NMI;
9501 } else if (fault->error_code & PFERR_RSVD_MASK)
9502 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9504 exit_reason = EXIT_REASON_EPT_VIOLATION;
9506 nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
9507 vmcs12->guest_physical_address = fault->address;
9510 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu)
9512 return nested_ept_get_cr3(vcpu) & VMX_EPT_AD_ENABLE_BIT;
9515 /* Callbacks for nested_ept_init_mmu_context: */
9517 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9519 /* return the page table to be shadowed - in our case, EPT12 */
9520 return get_vmcs12(vcpu)->ept_pointer;
9523 static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9527 WARN_ON(mmu_is_nested(vcpu));
9528 wants_ad = nested_ept_ad_enabled(vcpu);
9529 if (wants_ad && !enable_ept_ad_bits)
9532 kvm_mmu_unload(vcpu);
9533 kvm_init_shadow_ept_mmu(vcpu,
9534 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
9535 VMX_EPT_EXECUTE_ONLY_BIT,
9537 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
9538 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
9539 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
9541 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
9545 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
9547 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
9550 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9553 bool inequality, bit;
9555 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9557 (error_code & vmcs12->page_fault_error_code_mask) !=
9558 vmcs12->page_fault_error_code_match;
9559 return inequality ^ bit;
9562 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9563 struct x86_exception *fault)
9565 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9567 WARN_ON(!is_guest_mode(vcpu));
9569 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code)) {
9570 vmcs12->vm_exit_intr_error_code = fault->error_code;
9571 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
9572 PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
9573 INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
9576 kvm_inject_page_fault(vcpu, fault);
9580 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9581 struct vmcs12 *vmcs12);
9583 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9584 struct vmcs12 *vmcs12)
9586 struct vcpu_vmx *vmx = to_vmx(vcpu);
9589 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9591 * Translate L1 physical address to host physical
9592 * address for vmcs02. Keep the page pinned, so this
9593 * physical address remains valid. We keep a reference
9594 * to it so we can release it later.
9596 if (vmx->nested.apic_access_page) /* shouldn't happen */
9597 nested_release_page(vmx->nested.apic_access_page);
9598 vmx->nested.apic_access_page =
9599 nested_get_page(vcpu, vmcs12->apic_access_addr);
9601 * If translation failed, no matter: This feature asks
9602 * to exit when accessing the given address, and if it
9603 * can never be accessed, this feature won't do
9606 if (vmx->nested.apic_access_page) {
9607 hpa = page_to_phys(vmx->nested.apic_access_page);
9608 vmcs_write64(APIC_ACCESS_ADDR, hpa);
9610 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
9611 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9613 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9614 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9615 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
9616 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9617 kvm_vcpu_reload_apic_access_page(vcpu);
9620 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9621 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
9622 nested_release_page(vmx->nested.virtual_apic_page);
9623 vmx->nested.virtual_apic_page =
9624 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
9627 * If translation failed, VM entry will fail because
9628 * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
9629 * Failing the vm entry is _not_ what the processor
9630 * does but it's basically the only possibility we
9631 * have. We could still enter the guest if CR8 load
9632 * exits are enabled, CR8 store exits are enabled, and
9633 * virtualize APIC access is disabled; in this case
9634 * the processor would never use the TPR shadow and we
9635 * could simply clear the bit from the execution
9636 * control. But such a configuration is useless, so
9637 * let's keep the code simple.
9639 if (vmx->nested.virtual_apic_page) {
9640 hpa = page_to_phys(vmx->nested.virtual_apic_page);
9641 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
9645 if (nested_cpu_has_posted_intr(vmcs12)) {
9646 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9647 kunmap(vmx->nested.pi_desc_page);
9648 nested_release_page(vmx->nested.pi_desc_page);
9650 vmx->nested.pi_desc_page =
9651 nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
9652 vmx->nested.pi_desc =
9653 (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
9654 if (!vmx->nested.pi_desc) {
9655 nested_release_page_clean(vmx->nested.pi_desc_page);
9658 vmx->nested.pi_desc =
9659 (struct pi_desc *)((void *)vmx->nested.pi_desc +
9660 (unsigned long)(vmcs12->posted_intr_desc_addr &
9662 vmcs_write64(POSTED_INTR_DESC_ADDR,
9663 page_to_phys(vmx->nested.pi_desc_page) +
9664 (unsigned long)(vmcs12->posted_intr_desc_addr &
9667 if (cpu_has_vmx_msr_bitmap() &&
9668 nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS) &&
9669 nested_vmx_merge_msr_bitmap(vcpu, vmcs12))
9672 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
9673 CPU_BASED_USE_MSR_BITMAPS);
9676 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9678 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9679 struct vcpu_vmx *vmx = to_vmx(vcpu);
9681 if (vcpu->arch.virtual_tsc_khz == 0)
9684 /* Make sure short timeouts reliably trigger an immediate vmexit.
9685 * hrtimer_start does not guarantee this. */
9686 if (preemption_timeout <= 1) {
9687 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9691 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9692 preemption_timeout *= 1000000;
9693 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9694 hrtimer_start(&vmx->nested.preemption_timer,
9695 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9698 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
9699 struct vmcs12 *vmcs12)
9701 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
9704 if (!page_address_valid(vcpu, vmcs12->io_bitmap_a) ||
9705 !page_address_valid(vcpu, vmcs12->io_bitmap_b))
9711 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9712 struct vmcs12 *vmcs12)
9714 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9717 if (!page_address_valid(vcpu, vmcs12->msr_bitmap))
9724 * Merge L0's and L1's MSR bitmap, return false to indicate that
9725 * we do not use the hardware.
9727 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9728 struct vmcs12 *vmcs12)
9732 unsigned long *msr_bitmap_l1;
9733 unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.msr_bitmap;
9735 /* This shortcut is ok because we support only x2APIC MSRs so far. */
9736 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
9739 page = nested_get_page(vcpu, vmcs12->msr_bitmap);
9742 msr_bitmap_l1 = (unsigned long *)kmap(page);
9744 memset(msr_bitmap_l0, 0xff, PAGE_SIZE);
9746 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
9747 if (nested_cpu_has_apic_reg_virt(vmcs12))
9748 for (msr = 0x800; msr <= 0x8ff; msr++)
9749 nested_vmx_disable_intercept_for_msr(
9750 msr_bitmap_l1, msr_bitmap_l0,
9753 nested_vmx_disable_intercept_for_msr(
9754 msr_bitmap_l1, msr_bitmap_l0,
9755 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9756 MSR_TYPE_R | MSR_TYPE_W);
9758 if (nested_cpu_has_vid(vmcs12)) {
9759 nested_vmx_disable_intercept_for_msr(
9760 msr_bitmap_l1, msr_bitmap_l0,
9761 APIC_BASE_MSR + (APIC_EOI >> 4),
9763 nested_vmx_disable_intercept_for_msr(
9764 msr_bitmap_l1, msr_bitmap_l0,
9765 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9770 nested_release_page_clean(page);
9775 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
9776 struct vmcs12 *vmcs12)
9778 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9779 !nested_cpu_has_apic_reg_virt(vmcs12) &&
9780 !nested_cpu_has_vid(vmcs12) &&
9781 !nested_cpu_has_posted_intr(vmcs12))
9785 * If virtualize x2apic mode is enabled,
9786 * virtualize apic access must be disabled.
9788 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9789 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9793 * If virtual interrupt delivery is enabled,
9794 * we must exit on external interrupts.
9796 if (nested_cpu_has_vid(vmcs12) &&
9797 !nested_exit_on_intr(vcpu))
9801 * bits 15:8 should be zero in posted_intr_nv,
9802 * the descriptor address has been already checked
9803 * in nested_get_vmcs12_pages.
9805 if (nested_cpu_has_posted_intr(vmcs12) &&
9806 (!nested_cpu_has_vid(vmcs12) ||
9807 !nested_exit_intr_ack_set(vcpu) ||
9808 vmcs12->posted_intr_nv & 0xff00))
9811 /* tpr shadow is needed by all apicv features. */
9812 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9818 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9819 unsigned long count_field,
9820 unsigned long addr_field)
9825 if (vmcs12_read_any(vcpu, count_field, &count) ||
9826 vmcs12_read_any(vcpu, addr_field, &addr)) {
9832 maxphyaddr = cpuid_maxphyaddr(vcpu);
9833 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9834 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9835 pr_debug_ratelimited(
9836 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9837 addr_field, maxphyaddr, count, addr);
9843 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9844 struct vmcs12 *vmcs12)
9846 if (vmcs12->vm_exit_msr_load_count == 0 &&
9847 vmcs12->vm_exit_msr_store_count == 0 &&
9848 vmcs12->vm_entry_msr_load_count == 0)
9849 return 0; /* Fast path */
9850 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9851 VM_EXIT_MSR_LOAD_ADDR) ||
9852 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9853 VM_EXIT_MSR_STORE_ADDR) ||
9854 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9855 VM_ENTRY_MSR_LOAD_ADDR))
9860 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
9861 struct vmcs12 *vmcs12)
9863 u64 address = vmcs12->pml_address;
9864 int maxphyaddr = cpuid_maxphyaddr(vcpu);
9866 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML)) {
9867 if (!nested_cpu_has_ept(vmcs12) ||
9868 !IS_ALIGNED(address, 4096) ||
9869 address >> maxphyaddr)
9876 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9877 struct vmx_msr_entry *e)
9879 /* x2APIC MSR accesses are not allowed */
9880 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9882 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9883 e->index == MSR_IA32_UCODE_REV)
9885 if (e->reserved != 0)
9890 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9891 struct vmx_msr_entry *e)
9893 if (e->index == MSR_FS_BASE ||
9894 e->index == MSR_GS_BASE ||
9895 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9896 nested_vmx_msr_check_common(vcpu, e))
9901 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9902 struct vmx_msr_entry *e)
9904 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9905 nested_vmx_msr_check_common(vcpu, e))
9911 * Load guest's/host's msr at nested entry/exit.
9912 * return 0 for success, entry index for failure.
9914 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9917 struct vmx_msr_entry e;
9918 struct msr_data msr;
9920 msr.host_initiated = false;
9921 for (i = 0; i < count; i++) {
9922 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
9924 pr_debug_ratelimited(
9925 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9926 __func__, i, gpa + i * sizeof(e));
9929 if (nested_vmx_load_msr_check(vcpu, &e)) {
9930 pr_debug_ratelimited(
9931 "%s check failed (%u, 0x%x, 0x%x)\n",
9932 __func__, i, e.index, e.reserved);
9935 msr.index = e.index;
9937 if (kvm_set_msr(vcpu, &msr)) {
9938 pr_debug_ratelimited(
9939 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9940 __func__, i, e.index, e.value);
9949 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9952 struct vmx_msr_entry e;
9954 for (i = 0; i < count; i++) {
9955 struct msr_data msr_info;
9956 if (kvm_vcpu_read_guest(vcpu,
9957 gpa + i * sizeof(e),
9958 &e, 2 * sizeof(u32))) {
9959 pr_debug_ratelimited(
9960 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9961 __func__, i, gpa + i * sizeof(e));
9964 if (nested_vmx_store_msr_check(vcpu, &e)) {
9965 pr_debug_ratelimited(
9966 "%s check failed (%u, 0x%x, 0x%x)\n",
9967 __func__, i, e.index, e.reserved);
9970 msr_info.host_initiated = false;
9971 msr_info.index = e.index;
9972 if (kvm_get_msr(vcpu, &msr_info)) {
9973 pr_debug_ratelimited(
9974 "%s cannot read MSR (%u, 0x%x)\n",
9975 __func__, i, e.index);
9978 if (kvm_vcpu_write_guest(vcpu,
9979 gpa + i * sizeof(e) +
9980 offsetof(struct vmx_msr_entry, value),
9981 &msr_info.data, sizeof(msr_info.data))) {
9982 pr_debug_ratelimited(
9983 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9984 __func__, i, e.index, msr_info.data);
9991 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
9993 unsigned long invalid_mask;
9995 invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
9996 return (val & invalid_mask) == 0;
10000 * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
10001 * emulating VM entry into a guest with EPT enabled.
10002 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10003 * is assigned to entry_failure_code on failure.
10005 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
10006 u32 *entry_failure_code)
10008 if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
10009 if (!nested_cr3_valid(vcpu, cr3)) {
10010 *entry_failure_code = ENTRY_FAIL_DEFAULT;
10015 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
10016 * must not be dereferenced.
10018 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
10020 if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
10021 *entry_failure_code = ENTRY_FAIL_PDPTE;
10026 vcpu->arch.cr3 = cr3;
10027 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
10030 kvm_mmu_reset_context(vcpu);
10035 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
10036 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
10037 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
10038 * guest in a way that will both be appropriate to L1's requests, and our
10039 * needs. In addition to modifying the active vmcs (which is vmcs02), this
10040 * function also has additional necessary side-effects, like setting various
10041 * vcpu->arch fields.
10042 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10043 * is assigned to entry_failure_code on failure.
10045 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10046 bool from_vmentry, u32 *entry_failure_code)
10048 struct vcpu_vmx *vmx = to_vmx(vcpu);
10049 u32 exec_control, vmcs12_exec_ctrl;
10051 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
10052 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
10053 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
10054 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
10055 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
10056 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
10057 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
10058 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
10059 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
10060 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
10061 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
10062 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
10063 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
10064 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
10065 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
10066 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
10067 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
10068 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
10069 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
10070 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
10071 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
10072 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
10073 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
10074 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
10075 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
10076 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
10077 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
10078 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
10079 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
10080 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
10081 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
10082 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
10083 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
10084 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
10085 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
10086 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
10088 if (from_vmentry &&
10089 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
10090 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
10091 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
10093 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
10094 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
10096 if (from_vmentry) {
10097 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
10098 vmcs12->vm_entry_intr_info_field);
10099 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
10100 vmcs12->vm_entry_exception_error_code);
10101 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
10102 vmcs12->vm_entry_instruction_len);
10103 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
10104 vmcs12->guest_interruptibility_info);
10105 vmx->loaded_vmcs->nmi_known_unmasked =
10106 !(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
10108 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
10110 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
10111 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
10112 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
10113 vmcs12->guest_pending_dbg_exceptions);
10114 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
10115 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
10117 if (nested_cpu_has_xsaves(vmcs12))
10118 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
10119 vmcs_write64(VMCS_LINK_POINTER, -1ull);
10121 exec_control = vmcs12->pin_based_vm_exec_control;
10123 /* Preemption timer setting is only taken from vmcs01. */
10124 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10125 exec_control |= vmcs_config.pin_based_exec_ctrl;
10126 if (vmx->hv_deadline_tsc == -1)
10127 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10129 /* Posted interrupts setting is only taken from vmcs12. */
10130 if (nested_cpu_has_posted_intr(vmcs12)) {
10131 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
10132 vmx->nested.pi_pending = false;
10133 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
10135 exec_control &= ~PIN_BASED_POSTED_INTR;
10138 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
10140 vmx->nested.preemption_timer_expired = false;
10141 if (nested_cpu_has_preemption_timer(vmcs12))
10142 vmx_start_preemption_timer(vcpu);
10145 * Whether page-faults are trapped is determined by a combination of
10146 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
10147 * If enable_ept, L0 doesn't care about page faults and we should
10148 * set all of these to L1's desires. However, if !enable_ept, L0 does
10149 * care about (at least some) page faults, and because it is not easy
10150 * (if at all possible?) to merge L0 and L1's desires, we simply ask
10151 * to exit on each and every L2 page fault. This is done by setting
10152 * MASK=MATCH=0 and (see below) EB.PF=1.
10153 * Note that below we don't need special code to set EB.PF beyond the
10154 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
10155 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
10156 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
10158 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
10159 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
10160 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
10161 enable_ept ? vmcs12->page_fault_error_code_match : 0);
10163 if (cpu_has_secondary_exec_ctrls()) {
10164 exec_control = vmx_secondary_exec_control(vmx);
10166 /* Take the following fields only from vmcs12 */
10167 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
10168 SECONDARY_EXEC_RDTSCP |
10169 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
10170 SECONDARY_EXEC_APIC_REGISTER_VIRT);
10171 if (nested_cpu_has(vmcs12,
10172 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
10173 vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
10174 ~SECONDARY_EXEC_ENABLE_PML;
10175 exec_control |= vmcs12_exec_ctrl;
10178 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
10179 vmcs_write64(EOI_EXIT_BITMAP0,
10180 vmcs12->eoi_exit_bitmap0);
10181 vmcs_write64(EOI_EXIT_BITMAP1,
10182 vmcs12->eoi_exit_bitmap1);
10183 vmcs_write64(EOI_EXIT_BITMAP2,
10184 vmcs12->eoi_exit_bitmap2);
10185 vmcs_write64(EOI_EXIT_BITMAP3,
10186 vmcs12->eoi_exit_bitmap3);
10187 vmcs_write16(GUEST_INTR_STATUS,
10188 vmcs12->guest_intr_status);
10192 * Write an illegal value to APIC_ACCESS_ADDR. Later,
10193 * nested_get_vmcs12_pages will either fix it up or
10194 * remove the VM execution control.
10196 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
10197 vmcs_write64(APIC_ACCESS_ADDR, -1ull);
10199 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
10204 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
10205 * Some constant fields are set here by vmx_set_constant_host_state().
10206 * Other fields are different per CPU, and will be set later when
10207 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
10209 vmx_set_constant_host_state(vmx);
10212 * Set the MSR load/store lists to match L0's settings.
10214 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
10215 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10216 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
10217 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10218 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
10221 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
10222 * entry, but only if the current (host) sp changed from the value
10223 * we wrote last (vmx->host_rsp). This cache is no longer relevant
10224 * if we switch vmcs, and rather than hold a separate cache per vmcs,
10225 * here we just force the write to happen on entry.
10229 exec_control = vmx_exec_control(vmx); /* L0's desires */
10230 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
10231 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
10232 exec_control &= ~CPU_BASED_TPR_SHADOW;
10233 exec_control |= vmcs12->cpu_based_vm_exec_control;
10236 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
10237 * nested_get_vmcs12_pages can't fix it up, the illegal value
10238 * will result in a VM entry failure.
10240 if (exec_control & CPU_BASED_TPR_SHADOW) {
10241 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
10242 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
10246 * Merging of IO bitmap not currently supported.
10247 * Rather, exit every time.
10249 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
10250 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
10252 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
10254 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
10255 * bitwise-or of what L1 wants to trap for L2, and what we want to
10256 * trap. Note that CR0.TS also needs updating - we do this later.
10258 update_exception_bitmap(vcpu);
10259 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
10260 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10262 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
10263 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
10264 * bits are further modified by vmx_set_efer() below.
10266 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
10268 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
10269 * emulated by vmx_set_efer(), below.
10271 vm_entry_controls_init(vmx,
10272 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
10273 ~VM_ENTRY_IA32E_MODE) |
10274 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
10276 if (from_vmentry &&
10277 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
10278 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
10279 vcpu->arch.pat = vmcs12->guest_ia32_pat;
10280 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
10281 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
10284 set_cr4_guest_host_mask(vmx);
10286 if (from_vmentry &&
10287 vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
10288 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
10290 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
10291 vmcs_write64(TSC_OFFSET,
10292 vcpu->arch.tsc_offset + vmcs12->tsc_offset);
10294 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
10295 if (kvm_has_tsc_control)
10296 decache_tsc_multiplier(vmx);
10300 * There is no direct mapping between vpid02 and vpid12, the
10301 * vpid02 is per-vCPU for L0 and reused while the value of
10302 * vpid12 is changed w/ one invvpid during nested vmentry.
10303 * The vpid12 is allocated by L1 for L2, so it will not
10304 * influence global bitmap(for vpid01 and vpid02 allocation)
10305 * even if spawn a lot of nested vCPUs.
10307 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
10308 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
10309 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
10310 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
10311 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
10314 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
10315 vmx_flush_tlb(vcpu);
10322 * Conceptually we want to copy the PML address and index from
10323 * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
10324 * since we always flush the log on each vmexit, this happens
10325 * to be equivalent to simply resetting the fields in vmcs02.
10327 ASSERT(vmx->pml_pg);
10328 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
10329 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
10332 if (nested_cpu_has_ept(vmcs12)) {
10333 if (nested_ept_init_mmu_context(vcpu)) {
10334 *entry_failure_code = ENTRY_FAIL_DEFAULT;
10337 } else if (nested_cpu_has2(vmcs12,
10338 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10339 vmx_flush_tlb_ept_only(vcpu);
10343 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
10344 * bits which we consider mandatory enabled.
10345 * The CR0_READ_SHADOW is what L2 should have expected to read given
10346 * the specifications by L1; It's not enough to take
10347 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10348 * have more bits than L1 expected.
10350 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
10351 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
10353 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
10354 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
10356 if (from_vmentry &&
10357 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
10358 vcpu->arch.efer = vmcs12->guest_ia32_efer;
10359 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
10360 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10362 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10363 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10364 vmx_set_efer(vcpu, vcpu->arch.efer);
10366 /* Shadow page tables on either EPT or shadow page tables. */
10367 if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
10368 entry_failure_code))
10372 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
10375 * L1 may access the L2's PDPTR, so save them to construct vmcs12
10378 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
10379 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
10380 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
10381 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
10384 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
10385 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
10389 static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10391 struct vcpu_vmx *vmx = to_vmx(vcpu);
10393 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
10394 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
10395 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10397 if (nested_vmx_check_io_bitmap_controls(vcpu, vmcs12))
10398 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10400 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12))
10401 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10403 if (nested_vmx_check_apicv_controls(vcpu, vmcs12))
10404 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10406 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12))
10407 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10409 if (nested_vmx_check_pml_controls(vcpu, vmcs12))
10410 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10412 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
10413 vmx->nested.nested_vmx_procbased_ctls_low,
10414 vmx->nested.nested_vmx_procbased_ctls_high) ||
10415 (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
10416 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
10417 vmx->nested.nested_vmx_secondary_ctls_low,
10418 vmx->nested.nested_vmx_secondary_ctls_high)) ||
10419 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
10420 vmx->nested.nested_vmx_pinbased_ctls_low,
10421 vmx->nested.nested_vmx_pinbased_ctls_high) ||
10422 !vmx_control_verify(vmcs12->vm_exit_controls,
10423 vmx->nested.nested_vmx_exit_ctls_low,
10424 vmx->nested.nested_vmx_exit_ctls_high) ||
10425 !vmx_control_verify(vmcs12->vm_entry_controls,
10426 vmx->nested.nested_vmx_entry_ctls_low,
10427 vmx->nested.nested_vmx_entry_ctls_high))
10428 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10430 if (vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu))
10431 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10433 if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
10434 !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
10435 !nested_cr3_valid(vcpu, vmcs12->host_cr3))
10436 return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
10441 static int check_vmentry_postreqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10446 *exit_qual = ENTRY_FAIL_DEFAULT;
10448 if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
10449 !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
10452 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_SHADOW_VMCS) &&
10453 vmcs12->vmcs_link_pointer != -1ull) {
10454 *exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
10459 * If the load IA32_EFER VM-entry control is 1, the following checks
10460 * are performed on the field for the IA32_EFER MSR:
10461 * - Bits reserved in the IA32_EFER MSR must be 0.
10462 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
10463 * the IA-32e mode guest VM-exit control. It must also be identical
10464 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
10467 if (to_vmx(vcpu)->nested.nested_run_pending &&
10468 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
10469 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
10470 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
10471 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
10472 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
10473 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
10478 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
10479 * IA32_EFER MSR must be 0 in the field for that register. In addition,
10480 * the values of the LMA and LME bits in the field must each be that of
10481 * the host address-space size VM-exit control.
10483 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
10484 ia32e = (vmcs12->vm_exit_controls &
10485 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
10486 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
10487 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
10488 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
10495 static int enter_vmx_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
10497 struct vcpu_vmx *vmx = to_vmx(vcpu);
10498 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10499 struct loaded_vmcs *vmcs02;
10503 vmcs02 = nested_get_current_vmcs02(vmx);
10507 enter_guest_mode(vcpu);
10509 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
10510 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10512 vmx_switch_vmcs(vcpu, vmcs02);
10513 vmx_segment_cache_clear(vmx);
10515 if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &exit_qual)) {
10516 leave_guest_mode(vcpu);
10517 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10518 nested_vmx_entry_failure(vcpu, vmcs12,
10519 EXIT_REASON_INVALID_STATE, exit_qual);
10523 nested_get_vmcs12_pages(vcpu, vmcs12);
10525 msr_entry_idx = nested_vmx_load_msr(vcpu,
10526 vmcs12->vm_entry_msr_load_addr,
10527 vmcs12->vm_entry_msr_load_count);
10528 if (msr_entry_idx) {
10529 leave_guest_mode(vcpu);
10530 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10531 nested_vmx_entry_failure(vcpu, vmcs12,
10532 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
10537 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10538 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10539 * returned as far as L1 is concerned. It will only return (and set
10540 * the success flag) when L2 exits (see nested_vmx_vmexit()).
10546 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
10547 * for running an L2 nested guest.
10549 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
10551 struct vmcs12 *vmcs12;
10552 struct vcpu_vmx *vmx = to_vmx(vcpu);
10553 u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
10557 if (!nested_vmx_check_permission(vcpu))
10560 if (!nested_vmx_check_vmcs12(vcpu))
10563 vmcs12 = get_vmcs12(vcpu);
10565 if (enable_shadow_vmcs)
10566 copy_shadow_to_vmcs12(vmx);
10569 * The nested entry process starts with enforcing various prerequisites
10570 * on vmcs12 as required by the Intel SDM, and act appropriately when
10571 * they fail: As the SDM explains, some conditions should cause the
10572 * instruction to fail, while others will cause the instruction to seem
10573 * to succeed, but return an EXIT_REASON_INVALID_STATE.
10574 * To speed up the normal (success) code path, we should avoid checking
10575 * for misconfigurations which will anyway be caught by the processor
10576 * when using the merged vmcs02.
10578 if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS) {
10579 nested_vmx_failValid(vcpu,
10580 VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
10584 if (vmcs12->launch_state == launch) {
10585 nested_vmx_failValid(vcpu,
10586 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
10587 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
10591 ret = check_vmentry_prereqs(vcpu, vmcs12);
10593 nested_vmx_failValid(vcpu, ret);
10598 * After this point, the trap flag no longer triggers a singlestep trap
10599 * on the vm entry instructions; don't call kvm_skip_emulated_instruction.
10600 * This is not 100% correct; for performance reasons, we delegate most
10601 * of the checks on host state to the processor. If those fail,
10602 * the singlestep trap is missed.
10604 skip_emulated_instruction(vcpu);
10606 ret = check_vmentry_postreqs(vcpu, vmcs12, &exit_qual);
10608 nested_vmx_entry_failure(vcpu, vmcs12,
10609 EXIT_REASON_INVALID_STATE, exit_qual);
10614 * We're finally done with prerequisite checking, and can start with
10615 * the nested entry.
10618 ret = enter_vmx_non_root_mode(vcpu, true);
10622 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
10623 return kvm_vcpu_halt(vcpu);
10625 vmx->nested.nested_run_pending = 1;
10630 return kvm_skip_emulated_instruction(vcpu);
10634 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
10635 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
10636 * This function returns the new value we should put in vmcs12.guest_cr0.
10637 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
10638 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
10639 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
10640 * didn't trap the bit, because if L1 did, so would L0).
10641 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
10642 * been modified by L2, and L1 knows it. So just leave the old value of
10643 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
10644 * isn't relevant, because if L0 traps this bit it can set it to anything.
10645 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
10646 * changed these bits, and therefore they need to be updated, but L0
10647 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
10648 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
10650 static inline unsigned long
10651 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10654 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
10655 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
10656 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
10657 vcpu->arch.cr0_guest_owned_bits));
10660 static inline unsigned long
10661 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10664 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
10665 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
10666 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
10667 vcpu->arch.cr4_guest_owned_bits));
10670 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
10671 struct vmcs12 *vmcs12)
10676 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
10677 nr = vcpu->arch.exception.nr;
10678 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10680 if (kvm_exception_is_soft(nr)) {
10681 vmcs12->vm_exit_instruction_len =
10682 vcpu->arch.event_exit_inst_len;
10683 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
10685 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
10687 if (vcpu->arch.exception.has_error_code) {
10688 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
10689 vmcs12->idt_vectoring_error_code =
10690 vcpu->arch.exception.error_code;
10693 vmcs12->idt_vectoring_info_field = idt_vectoring;
10694 } else if (vcpu->arch.nmi_injected) {
10695 vmcs12->idt_vectoring_info_field =
10696 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
10697 } else if (vcpu->arch.interrupt.pending) {
10698 nr = vcpu->arch.interrupt.nr;
10699 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10701 if (vcpu->arch.interrupt.soft) {
10702 idt_vectoring |= INTR_TYPE_SOFT_INTR;
10703 vmcs12->vm_entry_instruction_len =
10704 vcpu->arch.event_exit_inst_len;
10706 idt_vectoring |= INTR_TYPE_EXT_INTR;
10708 vmcs12->idt_vectoring_info_field = idt_vectoring;
10712 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
10714 struct vcpu_vmx *vmx = to_vmx(vcpu);
10716 if (vcpu->arch.exception.pending ||
10717 vcpu->arch.nmi_injected ||
10718 vcpu->arch.interrupt.pending)
10721 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
10722 vmx->nested.preemption_timer_expired) {
10723 if (vmx->nested.nested_run_pending)
10725 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
10729 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
10730 if (vmx->nested.nested_run_pending)
10732 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10733 NMI_VECTOR | INTR_TYPE_NMI_INTR |
10734 INTR_INFO_VALID_MASK, 0);
10736 * The NMI-triggered VM exit counts as injection:
10737 * clear this one and block further NMIs.
10739 vcpu->arch.nmi_pending = 0;
10740 vmx_set_nmi_mask(vcpu, true);
10744 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
10745 nested_exit_on_intr(vcpu)) {
10746 if (vmx->nested.nested_run_pending)
10748 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
10752 vmx_complete_nested_posted_interrupt(vcpu);
10756 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
10758 ktime_t remaining =
10759 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
10762 if (ktime_to_ns(remaining) <= 0)
10765 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
10766 do_div(value, 1000000);
10767 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10771 * Update the guest state fields of vmcs12 to reflect changes that
10772 * occurred while L2 was running. (The "IA-32e mode guest" bit of the
10773 * VM-entry controls is also updated, since this is really a guest
10776 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10778 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
10779 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
10781 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
10782 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
10783 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
10785 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
10786 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
10787 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
10788 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
10789 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
10790 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
10791 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
10792 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
10793 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
10794 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
10795 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
10796 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
10797 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
10798 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
10799 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
10800 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
10801 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
10802 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
10803 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
10804 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
10805 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
10806 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
10807 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
10808 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
10809 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
10810 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
10811 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
10812 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
10813 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
10814 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
10815 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
10816 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
10817 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
10818 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
10819 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
10820 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
10822 vmcs12->guest_interruptibility_info =
10823 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
10824 vmcs12->guest_pending_dbg_exceptions =
10825 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
10826 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10827 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
10829 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
10831 if (nested_cpu_has_preemption_timer(vmcs12)) {
10832 if (vmcs12->vm_exit_controls &
10833 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
10834 vmcs12->vmx_preemption_timer_value =
10835 vmx_get_preemption_timer_value(vcpu);
10836 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
10840 * In some cases (usually, nested EPT), L2 is allowed to change its
10841 * own CR3 without exiting. If it has changed it, we must keep it.
10842 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10843 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10845 * Additionally, restore L2's PDPTR to vmcs12.
10848 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
10849 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
10850 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
10851 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
10852 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
10855 vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
10857 if (nested_cpu_has_vid(vmcs12))
10858 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
10860 vmcs12->vm_entry_controls =
10861 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
10862 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
10864 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
10865 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
10866 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10869 /* TODO: These cannot have changed unless we have MSR bitmaps and
10870 * the relevant bit asks not to trap the change */
10871 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
10872 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
10873 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
10874 vmcs12->guest_ia32_efer = vcpu->arch.efer;
10875 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
10876 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
10877 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
10878 if (kvm_mpx_supported())
10879 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
10883 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10884 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10885 * and this function updates it to reflect the changes to the guest state while
10886 * L2 was running (and perhaps made some exits which were handled directly by L0
10887 * without going back to L1), and to reflect the exit reason.
10888 * Note that we do not have to copy here all VMCS fields, just those that
10889 * could have changed by the L2 guest or the exit - i.e., the guest-state and
10890 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10891 * which already writes to vmcs12 directly.
10893 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10894 u32 exit_reason, u32 exit_intr_info,
10895 unsigned long exit_qualification)
10897 /* update guest state fields: */
10898 sync_vmcs12(vcpu, vmcs12);
10900 /* update exit information fields: */
10902 vmcs12->vm_exit_reason = exit_reason;
10903 vmcs12->exit_qualification = exit_qualification;
10904 vmcs12->vm_exit_intr_info = exit_intr_info;
10906 vmcs12->idt_vectoring_info_field = 0;
10907 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
10908 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
10910 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
10911 vmcs12->launch_state = 1;
10913 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10914 * instead of reading the real value. */
10915 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
10918 * Transfer the event that L0 or L1 may wanted to inject into
10919 * L2 to IDT_VECTORING_INFO_FIELD.
10921 vmcs12_save_pending_event(vcpu, vmcs12);
10925 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10926 * preserved above and would only end up incorrectly in L1.
10928 vcpu->arch.nmi_injected = false;
10929 kvm_clear_exception_queue(vcpu);
10930 kvm_clear_interrupt_queue(vcpu);
10934 * A part of what we need to when the nested L2 guest exits and we want to
10935 * run its L1 parent, is to reset L1's guest state to the host state specified
10937 * This function is to be called not only on normal nested exit, but also on
10938 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10939 * Failures During or After Loading Guest State").
10940 * This function should be called when the active VMCS is L1's (vmcs01).
10942 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10943 struct vmcs12 *vmcs12)
10945 struct kvm_segment seg;
10946 u32 entry_failure_code;
10948 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10949 vcpu->arch.efer = vmcs12->host_ia32_efer;
10950 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10951 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10953 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10954 vmx_set_efer(vcpu, vcpu->arch.efer);
10956 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10957 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10958 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10960 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10961 * actually changed, because vmx_set_cr0 refers to efer set above.
10963 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
10964 * (KVM doesn't change it);
10966 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
10967 vmx_set_cr0(vcpu, vmcs12->host_cr0);
10969 /* Same as above - no reason to call set_cr4_guest_host_mask(). */
10970 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
10971 kvm_set_cr4(vcpu, vmcs12->host_cr4);
10973 nested_ept_uninit_mmu_context(vcpu);
10976 * Only PDPTE load can fail as the value of cr3 was checked on entry and
10977 * couldn't have changed.
10979 if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
10980 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
10983 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
10987 * Trivially support vpid by letting L2s share their parent
10988 * L1's vpid. TODO: move to a more elaborate solution, giving
10989 * each L2 its own vpid and exposing the vpid feature to L1.
10991 vmx_flush_tlb(vcpu);
10993 /* Restore posted intr vector. */
10994 if (nested_cpu_has_posted_intr(vmcs12))
10995 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
10997 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
10998 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
10999 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
11000 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
11001 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
11003 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
11004 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
11005 vmcs_write64(GUEST_BNDCFGS, 0);
11007 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
11008 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
11009 vcpu->arch.pat = vmcs12->host_ia32_pat;
11011 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
11012 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
11013 vmcs12->host_ia32_perf_global_ctrl);
11015 /* Set L1 segment info according to Intel SDM
11016 27.5.2 Loading Host Segment and Descriptor-Table Registers */
11017 seg = (struct kvm_segment) {
11019 .limit = 0xFFFFFFFF,
11020 .selector = vmcs12->host_cs_selector,
11026 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
11030 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
11031 seg = (struct kvm_segment) {
11033 .limit = 0xFFFFFFFF,
11040 seg.selector = vmcs12->host_ds_selector;
11041 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
11042 seg.selector = vmcs12->host_es_selector;
11043 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
11044 seg.selector = vmcs12->host_ss_selector;
11045 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
11046 seg.selector = vmcs12->host_fs_selector;
11047 seg.base = vmcs12->host_fs_base;
11048 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
11049 seg.selector = vmcs12->host_gs_selector;
11050 seg.base = vmcs12->host_gs_base;
11051 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
11052 seg = (struct kvm_segment) {
11053 .base = vmcs12->host_tr_base,
11055 .selector = vmcs12->host_tr_selector,
11059 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
11061 kvm_set_dr(vcpu, 7, 0x400);
11062 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
11064 if (cpu_has_vmx_msr_bitmap())
11065 vmx_set_msr_bitmap(vcpu);
11067 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
11068 vmcs12->vm_exit_msr_load_count))
11069 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
11073 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
11074 * and modify vmcs12 to make it see what it would expect to see there if
11075 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
11077 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
11078 u32 exit_intr_info,
11079 unsigned long exit_qualification)
11081 struct vcpu_vmx *vmx = to_vmx(vcpu);
11082 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
11083 u32 vm_inst_error = 0;
11085 /* trying to cancel vmlaunch/vmresume is a bug */
11086 WARN_ON_ONCE(vmx->nested.nested_run_pending);
11088 leave_guest_mode(vcpu);
11089 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
11090 exit_qualification);
11092 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
11093 vmcs12->vm_exit_msr_store_count))
11094 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
11096 if (unlikely(vmx->fail))
11097 vm_inst_error = vmcs_read32(VM_INSTRUCTION_ERROR);
11099 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
11101 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
11102 && nested_exit_intr_ack_set(vcpu)) {
11103 int irq = kvm_cpu_get_interrupt(vcpu);
11105 vmcs12->vm_exit_intr_info = irq |
11106 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
11109 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
11110 vmcs12->exit_qualification,
11111 vmcs12->idt_vectoring_info_field,
11112 vmcs12->vm_exit_intr_info,
11113 vmcs12->vm_exit_intr_error_code,
11116 vm_entry_controls_reset_shadow(vmx);
11117 vm_exit_controls_reset_shadow(vmx);
11118 vmx_segment_cache_clear(vmx);
11120 /* if no vmcs02 cache requested, remove the one we used */
11121 if (VMCS02_POOL_SIZE == 0)
11122 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
11124 load_vmcs12_host_state(vcpu, vmcs12);
11126 /* Update any VMCS fields that might have changed while L2 ran */
11127 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11128 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11129 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
11130 if (vmx->hv_deadline_tsc == -1)
11131 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11132 PIN_BASED_VMX_PREEMPTION_TIMER);
11134 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11135 PIN_BASED_VMX_PREEMPTION_TIMER);
11136 if (kvm_has_tsc_control)
11137 decache_tsc_multiplier(vmx);
11139 if (vmx->nested.change_vmcs01_virtual_x2apic_mode) {
11140 vmx->nested.change_vmcs01_virtual_x2apic_mode = false;
11141 vmx_set_virtual_x2apic_mode(vcpu,
11142 vcpu->arch.apic_base & X2APIC_ENABLE);
11143 } else if (!nested_cpu_has_ept(vmcs12) &&
11144 nested_cpu_has2(vmcs12,
11145 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
11146 vmx_flush_tlb_ept_only(vcpu);
11149 /* This is needed for same reason as it was needed in prepare_vmcs02 */
11152 /* Unpin physical memory we referred to in vmcs02 */
11153 if (vmx->nested.apic_access_page) {
11154 nested_release_page(vmx->nested.apic_access_page);
11155 vmx->nested.apic_access_page = NULL;
11157 if (vmx->nested.virtual_apic_page) {
11158 nested_release_page(vmx->nested.virtual_apic_page);
11159 vmx->nested.virtual_apic_page = NULL;
11161 if (vmx->nested.pi_desc_page) {
11162 kunmap(vmx->nested.pi_desc_page);
11163 nested_release_page(vmx->nested.pi_desc_page);
11164 vmx->nested.pi_desc_page = NULL;
11165 vmx->nested.pi_desc = NULL;
11169 * We are now running in L2, mmu_notifier will force to reload the
11170 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
11172 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
11175 * Exiting from L2 to L1, we're now back to L1 which thinks it just
11176 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
11177 * success or failure flag accordingly.
11179 if (unlikely(vmx->fail)) {
11181 nested_vmx_failValid(vcpu, vm_inst_error);
11183 nested_vmx_succeed(vcpu);
11184 if (enable_shadow_vmcs)
11185 vmx->nested.sync_shadow_vmcs = true;
11187 /* in case we halted in L2 */
11188 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11192 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
11194 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
11196 if (is_guest_mode(vcpu)) {
11197 to_vmx(vcpu)->nested.nested_run_pending = 0;
11198 nested_vmx_vmexit(vcpu, -1, 0, 0);
11200 free_nested(to_vmx(vcpu));
11204 * L1's failure to enter L2 is a subset of a normal exit, as explained in
11205 * 23.7 "VM-entry failures during or after loading guest state" (this also
11206 * lists the acceptable exit-reason and exit-qualification parameters).
11207 * It should only be called before L2 actually succeeded to run, and when
11208 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
11210 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
11211 struct vmcs12 *vmcs12,
11212 u32 reason, unsigned long qualification)
11214 load_vmcs12_host_state(vcpu, vmcs12);
11215 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
11216 vmcs12->exit_qualification = qualification;
11217 nested_vmx_succeed(vcpu);
11218 if (enable_shadow_vmcs)
11219 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
11222 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
11223 struct x86_instruction_info *info,
11224 enum x86_intercept_stage stage)
11226 return X86EMUL_CONTINUE;
11229 #ifdef CONFIG_X86_64
11230 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
11231 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
11232 u64 divisor, u64 *result)
11234 u64 low = a << shift, high = a >> (64 - shift);
11236 /* To avoid the overflow on divq */
11237 if (high >= divisor)
11240 /* Low hold the result, high hold rem which is discarded */
11241 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
11242 "rm" (divisor), "0" (low), "1" (high));
11248 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
11250 struct vcpu_vmx *vmx = to_vmx(vcpu);
11251 u64 tscl = rdtsc();
11252 u64 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
11253 u64 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
11255 /* Convert to host delta tsc if tsc scaling is enabled */
11256 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
11257 u64_shl_div_u64(delta_tsc,
11258 kvm_tsc_scaling_ratio_frac_bits,
11259 vcpu->arch.tsc_scaling_ratio,
11264 * If the delta tsc can't fit in the 32 bit after the multi shift,
11265 * we can't use the preemption timer.
11266 * It's possible that it fits on later vmentries, but checking
11267 * on every vmentry is costly so we just use an hrtimer.
11269 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
11272 vmx->hv_deadline_tsc = tscl + delta_tsc;
11273 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11274 PIN_BASED_VMX_PREEMPTION_TIMER);
11276 return delta_tsc == 0;
11279 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
11281 struct vcpu_vmx *vmx = to_vmx(vcpu);
11282 vmx->hv_deadline_tsc = -1;
11283 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11284 PIN_BASED_VMX_PREEMPTION_TIMER);
11288 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
11291 shrink_ple_window(vcpu);
11294 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
11295 struct kvm_memory_slot *slot)
11297 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
11298 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
11301 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
11302 struct kvm_memory_slot *slot)
11304 kvm_mmu_slot_set_dirty(kvm, slot);
11307 static void vmx_flush_log_dirty(struct kvm *kvm)
11309 kvm_flush_pml_buffers(kvm);
11312 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
11314 struct vmcs12 *vmcs12;
11315 struct vcpu_vmx *vmx = to_vmx(vcpu);
11317 struct page *page = NULL;
11320 if (is_guest_mode(vcpu)) {
11321 WARN_ON_ONCE(vmx->nested.pml_full);
11324 * Check if PML is enabled for the nested guest.
11325 * Whether eptp bit 6 is set is already checked
11326 * as part of A/D emulation.
11328 vmcs12 = get_vmcs12(vcpu);
11329 if (!nested_cpu_has_pml(vmcs12))
11332 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
11333 vmx->nested.pml_full = true;
11337 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
11339 page = nested_get_page(vcpu, vmcs12->pml_address);
11343 pml_address = kmap(page);
11344 pml_address[vmcs12->guest_pml_index--] = gpa;
11346 nested_release_page_clean(page);
11352 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
11353 struct kvm_memory_slot *memslot,
11354 gfn_t offset, unsigned long mask)
11356 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
11360 * This routine does the following things for vCPU which is going
11361 * to be blocked if VT-d PI is enabled.
11362 * - Store the vCPU to the wakeup list, so when interrupts happen
11363 * we can find the right vCPU to wake up.
11364 * - Change the Posted-interrupt descriptor as below:
11365 * 'NDST' <-- vcpu->pre_pcpu
11366 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
11367 * - If 'ON' is set during this process, which means at least one
11368 * interrupt is posted for this vCPU, we cannot block it, in
11369 * this case, return 1, otherwise, return 0.
11372 static int pi_pre_block(struct kvm_vcpu *vcpu)
11374 unsigned long flags;
11376 struct pi_desc old, new;
11377 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11379 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11380 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11381 !kvm_vcpu_apicv_active(vcpu))
11384 vcpu->pre_pcpu = vcpu->cpu;
11385 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
11386 vcpu->pre_pcpu), flags);
11387 list_add_tail(&vcpu->blocked_vcpu_list,
11388 &per_cpu(blocked_vcpu_on_cpu,
11390 spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
11391 vcpu->pre_pcpu), flags);
11394 old.control = new.control = pi_desc->control;
11397 * We should not block the vCPU if
11398 * an interrupt is posted for it.
11400 if (pi_test_on(pi_desc) == 1) {
11401 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
11402 vcpu->pre_pcpu), flags);
11403 list_del(&vcpu->blocked_vcpu_list);
11404 spin_unlock_irqrestore(
11405 &per_cpu(blocked_vcpu_on_cpu_lock,
11406 vcpu->pre_pcpu), flags);
11407 vcpu->pre_pcpu = -1;
11412 WARN((pi_desc->sn == 1),
11413 "Warning: SN field of posted-interrupts "
11414 "is set before blocking\n");
11417 * Since vCPU can be preempted during this process,
11418 * vcpu->cpu could be different with pre_pcpu, we
11419 * need to set pre_pcpu as the destination of wakeup
11420 * notification event, then we can find the right vCPU
11421 * to wakeup in wakeup handler if interrupts happen
11422 * when the vCPU is in blocked state.
11424 dest = cpu_physical_id(vcpu->pre_pcpu);
11426 if (x2apic_enabled())
11429 new.ndst = (dest << 8) & 0xFF00;
11431 /* set 'NV' to 'wakeup vector' */
11432 new.nv = POSTED_INTR_WAKEUP_VECTOR;
11433 } while (cmpxchg(&pi_desc->control, old.control,
11434 new.control) != old.control);
11439 static int vmx_pre_block(struct kvm_vcpu *vcpu)
11441 if (pi_pre_block(vcpu))
11444 if (kvm_lapic_hv_timer_in_use(vcpu))
11445 kvm_lapic_switch_to_sw_timer(vcpu);
11450 static void pi_post_block(struct kvm_vcpu *vcpu)
11452 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11453 struct pi_desc old, new;
11455 unsigned long flags;
11457 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11458 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11459 !kvm_vcpu_apicv_active(vcpu))
11463 old.control = new.control = pi_desc->control;
11465 dest = cpu_physical_id(vcpu->cpu);
11467 if (x2apic_enabled())
11470 new.ndst = (dest << 8) & 0xFF00;
11472 /* Allow posting non-urgent interrupts */
11475 /* set 'NV' to 'notification vector' */
11476 new.nv = POSTED_INTR_VECTOR;
11477 } while (cmpxchg(&pi_desc->control, old.control,
11478 new.control) != old.control);
11480 if(vcpu->pre_pcpu != -1) {
11482 &per_cpu(blocked_vcpu_on_cpu_lock,
11483 vcpu->pre_pcpu), flags);
11484 list_del(&vcpu->blocked_vcpu_list);
11485 spin_unlock_irqrestore(
11486 &per_cpu(blocked_vcpu_on_cpu_lock,
11487 vcpu->pre_pcpu), flags);
11488 vcpu->pre_pcpu = -1;
11492 static void vmx_post_block(struct kvm_vcpu *vcpu)
11494 if (kvm_x86_ops->set_hv_timer)
11495 kvm_lapic_switch_to_hv_timer(vcpu);
11497 pi_post_block(vcpu);
11501 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
11504 * @host_irq: host irq of the interrupt
11505 * @guest_irq: gsi of the interrupt
11506 * @set: set or unset PI
11507 * returns 0 on success, < 0 on failure
11509 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
11510 uint32_t guest_irq, bool set)
11512 struct kvm_kernel_irq_routing_entry *e;
11513 struct kvm_irq_routing_table *irq_rt;
11514 struct kvm_lapic_irq irq;
11515 struct kvm_vcpu *vcpu;
11516 struct vcpu_data vcpu_info;
11517 int idx, ret = -EINVAL;
11519 if (!kvm_arch_has_assigned_device(kvm) ||
11520 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11521 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
11524 idx = srcu_read_lock(&kvm->irq_srcu);
11525 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
11526 BUG_ON(guest_irq >= irq_rt->nr_rt_entries);
11528 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
11529 if (e->type != KVM_IRQ_ROUTING_MSI)
11532 * VT-d PI cannot support posting multicast/broadcast
11533 * interrupts to a vCPU, we still use interrupt remapping
11534 * for these kind of interrupts.
11536 * For lowest-priority interrupts, we only support
11537 * those with single CPU as the destination, e.g. user
11538 * configures the interrupts via /proc/irq or uses
11539 * irqbalance to make the interrupts single-CPU.
11541 * We will support full lowest-priority interrupt later.
11544 kvm_set_msi_irq(kvm, e, &irq);
11545 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
11547 * Make sure the IRTE is in remapped mode if
11548 * we don't handle it in posted mode.
11550 ret = irq_set_vcpu_affinity(host_irq, NULL);
11553 "failed to back to remapped mode, irq: %u\n",
11561 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
11562 vcpu_info.vector = irq.vector;
11564 trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
11565 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
11568 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
11570 /* suppress notification event before unposting */
11571 pi_set_sn(vcpu_to_pi_desc(vcpu));
11572 ret = irq_set_vcpu_affinity(host_irq, NULL);
11573 pi_clear_sn(vcpu_to_pi_desc(vcpu));
11577 printk(KERN_INFO "%s: failed to update PI IRTE\n",
11585 srcu_read_unlock(&kvm->irq_srcu, idx);
11589 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
11591 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
11592 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
11593 FEATURE_CONTROL_LMCE;
11595 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
11596 ~FEATURE_CONTROL_LMCE;
11599 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
11600 .cpu_has_kvm_support = cpu_has_kvm_support,
11601 .disabled_by_bios = vmx_disabled_by_bios,
11602 .hardware_setup = hardware_setup,
11603 .hardware_unsetup = hardware_unsetup,
11604 .check_processor_compatibility = vmx_check_processor_compat,
11605 .hardware_enable = hardware_enable,
11606 .hardware_disable = hardware_disable,
11607 .cpu_has_accelerated_tpr = report_flexpriority,
11608 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
11610 .vcpu_create = vmx_create_vcpu,
11611 .vcpu_free = vmx_free_vcpu,
11612 .vcpu_reset = vmx_vcpu_reset,
11614 .prepare_guest_switch = vmx_save_host_state,
11615 .vcpu_load = vmx_vcpu_load,
11616 .vcpu_put = vmx_vcpu_put,
11618 .update_bp_intercept = update_exception_bitmap,
11619 .get_msr = vmx_get_msr,
11620 .set_msr = vmx_set_msr,
11621 .get_segment_base = vmx_get_segment_base,
11622 .get_segment = vmx_get_segment,
11623 .set_segment = vmx_set_segment,
11624 .get_cpl = vmx_get_cpl,
11625 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
11626 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
11627 .decache_cr3 = vmx_decache_cr3,
11628 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
11629 .set_cr0 = vmx_set_cr0,
11630 .set_cr3 = vmx_set_cr3,
11631 .set_cr4 = vmx_set_cr4,
11632 .set_efer = vmx_set_efer,
11633 .get_idt = vmx_get_idt,
11634 .set_idt = vmx_set_idt,
11635 .get_gdt = vmx_get_gdt,
11636 .set_gdt = vmx_set_gdt,
11637 .get_dr6 = vmx_get_dr6,
11638 .set_dr6 = vmx_set_dr6,
11639 .set_dr7 = vmx_set_dr7,
11640 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
11641 .cache_reg = vmx_cache_reg,
11642 .get_rflags = vmx_get_rflags,
11643 .set_rflags = vmx_set_rflags,
11645 .get_pkru = vmx_get_pkru,
11647 .tlb_flush = vmx_flush_tlb,
11649 .run = vmx_vcpu_run,
11650 .handle_exit = vmx_handle_exit,
11651 .skip_emulated_instruction = skip_emulated_instruction,
11652 .set_interrupt_shadow = vmx_set_interrupt_shadow,
11653 .get_interrupt_shadow = vmx_get_interrupt_shadow,
11654 .patch_hypercall = vmx_patch_hypercall,
11655 .set_irq = vmx_inject_irq,
11656 .set_nmi = vmx_inject_nmi,
11657 .queue_exception = vmx_queue_exception,
11658 .cancel_injection = vmx_cancel_injection,
11659 .interrupt_allowed = vmx_interrupt_allowed,
11660 .nmi_allowed = vmx_nmi_allowed,
11661 .get_nmi_mask = vmx_get_nmi_mask,
11662 .set_nmi_mask = vmx_set_nmi_mask,
11663 .enable_nmi_window = enable_nmi_window,
11664 .enable_irq_window = enable_irq_window,
11665 .update_cr8_intercept = update_cr8_intercept,
11666 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
11667 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
11668 .get_enable_apicv = vmx_get_enable_apicv,
11669 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
11670 .load_eoi_exitmap = vmx_load_eoi_exitmap,
11671 .apicv_post_state_restore = vmx_apicv_post_state_restore,
11672 .hwapic_irr_update = vmx_hwapic_irr_update,
11673 .hwapic_isr_update = vmx_hwapic_isr_update,
11674 .sync_pir_to_irr = vmx_sync_pir_to_irr,
11675 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
11677 .set_tss_addr = vmx_set_tss_addr,
11678 .get_tdp_level = get_ept_level,
11679 .get_mt_mask = vmx_get_mt_mask,
11681 .get_exit_info = vmx_get_exit_info,
11683 .get_lpage_level = vmx_get_lpage_level,
11685 .cpuid_update = vmx_cpuid_update,
11687 .rdtscp_supported = vmx_rdtscp_supported,
11688 .invpcid_supported = vmx_invpcid_supported,
11690 .set_supported_cpuid = vmx_set_supported_cpuid,
11692 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
11694 .write_tsc_offset = vmx_write_tsc_offset,
11696 .set_tdp_cr3 = vmx_set_cr3,
11698 .check_intercept = vmx_check_intercept,
11699 .handle_external_intr = vmx_handle_external_intr,
11700 .mpx_supported = vmx_mpx_supported,
11701 .xsaves_supported = vmx_xsaves_supported,
11703 .check_nested_events = vmx_check_nested_events,
11705 .sched_in = vmx_sched_in,
11707 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
11708 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
11709 .flush_log_dirty = vmx_flush_log_dirty,
11710 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
11711 .write_log_dirty = vmx_write_pml_buffer,
11713 .pre_block = vmx_pre_block,
11714 .post_block = vmx_post_block,
11716 .pmu_ops = &intel_pmu_ops,
11718 .update_pi_irte = vmx_update_pi_irte,
11720 #ifdef CONFIG_X86_64
11721 .set_hv_timer = vmx_set_hv_timer,
11722 .cancel_hv_timer = vmx_cancel_hv_timer,
11725 .setup_mce = vmx_setup_mce,
11728 static int __init vmx_init(void)
11730 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
11731 __alignof__(struct vcpu_vmx), THIS_MODULE);
11735 #ifdef CONFIG_KEXEC_CORE
11736 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
11737 crash_vmclear_local_loaded_vmcss);
11743 static void __exit vmx_exit(void)
11745 #ifdef CONFIG_KEXEC_CORE
11746 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
11753 module_init(vmx_init)
11754 module_exit(vmx_exit)