2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
24 #include <linux/kvm_host.h>
25 #include <linux/module.h>
26 #include <linux/kernel.h>
28 #include <linux/highmem.h>
29 #include <linux/sched.h>
30 #include <linux/moduleparam.h>
31 #include <linux/mod_devicetable.h>
32 #include <linux/trace_events.h>
33 #include <linux/slab.h>
34 #include <linux/tboot.h>
35 #include <linux/hrtimer.h>
36 #include "kvm_cache_regs.h"
43 #include <asm/virtext.h>
45 #include <asm/fpu/internal.h>
46 #include <asm/perf_event.h>
47 #include <asm/debugreg.h>
48 #include <asm/kexec.h>
50 #include <asm/irq_remapping.h>
55 #define __ex(x) __kvm_handle_fault_on_reboot(x)
56 #define __ex_clear(x, reg) \
57 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
59 MODULE_AUTHOR("Qumranet");
60 MODULE_LICENSE("GPL");
62 static const struct x86_cpu_id vmx_cpu_id[] = {
63 X86_FEATURE_MATCH(X86_FEATURE_VMX),
66 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
68 static bool __read_mostly enable_vpid = 1;
69 module_param_named(vpid, enable_vpid, bool, 0444);
71 static bool __read_mostly flexpriority_enabled = 1;
72 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
74 static bool __read_mostly enable_ept = 1;
75 module_param_named(ept, enable_ept, bool, S_IRUGO);
77 static bool __read_mostly enable_unrestricted_guest = 1;
78 module_param_named(unrestricted_guest,
79 enable_unrestricted_guest, bool, S_IRUGO);
81 static bool __read_mostly enable_ept_ad_bits = 1;
82 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
84 static bool __read_mostly emulate_invalid_guest_state = true;
85 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
87 static bool __read_mostly vmm_exclusive = 1;
88 module_param(vmm_exclusive, bool, S_IRUGO);
90 static bool __read_mostly fasteoi = 1;
91 module_param(fasteoi, bool, S_IRUGO);
93 static bool __read_mostly enable_apicv = 1;
94 module_param(enable_apicv, bool, S_IRUGO);
96 static bool __read_mostly enable_shadow_vmcs = 1;
97 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
99 * If nested=1, nested virtualization is supported, i.e., guests may use
100 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
101 * use VMX instructions.
103 static bool __read_mostly nested = 0;
104 module_param(nested, bool, S_IRUGO);
106 static u64 __read_mostly host_xss;
108 static bool __read_mostly enable_pml = 1;
109 module_param_named(pml, enable_pml, bool, S_IRUGO);
111 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
113 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
114 static int __read_mostly cpu_preemption_timer_multi;
115 static bool __read_mostly enable_preemption_timer = 1;
117 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
120 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
121 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
122 #define KVM_VM_CR0_ALWAYS_ON \
123 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
124 #define KVM_CR4_GUEST_OWNED_BITS \
125 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
126 | X86_CR4_OSXMMEXCPT | X86_CR4_TSD)
128 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
129 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
131 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
133 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
136 * Hyper-V requires all of these, so mark them as supported even though
137 * they are just treated the same as all-context.
139 #define VMX_VPID_EXTENT_SUPPORTED_MASK \
140 (VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT | \
141 VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT | \
142 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT | \
143 VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
146 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
147 * ple_gap: upper bound on the amount of time between two successive
148 * executions of PAUSE in a loop. Also indicate if ple enabled.
149 * According to test, this time is usually smaller than 128 cycles.
150 * ple_window: upper bound on the amount of time a guest is allowed to execute
151 * in a PAUSE loop. Tests indicate that most spinlocks are held for
152 * less than 2^12 cycles
153 * Time is measured based on a counter that runs at the same rate as the TSC,
154 * refer SDM volume 3b section 21.6.13 & 22.1.3.
156 #define KVM_VMX_DEFAULT_PLE_GAP 128
157 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
158 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
159 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
160 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
161 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
163 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
164 module_param(ple_gap, int, S_IRUGO);
166 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
167 module_param(ple_window, int, S_IRUGO);
169 /* Default doubles per-vcpu window every exit. */
170 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
171 module_param(ple_window_grow, int, S_IRUGO);
173 /* Default resets per-vcpu window every exit to ple_window. */
174 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
175 module_param(ple_window_shrink, int, S_IRUGO);
177 /* Default is to compute the maximum so we can never overflow. */
178 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
179 static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
180 module_param(ple_window_max, int, S_IRUGO);
182 extern const ulong vmx_return;
184 #define NR_AUTOLOAD_MSRS 8
185 #define VMCS02_POOL_SIZE 1
194 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
195 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
196 * loaded on this CPU (so we can clear them if the CPU goes down).
200 struct vmcs *shadow_vmcs;
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;
254 u64 guest_ia32_debugctl;
257 u64 guest_ia32_perf_global_ctrl;
265 u64 host_ia32_perf_global_ctrl;
266 u64 padding64[8]; /* room for future expansion */
268 * To allow migration of L1 (complete with its L2 guests) between
269 * machines of different natural widths (32 or 64 bit), we cannot have
270 * unsigned long fields with no explict size. We use u64 (aliased
271 * natural_width) instead. Luckily, x86 is little-endian.
273 natural_width cr0_guest_host_mask;
274 natural_width cr4_guest_host_mask;
275 natural_width cr0_read_shadow;
276 natural_width cr4_read_shadow;
277 natural_width cr3_target_value0;
278 natural_width cr3_target_value1;
279 natural_width cr3_target_value2;
280 natural_width cr3_target_value3;
281 natural_width exit_qualification;
282 natural_width guest_linear_address;
283 natural_width guest_cr0;
284 natural_width guest_cr3;
285 natural_width guest_cr4;
286 natural_width guest_es_base;
287 natural_width guest_cs_base;
288 natural_width guest_ss_base;
289 natural_width guest_ds_base;
290 natural_width guest_fs_base;
291 natural_width guest_gs_base;
292 natural_width guest_ldtr_base;
293 natural_width guest_tr_base;
294 natural_width guest_gdtr_base;
295 natural_width guest_idtr_base;
296 natural_width guest_dr7;
297 natural_width guest_rsp;
298 natural_width guest_rip;
299 natural_width guest_rflags;
300 natural_width guest_pending_dbg_exceptions;
301 natural_width guest_sysenter_esp;
302 natural_width guest_sysenter_eip;
303 natural_width host_cr0;
304 natural_width host_cr3;
305 natural_width host_cr4;
306 natural_width host_fs_base;
307 natural_width host_gs_base;
308 natural_width host_tr_base;
309 natural_width host_gdtr_base;
310 natural_width host_idtr_base;
311 natural_width host_ia32_sysenter_esp;
312 natural_width host_ia32_sysenter_eip;
313 natural_width host_rsp;
314 natural_width host_rip;
315 natural_width paddingl[8]; /* room for future expansion */
316 u32 pin_based_vm_exec_control;
317 u32 cpu_based_vm_exec_control;
318 u32 exception_bitmap;
319 u32 page_fault_error_code_mask;
320 u32 page_fault_error_code_match;
321 u32 cr3_target_count;
322 u32 vm_exit_controls;
323 u32 vm_exit_msr_store_count;
324 u32 vm_exit_msr_load_count;
325 u32 vm_entry_controls;
326 u32 vm_entry_msr_load_count;
327 u32 vm_entry_intr_info_field;
328 u32 vm_entry_exception_error_code;
329 u32 vm_entry_instruction_len;
331 u32 secondary_vm_exec_control;
332 u32 vm_instruction_error;
334 u32 vm_exit_intr_info;
335 u32 vm_exit_intr_error_code;
336 u32 idt_vectoring_info_field;
337 u32 idt_vectoring_error_code;
338 u32 vm_exit_instruction_len;
339 u32 vmx_instruction_info;
346 u32 guest_ldtr_limit;
348 u32 guest_gdtr_limit;
349 u32 guest_idtr_limit;
350 u32 guest_es_ar_bytes;
351 u32 guest_cs_ar_bytes;
352 u32 guest_ss_ar_bytes;
353 u32 guest_ds_ar_bytes;
354 u32 guest_fs_ar_bytes;
355 u32 guest_gs_ar_bytes;
356 u32 guest_ldtr_ar_bytes;
357 u32 guest_tr_ar_bytes;
358 u32 guest_interruptibility_info;
359 u32 guest_activity_state;
360 u32 guest_sysenter_cs;
361 u32 host_ia32_sysenter_cs;
362 u32 vmx_preemption_timer_value;
363 u32 padding32[7]; /* room for future expansion */
364 u16 virtual_processor_id;
366 u16 guest_es_selector;
367 u16 guest_cs_selector;
368 u16 guest_ss_selector;
369 u16 guest_ds_selector;
370 u16 guest_fs_selector;
371 u16 guest_gs_selector;
372 u16 guest_ldtr_selector;
373 u16 guest_tr_selector;
374 u16 guest_intr_status;
375 u16 host_es_selector;
376 u16 host_cs_selector;
377 u16 host_ss_selector;
378 u16 host_ds_selector;
379 u16 host_fs_selector;
380 u16 host_gs_selector;
381 u16 host_tr_selector;
385 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
386 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
387 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
389 #define VMCS12_REVISION 0x11e57ed0
392 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
393 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
394 * current implementation, 4K are reserved to avoid future complications.
396 #define VMCS12_SIZE 0x1000
398 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
400 struct list_head list;
402 struct loaded_vmcs vmcs02;
406 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
407 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
410 /* Has the level1 guest done vmxon? */
414 /* The guest-physical address of the current VMCS L1 keeps for L2 */
416 /* The host-usable pointer to the above */
417 struct page *current_vmcs12_page;
418 struct vmcs12 *current_vmcs12;
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 VMXOFF, VMCLEAR, 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;
563 bool nmi_known_unmasked;
565 u32 idt_vectoring_info;
567 struct shared_msr_entry *guest_msrs;
570 unsigned long host_idt_base;
572 u64 msr_host_kernel_gs_base;
573 u64 msr_guest_kernel_gs_base;
575 u32 vm_entry_controls_shadow;
576 u32 vm_exit_controls_shadow;
578 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
579 * non-nested (L1) guest, it always points to vmcs01. For a nested
580 * guest (L2), it points to a different VMCS.
582 struct loaded_vmcs vmcs01;
583 struct loaded_vmcs *loaded_vmcs;
584 bool __launched; /* temporary, used in vmx_vcpu_run */
585 struct msr_autoload {
587 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
588 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
592 u16 fs_sel, gs_sel, ldt_sel;
596 int gs_ldt_reload_needed;
597 int fs_reload_needed;
598 u64 msr_host_bndcfgs;
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(HOST_ES_SELECTOR, host_es_selector),
749 FIELD(HOST_CS_SELECTOR, host_cs_selector),
750 FIELD(HOST_SS_SELECTOR, host_ss_selector),
751 FIELD(HOST_DS_SELECTOR, host_ds_selector),
752 FIELD(HOST_FS_SELECTOR, host_fs_selector),
753 FIELD(HOST_GS_SELECTOR, host_gs_selector),
754 FIELD(HOST_TR_SELECTOR, host_tr_selector),
755 FIELD64(IO_BITMAP_A, io_bitmap_a),
756 FIELD64(IO_BITMAP_B, io_bitmap_b),
757 FIELD64(MSR_BITMAP, msr_bitmap),
758 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
759 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
760 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
761 FIELD64(TSC_OFFSET, tsc_offset),
762 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
763 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
764 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
765 FIELD64(EPT_POINTER, ept_pointer),
766 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
767 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
768 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
769 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
770 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
771 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
772 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
773 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
774 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
775 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
776 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
777 FIELD64(GUEST_PDPTR0, guest_pdptr0),
778 FIELD64(GUEST_PDPTR1, guest_pdptr1),
779 FIELD64(GUEST_PDPTR2, guest_pdptr2),
780 FIELD64(GUEST_PDPTR3, guest_pdptr3),
781 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
782 FIELD64(HOST_IA32_PAT, host_ia32_pat),
783 FIELD64(HOST_IA32_EFER, host_ia32_efer),
784 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
785 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
786 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
787 FIELD(EXCEPTION_BITMAP, exception_bitmap),
788 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
789 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
790 FIELD(CR3_TARGET_COUNT, cr3_target_count),
791 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
792 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
793 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
794 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
795 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
796 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
797 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
798 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
799 FIELD(TPR_THRESHOLD, tpr_threshold),
800 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
801 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
802 FIELD(VM_EXIT_REASON, vm_exit_reason),
803 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
804 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
805 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
806 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
807 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
808 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
809 FIELD(GUEST_ES_LIMIT, guest_es_limit),
810 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
811 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
812 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
813 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
814 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
815 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
816 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
817 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
818 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
819 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
820 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
821 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
822 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
823 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
824 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
825 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
826 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
827 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
828 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
829 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
830 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
831 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
832 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
833 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
834 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
835 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
836 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
837 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
838 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
839 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
840 FIELD(EXIT_QUALIFICATION, exit_qualification),
841 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
842 FIELD(GUEST_CR0, guest_cr0),
843 FIELD(GUEST_CR3, guest_cr3),
844 FIELD(GUEST_CR4, guest_cr4),
845 FIELD(GUEST_ES_BASE, guest_es_base),
846 FIELD(GUEST_CS_BASE, guest_cs_base),
847 FIELD(GUEST_SS_BASE, guest_ss_base),
848 FIELD(GUEST_DS_BASE, guest_ds_base),
849 FIELD(GUEST_FS_BASE, guest_fs_base),
850 FIELD(GUEST_GS_BASE, guest_gs_base),
851 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
852 FIELD(GUEST_TR_BASE, guest_tr_base),
853 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
854 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
855 FIELD(GUEST_DR7, guest_dr7),
856 FIELD(GUEST_RSP, guest_rsp),
857 FIELD(GUEST_RIP, guest_rip),
858 FIELD(GUEST_RFLAGS, guest_rflags),
859 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
860 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
861 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
862 FIELD(HOST_CR0, host_cr0),
863 FIELD(HOST_CR3, host_cr3),
864 FIELD(HOST_CR4, host_cr4),
865 FIELD(HOST_FS_BASE, host_fs_base),
866 FIELD(HOST_GS_BASE, host_gs_base),
867 FIELD(HOST_TR_BASE, host_tr_base),
868 FIELD(HOST_GDTR_BASE, host_gdtr_base),
869 FIELD(HOST_IDTR_BASE, host_idtr_base),
870 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
871 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
872 FIELD(HOST_RSP, host_rsp),
873 FIELD(HOST_RIP, host_rip),
876 static inline short vmcs_field_to_offset(unsigned long field)
878 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
880 if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
881 vmcs_field_to_offset_table[field] == 0)
884 return vmcs_field_to_offset_table[field];
887 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
889 return to_vmx(vcpu)->nested.cached_vmcs12;
892 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
894 struct page *page = kvm_vcpu_gfn_to_page(vcpu, addr >> PAGE_SHIFT);
895 if (is_error_page(page))
901 static void nested_release_page(struct page *page)
903 kvm_release_page_dirty(page);
906 static void nested_release_page_clean(struct page *page)
908 kvm_release_page_clean(page);
911 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
912 static u64 construct_eptp(unsigned long root_hpa);
913 static void kvm_cpu_vmxon(u64 addr);
914 static void kvm_cpu_vmxoff(void);
915 static bool vmx_xsaves_supported(void);
916 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
917 static void vmx_set_segment(struct kvm_vcpu *vcpu,
918 struct kvm_segment *var, int seg);
919 static void vmx_get_segment(struct kvm_vcpu *vcpu,
920 struct kvm_segment *var, int seg);
921 static bool guest_state_valid(struct kvm_vcpu *vcpu);
922 static u32 vmx_segment_access_rights(struct kvm_segment *var);
923 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
924 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
925 static int alloc_identity_pagetable(struct kvm *kvm);
927 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
928 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
930 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
931 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
933 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
934 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
937 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
938 * can find which vCPU should be waken up.
940 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
941 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
946 VMX_MSR_BITMAP_LEGACY,
947 VMX_MSR_BITMAP_LONGMODE,
948 VMX_MSR_BITMAP_LEGACY_X2APIC_APICV,
949 VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV,
950 VMX_MSR_BITMAP_LEGACY_X2APIC,
951 VMX_MSR_BITMAP_LONGMODE_X2APIC,
957 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
959 #define vmx_io_bitmap_a (vmx_bitmap[VMX_IO_BITMAP_A])
960 #define vmx_io_bitmap_b (vmx_bitmap[VMX_IO_BITMAP_B])
961 #define vmx_msr_bitmap_legacy (vmx_bitmap[VMX_MSR_BITMAP_LEGACY])
962 #define vmx_msr_bitmap_longmode (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE])
963 #define vmx_msr_bitmap_legacy_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC_APICV])
964 #define vmx_msr_bitmap_longmode_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV])
965 #define vmx_msr_bitmap_legacy_x2apic (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC])
966 #define vmx_msr_bitmap_longmode_x2apic (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC])
967 #define vmx_vmread_bitmap (vmx_bitmap[VMX_VMREAD_BITMAP])
968 #define vmx_vmwrite_bitmap (vmx_bitmap[VMX_VMWRITE_BITMAP])
970 static bool cpu_has_load_ia32_efer;
971 static bool cpu_has_load_perf_global_ctrl;
973 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
974 static DEFINE_SPINLOCK(vmx_vpid_lock);
976 static struct vmcs_config {
981 u32 pin_based_exec_ctrl;
982 u32 cpu_based_exec_ctrl;
983 u32 cpu_based_2nd_exec_ctrl;
988 static struct vmx_capability {
993 #define VMX_SEGMENT_FIELD(seg) \
994 [VCPU_SREG_##seg] = { \
995 .selector = GUEST_##seg##_SELECTOR, \
996 .base = GUEST_##seg##_BASE, \
997 .limit = GUEST_##seg##_LIMIT, \
998 .ar_bytes = GUEST_##seg##_AR_BYTES, \
1001 static const struct kvm_vmx_segment_field {
1006 } kvm_vmx_segment_fields[] = {
1007 VMX_SEGMENT_FIELD(CS),
1008 VMX_SEGMENT_FIELD(DS),
1009 VMX_SEGMENT_FIELD(ES),
1010 VMX_SEGMENT_FIELD(FS),
1011 VMX_SEGMENT_FIELD(GS),
1012 VMX_SEGMENT_FIELD(SS),
1013 VMX_SEGMENT_FIELD(TR),
1014 VMX_SEGMENT_FIELD(LDTR),
1017 static u64 host_efer;
1019 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
1022 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
1023 * away by decrementing the array size.
1025 static const u32 vmx_msr_index[] = {
1026 #ifdef CONFIG_X86_64
1027 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
1029 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
1032 static inline bool is_exception_n(u32 intr_info, u8 vector)
1034 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1035 INTR_INFO_VALID_MASK)) ==
1036 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
1039 static inline bool is_debug(u32 intr_info)
1041 return is_exception_n(intr_info, DB_VECTOR);
1044 static inline bool is_breakpoint(u32 intr_info)
1046 return is_exception_n(intr_info, BP_VECTOR);
1049 static inline bool is_page_fault(u32 intr_info)
1051 return is_exception_n(intr_info, PF_VECTOR);
1054 static inline bool is_no_device(u32 intr_info)
1056 return is_exception_n(intr_info, NM_VECTOR);
1059 static inline bool is_invalid_opcode(u32 intr_info)
1061 return is_exception_n(intr_info, UD_VECTOR);
1064 static inline bool is_external_interrupt(u32 intr_info)
1066 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1067 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1070 static inline bool is_machine_check(u32 intr_info)
1072 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1073 INTR_INFO_VALID_MASK)) ==
1074 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1077 static inline bool cpu_has_vmx_msr_bitmap(void)
1079 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1082 static inline bool cpu_has_vmx_tpr_shadow(void)
1084 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1087 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1089 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1092 static inline bool cpu_has_secondary_exec_ctrls(void)
1094 return vmcs_config.cpu_based_exec_ctrl &
1095 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1098 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1100 return vmcs_config.cpu_based_2nd_exec_ctrl &
1101 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1104 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1106 return vmcs_config.cpu_based_2nd_exec_ctrl &
1107 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1110 static inline bool cpu_has_vmx_apic_register_virt(void)
1112 return vmcs_config.cpu_based_2nd_exec_ctrl &
1113 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1116 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1118 return vmcs_config.cpu_based_2nd_exec_ctrl &
1119 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1123 * Comment's format: document - errata name - stepping - processor name.
1125 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1127 static u32 vmx_preemption_cpu_tfms[] = {
1128 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
1130 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
1131 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1132 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1134 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1136 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
1137 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
1139 * 320767.pdf - AAP86 - B1 -
1140 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1143 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1145 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1147 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1149 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1150 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1151 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1155 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1157 u32 eax = cpuid_eax(0x00000001), i;
1159 /* Clear the reserved bits */
1160 eax &= ~(0x3U << 14 | 0xfU << 28);
1161 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
1162 if (eax == vmx_preemption_cpu_tfms[i])
1168 static inline bool cpu_has_vmx_preemption_timer(void)
1170 return vmcs_config.pin_based_exec_ctrl &
1171 PIN_BASED_VMX_PREEMPTION_TIMER;
1174 static inline bool cpu_has_vmx_posted_intr(void)
1176 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1177 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1180 static inline bool cpu_has_vmx_apicv(void)
1182 return cpu_has_vmx_apic_register_virt() &&
1183 cpu_has_vmx_virtual_intr_delivery() &&
1184 cpu_has_vmx_posted_intr();
1187 static inline bool cpu_has_vmx_flexpriority(void)
1189 return cpu_has_vmx_tpr_shadow() &&
1190 cpu_has_vmx_virtualize_apic_accesses();
1193 static inline bool cpu_has_vmx_ept_execute_only(void)
1195 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1198 static inline bool cpu_has_vmx_ept_2m_page(void)
1200 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1203 static inline bool cpu_has_vmx_ept_1g_page(void)
1205 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1208 static inline bool cpu_has_vmx_ept_4levels(void)
1210 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1213 static inline bool cpu_has_vmx_ept_ad_bits(void)
1215 return vmx_capability.ept & VMX_EPT_AD_BIT;
1218 static inline bool cpu_has_vmx_invept_context(void)
1220 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1223 static inline bool cpu_has_vmx_invept_global(void)
1225 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1228 static inline bool cpu_has_vmx_invvpid_single(void)
1230 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1233 static inline bool cpu_has_vmx_invvpid_global(void)
1235 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1238 static inline bool cpu_has_vmx_invvpid(void)
1240 return vmx_capability.vpid & VMX_VPID_INVVPID_BIT;
1243 static inline bool cpu_has_vmx_ept(void)
1245 return vmcs_config.cpu_based_2nd_exec_ctrl &
1246 SECONDARY_EXEC_ENABLE_EPT;
1249 static inline bool cpu_has_vmx_unrestricted_guest(void)
1251 return vmcs_config.cpu_based_2nd_exec_ctrl &
1252 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1255 static inline bool cpu_has_vmx_ple(void)
1257 return vmcs_config.cpu_based_2nd_exec_ctrl &
1258 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1261 static inline bool cpu_has_vmx_basic_inout(void)
1263 return (((u64)vmcs_config.basic_cap << 32) & VMX_BASIC_INOUT);
1266 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1268 return flexpriority_enabled && lapic_in_kernel(vcpu);
1271 static inline bool cpu_has_vmx_vpid(void)
1273 return vmcs_config.cpu_based_2nd_exec_ctrl &
1274 SECONDARY_EXEC_ENABLE_VPID;
1277 static inline bool cpu_has_vmx_rdtscp(void)
1279 return vmcs_config.cpu_based_2nd_exec_ctrl &
1280 SECONDARY_EXEC_RDTSCP;
1283 static inline bool cpu_has_vmx_invpcid(void)
1285 return vmcs_config.cpu_based_2nd_exec_ctrl &
1286 SECONDARY_EXEC_ENABLE_INVPCID;
1289 static inline bool cpu_has_vmx_wbinvd_exit(void)
1291 return vmcs_config.cpu_based_2nd_exec_ctrl &
1292 SECONDARY_EXEC_WBINVD_EXITING;
1295 static inline bool cpu_has_vmx_shadow_vmcs(void)
1298 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1299 /* check if the cpu supports writing r/o exit information fields */
1300 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1303 return vmcs_config.cpu_based_2nd_exec_ctrl &
1304 SECONDARY_EXEC_SHADOW_VMCS;
1307 static inline bool cpu_has_vmx_pml(void)
1309 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1312 static inline bool cpu_has_vmx_tsc_scaling(void)
1314 return vmcs_config.cpu_based_2nd_exec_ctrl &
1315 SECONDARY_EXEC_TSC_SCALING;
1318 static inline bool report_flexpriority(void)
1320 return flexpriority_enabled;
1323 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1325 return vmcs12->cpu_based_vm_exec_control & bit;
1328 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1330 return (vmcs12->cpu_based_vm_exec_control &
1331 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1332 (vmcs12->secondary_vm_exec_control & bit);
1335 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1337 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1340 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1342 return vmcs12->pin_based_vm_exec_control &
1343 PIN_BASED_VMX_PREEMPTION_TIMER;
1346 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1348 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1351 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1353 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) &&
1354 vmx_xsaves_supported();
1357 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1359 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1362 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1364 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1367 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1369 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1372 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1374 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1377 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1379 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1382 static inline bool is_nmi(u32 intr_info)
1384 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1385 == (INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK);
1388 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1390 unsigned long exit_qualification);
1391 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1392 struct vmcs12 *vmcs12,
1393 u32 reason, unsigned long qualification);
1395 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1399 for (i = 0; i < vmx->nmsrs; ++i)
1400 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1405 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1411 } operand = { vpid, 0, gva };
1413 asm volatile (__ex(ASM_VMX_INVVPID)
1414 /* CF==1 or ZF==1 --> rc = -1 */
1415 "; ja 1f ; ud2 ; 1:"
1416 : : "a"(&operand), "c"(ext) : "cc", "memory");
1419 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1423 } operand = {eptp, gpa};
1425 asm volatile (__ex(ASM_VMX_INVEPT)
1426 /* CF==1 or ZF==1 --> rc = -1 */
1427 "; ja 1f ; ud2 ; 1:\n"
1428 : : "a" (&operand), "c" (ext) : "cc", "memory");
1431 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1435 i = __find_msr_index(vmx, msr);
1437 return &vmx->guest_msrs[i];
1441 static void vmcs_clear(struct vmcs *vmcs)
1443 u64 phys_addr = __pa(vmcs);
1446 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1447 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1450 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1454 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1456 vmcs_clear(loaded_vmcs->vmcs);
1457 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
1458 vmcs_clear(loaded_vmcs->shadow_vmcs);
1459 loaded_vmcs->cpu = -1;
1460 loaded_vmcs->launched = 0;
1463 static void vmcs_load(struct vmcs *vmcs)
1465 u64 phys_addr = __pa(vmcs);
1468 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1469 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1472 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1476 #ifdef CONFIG_KEXEC_CORE
1478 * This bitmap is used to indicate whether the vmclear
1479 * operation is enabled on all cpus. All disabled by
1482 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1484 static inline void crash_enable_local_vmclear(int cpu)
1486 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1489 static inline void crash_disable_local_vmclear(int cpu)
1491 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1494 static inline int crash_local_vmclear_enabled(int cpu)
1496 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1499 static void crash_vmclear_local_loaded_vmcss(void)
1501 int cpu = raw_smp_processor_id();
1502 struct loaded_vmcs *v;
1504 if (!crash_local_vmclear_enabled(cpu))
1507 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1508 loaded_vmcss_on_cpu_link)
1509 vmcs_clear(v->vmcs);
1512 static inline void crash_enable_local_vmclear(int cpu) { }
1513 static inline void crash_disable_local_vmclear(int cpu) { }
1514 #endif /* CONFIG_KEXEC_CORE */
1516 static void __loaded_vmcs_clear(void *arg)
1518 struct loaded_vmcs *loaded_vmcs = arg;
1519 int cpu = raw_smp_processor_id();
1521 if (loaded_vmcs->cpu != cpu)
1522 return; /* vcpu migration can race with cpu offline */
1523 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1524 per_cpu(current_vmcs, cpu) = NULL;
1525 crash_disable_local_vmclear(cpu);
1526 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1529 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1530 * is before setting loaded_vmcs->vcpu to -1 which is done in
1531 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1532 * then adds the vmcs into percpu list before it is deleted.
1536 loaded_vmcs_init(loaded_vmcs);
1537 crash_enable_local_vmclear(cpu);
1540 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1542 int cpu = loaded_vmcs->cpu;
1545 smp_call_function_single(cpu,
1546 __loaded_vmcs_clear, loaded_vmcs, 1);
1549 static inline void vpid_sync_vcpu_single(int vpid)
1554 if (cpu_has_vmx_invvpid_single())
1555 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1558 static inline void vpid_sync_vcpu_global(void)
1560 if (cpu_has_vmx_invvpid_global())
1561 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1564 static inline void vpid_sync_context(int vpid)
1566 if (cpu_has_vmx_invvpid_single())
1567 vpid_sync_vcpu_single(vpid);
1569 vpid_sync_vcpu_global();
1572 static inline void ept_sync_global(void)
1574 if (cpu_has_vmx_invept_global())
1575 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1578 static inline void ept_sync_context(u64 eptp)
1581 if (cpu_has_vmx_invept_context())
1582 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1588 static __always_inline void vmcs_check16(unsigned long field)
1590 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1591 "16-bit accessor invalid for 64-bit field");
1592 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1593 "16-bit accessor invalid for 64-bit high field");
1594 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1595 "16-bit accessor invalid for 32-bit high field");
1596 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1597 "16-bit accessor invalid for natural width field");
1600 static __always_inline void vmcs_check32(unsigned long field)
1602 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1603 "32-bit accessor invalid for 16-bit field");
1604 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1605 "32-bit accessor invalid for natural width field");
1608 static __always_inline void vmcs_check64(unsigned long field)
1610 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1611 "64-bit accessor invalid for 16-bit field");
1612 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1613 "64-bit accessor invalid for 64-bit high field");
1614 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1615 "64-bit accessor invalid for 32-bit field");
1616 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1617 "64-bit accessor invalid for natural width field");
1620 static __always_inline void vmcs_checkl(unsigned long field)
1622 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1623 "Natural width accessor invalid for 16-bit field");
1624 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1625 "Natural width accessor invalid for 64-bit field");
1626 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1627 "Natural width accessor invalid for 64-bit high field");
1628 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1629 "Natural width accessor invalid for 32-bit field");
1632 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1634 unsigned long value;
1636 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1637 : "=a"(value) : "d"(field) : "cc");
1641 static __always_inline u16 vmcs_read16(unsigned long field)
1643 vmcs_check16(field);
1644 return __vmcs_readl(field);
1647 static __always_inline u32 vmcs_read32(unsigned long field)
1649 vmcs_check32(field);
1650 return __vmcs_readl(field);
1653 static __always_inline u64 vmcs_read64(unsigned long field)
1655 vmcs_check64(field);
1656 #ifdef CONFIG_X86_64
1657 return __vmcs_readl(field);
1659 return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1663 static __always_inline unsigned long vmcs_readl(unsigned long field)
1666 return __vmcs_readl(field);
1669 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1671 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1672 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1676 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1680 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1681 : "=q"(error) : "a"(value), "d"(field) : "cc");
1682 if (unlikely(error))
1683 vmwrite_error(field, value);
1686 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1688 vmcs_check16(field);
1689 __vmcs_writel(field, value);
1692 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1694 vmcs_check32(field);
1695 __vmcs_writel(field, value);
1698 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1700 vmcs_check64(field);
1701 __vmcs_writel(field, value);
1702 #ifndef CONFIG_X86_64
1704 __vmcs_writel(field+1, value >> 32);
1708 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1711 __vmcs_writel(field, value);
1714 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1716 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1717 "vmcs_clear_bits does not support 64-bit fields");
1718 __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1721 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1723 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1724 "vmcs_set_bits does not support 64-bit fields");
1725 __vmcs_writel(field, __vmcs_readl(field) | mask);
1728 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
1730 vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
1733 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1735 vmcs_write32(VM_ENTRY_CONTROLS, val);
1736 vmx->vm_entry_controls_shadow = val;
1739 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1741 if (vmx->vm_entry_controls_shadow != val)
1742 vm_entry_controls_init(vmx, val);
1745 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1747 return vmx->vm_entry_controls_shadow;
1751 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1753 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1756 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1758 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1761 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
1763 vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
1766 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1768 vmcs_write32(VM_EXIT_CONTROLS, val);
1769 vmx->vm_exit_controls_shadow = val;
1772 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1774 if (vmx->vm_exit_controls_shadow != val)
1775 vm_exit_controls_init(vmx, val);
1778 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1780 return vmx->vm_exit_controls_shadow;
1784 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1786 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1789 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1791 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1794 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1796 vmx->segment_cache.bitmask = 0;
1799 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1803 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1805 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1806 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1807 vmx->segment_cache.bitmask = 0;
1809 ret = vmx->segment_cache.bitmask & mask;
1810 vmx->segment_cache.bitmask |= mask;
1814 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1816 u16 *p = &vmx->segment_cache.seg[seg].selector;
1818 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1819 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1823 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1825 ulong *p = &vmx->segment_cache.seg[seg].base;
1827 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1828 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1832 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1834 u32 *p = &vmx->segment_cache.seg[seg].limit;
1836 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1837 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1841 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1843 u32 *p = &vmx->segment_cache.seg[seg].ar;
1845 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1846 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1850 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1854 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1855 (1u << DB_VECTOR) | (1u << AC_VECTOR);
1856 if ((vcpu->guest_debug &
1857 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1858 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1859 eb |= 1u << BP_VECTOR;
1860 if (to_vmx(vcpu)->rmode.vm86_active)
1863 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1865 /* When we are running a nested L2 guest and L1 specified for it a
1866 * certain exception bitmap, we must trap the same exceptions and pass
1867 * them to L1. When running L2, we will only handle the exceptions
1868 * specified above if L1 did not want them.
1870 if (is_guest_mode(vcpu))
1871 eb |= get_vmcs12(vcpu)->exception_bitmap;
1873 vmcs_write32(EXCEPTION_BITMAP, eb);
1876 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1877 unsigned long entry, unsigned long exit)
1879 vm_entry_controls_clearbit(vmx, entry);
1880 vm_exit_controls_clearbit(vmx, exit);
1883 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1886 struct msr_autoload *m = &vmx->msr_autoload;
1890 if (cpu_has_load_ia32_efer) {
1891 clear_atomic_switch_msr_special(vmx,
1892 VM_ENTRY_LOAD_IA32_EFER,
1893 VM_EXIT_LOAD_IA32_EFER);
1897 case MSR_CORE_PERF_GLOBAL_CTRL:
1898 if (cpu_has_load_perf_global_ctrl) {
1899 clear_atomic_switch_msr_special(vmx,
1900 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1901 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1907 for (i = 0; i < m->nr; ++i)
1908 if (m->guest[i].index == msr)
1914 m->guest[i] = m->guest[m->nr];
1915 m->host[i] = m->host[m->nr];
1916 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1917 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1920 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1921 unsigned long entry, unsigned long exit,
1922 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1923 u64 guest_val, u64 host_val)
1925 vmcs_write64(guest_val_vmcs, guest_val);
1926 vmcs_write64(host_val_vmcs, host_val);
1927 vm_entry_controls_setbit(vmx, entry);
1928 vm_exit_controls_setbit(vmx, exit);
1931 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1932 u64 guest_val, u64 host_val)
1935 struct msr_autoload *m = &vmx->msr_autoload;
1939 if (cpu_has_load_ia32_efer) {
1940 add_atomic_switch_msr_special(vmx,
1941 VM_ENTRY_LOAD_IA32_EFER,
1942 VM_EXIT_LOAD_IA32_EFER,
1945 guest_val, host_val);
1949 case MSR_CORE_PERF_GLOBAL_CTRL:
1950 if (cpu_has_load_perf_global_ctrl) {
1951 add_atomic_switch_msr_special(vmx,
1952 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1953 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1954 GUEST_IA32_PERF_GLOBAL_CTRL,
1955 HOST_IA32_PERF_GLOBAL_CTRL,
1956 guest_val, host_val);
1960 case MSR_IA32_PEBS_ENABLE:
1961 /* PEBS needs a quiescent period after being disabled (to write
1962 * a record). Disabling PEBS through VMX MSR swapping doesn't
1963 * provide that period, so a CPU could write host's record into
1966 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1969 for (i = 0; i < m->nr; ++i)
1970 if (m->guest[i].index == msr)
1973 if (i == NR_AUTOLOAD_MSRS) {
1974 printk_once(KERN_WARNING "Not enough msr switch entries. "
1975 "Can't add msr %x\n", msr);
1977 } else if (i == m->nr) {
1979 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1980 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1983 m->guest[i].index = msr;
1984 m->guest[i].value = guest_val;
1985 m->host[i].index = msr;
1986 m->host[i].value = host_val;
1989 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1991 u64 guest_efer = vmx->vcpu.arch.efer;
1992 u64 ignore_bits = 0;
1996 * NX is needed to handle CR0.WP=1, CR4.SMEP=1. Testing
1997 * host CPUID is more efficient than testing guest CPUID
1998 * or CR4. Host SMEP is anyway a requirement for guest SMEP.
2000 if (boot_cpu_has(X86_FEATURE_SMEP))
2001 guest_efer |= EFER_NX;
2002 else if (!(guest_efer & EFER_NX))
2003 ignore_bits |= EFER_NX;
2007 * LMA and LME handled by hardware; SCE meaningless outside long mode.
2009 ignore_bits |= EFER_SCE;
2010 #ifdef CONFIG_X86_64
2011 ignore_bits |= EFER_LMA | EFER_LME;
2012 /* SCE is meaningful only in long mode on Intel */
2013 if (guest_efer & EFER_LMA)
2014 ignore_bits &= ~(u64)EFER_SCE;
2017 clear_atomic_switch_msr(vmx, MSR_EFER);
2020 * On EPT, we can't emulate NX, so we must switch EFER atomically.
2021 * On CPUs that support "load IA32_EFER", always switch EFER
2022 * atomically, since it's faster than switching it manually.
2024 if (cpu_has_load_ia32_efer ||
2025 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
2026 if (!(guest_efer & EFER_LMA))
2027 guest_efer &= ~EFER_LME;
2028 if (guest_efer != host_efer)
2029 add_atomic_switch_msr(vmx, MSR_EFER,
2030 guest_efer, host_efer);
2033 guest_efer &= ~ignore_bits;
2034 guest_efer |= host_efer & ignore_bits;
2036 vmx->guest_msrs[efer_offset].data = guest_efer;
2037 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
2043 #ifdef CONFIG_X86_32
2045 * On 32-bit kernels, VM exits still load the FS and GS bases from the
2046 * VMCS rather than the segment table. KVM uses this helper to figure
2047 * out the current bases to poke them into the VMCS before entry.
2049 static unsigned long segment_base(u16 selector)
2051 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2052 struct desc_struct *table;
2055 if (!(selector & ~SEGMENT_RPL_MASK))
2058 table = (struct desc_struct *)gdt->address;
2060 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
2061 u16 ldt_selector = kvm_read_ldt();
2063 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
2066 table = (struct desc_struct *)segment_base(ldt_selector);
2068 v = get_desc_base(&table[selector >> 3]);
2073 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
2075 struct vcpu_vmx *vmx = to_vmx(vcpu);
2078 if (vmx->host_state.loaded)
2081 vmx->host_state.loaded = 1;
2083 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
2084 * allow segment selectors with cpl > 0 or ti == 1.
2086 vmx->host_state.ldt_sel = kvm_read_ldt();
2087 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
2088 savesegment(fs, vmx->host_state.fs_sel);
2089 if (!(vmx->host_state.fs_sel & 7)) {
2090 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
2091 vmx->host_state.fs_reload_needed = 0;
2093 vmcs_write16(HOST_FS_SELECTOR, 0);
2094 vmx->host_state.fs_reload_needed = 1;
2096 savesegment(gs, vmx->host_state.gs_sel);
2097 if (!(vmx->host_state.gs_sel & 7))
2098 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
2100 vmcs_write16(HOST_GS_SELECTOR, 0);
2101 vmx->host_state.gs_ldt_reload_needed = 1;
2104 #ifdef CONFIG_X86_64
2105 savesegment(ds, vmx->host_state.ds_sel);
2106 savesegment(es, vmx->host_state.es_sel);
2109 #ifdef CONFIG_X86_64
2110 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
2111 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
2113 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
2114 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
2117 #ifdef CONFIG_X86_64
2118 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2119 if (is_long_mode(&vmx->vcpu))
2120 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2122 if (boot_cpu_has(X86_FEATURE_MPX))
2123 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2124 for (i = 0; i < vmx->save_nmsrs; ++i)
2125 kvm_set_shared_msr(vmx->guest_msrs[i].index,
2126 vmx->guest_msrs[i].data,
2127 vmx->guest_msrs[i].mask);
2130 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
2132 if (!vmx->host_state.loaded)
2135 ++vmx->vcpu.stat.host_state_reload;
2136 vmx->host_state.loaded = 0;
2137 #ifdef CONFIG_X86_64
2138 if (is_long_mode(&vmx->vcpu))
2139 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2141 if (vmx->host_state.gs_ldt_reload_needed) {
2142 kvm_load_ldt(vmx->host_state.ldt_sel);
2143 #ifdef CONFIG_X86_64
2144 load_gs_index(vmx->host_state.gs_sel);
2146 loadsegment(gs, vmx->host_state.gs_sel);
2149 if (vmx->host_state.fs_reload_needed)
2150 loadsegment(fs, vmx->host_state.fs_sel);
2151 #ifdef CONFIG_X86_64
2152 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2153 loadsegment(ds, vmx->host_state.ds_sel);
2154 loadsegment(es, vmx->host_state.es_sel);
2157 invalidate_tss_limit();
2158 #ifdef CONFIG_X86_64
2159 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2161 if (vmx->host_state.msr_host_bndcfgs)
2162 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2163 load_gdt(this_cpu_ptr(&host_gdt));
2166 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2169 __vmx_load_host_state(vmx);
2173 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2175 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2176 struct pi_desc old, new;
2179 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2180 !irq_remapping_cap(IRQ_POSTING_CAP) ||
2181 !kvm_vcpu_apicv_active(vcpu))
2185 old.control = new.control = pi_desc->control;
2188 * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
2189 * are two possible cases:
2190 * 1. After running 'pre_block', context switch
2191 * happened. For this case, 'sn' was set in
2192 * vmx_vcpu_put(), so we need to clear it here.
2193 * 2. After running 'pre_block', we were blocked,
2194 * and woken up by some other guy. For this case,
2195 * we don't need to do anything, 'pi_post_block'
2196 * will do everything for us. However, we cannot
2197 * check whether it is case #1 or case #2 here
2198 * (maybe, not needed), so we also clear sn here,
2199 * I think it is not a big deal.
2201 if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) {
2202 if (vcpu->cpu != cpu) {
2203 dest = cpu_physical_id(cpu);
2205 if (x2apic_enabled())
2208 new.ndst = (dest << 8) & 0xFF00;
2211 /* set 'NV' to 'notification vector' */
2212 new.nv = POSTED_INTR_VECTOR;
2215 /* Allow posting non-urgent interrupts */
2217 } while (cmpxchg(&pi_desc->control, old.control,
2218 new.control) != old.control);
2221 static void decache_tsc_multiplier(struct vcpu_vmx *vmx)
2223 vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
2224 vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2228 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2229 * vcpu mutex is already taken.
2231 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2233 struct vcpu_vmx *vmx = to_vmx(vcpu);
2234 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2235 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
2238 kvm_cpu_vmxon(phys_addr);
2239 else if (!already_loaded)
2240 loaded_vmcs_clear(vmx->loaded_vmcs);
2242 if (!already_loaded) {
2243 local_irq_disable();
2244 crash_disable_local_vmclear(cpu);
2247 * Read loaded_vmcs->cpu should be before fetching
2248 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2249 * See the comments in __loaded_vmcs_clear().
2253 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2254 &per_cpu(loaded_vmcss_on_cpu, cpu));
2255 crash_enable_local_vmclear(cpu);
2259 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2260 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2261 vmcs_load(vmx->loaded_vmcs->vmcs);
2264 if (!already_loaded) {
2265 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2266 unsigned long sysenter_esp;
2268 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2271 * Linux uses per-cpu TSS and GDT, so set these when switching
2272 * processors. See 22.2.4.
2274 vmcs_writel(HOST_TR_BASE,
2275 (unsigned long)this_cpu_ptr(&cpu_tss));
2276 vmcs_writel(HOST_GDTR_BASE, gdt->address);
2279 * VM exits change the host TR limit to 0x67 after a VM
2280 * exit. This is okay, since 0x67 covers everything except
2281 * the IO bitmap and have have code to handle the IO bitmap
2282 * being lost after a VM exit.
2284 BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
2286 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2287 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2289 vmx->loaded_vmcs->cpu = cpu;
2292 /* Setup TSC multiplier */
2293 if (kvm_has_tsc_control &&
2294 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
2295 decache_tsc_multiplier(vmx);
2297 vmx_vcpu_pi_load(vcpu, cpu);
2298 vmx->host_pkru = read_pkru();
2301 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2303 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2305 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2306 !irq_remapping_cap(IRQ_POSTING_CAP) ||
2307 !kvm_vcpu_apicv_active(vcpu))
2310 /* Set SN when the vCPU is preempted */
2311 if (vcpu->preempted)
2315 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2317 vmx_vcpu_pi_put(vcpu);
2319 __vmx_load_host_state(to_vmx(vcpu));
2320 if (!vmm_exclusive) {
2321 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
2327 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2330 * Return the cr0 value that a nested guest would read. This is a combination
2331 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2332 * its hypervisor (cr0_read_shadow).
2334 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2336 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2337 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2339 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2341 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2342 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2345 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2347 unsigned long rflags, save_rflags;
2349 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2350 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2351 rflags = vmcs_readl(GUEST_RFLAGS);
2352 if (to_vmx(vcpu)->rmode.vm86_active) {
2353 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2354 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2355 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2357 to_vmx(vcpu)->rflags = rflags;
2359 return to_vmx(vcpu)->rflags;
2362 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2364 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2365 to_vmx(vcpu)->rflags = rflags;
2366 if (to_vmx(vcpu)->rmode.vm86_active) {
2367 to_vmx(vcpu)->rmode.save_rflags = rflags;
2368 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2370 vmcs_writel(GUEST_RFLAGS, rflags);
2373 static u32 vmx_get_pkru(struct kvm_vcpu *vcpu)
2375 return to_vmx(vcpu)->guest_pkru;
2378 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2380 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2383 if (interruptibility & GUEST_INTR_STATE_STI)
2384 ret |= KVM_X86_SHADOW_INT_STI;
2385 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2386 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2391 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2393 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2394 u32 interruptibility = interruptibility_old;
2396 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2398 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2399 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2400 else if (mask & KVM_X86_SHADOW_INT_STI)
2401 interruptibility |= GUEST_INTR_STATE_STI;
2403 if ((interruptibility != interruptibility_old))
2404 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2407 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2411 rip = kvm_rip_read(vcpu);
2412 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2413 kvm_rip_write(vcpu, rip);
2415 /* skipping an emulated instruction also counts */
2416 vmx_set_interrupt_shadow(vcpu, 0);
2420 * KVM wants to inject page-faults which it got to the guest. This function
2421 * checks whether in a nested guest, we need to inject them to L1 or L2.
2423 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
2425 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2427 if (!(vmcs12->exception_bitmap & (1u << nr)))
2430 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
2431 vmcs_read32(VM_EXIT_INTR_INFO),
2432 vmcs_readl(EXIT_QUALIFICATION));
2436 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2437 bool has_error_code, u32 error_code,
2440 struct vcpu_vmx *vmx = to_vmx(vcpu);
2441 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2443 if (!reinject && is_guest_mode(vcpu) &&
2444 nested_vmx_check_exception(vcpu, nr))
2447 if (has_error_code) {
2448 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2449 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2452 if (vmx->rmode.vm86_active) {
2454 if (kvm_exception_is_soft(nr))
2455 inc_eip = vcpu->arch.event_exit_inst_len;
2456 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2457 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2461 if (kvm_exception_is_soft(nr)) {
2462 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2463 vmx->vcpu.arch.event_exit_inst_len);
2464 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2466 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2468 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2471 static bool vmx_rdtscp_supported(void)
2473 return cpu_has_vmx_rdtscp();
2476 static bool vmx_invpcid_supported(void)
2478 return cpu_has_vmx_invpcid() && enable_ept;
2482 * Swap MSR entry in host/guest MSR entry array.
2484 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2486 struct shared_msr_entry tmp;
2488 tmp = vmx->guest_msrs[to];
2489 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2490 vmx->guest_msrs[from] = tmp;
2493 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2495 unsigned long *msr_bitmap;
2497 if (is_guest_mode(vcpu))
2498 msr_bitmap = to_vmx(vcpu)->nested.msr_bitmap;
2499 else if (cpu_has_secondary_exec_ctrls() &&
2500 (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
2501 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
2502 if (enable_apicv && kvm_vcpu_apicv_active(vcpu)) {
2503 if (is_long_mode(vcpu))
2504 msr_bitmap = vmx_msr_bitmap_longmode_x2apic_apicv;
2506 msr_bitmap = vmx_msr_bitmap_legacy_x2apic_apicv;
2508 if (is_long_mode(vcpu))
2509 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2511 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2514 if (is_long_mode(vcpu))
2515 msr_bitmap = vmx_msr_bitmap_longmode;
2517 msr_bitmap = vmx_msr_bitmap_legacy;
2520 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2524 * Set up the vmcs to automatically save and restore system
2525 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2526 * mode, as fiddling with msrs is very expensive.
2528 static void setup_msrs(struct vcpu_vmx *vmx)
2530 int save_nmsrs, index;
2533 #ifdef CONFIG_X86_64
2534 if (is_long_mode(&vmx->vcpu)) {
2535 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2537 move_msr_up(vmx, index, save_nmsrs++);
2538 index = __find_msr_index(vmx, MSR_LSTAR);
2540 move_msr_up(vmx, index, save_nmsrs++);
2541 index = __find_msr_index(vmx, MSR_CSTAR);
2543 move_msr_up(vmx, index, save_nmsrs++);
2544 index = __find_msr_index(vmx, MSR_TSC_AUX);
2545 if (index >= 0 && guest_cpuid_has_rdtscp(&vmx->vcpu))
2546 move_msr_up(vmx, index, save_nmsrs++);
2548 * MSR_STAR is only needed on long mode guests, and only
2549 * if efer.sce is enabled.
2551 index = __find_msr_index(vmx, MSR_STAR);
2552 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2553 move_msr_up(vmx, index, save_nmsrs++);
2556 index = __find_msr_index(vmx, MSR_EFER);
2557 if (index >= 0 && update_transition_efer(vmx, index))
2558 move_msr_up(vmx, index, save_nmsrs++);
2560 vmx->save_nmsrs = save_nmsrs;
2562 if (cpu_has_vmx_msr_bitmap())
2563 vmx_set_msr_bitmap(&vmx->vcpu);
2567 * reads and returns guest's timestamp counter "register"
2568 * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2569 * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2571 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2573 u64 host_tsc, tsc_offset;
2576 tsc_offset = vmcs_read64(TSC_OFFSET);
2577 return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2581 * writes 'offset' into guest's timestamp counter offset register
2583 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2585 if (is_guest_mode(vcpu)) {
2587 * We're here if L1 chose not to trap WRMSR to TSC. According
2588 * to the spec, this should set L1's TSC; The offset that L1
2589 * set for L2 remains unchanged, and still needs to be added
2590 * to the newly set TSC to get L2's TSC.
2592 struct vmcs12 *vmcs12;
2593 /* recalculate vmcs02.TSC_OFFSET: */
2594 vmcs12 = get_vmcs12(vcpu);
2595 vmcs_write64(TSC_OFFSET, offset +
2596 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2597 vmcs12->tsc_offset : 0));
2599 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2600 vmcs_read64(TSC_OFFSET), offset);
2601 vmcs_write64(TSC_OFFSET, offset);
2605 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2607 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2608 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2612 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2613 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2614 * all guests if the "nested" module option is off, and can also be disabled
2615 * for a single guest by disabling its VMX cpuid bit.
2617 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2619 return nested && guest_cpuid_has_vmx(vcpu);
2623 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2624 * returned for the various VMX controls MSRs when nested VMX is enabled.
2625 * The same values should also be used to verify that vmcs12 control fields are
2626 * valid during nested entry from L1 to L2.
2627 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2628 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2629 * bit in the high half is on if the corresponding bit in the control field
2630 * may be on. See also vmx_control_verify().
2632 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2635 * Note that as a general rule, the high half of the MSRs (bits in
2636 * the control fields which may be 1) should be initialized by the
2637 * intersection of the underlying hardware's MSR (i.e., features which
2638 * can be supported) and the list of features we want to expose -
2639 * because they are known to be properly supported in our code.
2640 * Also, usually, the low half of the MSRs (bits which must be 1) can
2641 * be set to 0, meaning that L1 may turn off any of these bits. The
2642 * reason is that if one of these bits is necessary, it will appear
2643 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2644 * fields of vmcs01 and vmcs02, will turn these bits off - and
2645 * nested_vmx_exit_handled() will not pass related exits to L1.
2646 * These rules have exceptions below.
2649 /* pin-based controls */
2650 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2651 vmx->nested.nested_vmx_pinbased_ctls_low,
2652 vmx->nested.nested_vmx_pinbased_ctls_high);
2653 vmx->nested.nested_vmx_pinbased_ctls_low |=
2654 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2655 vmx->nested.nested_vmx_pinbased_ctls_high &=
2656 PIN_BASED_EXT_INTR_MASK |
2657 PIN_BASED_NMI_EXITING |
2658 PIN_BASED_VIRTUAL_NMIS;
2659 vmx->nested.nested_vmx_pinbased_ctls_high |=
2660 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2661 PIN_BASED_VMX_PREEMPTION_TIMER;
2662 if (kvm_vcpu_apicv_active(&vmx->vcpu))
2663 vmx->nested.nested_vmx_pinbased_ctls_high |=
2664 PIN_BASED_POSTED_INTR;
2667 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2668 vmx->nested.nested_vmx_exit_ctls_low,
2669 vmx->nested.nested_vmx_exit_ctls_high);
2670 vmx->nested.nested_vmx_exit_ctls_low =
2671 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2673 vmx->nested.nested_vmx_exit_ctls_high &=
2674 #ifdef CONFIG_X86_64
2675 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2677 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2678 vmx->nested.nested_vmx_exit_ctls_high |=
2679 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2680 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2681 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2683 if (kvm_mpx_supported())
2684 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2686 /* We support free control of debug control saving. */
2687 vmx->nested.nested_vmx_exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2689 /* entry controls */
2690 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2691 vmx->nested.nested_vmx_entry_ctls_low,
2692 vmx->nested.nested_vmx_entry_ctls_high);
2693 vmx->nested.nested_vmx_entry_ctls_low =
2694 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2695 vmx->nested.nested_vmx_entry_ctls_high &=
2696 #ifdef CONFIG_X86_64
2697 VM_ENTRY_IA32E_MODE |
2699 VM_ENTRY_LOAD_IA32_PAT;
2700 vmx->nested.nested_vmx_entry_ctls_high |=
2701 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2702 if (kvm_mpx_supported())
2703 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2705 /* We support free control of debug control loading. */
2706 vmx->nested.nested_vmx_entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2708 /* cpu-based controls */
2709 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2710 vmx->nested.nested_vmx_procbased_ctls_low,
2711 vmx->nested.nested_vmx_procbased_ctls_high);
2712 vmx->nested.nested_vmx_procbased_ctls_low =
2713 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2714 vmx->nested.nested_vmx_procbased_ctls_high &=
2715 CPU_BASED_VIRTUAL_INTR_PENDING |
2716 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2717 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2718 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2719 CPU_BASED_CR3_STORE_EXITING |
2720 #ifdef CONFIG_X86_64
2721 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2723 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2724 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2725 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2726 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2727 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2729 * We can allow some features even when not supported by the
2730 * hardware. For example, L1 can specify an MSR bitmap - and we
2731 * can use it to avoid exits to L1 - even when L0 runs L2
2732 * without MSR bitmaps.
2734 vmx->nested.nested_vmx_procbased_ctls_high |=
2735 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2736 CPU_BASED_USE_MSR_BITMAPS;
2738 /* We support free control of CR3 access interception. */
2739 vmx->nested.nested_vmx_procbased_ctls_low &=
2740 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2742 /* secondary cpu-based controls */
2743 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2744 vmx->nested.nested_vmx_secondary_ctls_low,
2745 vmx->nested.nested_vmx_secondary_ctls_high);
2746 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2747 vmx->nested.nested_vmx_secondary_ctls_high &=
2748 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2749 SECONDARY_EXEC_RDTSCP |
2750 SECONDARY_EXEC_DESC |
2751 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2752 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2753 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2754 SECONDARY_EXEC_WBINVD_EXITING |
2755 SECONDARY_EXEC_XSAVES;
2758 /* nested EPT: emulate EPT also to L1 */
2759 vmx->nested.nested_vmx_secondary_ctls_high |=
2760 SECONDARY_EXEC_ENABLE_EPT;
2761 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2762 VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
2763 if (cpu_has_vmx_ept_execute_only())
2764 vmx->nested.nested_vmx_ept_caps |=
2765 VMX_EPT_EXECUTE_ONLY_BIT;
2766 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2767 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
2768 VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
2769 VMX_EPT_1GB_PAGE_BIT;
2770 if (enable_ept_ad_bits)
2771 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_AD_BIT;
2773 vmx->nested.nested_vmx_ept_caps = 0;
2776 * Old versions of KVM use the single-context version without
2777 * checking for support, so declare that it is supported even
2778 * though it is treated as global context. The alternative is
2779 * not failing the single-context invvpid, and it is worse.
2782 vmx->nested.nested_vmx_secondary_ctls_high |=
2783 SECONDARY_EXEC_ENABLE_VPID;
2784 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2785 VMX_VPID_EXTENT_SUPPORTED_MASK;
2787 vmx->nested.nested_vmx_vpid_caps = 0;
2789 if (enable_unrestricted_guest)
2790 vmx->nested.nested_vmx_secondary_ctls_high |=
2791 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2793 /* miscellaneous data */
2794 rdmsr(MSR_IA32_VMX_MISC,
2795 vmx->nested.nested_vmx_misc_low,
2796 vmx->nested.nested_vmx_misc_high);
2797 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2798 vmx->nested.nested_vmx_misc_low |=
2799 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2800 VMX_MISC_ACTIVITY_HLT;
2801 vmx->nested.nested_vmx_misc_high = 0;
2804 * This MSR reports some information about VMX support. We
2805 * should return information about the VMX we emulate for the
2806 * guest, and the VMCS structure we give it - not about the
2807 * VMX support of the underlying hardware.
2809 vmx->nested.nested_vmx_basic =
2811 VMX_BASIC_TRUE_CTLS |
2812 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2813 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2815 if (cpu_has_vmx_basic_inout())
2816 vmx->nested.nested_vmx_basic |= VMX_BASIC_INOUT;
2819 * These MSRs specify bits which the guest must keep fixed on
2820 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2821 * We picked the standard core2 setting.
2823 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2824 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2825 vmx->nested.nested_vmx_cr0_fixed0 = VMXON_CR0_ALWAYSON;
2826 vmx->nested.nested_vmx_cr4_fixed0 = VMXON_CR4_ALWAYSON;
2828 /* These MSRs specify bits which the guest must keep fixed off. */
2829 rdmsrl(MSR_IA32_VMX_CR0_FIXED1, vmx->nested.nested_vmx_cr0_fixed1);
2830 rdmsrl(MSR_IA32_VMX_CR4_FIXED1, vmx->nested.nested_vmx_cr4_fixed1);
2832 /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2833 vmx->nested.nested_vmx_vmcs_enum = 0x2e;
2837 * if fixed0[i] == 1: val[i] must be 1
2838 * if fixed1[i] == 0: val[i] must be 0
2840 static inline bool fixed_bits_valid(u64 val, u64 fixed0, u64 fixed1)
2842 return ((val & fixed1) | fixed0) == val;
2845 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2847 return fixed_bits_valid(control, low, high);
2850 static inline u64 vmx_control_msr(u32 low, u32 high)
2852 return low | ((u64)high << 32);
2855 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
2860 return (superset | subset) == superset;
2863 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
2865 const u64 feature_and_reserved =
2866 /* feature (except bit 48; see below) */
2867 BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
2869 BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
2870 u64 vmx_basic = vmx->nested.nested_vmx_basic;
2872 if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
2876 * KVM does not emulate a version of VMX that constrains physical
2877 * addresses of VMX structures (e.g. VMCS) to 32-bits.
2879 if (data & BIT_ULL(48))
2882 if (vmx_basic_vmcs_revision_id(vmx_basic) !=
2883 vmx_basic_vmcs_revision_id(data))
2886 if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
2889 vmx->nested.nested_vmx_basic = data;
2894 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
2899 switch (msr_index) {
2900 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2901 lowp = &vmx->nested.nested_vmx_pinbased_ctls_low;
2902 highp = &vmx->nested.nested_vmx_pinbased_ctls_high;
2904 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2905 lowp = &vmx->nested.nested_vmx_procbased_ctls_low;
2906 highp = &vmx->nested.nested_vmx_procbased_ctls_high;
2908 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2909 lowp = &vmx->nested.nested_vmx_exit_ctls_low;
2910 highp = &vmx->nested.nested_vmx_exit_ctls_high;
2912 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2913 lowp = &vmx->nested.nested_vmx_entry_ctls_low;
2914 highp = &vmx->nested.nested_vmx_entry_ctls_high;
2916 case MSR_IA32_VMX_PROCBASED_CTLS2:
2917 lowp = &vmx->nested.nested_vmx_secondary_ctls_low;
2918 highp = &vmx->nested.nested_vmx_secondary_ctls_high;
2924 supported = vmx_control_msr(*lowp, *highp);
2926 /* Check must-be-1 bits are still 1. */
2927 if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
2930 /* Check must-be-0 bits are still 0. */
2931 if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
2935 *highp = data >> 32;
2939 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
2941 const u64 feature_and_reserved_bits =
2943 BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
2944 BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
2946 GENMASK_ULL(13, 9) | BIT_ULL(31);
2949 vmx_misc = vmx_control_msr(vmx->nested.nested_vmx_misc_low,
2950 vmx->nested.nested_vmx_misc_high);
2952 if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
2955 if ((vmx->nested.nested_vmx_pinbased_ctls_high &
2956 PIN_BASED_VMX_PREEMPTION_TIMER) &&
2957 vmx_misc_preemption_timer_rate(data) !=
2958 vmx_misc_preemption_timer_rate(vmx_misc))
2961 if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
2964 if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
2967 if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
2970 vmx->nested.nested_vmx_misc_low = data;
2971 vmx->nested.nested_vmx_misc_high = data >> 32;
2975 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
2977 u64 vmx_ept_vpid_cap;
2979 vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.nested_vmx_ept_caps,
2980 vmx->nested.nested_vmx_vpid_caps);
2982 /* Every bit is either reserved or a feature bit. */
2983 if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
2986 vmx->nested.nested_vmx_ept_caps = data;
2987 vmx->nested.nested_vmx_vpid_caps = data >> 32;
2991 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
2995 switch (msr_index) {
2996 case MSR_IA32_VMX_CR0_FIXED0:
2997 msr = &vmx->nested.nested_vmx_cr0_fixed0;
2999 case MSR_IA32_VMX_CR4_FIXED0:
3000 msr = &vmx->nested.nested_vmx_cr4_fixed0;
3007 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
3008 * must be 1 in the restored value.
3010 if (!is_bitwise_subset(data, *msr, -1ULL))
3018 * Called when userspace is restoring VMX MSRs.
3020 * Returns 0 on success, non-0 otherwise.
3022 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
3024 struct vcpu_vmx *vmx = to_vmx(vcpu);
3026 switch (msr_index) {
3027 case MSR_IA32_VMX_BASIC:
3028 return vmx_restore_vmx_basic(vmx, data);
3029 case MSR_IA32_VMX_PINBASED_CTLS:
3030 case MSR_IA32_VMX_PROCBASED_CTLS:
3031 case MSR_IA32_VMX_EXIT_CTLS:
3032 case MSR_IA32_VMX_ENTRY_CTLS:
3034 * The "non-true" VMX capability MSRs are generated from the
3035 * "true" MSRs, so we do not support restoring them directly.
3037 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
3038 * should restore the "true" MSRs with the must-be-1 bits
3039 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
3040 * DEFAULT SETTINGS".
3043 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3044 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3045 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3046 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3047 case MSR_IA32_VMX_PROCBASED_CTLS2:
3048 return vmx_restore_control_msr(vmx, msr_index, data);
3049 case MSR_IA32_VMX_MISC:
3050 return vmx_restore_vmx_misc(vmx, data);
3051 case MSR_IA32_VMX_CR0_FIXED0:
3052 case MSR_IA32_VMX_CR4_FIXED0:
3053 return vmx_restore_fixed0_msr(vmx, msr_index, data);
3054 case MSR_IA32_VMX_CR0_FIXED1:
3055 case MSR_IA32_VMX_CR4_FIXED1:
3057 * These MSRs are generated based on the vCPU's CPUID, so we
3058 * do not support restoring them directly.
3061 case MSR_IA32_VMX_EPT_VPID_CAP:
3062 return vmx_restore_vmx_ept_vpid_cap(vmx, data);
3063 case MSR_IA32_VMX_VMCS_ENUM:
3064 vmx->nested.nested_vmx_vmcs_enum = data;
3068 * The rest of the VMX capability MSRs do not support restore.
3074 /* Returns 0 on success, non-0 otherwise. */
3075 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
3077 struct vcpu_vmx *vmx = to_vmx(vcpu);
3079 switch (msr_index) {
3080 case MSR_IA32_VMX_BASIC:
3081 *pdata = vmx->nested.nested_vmx_basic;
3083 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3084 case MSR_IA32_VMX_PINBASED_CTLS:
3085 *pdata = vmx_control_msr(
3086 vmx->nested.nested_vmx_pinbased_ctls_low,
3087 vmx->nested.nested_vmx_pinbased_ctls_high);
3088 if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
3089 *pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3091 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3092 case MSR_IA32_VMX_PROCBASED_CTLS:
3093 *pdata = vmx_control_msr(
3094 vmx->nested.nested_vmx_procbased_ctls_low,
3095 vmx->nested.nested_vmx_procbased_ctls_high);
3096 if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
3097 *pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3099 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3100 case MSR_IA32_VMX_EXIT_CTLS:
3101 *pdata = vmx_control_msr(
3102 vmx->nested.nested_vmx_exit_ctls_low,
3103 vmx->nested.nested_vmx_exit_ctls_high);
3104 if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
3105 *pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
3107 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3108 case MSR_IA32_VMX_ENTRY_CTLS:
3109 *pdata = vmx_control_msr(
3110 vmx->nested.nested_vmx_entry_ctls_low,
3111 vmx->nested.nested_vmx_entry_ctls_high);
3112 if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
3113 *pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
3115 case MSR_IA32_VMX_MISC:
3116 *pdata = vmx_control_msr(
3117 vmx->nested.nested_vmx_misc_low,
3118 vmx->nested.nested_vmx_misc_high);
3120 case MSR_IA32_VMX_CR0_FIXED0:
3121 *pdata = vmx->nested.nested_vmx_cr0_fixed0;
3123 case MSR_IA32_VMX_CR0_FIXED1:
3124 *pdata = vmx->nested.nested_vmx_cr0_fixed1;
3126 case MSR_IA32_VMX_CR4_FIXED0:
3127 *pdata = vmx->nested.nested_vmx_cr4_fixed0;
3129 case MSR_IA32_VMX_CR4_FIXED1:
3130 *pdata = vmx->nested.nested_vmx_cr4_fixed1;
3132 case MSR_IA32_VMX_VMCS_ENUM:
3133 *pdata = vmx->nested.nested_vmx_vmcs_enum;
3135 case MSR_IA32_VMX_PROCBASED_CTLS2:
3136 *pdata = vmx_control_msr(
3137 vmx->nested.nested_vmx_secondary_ctls_low,
3138 vmx->nested.nested_vmx_secondary_ctls_high);
3140 case MSR_IA32_VMX_EPT_VPID_CAP:
3141 *pdata = vmx->nested.nested_vmx_ept_caps |
3142 ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
3151 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
3154 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
3156 return !(val & ~valid_bits);
3160 * Reads an msr value (of 'msr_index') into 'pdata'.
3161 * Returns 0 on success, non-0 otherwise.
3162 * Assumes vcpu_load() was already called.
3164 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3166 struct shared_msr_entry *msr;
3168 switch (msr_info->index) {
3169 #ifdef CONFIG_X86_64
3171 msr_info->data = vmcs_readl(GUEST_FS_BASE);
3174 msr_info->data = vmcs_readl(GUEST_GS_BASE);
3176 case MSR_KERNEL_GS_BASE:
3177 vmx_load_host_state(to_vmx(vcpu));
3178 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
3182 return kvm_get_msr_common(vcpu, msr_info);
3184 msr_info->data = guest_read_tsc(vcpu);
3186 case MSR_IA32_SYSENTER_CS:
3187 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
3189 case MSR_IA32_SYSENTER_EIP:
3190 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
3192 case MSR_IA32_SYSENTER_ESP:
3193 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
3195 case MSR_IA32_BNDCFGS:
3196 if (!kvm_mpx_supported())
3198 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
3200 case MSR_IA32_MCG_EXT_CTL:
3201 if (!msr_info->host_initiated &&
3202 !(to_vmx(vcpu)->msr_ia32_feature_control &
3203 FEATURE_CONTROL_LMCE))
3205 msr_info->data = vcpu->arch.mcg_ext_ctl;
3207 case MSR_IA32_FEATURE_CONTROL:
3208 msr_info->data = to_vmx(vcpu)->msr_ia32_feature_control;
3210 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3211 if (!nested_vmx_allowed(vcpu))
3213 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
3215 if (!vmx_xsaves_supported())
3217 msr_info->data = vcpu->arch.ia32_xss;
3220 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3222 /* Otherwise falls through */
3224 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
3226 msr_info->data = msr->data;
3229 return kvm_get_msr_common(vcpu, msr_info);
3235 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3238 * Writes msr value into into the appropriate "register".
3239 * Returns 0 on success, non-0 otherwise.
3240 * Assumes vcpu_load() was already called.
3242 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3244 struct vcpu_vmx *vmx = to_vmx(vcpu);
3245 struct shared_msr_entry *msr;
3247 u32 msr_index = msr_info->index;
3248 u64 data = msr_info->data;
3250 switch (msr_index) {
3252 ret = kvm_set_msr_common(vcpu, msr_info);
3254 #ifdef CONFIG_X86_64
3256 vmx_segment_cache_clear(vmx);
3257 vmcs_writel(GUEST_FS_BASE, data);
3260 vmx_segment_cache_clear(vmx);
3261 vmcs_writel(GUEST_GS_BASE, data);
3263 case MSR_KERNEL_GS_BASE:
3264 vmx_load_host_state(vmx);
3265 vmx->msr_guest_kernel_gs_base = data;
3268 case MSR_IA32_SYSENTER_CS:
3269 vmcs_write32(GUEST_SYSENTER_CS, data);
3271 case MSR_IA32_SYSENTER_EIP:
3272 vmcs_writel(GUEST_SYSENTER_EIP, data);
3274 case MSR_IA32_SYSENTER_ESP:
3275 vmcs_writel(GUEST_SYSENTER_ESP, data);
3277 case MSR_IA32_BNDCFGS:
3278 if (!kvm_mpx_supported())
3280 vmcs_write64(GUEST_BNDCFGS, data);
3283 kvm_write_tsc(vcpu, msr_info);
3285 case MSR_IA32_CR_PAT:
3286 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3287 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3289 vmcs_write64(GUEST_IA32_PAT, data);
3290 vcpu->arch.pat = data;
3293 ret = kvm_set_msr_common(vcpu, msr_info);
3295 case MSR_IA32_TSC_ADJUST:
3296 ret = kvm_set_msr_common(vcpu, msr_info);
3298 case MSR_IA32_MCG_EXT_CTL:
3299 if ((!msr_info->host_initiated &&
3300 !(to_vmx(vcpu)->msr_ia32_feature_control &
3301 FEATURE_CONTROL_LMCE)) ||
3302 (data & ~MCG_EXT_CTL_LMCE_EN))
3304 vcpu->arch.mcg_ext_ctl = data;
3306 case MSR_IA32_FEATURE_CONTROL:
3307 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3308 (to_vmx(vcpu)->msr_ia32_feature_control &
3309 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3311 vmx->msr_ia32_feature_control = data;
3312 if (msr_info->host_initiated && data == 0)
3313 vmx_leave_nested(vcpu);
3315 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3316 if (!msr_info->host_initiated)
3317 return 1; /* they are read-only */
3318 if (!nested_vmx_allowed(vcpu))
3320 return vmx_set_vmx_msr(vcpu, msr_index, data);
3322 if (!vmx_xsaves_supported())
3325 * The only supported bit as of Skylake is bit 8, but
3326 * it is not supported on KVM.
3330 vcpu->arch.ia32_xss = data;
3331 if (vcpu->arch.ia32_xss != host_xss)
3332 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3333 vcpu->arch.ia32_xss, host_xss);
3335 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3338 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3340 /* Check reserved bit, higher 32 bits should be zero */
3341 if ((data >> 32) != 0)
3343 /* Otherwise falls through */
3345 msr = find_msr_entry(vmx, msr_index);
3347 u64 old_msr_data = msr->data;
3349 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3351 ret = kvm_set_shared_msr(msr->index, msr->data,
3355 msr->data = old_msr_data;
3359 ret = kvm_set_msr_common(vcpu, msr_info);
3365 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3367 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3370 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3373 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3375 case VCPU_EXREG_PDPTR:
3377 ept_save_pdptrs(vcpu);
3384 static __init int cpu_has_kvm_support(void)
3386 return cpu_has_vmx();
3389 static __init int vmx_disabled_by_bios(void)
3393 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3394 if (msr & FEATURE_CONTROL_LOCKED) {
3395 /* launched w/ TXT and VMX disabled */
3396 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3399 /* launched w/o TXT and VMX only enabled w/ TXT */
3400 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3401 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3402 && !tboot_enabled()) {
3403 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3404 "activate TXT before enabling KVM\n");
3407 /* launched w/o TXT and VMX disabled */
3408 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3409 && !tboot_enabled())
3416 static void kvm_cpu_vmxon(u64 addr)
3418 intel_pt_handle_vmx(1);
3420 asm volatile (ASM_VMX_VMXON_RAX
3421 : : "a"(&addr), "m"(addr)
3425 static int hardware_enable(void)
3427 int cpu = raw_smp_processor_id();
3428 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3431 if (cr4_read_shadow() & X86_CR4_VMXE)
3434 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3435 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3436 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3439 * Now we can enable the vmclear operation in kdump
3440 * since the loaded_vmcss_on_cpu list on this cpu
3441 * has been initialized.
3443 * Though the cpu is not in VMX operation now, there
3444 * is no problem to enable the vmclear operation
3445 * for the loaded_vmcss_on_cpu list is empty!
3447 crash_enable_local_vmclear(cpu);
3449 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3451 test_bits = FEATURE_CONTROL_LOCKED;
3452 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3453 if (tboot_enabled())
3454 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3456 if ((old & test_bits) != test_bits) {
3457 /* enable and lock */
3458 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3460 cr4_set_bits(X86_CR4_VMXE);
3462 if (vmm_exclusive) {
3463 kvm_cpu_vmxon(phys_addr);
3467 native_store_gdt(this_cpu_ptr(&host_gdt));
3472 static void vmclear_local_loaded_vmcss(void)
3474 int cpu = raw_smp_processor_id();
3475 struct loaded_vmcs *v, *n;
3477 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3478 loaded_vmcss_on_cpu_link)
3479 __loaded_vmcs_clear(v);
3483 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3486 static void kvm_cpu_vmxoff(void)
3488 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3490 intel_pt_handle_vmx(0);
3493 static void hardware_disable(void)
3495 if (vmm_exclusive) {
3496 vmclear_local_loaded_vmcss();
3499 cr4_clear_bits(X86_CR4_VMXE);
3502 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3503 u32 msr, u32 *result)
3505 u32 vmx_msr_low, vmx_msr_high;
3506 u32 ctl = ctl_min | ctl_opt;
3508 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3510 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3511 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3513 /* Ensure minimum (required) set of control bits are supported. */
3521 static __init bool allow_1_setting(u32 msr, u32 ctl)
3523 u32 vmx_msr_low, vmx_msr_high;
3525 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3526 return vmx_msr_high & ctl;
3529 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3531 u32 vmx_msr_low, vmx_msr_high;
3532 u32 min, opt, min2, opt2;
3533 u32 _pin_based_exec_control = 0;
3534 u32 _cpu_based_exec_control = 0;
3535 u32 _cpu_based_2nd_exec_control = 0;
3536 u32 _vmexit_control = 0;
3537 u32 _vmentry_control = 0;
3539 min = CPU_BASED_HLT_EXITING |
3540 #ifdef CONFIG_X86_64
3541 CPU_BASED_CR8_LOAD_EXITING |
3542 CPU_BASED_CR8_STORE_EXITING |
3544 CPU_BASED_CR3_LOAD_EXITING |
3545 CPU_BASED_CR3_STORE_EXITING |
3546 CPU_BASED_USE_IO_BITMAPS |
3547 CPU_BASED_MOV_DR_EXITING |
3548 CPU_BASED_USE_TSC_OFFSETING |
3549 CPU_BASED_MWAIT_EXITING |
3550 CPU_BASED_MONITOR_EXITING |
3551 CPU_BASED_INVLPG_EXITING |
3552 CPU_BASED_RDPMC_EXITING;
3554 opt = CPU_BASED_TPR_SHADOW |
3555 CPU_BASED_USE_MSR_BITMAPS |
3556 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3557 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3558 &_cpu_based_exec_control) < 0)
3560 #ifdef CONFIG_X86_64
3561 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3562 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3563 ~CPU_BASED_CR8_STORE_EXITING;
3565 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3567 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3568 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3569 SECONDARY_EXEC_WBINVD_EXITING |
3570 SECONDARY_EXEC_ENABLE_VPID |
3571 SECONDARY_EXEC_ENABLE_EPT |
3572 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3573 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3574 SECONDARY_EXEC_RDTSCP |
3575 SECONDARY_EXEC_ENABLE_INVPCID |
3576 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3577 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3578 SECONDARY_EXEC_SHADOW_VMCS |
3579 SECONDARY_EXEC_XSAVES |
3580 SECONDARY_EXEC_ENABLE_PML |
3581 SECONDARY_EXEC_TSC_SCALING;
3582 if (adjust_vmx_controls(min2, opt2,
3583 MSR_IA32_VMX_PROCBASED_CTLS2,
3584 &_cpu_based_2nd_exec_control) < 0)
3587 #ifndef CONFIG_X86_64
3588 if (!(_cpu_based_2nd_exec_control &
3589 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3590 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3593 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3594 _cpu_based_2nd_exec_control &= ~(
3595 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3596 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3597 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3599 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3600 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3602 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3603 CPU_BASED_CR3_STORE_EXITING |
3604 CPU_BASED_INVLPG_EXITING);
3605 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3606 vmx_capability.ept, vmx_capability.vpid);
3609 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
3610 #ifdef CONFIG_X86_64
3611 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3613 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3614 VM_EXIT_CLEAR_BNDCFGS;
3615 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3616 &_vmexit_control) < 0)
3619 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
3620 PIN_BASED_VIRTUAL_NMIS;
3621 opt = PIN_BASED_POSTED_INTR | PIN_BASED_VMX_PREEMPTION_TIMER;
3622 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3623 &_pin_based_exec_control) < 0)
3626 if (cpu_has_broken_vmx_preemption_timer())
3627 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3628 if (!(_cpu_based_2nd_exec_control &
3629 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
3630 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3632 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3633 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3634 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3635 &_vmentry_control) < 0)
3638 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3640 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3641 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3644 #ifdef CONFIG_X86_64
3645 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3646 if (vmx_msr_high & (1u<<16))
3650 /* Require Write-Back (WB) memory type for VMCS accesses. */
3651 if (((vmx_msr_high >> 18) & 15) != 6)
3654 vmcs_conf->size = vmx_msr_high & 0x1fff;
3655 vmcs_conf->order = get_order(vmcs_conf->size);
3656 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
3657 vmcs_conf->revision_id = vmx_msr_low;
3659 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3660 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3661 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3662 vmcs_conf->vmexit_ctrl = _vmexit_control;
3663 vmcs_conf->vmentry_ctrl = _vmentry_control;
3665 cpu_has_load_ia32_efer =
3666 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3667 VM_ENTRY_LOAD_IA32_EFER)
3668 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3669 VM_EXIT_LOAD_IA32_EFER);
3671 cpu_has_load_perf_global_ctrl =
3672 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3673 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3674 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3675 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3678 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3679 * but due to errata below it can't be used. Workaround is to use
3680 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3682 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3687 * BC86,AAY89,BD102 (model 44)
3691 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3692 switch (boot_cpu_data.x86_model) {
3698 cpu_has_load_perf_global_ctrl = false;
3699 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3700 "does not work properly. Using workaround\n");
3707 if (boot_cpu_has(X86_FEATURE_XSAVES))
3708 rdmsrl(MSR_IA32_XSS, host_xss);
3713 static struct vmcs *alloc_vmcs_cpu(int cpu)
3715 int node = cpu_to_node(cpu);
3719 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3722 vmcs = page_address(pages);
3723 memset(vmcs, 0, vmcs_config.size);
3724 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3728 static struct vmcs *alloc_vmcs(void)
3730 return alloc_vmcs_cpu(raw_smp_processor_id());
3733 static void free_vmcs(struct vmcs *vmcs)
3735 free_pages((unsigned long)vmcs, vmcs_config.order);
3739 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3741 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3743 if (!loaded_vmcs->vmcs)
3745 loaded_vmcs_clear(loaded_vmcs);
3746 free_vmcs(loaded_vmcs->vmcs);
3747 loaded_vmcs->vmcs = NULL;
3748 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
3751 static void free_kvm_area(void)
3755 for_each_possible_cpu(cpu) {
3756 free_vmcs(per_cpu(vmxarea, cpu));
3757 per_cpu(vmxarea, cpu) = NULL;
3761 static void init_vmcs_shadow_fields(void)
3765 /* No checks for read only fields yet */
3767 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3768 switch (shadow_read_write_fields[i]) {
3770 if (!kvm_mpx_supported())
3778 shadow_read_write_fields[j] =
3779 shadow_read_write_fields[i];
3782 max_shadow_read_write_fields = j;
3784 /* shadowed fields guest access without vmexit */
3785 for (i = 0; i < max_shadow_read_write_fields; i++) {
3786 clear_bit(shadow_read_write_fields[i],
3787 vmx_vmwrite_bitmap);
3788 clear_bit(shadow_read_write_fields[i],
3791 for (i = 0; i < max_shadow_read_only_fields; i++)
3792 clear_bit(shadow_read_only_fields[i],
3796 static __init int alloc_kvm_area(void)
3800 for_each_possible_cpu(cpu) {
3803 vmcs = alloc_vmcs_cpu(cpu);
3809 per_cpu(vmxarea, cpu) = vmcs;
3814 static bool emulation_required(struct kvm_vcpu *vcpu)
3816 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3819 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3820 struct kvm_segment *save)
3822 if (!emulate_invalid_guest_state) {
3824 * CS and SS RPL should be equal during guest entry according
3825 * to VMX spec, but in reality it is not always so. Since vcpu
3826 * is in the middle of the transition from real mode to
3827 * protected mode it is safe to assume that RPL 0 is a good
3830 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3831 save->selector &= ~SEGMENT_RPL_MASK;
3832 save->dpl = save->selector & SEGMENT_RPL_MASK;
3835 vmx_set_segment(vcpu, save, seg);
3838 static void enter_pmode(struct kvm_vcpu *vcpu)
3840 unsigned long flags;
3841 struct vcpu_vmx *vmx = to_vmx(vcpu);
3844 * Update real mode segment cache. It may be not up-to-date if sement
3845 * register was written while vcpu was in a guest mode.
3847 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3848 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3849 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3850 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3851 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3852 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3854 vmx->rmode.vm86_active = 0;
3856 vmx_segment_cache_clear(vmx);
3858 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3860 flags = vmcs_readl(GUEST_RFLAGS);
3861 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3862 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3863 vmcs_writel(GUEST_RFLAGS, flags);
3865 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3866 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3868 update_exception_bitmap(vcpu);
3870 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3871 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3872 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3873 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3874 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3875 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3878 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3880 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3881 struct kvm_segment var = *save;
3884 if (seg == VCPU_SREG_CS)
3887 if (!emulate_invalid_guest_state) {
3888 var.selector = var.base >> 4;
3889 var.base = var.base & 0xffff0;
3899 if (save->base & 0xf)
3900 printk_once(KERN_WARNING "kvm: segment base is not "
3901 "paragraph aligned when entering "
3902 "protected mode (seg=%d)", seg);
3905 vmcs_write16(sf->selector, var.selector);
3906 vmcs_writel(sf->base, var.base);
3907 vmcs_write32(sf->limit, var.limit);
3908 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3911 static void enter_rmode(struct kvm_vcpu *vcpu)
3913 unsigned long flags;
3914 struct vcpu_vmx *vmx = to_vmx(vcpu);
3916 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3917 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3918 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3919 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3920 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3921 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3922 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3924 vmx->rmode.vm86_active = 1;
3927 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3928 * vcpu. Warn the user that an update is overdue.
3930 if (!vcpu->kvm->arch.tss_addr)
3931 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3932 "called before entering vcpu\n");
3934 vmx_segment_cache_clear(vmx);
3936 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3937 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3938 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3940 flags = vmcs_readl(GUEST_RFLAGS);
3941 vmx->rmode.save_rflags = flags;
3943 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3945 vmcs_writel(GUEST_RFLAGS, flags);
3946 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3947 update_exception_bitmap(vcpu);
3949 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3950 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3951 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3952 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3953 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3954 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3956 kvm_mmu_reset_context(vcpu);
3959 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3961 struct vcpu_vmx *vmx = to_vmx(vcpu);
3962 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3968 * Force kernel_gs_base reloading before EFER changes, as control
3969 * of this msr depends on is_long_mode().
3971 vmx_load_host_state(to_vmx(vcpu));
3972 vcpu->arch.efer = efer;
3973 if (efer & EFER_LMA) {
3974 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3977 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3979 msr->data = efer & ~EFER_LME;
3984 #ifdef CONFIG_X86_64
3986 static void enter_lmode(struct kvm_vcpu *vcpu)
3990 vmx_segment_cache_clear(to_vmx(vcpu));
3992 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3993 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
3994 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3996 vmcs_write32(GUEST_TR_AR_BYTES,
3997 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
3998 | VMX_AR_TYPE_BUSY_64_TSS);
4000 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
4003 static void exit_lmode(struct kvm_vcpu *vcpu)
4005 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4006 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
4011 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
4014 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
4016 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
4018 vpid_sync_context(vpid);
4022 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
4024 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
4027 static void vmx_flush_tlb_ept_only(struct kvm_vcpu *vcpu)
4030 vmx_flush_tlb(vcpu);
4033 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
4035 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
4037 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
4038 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
4041 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
4043 if (enable_ept && is_paging(vcpu))
4044 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4045 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
4048 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
4050 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
4052 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
4053 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
4056 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
4058 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4060 if (!test_bit(VCPU_EXREG_PDPTR,
4061 (unsigned long *)&vcpu->arch.regs_dirty))
4064 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4065 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
4066 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
4067 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
4068 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
4072 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
4074 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4076 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4077 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
4078 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
4079 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
4080 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
4083 __set_bit(VCPU_EXREG_PDPTR,
4084 (unsigned long *)&vcpu->arch.regs_avail);
4085 __set_bit(VCPU_EXREG_PDPTR,
4086 (unsigned long *)&vcpu->arch.regs_dirty);
4089 static bool nested_guest_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4091 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4092 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4093 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4095 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
4096 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4097 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4098 fixed0 &= ~(X86_CR0_PE | X86_CR0_PG);
4100 return fixed_bits_valid(val, fixed0, fixed1);
4103 static bool nested_host_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4105 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4106 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4108 return fixed_bits_valid(val, fixed0, fixed1);
4111 static bool nested_cr4_valid(struct kvm_vcpu *vcpu, unsigned long val)
4113 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed0;
4114 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed1;
4116 return fixed_bits_valid(val, fixed0, fixed1);
4119 /* No difference in the restrictions on guest and host CR4 in VMX operation. */
4120 #define nested_guest_cr4_valid nested_cr4_valid
4121 #define nested_host_cr4_valid nested_cr4_valid
4123 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
4125 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
4127 struct kvm_vcpu *vcpu)
4129 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
4130 vmx_decache_cr3(vcpu);
4131 if (!(cr0 & X86_CR0_PG)) {
4132 /* From paging/starting to nonpaging */
4133 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4134 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
4135 (CPU_BASED_CR3_LOAD_EXITING |
4136 CPU_BASED_CR3_STORE_EXITING));
4137 vcpu->arch.cr0 = cr0;
4138 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4139 } else if (!is_paging(vcpu)) {
4140 /* From nonpaging to paging */
4141 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4142 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
4143 ~(CPU_BASED_CR3_LOAD_EXITING |
4144 CPU_BASED_CR3_STORE_EXITING));
4145 vcpu->arch.cr0 = cr0;
4146 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4149 if (!(cr0 & X86_CR0_WP))
4150 *hw_cr0 &= ~X86_CR0_WP;
4153 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
4155 struct vcpu_vmx *vmx = to_vmx(vcpu);
4156 unsigned long hw_cr0;
4158 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
4159 if (enable_unrestricted_guest)
4160 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
4162 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
4164 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
4167 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
4171 #ifdef CONFIG_X86_64
4172 if (vcpu->arch.efer & EFER_LME) {
4173 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
4175 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
4181 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
4183 vmcs_writel(CR0_READ_SHADOW, cr0);
4184 vmcs_writel(GUEST_CR0, hw_cr0);
4185 vcpu->arch.cr0 = cr0;
4187 /* depends on vcpu->arch.cr0 to be set to a new value */
4188 vmx->emulation_required = emulation_required(vcpu);
4191 static u64 construct_eptp(unsigned long root_hpa)
4195 /* TODO write the value reading from MSR */
4196 eptp = VMX_EPT_DEFAULT_MT |
4197 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
4198 if (enable_ept_ad_bits)
4199 eptp |= VMX_EPT_AD_ENABLE_BIT;
4200 eptp |= (root_hpa & PAGE_MASK);
4205 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
4207 unsigned long guest_cr3;
4212 eptp = construct_eptp(cr3);
4213 vmcs_write64(EPT_POINTER, eptp);
4214 if (is_paging(vcpu) || is_guest_mode(vcpu))
4215 guest_cr3 = kvm_read_cr3(vcpu);
4217 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
4218 ept_load_pdptrs(vcpu);
4221 vmx_flush_tlb(vcpu);
4222 vmcs_writel(GUEST_CR3, guest_cr3);
4225 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
4228 * Pass through host's Machine Check Enable value to hw_cr4, which
4229 * is in force while we are in guest mode. Do not let guests control
4230 * this bit, even if host CR4.MCE == 0.
4232 unsigned long hw_cr4 =
4233 (cr4_read_shadow() & X86_CR4_MCE) |
4234 (cr4 & ~X86_CR4_MCE) |
4235 (to_vmx(vcpu)->rmode.vm86_active ?
4236 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
4238 if (cr4 & X86_CR4_VMXE) {
4240 * To use VMXON (and later other VMX instructions), a guest
4241 * must first be able to turn on cr4.VMXE (see handle_vmon()).
4242 * So basically the check on whether to allow nested VMX
4245 if (!nested_vmx_allowed(vcpu))
4249 if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
4252 vcpu->arch.cr4 = cr4;
4254 if (!is_paging(vcpu)) {
4255 hw_cr4 &= ~X86_CR4_PAE;
4256 hw_cr4 |= X86_CR4_PSE;
4257 } else if (!(cr4 & X86_CR4_PAE)) {
4258 hw_cr4 &= ~X86_CR4_PAE;
4262 if (!enable_unrestricted_guest && !is_paging(vcpu))
4264 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4265 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
4266 * to be manually disabled when guest switches to non-paging
4269 * If !enable_unrestricted_guest, the CPU is always running
4270 * with CR0.PG=1 and CR4 needs to be modified.
4271 * If enable_unrestricted_guest, the CPU automatically
4272 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4274 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
4276 vmcs_writel(CR4_READ_SHADOW, cr4);
4277 vmcs_writel(GUEST_CR4, hw_cr4);
4281 static void vmx_get_segment(struct kvm_vcpu *vcpu,
4282 struct kvm_segment *var, int seg)
4284 struct vcpu_vmx *vmx = to_vmx(vcpu);
4287 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4288 *var = vmx->rmode.segs[seg];
4289 if (seg == VCPU_SREG_TR
4290 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
4292 var->base = vmx_read_guest_seg_base(vmx, seg);
4293 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4296 var->base = vmx_read_guest_seg_base(vmx, seg);
4297 var->limit = vmx_read_guest_seg_limit(vmx, seg);
4298 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4299 ar = vmx_read_guest_seg_ar(vmx, seg);
4300 var->unusable = (ar >> 16) & 1;
4301 var->type = ar & 15;
4302 var->s = (ar >> 4) & 1;
4303 var->dpl = (ar >> 5) & 3;
4305 * Some userspaces do not preserve unusable property. Since usable
4306 * segment has to be present according to VMX spec we can use present
4307 * property to amend userspace bug by making unusable segment always
4308 * nonpresent. vmx_segment_access_rights() already marks nonpresent
4309 * segment as unusable.
4311 var->present = !var->unusable;
4312 var->avl = (ar >> 12) & 1;
4313 var->l = (ar >> 13) & 1;
4314 var->db = (ar >> 14) & 1;
4315 var->g = (ar >> 15) & 1;
4318 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
4320 struct kvm_segment s;
4322 if (to_vmx(vcpu)->rmode.vm86_active) {
4323 vmx_get_segment(vcpu, &s, seg);
4326 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
4329 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
4331 struct vcpu_vmx *vmx = to_vmx(vcpu);
4333 if (unlikely(vmx->rmode.vm86_active))
4336 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
4337 return VMX_AR_DPL(ar);
4341 static u32 vmx_segment_access_rights(struct kvm_segment *var)
4345 if (var->unusable || !var->present)
4348 ar = var->type & 15;
4349 ar |= (var->s & 1) << 4;
4350 ar |= (var->dpl & 3) << 5;
4351 ar |= (var->present & 1) << 7;
4352 ar |= (var->avl & 1) << 12;
4353 ar |= (var->l & 1) << 13;
4354 ar |= (var->db & 1) << 14;
4355 ar |= (var->g & 1) << 15;
4361 static void vmx_set_segment(struct kvm_vcpu *vcpu,
4362 struct kvm_segment *var, int seg)
4364 struct vcpu_vmx *vmx = to_vmx(vcpu);
4365 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4367 vmx_segment_cache_clear(vmx);
4369 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4370 vmx->rmode.segs[seg] = *var;
4371 if (seg == VCPU_SREG_TR)
4372 vmcs_write16(sf->selector, var->selector);
4374 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
4378 vmcs_writel(sf->base, var->base);
4379 vmcs_write32(sf->limit, var->limit);
4380 vmcs_write16(sf->selector, var->selector);
4383 * Fix the "Accessed" bit in AR field of segment registers for older
4385 * IA32 arch specifies that at the time of processor reset the
4386 * "Accessed" bit in the AR field of segment registers is 1. And qemu
4387 * is setting it to 0 in the userland code. This causes invalid guest
4388 * state vmexit when "unrestricted guest" mode is turned on.
4389 * Fix for this setup issue in cpu_reset is being pushed in the qemu
4390 * tree. Newer qemu binaries with that qemu fix would not need this
4393 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
4394 var->type |= 0x1; /* Accessed */
4396 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4399 vmx->emulation_required = emulation_required(vcpu);
4402 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4404 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4406 *db = (ar >> 14) & 1;
4407 *l = (ar >> 13) & 1;
4410 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4412 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4413 dt->address = vmcs_readl(GUEST_IDTR_BASE);
4416 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4418 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4419 vmcs_writel(GUEST_IDTR_BASE, dt->address);
4422 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4424 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4425 dt->address = vmcs_readl(GUEST_GDTR_BASE);
4428 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4430 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4431 vmcs_writel(GUEST_GDTR_BASE, dt->address);
4434 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4436 struct kvm_segment var;
4439 vmx_get_segment(vcpu, &var, seg);
4441 if (seg == VCPU_SREG_CS)
4443 ar = vmx_segment_access_rights(&var);
4445 if (var.base != (var.selector << 4))
4447 if (var.limit != 0xffff)
4455 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4457 struct kvm_segment cs;
4458 unsigned int cs_rpl;
4460 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4461 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4465 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4469 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4470 if (cs.dpl > cs_rpl)
4473 if (cs.dpl != cs_rpl)
4479 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4483 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4485 struct kvm_segment ss;
4486 unsigned int ss_rpl;
4488 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4489 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4493 if (ss.type != 3 && ss.type != 7)
4497 if (ss.dpl != ss_rpl) /* DPL != RPL */
4505 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4507 struct kvm_segment var;
4510 vmx_get_segment(vcpu, &var, seg);
4511 rpl = var.selector & SEGMENT_RPL_MASK;
4519 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4520 if (var.dpl < rpl) /* DPL < RPL */
4524 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4530 static bool tr_valid(struct kvm_vcpu *vcpu)
4532 struct kvm_segment tr;
4534 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4538 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4540 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4548 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4550 struct kvm_segment ldtr;
4552 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4556 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4566 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4568 struct kvm_segment cs, ss;
4570 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4571 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4573 return ((cs.selector & SEGMENT_RPL_MASK) ==
4574 (ss.selector & SEGMENT_RPL_MASK));
4578 * Check if guest state is valid. Returns true if valid, false if
4580 * We assume that registers are always usable
4582 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4584 if (enable_unrestricted_guest)
4587 /* real mode guest state checks */
4588 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4589 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4591 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4593 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4595 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4597 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4599 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4602 /* protected mode guest state checks */
4603 if (!cs_ss_rpl_check(vcpu))
4605 if (!code_segment_valid(vcpu))
4607 if (!stack_segment_valid(vcpu))
4609 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4611 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4613 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4615 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4617 if (!tr_valid(vcpu))
4619 if (!ldtr_valid(vcpu))
4623 * - Add checks on RIP
4624 * - Add checks on RFLAGS
4630 static int init_rmode_tss(struct kvm *kvm)
4636 idx = srcu_read_lock(&kvm->srcu);
4637 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4638 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4641 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4642 r = kvm_write_guest_page(kvm, fn++, &data,
4643 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4646 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4649 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4653 r = kvm_write_guest_page(kvm, fn, &data,
4654 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4657 srcu_read_unlock(&kvm->srcu, idx);
4661 static int init_rmode_identity_map(struct kvm *kvm)
4664 kvm_pfn_t identity_map_pfn;
4670 /* Protect kvm->arch.ept_identity_pagetable_done. */
4671 mutex_lock(&kvm->slots_lock);
4673 if (likely(kvm->arch.ept_identity_pagetable_done))
4676 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4678 r = alloc_identity_pagetable(kvm);
4682 idx = srcu_read_lock(&kvm->srcu);
4683 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4686 /* Set up identity-mapping pagetable for EPT in real mode */
4687 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4688 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4689 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4690 r = kvm_write_guest_page(kvm, identity_map_pfn,
4691 &tmp, i * sizeof(tmp), sizeof(tmp));
4695 kvm->arch.ept_identity_pagetable_done = true;
4698 srcu_read_unlock(&kvm->srcu, idx);
4701 mutex_unlock(&kvm->slots_lock);
4705 static void seg_setup(int seg)
4707 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4710 vmcs_write16(sf->selector, 0);
4711 vmcs_writel(sf->base, 0);
4712 vmcs_write32(sf->limit, 0xffff);
4714 if (seg == VCPU_SREG_CS)
4715 ar |= 0x08; /* code segment */
4717 vmcs_write32(sf->ar_bytes, ar);
4720 static int alloc_apic_access_page(struct kvm *kvm)
4725 mutex_lock(&kvm->slots_lock);
4726 if (kvm->arch.apic_access_page_done)
4728 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4729 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4733 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4734 if (is_error_page(page)) {
4740 * Do not pin the page in memory, so that memory hot-unplug
4741 * is able to migrate it.
4744 kvm->arch.apic_access_page_done = true;
4746 mutex_unlock(&kvm->slots_lock);
4750 static int alloc_identity_pagetable(struct kvm *kvm)
4752 /* Called with kvm->slots_lock held. */
4756 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4758 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4759 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4764 static int allocate_vpid(void)
4770 spin_lock(&vmx_vpid_lock);
4771 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4772 if (vpid < VMX_NR_VPIDS)
4773 __set_bit(vpid, vmx_vpid_bitmap);
4776 spin_unlock(&vmx_vpid_lock);
4780 static void free_vpid(int vpid)
4782 if (!enable_vpid || vpid == 0)
4784 spin_lock(&vmx_vpid_lock);
4785 __clear_bit(vpid, vmx_vpid_bitmap);
4786 spin_unlock(&vmx_vpid_lock);
4789 #define MSR_TYPE_R 1
4790 #define MSR_TYPE_W 2
4791 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4794 int f = sizeof(unsigned long);
4796 if (!cpu_has_vmx_msr_bitmap())
4800 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4801 * have the write-low and read-high bitmap offsets the wrong way round.
4802 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4804 if (msr <= 0x1fff) {
4805 if (type & MSR_TYPE_R)
4807 __clear_bit(msr, msr_bitmap + 0x000 / f);
4809 if (type & MSR_TYPE_W)
4811 __clear_bit(msr, msr_bitmap + 0x800 / f);
4813 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4815 if (type & MSR_TYPE_R)
4817 __clear_bit(msr, msr_bitmap + 0x400 / f);
4819 if (type & MSR_TYPE_W)
4821 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4827 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4828 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4830 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4831 unsigned long *msr_bitmap_nested,
4834 int f = sizeof(unsigned long);
4836 if (!cpu_has_vmx_msr_bitmap()) {
4842 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4843 * have the write-low and read-high bitmap offsets the wrong way round.
4844 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4846 if (msr <= 0x1fff) {
4847 if (type & MSR_TYPE_R &&
4848 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4850 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4852 if (type & MSR_TYPE_W &&
4853 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4855 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4857 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4859 if (type & MSR_TYPE_R &&
4860 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4862 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4864 if (type & MSR_TYPE_W &&
4865 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4867 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4872 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4875 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4876 msr, MSR_TYPE_R | MSR_TYPE_W);
4877 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4878 msr, MSR_TYPE_R | MSR_TYPE_W);
4881 static void vmx_disable_intercept_msr_x2apic(u32 msr, int type, bool apicv_active)
4884 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic_apicv,
4886 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic_apicv,
4889 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4891 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4896 static bool vmx_get_enable_apicv(void)
4898 return enable_apicv;
4901 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
4903 struct vcpu_vmx *vmx = to_vmx(vcpu);
4908 if (vmx->nested.pi_desc &&
4909 vmx->nested.pi_pending) {
4910 vmx->nested.pi_pending = false;
4911 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
4914 max_irr = find_last_bit(
4915 (unsigned long *)vmx->nested.pi_desc->pir, 256);
4920 vapic_page = kmap(vmx->nested.virtual_apic_page);
4921 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
4922 kunmap(vmx->nested.virtual_apic_page);
4924 status = vmcs_read16(GUEST_INTR_STATUS);
4925 if ((u8)max_irr > ((u8)status & 0xff)) {
4927 status |= (u8)max_irr;
4928 vmcs_write16(GUEST_INTR_STATUS, status);
4933 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu)
4936 if (vcpu->mode == IN_GUEST_MODE) {
4937 struct vcpu_vmx *vmx = to_vmx(vcpu);
4940 * Currently, we don't support urgent interrupt,
4941 * all interrupts are recognized as non-urgent
4942 * interrupt, so we cannot post interrupts when
4945 * If the vcpu is in guest mode, it means it is
4946 * running instead of being scheduled out and
4947 * waiting in the run queue, and that's the only
4948 * case when 'SN' is set currently, warning if
4951 WARN_ON_ONCE(pi_test_sn(&vmx->pi_desc));
4953 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4954 POSTED_INTR_VECTOR);
4961 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4964 struct vcpu_vmx *vmx = to_vmx(vcpu);
4966 if (is_guest_mode(vcpu) &&
4967 vector == vmx->nested.posted_intr_nv) {
4968 /* the PIR and ON have been set by L1. */
4969 kvm_vcpu_trigger_posted_interrupt(vcpu);
4971 * If a posted intr is not recognized by hardware,
4972 * we will accomplish it in the next vmentry.
4974 vmx->nested.pi_pending = true;
4975 kvm_make_request(KVM_REQ_EVENT, vcpu);
4981 * Send interrupt to vcpu via posted interrupt way.
4982 * 1. If target vcpu is running(non-root mode), send posted interrupt
4983 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4984 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4985 * interrupt from PIR in next vmentry.
4987 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4989 struct vcpu_vmx *vmx = to_vmx(vcpu);
4992 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4996 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4999 /* If a previous notification has sent the IPI, nothing to do. */
5000 if (pi_test_and_set_on(&vmx->pi_desc))
5003 if (!kvm_vcpu_trigger_posted_interrupt(vcpu))
5004 kvm_vcpu_kick(vcpu);
5008 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
5009 * will not change in the lifetime of the guest.
5010 * Note that host-state that does change is set elsewhere. E.g., host-state
5011 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
5013 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
5018 unsigned long cr0, cr4;
5021 WARN_ON(cr0 & X86_CR0_TS);
5022 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
5023 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
5025 /* Save the most likely value for this task's CR4 in the VMCS. */
5026 cr4 = cr4_read_shadow();
5027 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
5028 vmx->host_state.vmcs_host_cr4 = cr4;
5030 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
5031 #ifdef CONFIG_X86_64
5033 * Load null selectors, so we can avoid reloading them in
5034 * __vmx_load_host_state(), in case userspace uses the null selectors
5035 * too (the expected case).
5037 vmcs_write16(HOST_DS_SELECTOR, 0);
5038 vmcs_write16(HOST_ES_SELECTOR, 0);
5040 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5041 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5043 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5044 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
5046 native_store_idt(&dt);
5047 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
5048 vmx->host_idt_base = dt.address;
5050 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
5052 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
5053 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
5054 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
5055 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
5057 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
5058 rdmsr(MSR_IA32_CR_PAT, low32, high32);
5059 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
5063 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
5065 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
5067 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
5068 if (is_guest_mode(&vmx->vcpu))
5069 vmx->vcpu.arch.cr4_guest_owned_bits &=
5070 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
5071 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
5074 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
5076 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
5078 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5079 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
5080 /* Enable the preemption timer dynamically */
5081 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
5082 return pin_based_exec_ctrl;
5085 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5087 struct vcpu_vmx *vmx = to_vmx(vcpu);
5089 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5090 if (cpu_has_secondary_exec_ctrls()) {
5091 if (kvm_vcpu_apicv_active(vcpu))
5092 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
5093 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5094 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5096 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
5097 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5098 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5101 if (cpu_has_vmx_msr_bitmap())
5102 vmx_set_msr_bitmap(vcpu);
5105 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
5107 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
5109 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
5110 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5112 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
5113 exec_control &= ~CPU_BASED_TPR_SHADOW;
5114 #ifdef CONFIG_X86_64
5115 exec_control |= CPU_BASED_CR8_STORE_EXITING |
5116 CPU_BASED_CR8_LOAD_EXITING;
5120 exec_control |= CPU_BASED_CR3_STORE_EXITING |
5121 CPU_BASED_CR3_LOAD_EXITING |
5122 CPU_BASED_INVLPG_EXITING;
5123 return exec_control;
5126 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
5128 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
5129 if (!cpu_need_virtualize_apic_accesses(&vmx->vcpu))
5130 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5132 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
5134 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
5135 enable_unrestricted_guest = 0;
5136 /* Enable INVPCID for non-ept guests may cause performance regression. */
5137 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5139 if (!enable_unrestricted_guest)
5140 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
5142 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
5143 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5144 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
5145 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5146 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
5147 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
5149 We can NOT enable shadow_vmcs here because we don't have yet
5152 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
5155 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
5157 return exec_control;
5160 static void ept_set_mmio_spte_mask(void)
5163 * EPT Misconfigurations can be generated if the value of bits 2:0
5164 * of an EPT paging-structure entry is 110b (write/execute).
5166 kvm_mmu_set_mmio_spte_mask(VMX_EPT_MISCONFIG_WX_VALUE);
5169 #define VMX_XSS_EXIT_BITMAP 0
5171 * Sets up the vmcs for emulated real mode.
5173 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
5175 #ifdef CONFIG_X86_64
5181 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
5182 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
5184 if (enable_shadow_vmcs) {
5185 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5186 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5188 if (cpu_has_vmx_msr_bitmap())
5189 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
5191 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
5194 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5195 vmx->hv_deadline_tsc = -1;
5197 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
5199 if (cpu_has_secondary_exec_ctrls()) {
5200 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5201 vmx_secondary_exec_control(vmx));
5204 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
5205 vmcs_write64(EOI_EXIT_BITMAP0, 0);
5206 vmcs_write64(EOI_EXIT_BITMAP1, 0);
5207 vmcs_write64(EOI_EXIT_BITMAP2, 0);
5208 vmcs_write64(EOI_EXIT_BITMAP3, 0);
5210 vmcs_write16(GUEST_INTR_STATUS, 0);
5212 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
5213 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
5217 vmcs_write32(PLE_GAP, ple_gap);
5218 vmx->ple_window = ple_window;
5219 vmx->ple_window_dirty = true;
5222 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
5223 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
5224 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
5226 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
5227 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
5228 vmx_set_constant_host_state(vmx);
5229 #ifdef CONFIG_X86_64
5230 rdmsrl(MSR_FS_BASE, a);
5231 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
5232 rdmsrl(MSR_GS_BASE, a);
5233 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
5235 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
5236 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
5239 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
5240 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
5241 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
5242 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
5243 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
5245 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
5246 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
5248 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
5249 u32 index = vmx_msr_index[i];
5250 u32 data_low, data_high;
5253 if (rdmsr_safe(index, &data_low, &data_high) < 0)
5255 if (wrmsr_safe(index, data_low, data_high) < 0)
5257 vmx->guest_msrs[j].index = i;
5258 vmx->guest_msrs[j].data = 0;
5259 vmx->guest_msrs[j].mask = -1ull;
5264 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
5266 /* 22.2.1, 20.8.1 */
5267 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
5269 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
5270 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
5272 set_cr4_guest_host_mask(vmx);
5274 if (vmx_xsaves_supported())
5275 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
5278 ASSERT(vmx->pml_pg);
5279 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
5280 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5286 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
5288 struct vcpu_vmx *vmx = to_vmx(vcpu);
5289 struct msr_data apic_base_msr;
5292 vmx->rmode.vm86_active = 0;
5294 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
5295 kvm_set_cr8(vcpu, 0);
5298 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
5299 MSR_IA32_APICBASE_ENABLE;
5300 if (kvm_vcpu_is_reset_bsp(vcpu))
5301 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
5302 apic_base_msr.host_initiated = true;
5303 kvm_set_apic_base(vcpu, &apic_base_msr);
5306 vmx_segment_cache_clear(vmx);
5308 seg_setup(VCPU_SREG_CS);
5309 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
5310 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
5312 seg_setup(VCPU_SREG_DS);
5313 seg_setup(VCPU_SREG_ES);
5314 seg_setup(VCPU_SREG_FS);
5315 seg_setup(VCPU_SREG_GS);
5316 seg_setup(VCPU_SREG_SS);
5318 vmcs_write16(GUEST_TR_SELECTOR, 0);
5319 vmcs_writel(GUEST_TR_BASE, 0);
5320 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
5321 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
5323 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
5324 vmcs_writel(GUEST_LDTR_BASE, 0);
5325 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
5326 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
5329 vmcs_write32(GUEST_SYSENTER_CS, 0);
5330 vmcs_writel(GUEST_SYSENTER_ESP, 0);
5331 vmcs_writel(GUEST_SYSENTER_EIP, 0);
5332 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
5335 vmcs_writel(GUEST_RFLAGS, 0x02);
5336 kvm_rip_write(vcpu, 0xfff0);
5338 vmcs_writel(GUEST_GDTR_BASE, 0);
5339 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
5341 vmcs_writel(GUEST_IDTR_BASE, 0);
5342 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
5344 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
5345 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
5346 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
5350 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
5352 if (cpu_has_vmx_tpr_shadow() && !init_event) {
5353 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
5354 if (cpu_need_tpr_shadow(vcpu))
5355 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
5356 __pa(vcpu->arch.apic->regs));
5357 vmcs_write32(TPR_THRESHOLD, 0);
5360 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5362 if (kvm_vcpu_apicv_active(vcpu))
5363 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
5366 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5368 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5369 vmx->vcpu.arch.cr0 = cr0;
5370 vmx_set_cr0(vcpu, cr0); /* enter rmode */
5371 vmx_set_cr4(vcpu, 0);
5372 vmx_set_efer(vcpu, 0);
5374 update_exception_bitmap(vcpu);
5376 vpid_sync_context(vmx->vpid);
5380 * In nested virtualization, check if L1 asked to exit on external interrupts.
5381 * For most existing hypervisors, this will always return true.
5383 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5385 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5386 PIN_BASED_EXT_INTR_MASK;
5390 * In nested virtualization, check if L1 has set
5391 * VM_EXIT_ACK_INTR_ON_EXIT
5393 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5395 return get_vmcs12(vcpu)->vm_exit_controls &
5396 VM_EXIT_ACK_INTR_ON_EXIT;
5399 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5401 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5402 PIN_BASED_NMI_EXITING;
5405 static void enable_irq_window(struct kvm_vcpu *vcpu)
5407 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5408 CPU_BASED_VIRTUAL_INTR_PENDING);
5411 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5413 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5414 enable_irq_window(vcpu);
5418 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5419 CPU_BASED_VIRTUAL_NMI_PENDING);
5422 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5424 struct vcpu_vmx *vmx = to_vmx(vcpu);
5426 int irq = vcpu->arch.interrupt.nr;
5428 trace_kvm_inj_virq(irq);
5430 ++vcpu->stat.irq_injections;
5431 if (vmx->rmode.vm86_active) {
5433 if (vcpu->arch.interrupt.soft)
5434 inc_eip = vcpu->arch.event_exit_inst_len;
5435 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5436 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5439 intr = irq | INTR_INFO_VALID_MASK;
5440 if (vcpu->arch.interrupt.soft) {
5441 intr |= INTR_TYPE_SOFT_INTR;
5442 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5443 vmx->vcpu.arch.event_exit_inst_len);
5445 intr |= INTR_TYPE_EXT_INTR;
5446 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5449 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5451 struct vcpu_vmx *vmx = to_vmx(vcpu);
5453 if (!is_guest_mode(vcpu)) {
5454 ++vcpu->stat.nmi_injections;
5455 vmx->nmi_known_unmasked = false;
5458 if (vmx->rmode.vm86_active) {
5459 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5460 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5464 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5465 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5468 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5470 if (to_vmx(vcpu)->nmi_known_unmasked)
5472 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5475 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5477 struct vcpu_vmx *vmx = to_vmx(vcpu);
5479 vmx->nmi_known_unmasked = !masked;
5481 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5482 GUEST_INTR_STATE_NMI);
5484 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5485 GUEST_INTR_STATE_NMI);
5488 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5490 if (to_vmx(vcpu)->nested.nested_run_pending)
5493 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5494 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5495 | GUEST_INTR_STATE_NMI));
5498 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5500 return (!to_vmx(vcpu)->nested.nested_run_pending &&
5501 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5502 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5503 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5506 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5510 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5514 kvm->arch.tss_addr = addr;
5515 return init_rmode_tss(kvm);
5518 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5523 * Update instruction length as we may reinject the exception
5524 * from user space while in guest debugging mode.
5526 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5527 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5528 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5532 if (vcpu->guest_debug &
5533 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5550 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5551 int vec, u32 err_code)
5554 * Instruction with address size override prefix opcode 0x67
5555 * Cause the #SS fault with 0 error code in VM86 mode.
5557 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5558 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5559 if (vcpu->arch.halt_request) {
5560 vcpu->arch.halt_request = 0;
5561 return kvm_vcpu_halt(vcpu);
5569 * Forward all other exceptions that are valid in real mode.
5570 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5571 * the required debugging infrastructure rework.
5573 kvm_queue_exception(vcpu, vec);
5578 * Trigger machine check on the host. We assume all the MSRs are already set up
5579 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5580 * We pass a fake environment to the machine check handler because we want
5581 * the guest to be always treated like user space, no matter what context
5582 * it used internally.
5584 static void kvm_machine_check(void)
5586 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5587 struct pt_regs regs = {
5588 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5589 .flags = X86_EFLAGS_IF,
5592 do_machine_check(®s, 0);
5596 static int handle_machine_check(struct kvm_vcpu *vcpu)
5598 /* already handled by vcpu_run */
5602 static int handle_exception(struct kvm_vcpu *vcpu)
5604 struct vcpu_vmx *vmx = to_vmx(vcpu);
5605 struct kvm_run *kvm_run = vcpu->run;
5606 u32 intr_info, ex_no, error_code;
5607 unsigned long cr2, rip, dr6;
5609 enum emulation_result er;
5611 vect_info = vmx->idt_vectoring_info;
5612 intr_info = vmx->exit_intr_info;
5614 if (is_machine_check(intr_info))
5615 return handle_machine_check(vcpu);
5617 if (is_nmi(intr_info))
5618 return 1; /* already handled by vmx_vcpu_run() */
5620 if (is_invalid_opcode(intr_info)) {
5621 if (is_guest_mode(vcpu)) {
5622 kvm_queue_exception(vcpu, UD_VECTOR);
5625 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5626 if (er != EMULATE_DONE)
5627 kvm_queue_exception(vcpu, UD_VECTOR);
5632 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5633 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5636 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5637 * MMIO, it is better to report an internal error.
5638 * See the comments in vmx_handle_exit.
5640 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5641 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5642 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5643 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5644 vcpu->run->internal.ndata = 3;
5645 vcpu->run->internal.data[0] = vect_info;
5646 vcpu->run->internal.data[1] = intr_info;
5647 vcpu->run->internal.data[2] = error_code;
5651 if (is_page_fault(intr_info)) {
5652 /* EPT won't cause page fault directly */
5654 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5655 trace_kvm_page_fault(cr2, error_code);
5657 if (kvm_event_needs_reinjection(vcpu))
5658 kvm_mmu_unprotect_page_virt(vcpu, cr2);
5659 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
5662 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5664 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5665 return handle_rmode_exception(vcpu, ex_no, error_code);
5669 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5672 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5673 if (!(vcpu->guest_debug &
5674 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5675 vcpu->arch.dr6 &= ~15;
5676 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5677 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5678 skip_emulated_instruction(vcpu);
5680 kvm_queue_exception(vcpu, DB_VECTOR);
5683 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5684 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5688 * Update instruction length as we may reinject #BP from
5689 * user space while in guest debugging mode. Reading it for
5690 * #DB as well causes no harm, it is not used in that case.
5692 vmx->vcpu.arch.event_exit_inst_len =
5693 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5694 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5695 rip = kvm_rip_read(vcpu);
5696 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5697 kvm_run->debug.arch.exception = ex_no;
5700 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5701 kvm_run->ex.exception = ex_no;
5702 kvm_run->ex.error_code = error_code;
5708 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5710 ++vcpu->stat.irq_exits;
5714 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5716 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5720 static int handle_io(struct kvm_vcpu *vcpu)
5722 unsigned long exit_qualification;
5723 int size, in, string, ret;
5726 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5727 string = (exit_qualification & 16) != 0;
5728 in = (exit_qualification & 8) != 0;
5730 ++vcpu->stat.io_exits;
5733 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5735 port = exit_qualification >> 16;
5736 size = (exit_qualification & 7) + 1;
5738 ret = kvm_skip_emulated_instruction(vcpu);
5741 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
5742 * KVM_EXIT_DEBUG here.
5744 return kvm_fast_pio_out(vcpu, size, port) && ret;
5748 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5751 * Patch in the VMCALL instruction:
5753 hypercall[0] = 0x0f;
5754 hypercall[1] = 0x01;
5755 hypercall[2] = 0xc1;
5758 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5759 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5761 if (is_guest_mode(vcpu)) {
5762 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5763 unsigned long orig_val = val;
5766 * We get here when L2 changed cr0 in a way that did not change
5767 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5768 * but did change L0 shadowed bits. So we first calculate the
5769 * effective cr0 value that L1 would like to write into the
5770 * hardware. It consists of the L2-owned bits from the new
5771 * value combined with the L1-owned bits from L1's guest_cr0.
5773 val = (val & ~vmcs12->cr0_guest_host_mask) |
5774 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5776 if (!nested_guest_cr0_valid(vcpu, val))
5779 if (kvm_set_cr0(vcpu, val))
5781 vmcs_writel(CR0_READ_SHADOW, orig_val);
5784 if (to_vmx(vcpu)->nested.vmxon &&
5785 !nested_host_cr0_valid(vcpu, val))
5788 return kvm_set_cr0(vcpu, val);
5792 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5794 if (is_guest_mode(vcpu)) {
5795 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5796 unsigned long orig_val = val;
5798 /* analogously to handle_set_cr0 */
5799 val = (val & ~vmcs12->cr4_guest_host_mask) |
5800 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5801 if (kvm_set_cr4(vcpu, val))
5803 vmcs_writel(CR4_READ_SHADOW, orig_val);
5806 return kvm_set_cr4(vcpu, val);
5809 static int handle_cr(struct kvm_vcpu *vcpu)
5811 unsigned long exit_qualification, val;
5817 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5818 cr = exit_qualification & 15;
5819 reg = (exit_qualification >> 8) & 15;
5820 switch ((exit_qualification >> 4) & 3) {
5821 case 0: /* mov to cr */
5822 val = kvm_register_readl(vcpu, reg);
5823 trace_kvm_cr_write(cr, val);
5826 err = handle_set_cr0(vcpu, val);
5827 return kvm_complete_insn_gp(vcpu, err);
5829 err = kvm_set_cr3(vcpu, val);
5830 return kvm_complete_insn_gp(vcpu, err);
5832 err = handle_set_cr4(vcpu, val);
5833 return kvm_complete_insn_gp(vcpu, err);
5835 u8 cr8_prev = kvm_get_cr8(vcpu);
5837 err = kvm_set_cr8(vcpu, cr8);
5838 ret = kvm_complete_insn_gp(vcpu, err);
5839 if (lapic_in_kernel(vcpu))
5841 if (cr8_prev <= cr8)
5844 * TODO: we might be squashing a
5845 * KVM_GUESTDBG_SINGLESTEP-triggered
5846 * KVM_EXIT_DEBUG here.
5848 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5854 WARN_ONCE(1, "Guest should always own CR0.TS");
5855 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5856 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5857 return kvm_skip_emulated_instruction(vcpu);
5858 case 1: /*mov from cr*/
5861 val = kvm_read_cr3(vcpu);
5862 kvm_register_write(vcpu, reg, val);
5863 trace_kvm_cr_read(cr, val);
5864 return kvm_skip_emulated_instruction(vcpu);
5866 val = kvm_get_cr8(vcpu);
5867 kvm_register_write(vcpu, reg, val);
5868 trace_kvm_cr_read(cr, val);
5869 return kvm_skip_emulated_instruction(vcpu);
5873 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5874 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5875 kvm_lmsw(vcpu, val);
5877 return kvm_skip_emulated_instruction(vcpu);
5881 vcpu->run->exit_reason = 0;
5882 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5883 (int)(exit_qualification >> 4) & 3, cr);
5887 static int handle_dr(struct kvm_vcpu *vcpu)
5889 unsigned long exit_qualification;
5892 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5893 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5895 /* First, if DR does not exist, trigger UD */
5896 if (!kvm_require_dr(vcpu, dr))
5899 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5900 if (!kvm_require_cpl(vcpu, 0))
5902 dr7 = vmcs_readl(GUEST_DR7);
5905 * As the vm-exit takes precedence over the debug trap, we
5906 * need to emulate the latter, either for the host or the
5907 * guest debugging itself.
5909 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5910 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5911 vcpu->run->debug.arch.dr7 = dr7;
5912 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5913 vcpu->run->debug.arch.exception = DB_VECTOR;
5914 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5917 vcpu->arch.dr6 &= ~15;
5918 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5919 kvm_queue_exception(vcpu, DB_VECTOR);
5924 if (vcpu->guest_debug == 0) {
5925 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
5926 CPU_BASED_MOV_DR_EXITING);
5929 * No more DR vmexits; force a reload of the debug registers
5930 * and reenter on this instruction. The next vmexit will
5931 * retrieve the full state of the debug registers.
5933 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5937 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5938 if (exit_qualification & TYPE_MOV_FROM_DR) {
5941 if (kvm_get_dr(vcpu, dr, &val))
5943 kvm_register_write(vcpu, reg, val);
5945 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5948 return kvm_skip_emulated_instruction(vcpu);
5951 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5953 return vcpu->arch.dr6;
5956 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5960 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5962 get_debugreg(vcpu->arch.db[0], 0);
5963 get_debugreg(vcpu->arch.db[1], 1);
5964 get_debugreg(vcpu->arch.db[2], 2);
5965 get_debugreg(vcpu->arch.db[3], 3);
5966 get_debugreg(vcpu->arch.dr6, 6);
5967 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5969 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5970 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
5973 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5975 vmcs_writel(GUEST_DR7, val);
5978 static int handle_cpuid(struct kvm_vcpu *vcpu)
5980 return kvm_emulate_cpuid(vcpu);
5983 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5985 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5986 struct msr_data msr_info;
5988 msr_info.index = ecx;
5989 msr_info.host_initiated = false;
5990 if (vmx_get_msr(vcpu, &msr_info)) {
5991 trace_kvm_msr_read_ex(ecx);
5992 kvm_inject_gp(vcpu, 0);
5996 trace_kvm_msr_read(ecx, msr_info.data);
5998 /* FIXME: handling of bits 32:63 of rax, rdx */
5999 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
6000 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
6001 return kvm_skip_emulated_instruction(vcpu);
6004 static int handle_wrmsr(struct kvm_vcpu *vcpu)
6006 struct msr_data msr;
6007 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6008 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
6009 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6013 msr.host_initiated = false;
6014 if (kvm_set_msr(vcpu, &msr) != 0) {
6015 trace_kvm_msr_write_ex(ecx, data);
6016 kvm_inject_gp(vcpu, 0);
6020 trace_kvm_msr_write(ecx, data);
6021 return kvm_skip_emulated_instruction(vcpu);
6024 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
6026 kvm_apic_update_ppr(vcpu);
6030 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
6032 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6033 CPU_BASED_VIRTUAL_INTR_PENDING);
6035 kvm_make_request(KVM_REQ_EVENT, vcpu);
6037 ++vcpu->stat.irq_window_exits;
6041 static int handle_halt(struct kvm_vcpu *vcpu)
6043 return kvm_emulate_halt(vcpu);
6046 static int handle_vmcall(struct kvm_vcpu *vcpu)
6048 return kvm_emulate_hypercall(vcpu);
6051 static int handle_invd(struct kvm_vcpu *vcpu)
6053 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6056 static int handle_invlpg(struct kvm_vcpu *vcpu)
6058 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6060 kvm_mmu_invlpg(vcpu, exit_qualification);
6061 return kvm_skip_emulated_instruction(vcpu);
6064 static int handle_rdpmc(struct kvm_vcpu *vcpu)
6068 err = kvm_rdpmc(vcpu);
6069 return kvm_complete_insn_gp(vcpu, err);
6072 static int handle_wbinvd(struct kvm_vcpu *vcpu)
6074 return kvm_emulate_wbinvd(vcpu);
6077 static int handle_xsetbv(struct kvm_vcpu *vcpu)
6079 u64 new_bv = kvm_read_edx_eax(vcpu);
6080 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
6082 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
6083 return kvm_skip_emulated_instruction(vcpu);
6087 static int handle_xsaves(struct kvm_vcpu *vcpu)
6089 kvm_skip_emulated_instruction(vcpu);
6090 WARN(1, "this should never happen\n");
6094 static int handle_xrstors(struct kvm_vcpu *vcpu)
6096 kvm_skip_emulated_instruction(vcpu);
6097 WARN(1, "this should never happen\n");
6101 static int handle_apic_access(struct kvm_vcpu *vcpu)
6103 if (likely(fasteoi)) {
6104 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6105 int access_type, offset;
6107 access_type = exit_qualification & APIC_ACCESS_TYPE;
6108 offset = exit_qualification & APIC_ACCESS_OFFSET;
6110 * Sane guest uses MOV to write EOI, with written value
6111 * not cared. So make a short-circuit here by avoiding
6112 * heavy instruction emulation.
6114 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
6115 (offset == APIC_EOI)) {
6116 kvm_lapic_set_eoi(vcpu);
6117 return kvm_skip_emulated_instruction(vcpu);
6120 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6123 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
6125 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6126 int vector = exit_qualification & 0xff;
6128 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
6129 kvm_apic_set_eoi_accelerated(vcpu, vector);
6133 static int handle_apic_write(struct kvm_vcpu *vcpu)
6135 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6136 u32 offset = exit_qualification & 0xfff;
6138 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6139 kvm_apic_write_nodecode(vcpu, offset);
6143 static int handle_task_switch(struct kvm_vcpu *vcpu)
6145 struct vcpu_vmx *vmx = to_vmx(vcpu);
6146 unsigned long exit_qualification;
6147 bool has_error_code = false;
6150 int reason, type, idt_v, idt_index;
6152 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
6153 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
6154 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
6156 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6158 reason = (u32)exit_qualification >> 30;
6159 if (reason == TASK_SWITCH_GATE && idt_v) {
6161 case INTR_TYPE_NMI_INTR:
6162 vcpu->arch.nmi_injected = false;
6163 vmx_set_nmi_mask(vcpu, true);
6165 case INTR_TYPE_EXT_INTR:
6166 case INTR_TYPE_SOFT_INTR:
6167 kvm_clear_interrupt_queue(vcpu);
6169 case INTR_TYPE_HARD_EXCEPTION:
6170 if (vmx->idt_vectoring_info &
6171 VECTORING_INFO_DELIVER_CODE_MASK) {
6172 has_error_code = true;
6174 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6177 case INTR_TYPE_SOFT_EXCEPTION:
6178 kvm_clear_exception_queue(vcpu);
6184 tss_selector = exit_qualification;
6186 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
6187 type != INTR_TYPE_EXT_INTR &&
6188 type != INTR_TYPE_NMI_INTR))
6189 skip_emulated_instruction(vcpu);
6191 if (kvm_task_switch(vcpu, tss_selector,
6192 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
6193 has_error_code, error_code) == EMULATE_FAIL) {
6194 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6195 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6196 vcpu->run->internal.ndata = 0;
6201 * TODO: What about debug traps on tss switch?
6202 * Are we supposed to inject them and update dr6?
6208 static int handle_ept_violation(struct kvm_vcpu *vcpu)
6210 unsigned long exit_qualification;
6214 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6216 if (is_guest_mode(vcpu)
6217 && !(exit_qualification & EPT_VIOLATION_GVA_TRANSLATED)) {
6219 * Fix up exit_qualification according to whether guest
6220 * page table accesses are reads or writes.
6222 u64 eptp = nested_ept_get_cr3(vcpu);
6223 if (eptp & VMX_EPT_AD_ENABLE_BIT)
6224 exit_qualification &= ~EPT_VIOLATION_ACC_WRITE;
6228 * EPT violation happened while executing iret from NMI,
6229 * "blocked by NMI" bit has to be set before next VM entry.
6230 * There are errata that may cause this bit to not be set:
6233 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
6234 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
6235 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
6237 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6238 trace_kvm_page_fault(gpa, exit_qualification);
6240 /* Is it a read fault? */
6241 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
6242 ? PFERR_USER_MASK : 0;
6243 /* Is it a write fault? */
6244 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
6245 ? PFERR_WRITE_MASK : 0;
6246 /* Is it a fetch fault? */
6247 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
6248 ? PFERR_FETCH_MASK : 0;
6249 /* ept page table entry is present? */
6250 error_code |= (exit_qualification &
6251 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
6252 EPT_VIOLATION_EXECUTABLE))
6253 ? PFERR_PRESENT_MASK : 0;
6255 vcpu->arch.gpa_available = true;
6256 vcpu->arch.exit_qualification = exit_qualification;
6258 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
6261 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
6266 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6267 if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
6268 trace_kvm_fast_mmio(gpa);
6269 return kvm_skip_emulated_instruction(vcpu);
6272 ret = handle_mmio_page_fault(vcpu, gpa, true);
6273 vcpu->arch.gpa_available = true;
6274 if (likely(ret == RET_MMIO_PF_EMULATE))
6275 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
6278 if (unlikely(ret == RET_MMIO_PF_INVALID))
6279 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
6281 if (unlikely(ret == RET_MMIO_PF_RETRY))
6284 /* It is the real ept misconfig */
6287 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6288 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6293 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6295 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6296 CPU_BASED_VIRTUAL_NMI_PENDING);
6297 ++vcpu->stat.nmi_window_exits;
6298 kvm_make_request(KVM_REQ_EVENT, vcpu);
6303 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6305 struct vcpu_vmx *vmx = to_vmx(vcpu);
6306 enum emulation_result err = EMULATE_DONE;
6309 bool intr_window_requested;
6310 unsigned count = 130;
6312 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6313 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6315 while (vmx->emulation_required && count-- != 0) {
6316 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6317 return handle_interrupt_window(&vmx->vcpu);
6319 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
6322 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
6324 if (err == EMULATE_USER_EXIT) {
6325 ++vcpu->stat.mmio_exits;
6330 if (err != EMULATE_DONE) {
6331 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6332 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6333 vcpu->run->internal.ndata = 0;
6337 if (vcpu->arch.halt_request) {
6338 vcpu->arch.halt_request = 0;
6339 ret = kvm_vcpu_halt(vcpu);
6343 if (signal_pending(current))
6353 static int __grow_ple_window(int val)
6355 if (ple_window_grow < 1)
6358 val = min(val, ple_window_actual_max);
6360 if (ple_window_grow < ple_window)
6361 val *= ple_window_grow;
6363 val += ple_window_grow;
6368 static int __shrink_ple_window(int val, int modifier, int minimum)
6373 if (modifier < ple_window)
6378 return max(val, minimum);
6381 static void grow_ple_window(struct kvm_vcpu *vcpu)
6383 struct vcpu_vmx *vmx = to_vmx(vcpu);
6384 int old = vmx->ple_window;
6386 vmx->ple_window = __grow_ple_window(old);
6388 if (vmx->ple_window != old)
6389 vmx->ple_window_dirty = true;
6391 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6394 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6396 struct vcpu_vmx *vmx = to_vmx(vcpu);
6397 int old = vmx->ple_window;
6399 vmx->ple_window = __shrink_ple_window(old,
6400 ple_window_shrink, ple_window);
6402 if (vmx->ple_window != old)
6403 vmx->ple_window_dirty = true;
6405 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6409 * ple_window_actual_max is computed to be one grow_ple_window() below
6410 * ple_window_max. (See __grow_ple_window for the reason.)
6411 * This prevents overflows, because ple_window_max is int.
6412 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6414 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6416 static void update_ple_window_actual_max(void)
6418 ple_window_actual_max =
6419 __shrink_ple_window(max(ple_window_max, ple_window),
6420 ple_window_grow, INT_MIN);
6424 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6426 static void wakeup_handler(void)
6428 struct kvm_vcpu *vcpu;
6429 int cpu = smp_processor_id();
6431 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6432 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6433 blocked_vcpu_list) {
6434 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6436 if (pi_test_on(pi_desc) == 1)
6437 kvm_vcpu_kick(vcpu);
6439 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6442 void vmx_enable_tdp(void)
6444 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
6445 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
6446 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
6447 0ull, VMX_EPT_EXECUTABLE_MASK,
6448 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
6449 enable_ept_ad_bits ? 0ull : VMX_EPT_RWX_MASK);
6451 ept_set_mmio_spte_mask();
6455 static __init int hardware_setup(void)
6457 int r = -ENOMEM, i, msr;
6459 rdmsrl_safe(MSR_EFER, &host_efer);
6461 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6462 kvm_define_shared_msr(i, vmx_msr_index[i]);
6464 for (i = 0; i < VMX_BITMAP_NR; i++) {
6465 vmx_bitmap[i] = (unsigned long *)__get_free_page(GFP_KERNEL);
6470 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6471 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6472 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6475 * Allow direct access to the PC debug port (it is often used for I/O
6476 * delays, but the vmexits simply slow things down).
6478 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6479 clear_bit(0x80, vmx_io_bitmap_a);
6481 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6483 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6484 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6486 if (setup_vmcs_config(&vmcs_config) < 0) {
6491 if (boot_cpu_has(X86_FEATURE_NX))
6492 kvm_enable_efer_bits(EFER_NX);
6494 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
6495 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
6498 if (!cpu_has_vmx_shadow_vmcs())
6499 enable_shadow_vmcs = 0;
6500 if (enable_shadow_vmcs)
6501 init_vmcs_shadow_fields();
6503 if (!cpu_has_vmx_ept() ||
6504 !cpu_has_vmx_ept_4levels()) {
6506 enable_unrestricted_guest = 0;
6507 enable_ept_ad_bits = 0;
6510 if (!cpu_has_vmx_ept_ad_bits())
6511 enable_ept_ad_bits = 0;
6513 if (!cpu_has_vmx_unrestricted_guest())
6514 enable_unrestricted_guest = 0;
6516 if (!cpu_has_vmx_flexpriority())
6517 flexpriority_enabled = 0;
6520 * set_apic_access_page_addr() is used to reload apic access
6521 * page upon invalidation. No need to do anything if not
6522 * using the APIC_ACCESS_ADDR VMCS field.
6524 if (!flexpriority_enabled)
6525 kvm_x86_ops->set_apic_access_page_addr = NULL;
6527 if (!cpu_has_vmx_tpr_shadow())
6528 kvm_x86_ops->update_cr8_intercept = NULL;
6530 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6531 kvm_disable_largepages();
6533 if (!cpu_has_vmx_ple())
6536 if (!cpu_has_vmx_apicv()) {
6538 kvm_x86_ops->sync_pir_to_irr = NULL;
6541 if (cpu_has_vmx_tsc_scaling()) {
6542 kvm_has_tsc_control = true;
6543 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6544 kvm_tsc_scaling_ratio_frac_bits = 48;
6547 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6548 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6549 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6550 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6551 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6552 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6553 vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
6555 memcpy(vmx_msr_bitmap_legacy_x2apic_apicv,
6556 vmx_msr_bitmap_legacy, PAGE_SIZE);
6557 memcpy(vmx_msr_bitmap_longmode_x2apic_apicv,
6558 vmx_msr_bitmap_longmode, PAGE_SIZE);
6559 memcpy(vmx_msr_bitmap_legacy_x2apic,
6560 vmx_msr_bitmap_legacy, PAGE_SIZE);
6561 memcpy(vmx_msr_bitmap_longmode_x2apic,
6562 vmx_msr_bitmap_longmode, PAGE_SIZE);
6564 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6566 for (msr = 0x800; msr <= 0x8ff; msr++) {
6567 if (msr == 0x839 /* TMCCT */)
6569 vmx_disable_intercept_msr_x2apic(msr, MSR_TYPE_R, true);
6573 * TPR reads and writes can be virtualized even if virtual interrupt
6574 * delivery is not in use.
6576 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_W, true);
6577 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_R | MSR_TYPE_W, false);
6580 vmx_disable_intercept_msr_x2apic(0x80b, MSR_TYPE_W, true);
6582 vmx_disable_intercept_msr_x2apic(0x83f, MSR_TYPE_W, true);
6589 update_ple_window_actual_max();
6592 * Only enable PML when hardware supports PML feature, and both EPT
6593 * and EPT A/D bit features are enabled -- PML depends on them to work.
6595 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6599 kvm_x86_ops->slot_enable_log_dirty = NULL;
6600 kvm_x86_ops->slot_disable_log_dirty = NULL;
6601 kvm_x86_ops->flush_log_dirty = NULL;
6602 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6605 if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
6608 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
6609 cpu_preemption_timer_multi =
6610 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
6612 kvm_x86_ops->set_hv_timer = NULL;
6613 kvm_x86_ops->cancel_hv_timer = NULL;
6616 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6618 kvm_mce_cap_supported |= MCG_LMCE_P;
6620 return alloc_kvm_area();
6623 for (i = 0; i < VMX_BITMAP_NR; i++)
6624 free_page((unsigned long)vmx_bitmap[i]);
6629 static __exit void hardware_unsetup(void)
6633 for (i = 0; i < VMX_BITMAP_NR; i++)
6634 free_page((unsigned long)vmx_bitmap[i]);
6640 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6641 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6643 static int handle_pause(struct kvm_vcpu *vcpu)
6646 grow_ple_window(vcpu);
6648 kvm_vcpu_on_spin(vcpu);
6649 return kvm_skip_emulated_instruction(vcpu);
6652 static int handle_nop(struct kvm_vcpu *vcpu)
6654 return kvm_skip_emulated_instruction(vcpu);
6657 static int handle_mwait(struct kvm_vcpu *vcpu)
6659 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6660 return handle_nop(vcpu);
6663 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6668 static int handle_monitor(struct kvm_vcpu *vcpu)
6670 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6671 return handle_nop(vcpu);
6675 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6676 * We could reuse a single VMCS for all the L2 guests, but we also want the
6677 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6678 * allows keeping them loaded on the processor, and in the future will allow
6679 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6680 * every entry if they never change.
6681 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6682 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6684 * The following functions allocate and free a vmcs02 in this pool.
6687 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6688 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6690 struct vmcs02_list *item;
6691 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6692 if (item->vmptr == vmx->nested.current_vmptr) {
6693 list_move(&item->list, &vmx->nested.vmcs02_pool);
6694 return &item->vmcs02;
6697 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6698 /* Recycle the least recently used VMCS. */
6699 item = list_last_entry(&vmx->nested.vmcs02_pool,
6700 struct vmcs02_list, list);
6701 item->vmptr = vmx->nested.current_vmptr;
6702 list_move(&item->list, &vmx->nested.vmcs02_pool);
6703 return &item->vmcs02;
6706 /* Create a new VMCS */
6707 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6710 item->vmcs02.vmcs = alloc_vmcs();
6711 item->vmcs02.shadow_vmcs = NULL;
6712 if (!item->vmcs02.vmcs) {
6716 loaded_vmcs_init(&item->vmcs02);
6717 item->vmptr = vmx->nested.current_vmptr;
6718 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6719 vmx->nested.vmcs02_num++;
6720 return &item->vmcs02;
6723 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6724 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6726 struct vmcs02_list *item;
6727 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6728 if (item->vmptr == vmptr) {
6729 free_loaded_vmcs(&item->vmcs02);
6730 list_del(&item->list);
6732 vmx->nested.vmcs02_num--;
6738 * Free all VMCSs saved for this vcpu, except the one pointed by
6739 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6740 * must be &vmx->vmcs01.
6742 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6744 struct vmcs02_list *item, *n;
6746 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6747 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6749 * Something will leak if the above WARN triggers. Better than
6752 if (vmx->loaded_vmcs == &item->vmcs02)
6755 free_loaded_vmcs(&item->vmcs02);
6756 list_del(&item->list);
6758 vmx->nested.vmcs02_num--;
6763 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6764 * set the success or error code of an emulated VMX instruction, as specified
6765 * by Vol 2B, VMX Instruction Reference, "Conventions".
6767 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6769 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6770 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6771 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6774 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6776 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6777 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6778 X86_EFLAGS_SF | X86_EFLAGS_OF))
6782 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6783 u32 vm_instruction_error)
6785 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6787 * failValid writes the error number to the current VMCS, which
6788 * can't be done there isn't a current VMCS.
6790 nested_vmx_failInvalid(vcpu);
6793 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6794 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6795 X86_EFLAGS_SF | X86_EFLAGS_OF))
6797 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6799 * We don't need to force a shadow sync because
6800 * VM_INSTRUCTION_ERROR is not shadowed
6804 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6806 /* TODO: not to reset guest simply here. */
6807 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6808 pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
6811 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6813 struct vcpu_vmx *vmx =
6814 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6816 vmx->nested.preemption_timer_expired = true;
6817 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6818 kvm_vcpu_kick(&vmx->vcpu);
6820 return HRTIMER_NORESTART;
6824 * Decode the memory-address operand of a vmx instruction, as recorded on an
6825 * exit caused by such an instruction (run by a guest hypervisor).
6826 * On success, returns 0. When the operand is invalid, returns 1 and throws
6829 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6830 unsigned long exit_qualification,
6831 u32 vmx_instruction_info, bool wr, gva_t *ret)
6835 struct kvm_segment s;
6838 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6839 * Execution", on an exit, vmx_instruction_info holds most of the
6840 * addressing components of the operand. Only the displacement part
6841 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6842 * For how an actual address is calculated from all these components,
6843 * refer to Vol. 1, "Operand Addressing".
6845 int scaling = vmx_instruction_info & 3;
6846 int addr_size = (vmx_instruction_info >> 7) & 7;
6847 bool is_reg = vmx_instruction_info & (1u << 10);
6848 int seg_reg = (vmx_instruction_info >> 15) & 7;
6849 int index_reg = (vmx_instruction_info >> 18) & 0xf;
6850 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6851 int base_reg = (vmx_instruction_info >> 23) & 0xf;
6852 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
6855 kvm_queue_exception(vcpu, UD_VECTOR);
6859 /* Addr = segment_base + offset */
6860 /* offset = base + [index * scale] + displacement */
6861 off = exit_qualification; /* holds the displacement */
6863 off += kvm_register_read(vcpu, base_reg);
6865 off += kvm_register_read(vcpu, index_reg)<<scaling;
6866 vmx_get_segment(vcpu, &s, seg_reg);
6867 *ret = s.base + off;
6869 if (addr_size == 1) /* 32 bit */
6872 /* Checks for #GP/#SS exceptions. */
6874 if (is_long_mode(vcpu)) {
6875 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6876 * non-canonical form. This is the only check on the memory
6877 * destination for long mode!
6879 exn = is_noncanonical_address(*ret);
6880 } else if (is_protmode(vcpu)) {
6881 /* Protected mode: apply checks for segment validity in the
6883 * - segment type check (#GP(0) may be thrown)
6884 * - usability check (#GP(0)/#SS(0))
6885 * - limit check (#GP(0)/#SS(0))
6888 /* #GP(0) if the destination operand is located in a
6889 * read-only data segment or any code segment.
6891 exn = ((s.type & 0xa) == 0 || (s.type & 8));
6893 /* #GP(0) if the source operand is located in an
6894 * execute-only code segment
6896 exn = ((s.type & 0xa) == 8);
6898 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6901 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
6903 exn = (s.unusable != 0);
6904 /* Protected mode: #GP(0)/#SS(0) if the memory
6905 * operand is outside the segment limit.
6907 exn = exn || (off + sizeof(u64) > s.limit);
6910 kvm_queue_exception_e(vcpu,
6911 seg_reg == VCPU_SREG_SS ?
6912 SS_VECTOR : GP_VECTOR,
6921 * This function performs the various checks including
6922 * - if it's 4KB aligned
6923 * - No bits beyond the physical address width are set
6924 * - Returns 0 on success or else 1
6925 * (Intel SDM Section 30.3)
6927 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6932 struct x86_exception e;
6934 struct vcpu_vmx *vmx = to_vmx(vcpu);
6935 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6937 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6938 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
6941 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6942 sizeof(vmptr), &e)) {
6943 kvm_inject_page_fault(vcpu, &e);
6947 switch (exit_reason) {
6948 case EXIT_REASON_VMON:
6951 * The first 4 bytes of VMXON region contain the supported
6952 * VMCS revision identifier
6954 * Note - IA32_VMX_BASIC[48] will never be 1
6955 * for the nested case;
6956 * which replaces physical address width with 32
6959 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6960 nested_vmx_failInvalid(vcpu);
6961 return kvm_skip_emulated_instruction(vcpu);
6964 page = nested_get_page(vcpu, vmptr);
6966 nested_vmx_failInvalid(vcpu);
6967 return kvm_skip_emulated_instruction(vcpu);
6969 if (*(u32 *)kmap(page) != VMCS12_REVISION) {
6971 nested_release_page_clean(page);
6972 nested_vmx_failInvalid(vcpu);
6973 return kvm_skip_emulated_instruction(vcpu);
6976 nested_release_page_clean(page);
6977 vmx->nested.vmxon_ptr = vmptr;
6979 case EXIT_REASON_VMCLEAR:
6980 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6981 nested_vmx_failValid(vcpu,
6982 VMXERR_VMCLEAR_INVALID_ADDRESS);
6983 return kvm_skip_emulated_instruction(vcpu);
6986 if (vmptr == vmx->nested.vmxon_ptr) {
6987 nested_vmx_failValid(vcpu,
6988 VMXERR_VMCLEAR_VMXON_POINTER);
6989 return kvm_skip_emulated_instruction(vcpu);
6992 case EXIT_REASON_VMPTRLD:
6993 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6994 nested_vmx_failValid(vcpu,
6995 VMXERR_VMPTRLD_INVALID_ADDRESS);
6996 return kvm_skip_emulated_instruction(vcpu);
6999 if (vmptr == vmx->nested.vmxon_ptr) {
7000 nested_vmx_failValid(vcpu,
7001 VMXERR_VMPTRLD_VMXON_POINTER);
7002 return kvm_skip_emulated_instruction(vcpu);
7006 return 1; /* shouldn't happen */
7014 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
7016 struct vcpu_vmx *vmx = to_vmx(vcpu);
7017 struct vmcs *shadow_vmcs;
7019 if (cpu_has_vmx_msr_bitmap()) {
7020 vmx->nested.msr_bitmap =
7021 (unsigned long *)__get_free_page(GFP_KERNEL);
7022 if (!vmx->nested.msr_bitmap)
7023 goto out_msr_bitmap;
7026 vmx->nested.cached_vmcs12 = kmalloc(VMCS12_SIZE, GFP_KERNEL);
7027 if (!vmx->nested.cached_vmcs12)
7028 goto out_cached_vmcs12;
7030 if (enable_shadow_vmcs) {
7031 shadow_vmcs = alloc_vmcs();
7033 goto out_shadow_vmcs;
7034 /* mark vmcs as shadow */
7035 shadow_vmcs->revision_id |= (1u << 31);
7036 /* init shadow vmcs */
7037 vmcs_clear(shadow_vmcs);
7038 vmx->vmcs01.shadow_vmcs = shadow_vmcs;
7041 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
7042 vmx->nested.vmcs02_num = 0;
7044 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
7045 HRTIMER_MODE_REL_PINNED);
7046 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
7048 vmx->nested.vmxon = true;
7052 kfree(vmx->nested.cached_vmcs12);
7055 free_page((unsigned long)vmx->nested.msr_bitmap);
7062 * Emulate the VMXON instruction.
7063 * Currently, we just remember that VMX is active, and do not save or even
7064 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
7065 * do not currently need to store anything in that guest-allocated memory
7066 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
7067 * argument is different from the VMXON pointer (which the spec says they do).
7069 static int handle_vmon(struct kvm_vcpu *vcpu)
7072 struct kvm_segment cs;
7073 struct vcpu_vmx *vmx = to_vmx(vcpu);
7074 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
7075 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7077 /* The Intel VMX Instruction Reference lists a bunch of bits that
7078 * are prerequisite to running VMXON, most notably cr4.VMXE must be
7079 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
7080 * Otherwise, we should fail with #UD. We test these now:
7082 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
7083 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
7084 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
7085 kvm_queue_exception(vcpu, UD_VECTOR);
7089 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
7090 if (is_long_mode(vcpu) && !cs.l) {
7091 kvm_queue_exception(vcpu, UD_VECTOR);
7095 if (vmx_get_cpl(vcpu)) {
7096 kvm_inject_gp(vcpu, 0);
7100 if (vmx->nested.vmxon) {
7101 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
7102 return kvm_skip_emulated_instruction(vcpu);
7105 if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
7106 != VMXON_NEEDED_FEATURES) {
7107 kvm_inject_gp(vcpu, 0);
7111 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
7114 ret = enter_vmx_operation(vcpu);
7118 nested_vmx_succeed(vcpu);
7119 return kvm_skip_emulated_instruction(vcpu);
7123 * Intel's VMX Instruction Reference specifies a common set of prerequisites
7124 * for running VMX instructions (except VMXON, whose prerequisites are
7125 * slightly different). It also specifies what exception to inject otherwise.
7127 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
7129 struct kvm_segment cs;
7130 struct vcpu_vmx *vmx = to_vmx(vcpu);
7132 if (!vmx->nested.vmxon) {
7133 kvm_queue_exception(vcpu, UD_VECTOR);
7137 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
7138 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
7139 (is_long_mode(vcpu) && !cs.l)) {
7140 kvm_queue_exception(vcpu, UD_VECTOR);
7144 if (vmx_get_cpl(vcpu)) {
7145 kvm_inject_gp(vcpu, 0);
7152 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
7154 if (vmx->nested.current_vmptr == -1ull)
7157 /* current_vmptr and current_vmcs12 are always set/reset together */
7158 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
7161 if (enable_shadow_vmcs) {
7162 /* copy to memory all shadowed fields in case
7163 they were modified */
7164 copy_shadow_to_vmcs12(vmx);
7165 vmx->nested.sync_shadow_vmcs = false;
7166 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
7167 SECONDARY_EXEC_SHADOW_VMCS);
7168 vmcs_write64(VMCS_LINK_POINTER, -1ull);
7170 vmx->nested.posted_intr_nv = -1;
7172 /* Flush VMCS12 to guest memory */
7173 memcpy(vmx->nested.current_vmcs12, vmx->nested.cached_vmcs12,
7176 kunmap(vmx->nested.current_vmcs12_page);
7177 nested_release_page(vmx->nested.current_vmcs12_page);
7178 vmx->nested.current_vmptr = -1ull;
7179 vmx->nested.current_vmcs12 = NULL;
7183 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7184 * just stops using VMX.
7186 static void free_nested(struct vcpu_vmx *vmx)
7188 if (!vmx->nested.vmxon)
7191 vmx->nested.vmxon = false;
7192 free_vpid(vmx->nested.vpid02);
7193 nested_release_vmcs12(vmx);
7194 if (vmx->nested.msr_bitmap) {
7195 free_page((unsigned long)vmx->nested.msr_bitmap);
7196 vmx->nested.msr_bitmap = NULL;
7198 if (enable_shadow_vmcs) {
7199 vmcs_clear(vmx->vmcs01.shadow_vmcs);
7200 free_vmcs(vmx->vmcs01.shadow_vmcs);
7201 vmx->vmcs01.shadow_vmcs = NULL;
7203 kfree(vmx->nested.cached_vmcs12);
7204 /* Unpin physical memory we referred to in current vmcs02 */
7205 if (vmx->nested.apic_access_page) {
7206 nested_release_page(vmx->nested.apic_access_page);
7207 vmx->nested.apic_access_page = NULL;
7209 if (vmx->nested.virtual_apic_page) {
7210 nested_release_page(vmx->nested.virtual_apic_page);
7211 vmx->nested.virtual_apic_page = NULL;
7213 if (vmx->nested.pi_desc_page) {
7214 kunmap(vmx->nested.pi_desc_page);
7215 nested_release_page(vmx->nested.pi_desc_page);
7216 vmx->nested.pi_desc_page = NULL;
7217 vmx->nested.pi_desc = NULL;
7220 nested_free_all_saved_vmcss(vmx);
7223 /* Emulate the VMXOFF instruction */
7224 static int handle_vmoff(struct kvm_vcpu *vcpu)
7226 if (!nested_vmx_check_permission(vcpu))
7228 free_nested(to_vmx(vcpu));
7229 nested_vmx_succeed(vcpu);
7230 return kvm_skip_emulated_instruction(vcpu);
7233 /* Emulate the VMCLEAR instruction */
7234 static int handle_vmclear(struct kvm_vcpu *vcpu)
7236 struct vcpu_vmx *vmx = to_vmx(vcpu);
7240 if (!nested_vmx_check_permission(vcpu))
7243 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
7246 if (vmptr == vmx->nested.current_vmptr)
7247 nested_release_vmcs12(vmx);
7249 kvm_vcpu_write_guest(vcpu,
7250 vmptr + offsetof(struct vmcs12, launch_state),
7251 &zero, sizeof(zero));
7253 nested_free_vmcs02(vmx, vmptr);
7255 nested_vmx_succeed(vcpu);
7256 return kvm_skip_emulated_instruction(vcpu);
7259 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
7261 /* Emulate the VMLAUNCH instruction */
7262 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
7264 return nested_vmx_run(vcpu, true);
7267 /* Emulate the VMRESUME instruction */
7268 static int handle_vmresume(struct kvm_vcpu *vcpu)
7271 return nested_vmx_run(vcpu, false);
7274 enum vmcs_field_type {
7275 VMCS_FIELD_TYPE_U16 = 0,
7276 VMCS_FIELD_TYPE_U64 = 1,
7277 VMCS_FIELD_TYPE_U32 = 2,
7278 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
7281 static inline int vmcs_field_type(unsigned long field)
7283 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
7284 return VMCS_FIELD_TYPE_U32;
7285 return (field >> 13) & 0x3 ;
7288 static inline int vmcs_field_readonly(unsigned long field)
7290 return (((field >> 10) & 0x3) == 1);
7294 * Read a vmcs12 field. Since these can have varying lengths and we return
7295 * one type, we chose the biggest type (u64) and zero-extend the return value
7296 * to that size. Note that the caller, handle_vmread, might need to use only
7297 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7298 * 64-bit fields are to be returned).
7300 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7301 unsigned long field, u64 *ret)
7303 short offset = vmcs_field_to_offset(field);
7309 p = ((char *)(get_vmcs12(vcpu))) + offset;
7311 switch (vmcs_field_type(field)) {
7312 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7313 *ret = *((natural_width *)p);
7315 case VMCS_FIELD_TYPE_U16:
7318 case VMCS_FIELD_TYPE_U32:
7321 case VMCS_FIELD_TYPE_U64:
7331 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7332 unsigned long field, u64 field_value){
7333 short offset = vmcs_field_to_offset(field);
7334 char *p = ((char *) get_vmcs12(vcpu)) + offset;
7338 switch (vmcs_field_type(field)) {
7339 case VMCS_FIELD_TYPE_U16:
7340 *(u16 *)p = field_value;
7342 case VMCS_FIELD_TYPE_U32:
7343 *(u32 *)p = field_value;
7345 case VMCS_FIELD_TYPE_U64:
7346 *(u64 *)p = field_value;
7348 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7349 *(natural_width *)p = field_value;
7358 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7361 unsigned long field;
7363 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7364 const unsigned long *fields = shadow_read_write_fields;
7365 const int num_fields = max_shadow_read_write_fields;
7369 vmcs_load(shadow_vmcs);
7371 for (i = 0; i < num_fields; i++) {
7373 switch (vmcs_field_type(field)) {
7374 case VMCS_FIELD_TYPE_U16:
7375 field_value = vmcs_read16(field);
7377 case VMCS_FIELD_TYPE_U32:
7378 field_value = vmcs_read32(field);
7380 case VMCS_FIELD_TYPE_U64:
7381 field_value = vmcs_read64(field);
7383 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7384 field_value = vmcs_readl(field);
7390 vmcs12_write_any(&vmx->vcpu, field, field_value);
7393 vmcs_clear(shadow_vmcs);
7394 vmcs_load(vmx->loaded_vmcs->vmcs);
7399 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7401 const unsigned long *fields[] = {
7402 shadow_read_write_fields,
7403 shadow_read_only_fields
7405 const int max_fields[] = {
7406 max_shadow_read_write_fields,
7407 max_shadow_read_only_fields
7410 unsigned long field;
7411 u64 field_value = 0;
7412 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7414 vmcs_load(shadow_vmcs);
7416 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7417 for (i = 0; i < max_fields[q]; i++) {
7418 field = fields[q][i];
7419 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7421 switch (vmcs_field_type(field)) {
7422 case VMCS_FIELD_TYPE_U16:
7423 vmcs_write16(field, (u16)field_value);
7425 case VMCS_FIELD_TYPE_U32:
7426 vmcs_write32(field, (u32)field_value);
7428 case VMCS_FIELD_TYPE_U64:
7429 vmcs_write64(field, (u64)field_value);
7431 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7432 vmcs_writel(field, (long)field_value);
7441 vmcs_clear(shadow_vmcs);
7442 vmcs_load(vmx->loaded_vmcs->vmcs);
7446 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7447 * used before) all generate the same failure when it is missing.
7449 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7451 struct vcpu_vmx *vmx = to_vmx(vcpu);
7452 if (vmx->nested.current_vmptr == -1ull) {
7453 nested_vmx_failInvalid(vcpu);
7459 static int handle_vmread(struct kvm_vcpu *vcpu)
7461 unsigned long field;
7463 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7464 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7467 if (!nested_vmx_check_permission(vcpu))
7470 if (!nested_vmx_check_vmcs12(vcpu))
7471 return kvm_skip_emulated_instruction(vcpu);
7473 /* Decode instruction info and find the field to read */
7474 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7475 /* Read the field, zero-extended to a u64 field_value */
7476 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7477 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7478 return kvm_skip_emulated_instruction(vcpu);
7481 * Now copy part of this value to register or memory, as requested.
7482 * Note that the number of bits actually copied is 32 or 64 depending
7483 * on the guest's mode (32 or 64 bit), not on the given field's length.
7485 if (vmx_instruction_info & (1u << 10)) {
7486 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7489 if (get_vmx_mem_address(vcpu, exit_qualification,
7490 vmx_instruction_info, true, &gva))
7492 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
7493 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7494 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7497 nested_vmx_succeed(vcpu);
7498 return kvm_skip_emulated_instruction(vcpu);
7502 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7504 unsigned long field;
7506 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7507 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7508 /* The value to write might be 32 or 64 bits, depending on L1's long
7509 * mode, and eventually we need to write that into a field of several
7510 * possible lengths. The code below first zero-extends the value to 64
7511 * bit (field_value), and then copies only the appropriate number of
7512 * bits into the vmcs12 field.
7514 u64 field_value = 0;
7515 struct x86_exception e;
7517 if (!nested_vmx_check_permission(vcpu))
7520 if (!nested_vmx_check_vmcs12(vcpu))
7521 return kvm_skip_emulated_instruction(vcpu);
7523 if (vmx_instruction_info & (1u << 10))
7524 field_value = kvm_register_readl(vcpu,
7525 (((vmx_instruction_info) >> 3) & 0xf));
7527 if (get_vmx_mem_address(vcpu, exit_qualification,
7528 vmx_instruction_info, false, &gva))
7530 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7531 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7532 kvm_inject_page_fault(vcpu, &e);
7538 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7539 if (vmcs_field_readonly(field)) {
7540 nested_vmx_failValid(vcpu,
7541 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7542 return kvm_skip_emulated_instruction(vcpu);
7545 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7546 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7547 return kvm_skip_emulated_instruction(vcpu);
7550 nested_vmx_succeed(vcpu);
7551 return kvm_skip_emulated_instruction(vcpu);
7554 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
7556 vmx->nested.current_vmptr = vmptr;
7557 if (enable_shadow_vmcs) {
7558 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7559 SECONDARY_EXEC_SHADOW_VMCS);
7560 vmcs_write64(VMCS_LINK_POINTER,
7561 __pa(vmx->vmcs01.shadow_vmcs));
7562 vmx->nested.sync_shadow_vmcs = true;
7566 /* Emulate the VMPTRLD instruction */
7567 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7569 struct vcpu_vmx *vmx = to_vmx(vcpu);
7572 if (!nested_vmx_check_permission(vcpu))
7575 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
7578 if (vmx->nested.current_vmptr != vmptr) {
7579 struct vmcs12 *new_vmcs12;
7581 page = nested_get_page(vcpu, vmptr);
7583 nested_vmx_failInvalid(vcpu);
7584 return kvm_skip_emulated_instruction(vcpu);
7586 new_vmcs12 = kmap(page);
7587 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7589 nested_release_page_clean(page);
7590 nested_vmx_failValid(vcpu,
7591 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7592 return kvm_skip_emulated_instruction(vcpu);
7595 nested_release_vmcs12(vmx);
7596 vmx->nested.current_vmcs12 = new_vmcs12;
7597 vmx->nested.current_vmcs12_page = page;
7599 * Load VMCS12 from guest memory since it is not already
7602 memcpy(vmx->nested.cached_vmcs12,
7603 vmx->nested.current_vmcs12, VMCS12_SIZE);
7604 set_current_vmptr(vmx, vmptr);
7607 nested_vmx_succeed(vcpu);
7608 return kvm_skip_emulated_instruction(vcpu);
7611 /* Emulate the VMPTRST instruction */
7612 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7614 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7615 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7617 struct x86_exception e;
7619 if (!nested_vmx_check_permission(vcpu))
7622 if (get_vmx_mem_address(vcpu, exit_qualification,
7623 vmx_instruction_info, true, &vmcs_gva))
7625 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
7626 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7627 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7629 kvm_inject_page_fault(vcpu, &e);
7632 nested_vmx_succeed(vcpu);
7633 return kvm_skip_emulated_instruction(vcpu);
7636 /* Emulate the INVEPT instruction */
7637 static int handle_invept(struct kvm_vcpu *vcpu)
7639 struct vcpu_vmx *vmx = to_vmx(vcpu);
7640 u32 vmx_instruction_info, types;
7643 struct x86_exception e;
7648 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7649 SECONDARY_EXEC_ENABLE_EPT) ||
7650 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7651 kvm_queue_exception(vcpu, UD_VECTOR);
7655 if (!nested_vmx_check_permission(vcpu))
7658 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
7659 kvm_queue_exception(vcpu, UD_VECTOR);
7663 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7664 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7666 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7668 if (type >= 32 || !(types & (1 << type))) {
7669 nested_vmx_failValid(vcpu,
7670 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7671 return kvm_skip_emulated_instruction(vcpu);
7674 /* According to the Intel VMX instruction reference, the memory
7675 * operand is read even if it isn't needed (e.g., for type==global)
7677 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7678 vmx_instruction_info, false, &gva))
7680 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7681 sizeof(operand), &e)) {
7682 kvm_inject_page_fault(vcpu, &e);
7687 case VMX_EPT_EXTENT_GLOBAL:
7689 * TODO: track mappings and invalidate
7690 * single context requests appropriately
7692 case VMX_EPT_EXTENT_CONTEXT:
7693 kvm_mmu_sync_roots(vcpu);
7694 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7695 nested_vmx_succeed(vcpu);
7702 return kvm_skip_emulated_instruction(vcpu);
7705 static int handle_invvpid(struct kvm_vcpu *vcpu)
7707 struct vcpu_vmx *vmx = to_vmx(vcpu);
7708 u32 vmx_instruction_info;
7709 unsigned long type, types;
7711 struct x86_exception e;
7714 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7715 SECONDARY_EXEC_ENABLE_VPID) ||
7716 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7717 kvm_queue_exception(vcpu, UD_VECTOR);
7721 if (!nested_vmx_check_permission(vcpu))
7724 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7725 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7727 types = (vmx->nested.nested_vmx_vpid_caps &
7728 VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
7730 if (type >= 32 || !(types & (1 << type))) {
7731 nested_vmx_failValid(vcpu,
7732 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7733 return kvm_skip_emulated_instruction(vcpu);
7736 /* according to the intel vmx instruction reference, the memory
7737 * operand is read even if it isn't needed (e.g., for type==global)
7739 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7740 vmx_instruction_info, false, &gva))
7742 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vpid,
7744 kvm_inject_page_fault(vcpu, &e);
7749 case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
7750 case VMX_VPID_EXTENT_SINGLE_CONTEXT:
7751 case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
7753 nested_vmx_failValid(vcpu,
7754 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7755 return kvm_skip_emulated_instruction(vcpu);
7758 case VMX_VPID_EXTENT_ALL_CONTEXT:
7762 return kvm_skip_emulated_instruction(vcpu);
7765 __vmx_flush_tlb(vcpu, vmx->nested.vpid02);
7766 nested_vmx_succeed(vcpu);
7768 return kvm_skip_emulated_instruction(vcpu);
7771 static int handle_pml_full(struct kvm_vcpu *vcpu)
7773 unsigned long exit_qualification;
7775 trace_kvm_pml_full(vcpu->vcpu_id);
7777 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7780 * PML buffer FULL happened while executing iret from NMI,
7781 * "blocked by NMI" bit has to be set before next VM entry.
7783 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7784 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7785 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7786 GUEST_INTR_STATE_NMI);
7789 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7790 * here.., and there's no userspace involvement needed for PML.
7795 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
7797 kvm_lapic_expired_hv_timer(vcpu);
7802 * The exit handlers return 1 if the exit was handled fully and guest execution
7803 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
7804 * to be done to userspace and return 0.
7806 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7807 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
7808 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
7809 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
7810 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
7811 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
7812 [EXIT_REASON_CR_ACCESS] = handle_cr,
7813 [EXIT_REASON_DR_ACCESS] = handle_dr,
7814 [EXIT_REASON_CPUID] = handle_cpuid,
7815 [EXIT_REASON_MSR_READ] = handle_rdmsr,
7816 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
7817 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
7818 [EXIT_REASON_HLT] = handle_halt,
7819 [EXIT_REASON_INVD] = handle_invd,
7820 [EXIT_REASON_INVLPG] = handle_invlpg,
7821 [EXIT_REASON_RDPMC] = handle_rdpmc,
7822 [EXIT_REASON_VMCALL] = handle_vmcall,
7823 [EXIT_REASON_VMCLEAR] = handle_vmclear,
7824 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
7825 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
7826 [EXIT_REASON_VMPTRST] = handle_vmptrst,
7827 [EXIT_REASON_VMREAD] = handle_vmread,
7828 [EXIT_REASON_VMRESUME] = handle_vmresume,
7829 [EXIT_REASON_VMWRITE] = handle_vmwrite,
7830 [EXIT_REASON_VMOFF] = handle_vmoff,
7831 [EXIT_REASON_VMON] = handle_vmon,
7832 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
7833 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
7834 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
7835 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
7836 [EXIT_REASON_WBINVD] = handle_wbinvd,
7837 [EXIT_REASON_XSETBV] = handle_xsetbv,
7838 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
7839 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
7840 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
7841 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
7842 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
7843 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
7844 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
7845 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
7846 [EXIT_REASON_INVEPT] = handle_invept,
7847 [EXIT_REASON_INVVPID] = handle_invvpid,
7848 [EXIT_REASON_XSAVES] = handle_xsaves,
7849 [EXIT_REASON_XRSTORS] = handle_xrstors,
7850 [EXIT_REASON_PML_FULL] = handle_pml_full,
7851 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
7854 static const int kvm_vmx_max_exit_handlers =
7855 ARRAY_SIZE(kvm_vmx_exit_handlers);
7857 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7858 struct vmcs12 *vmcs12)
7860 unsigned long exit_qualification;
7861 gpa_t bitmap, last_bitmap;
7866 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7867 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7869 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7871 port = exit_qualification >> 16;
7872 size = (exit_qualification & 7) + 1;
7874 last_bitmap = (gpa_t)-1;
7879 bitmap = vmcs12->io_bitmap_a;
7880 else if (port < 0x10000)
7881 bitmap = vmcs12->io_bitmap_b;
7884 bitmap += (port & 0x7fff) / 8;
7886 if (last_bitmap != bitmap)
7887 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
7889 if (b & (1 << (port & 7)))
7894 last_bitmap = bitmap;
7901 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7902 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7903 * disinterest in the current event (read or write a specific MSR) by using an
7904 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7906 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7907 struct vmcs12 *vmcs12, u32 exit_reason)
7909 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7912 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7916 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7917 * for the four combinations of read/write and low/high MSR numbers.
7918 * First we need to figure out which of the four to use:
7920 bitmap = vmcs12->msr_bitmap;
7921 if (exit_reason == EXIT_REASON_MSR_WRITE)
7923 if (msr_index >= 0xc0000000) {
7924 msr_index -= 0xc0000000;
7928 /* Then read the msr_index'th bit from this bitmap: */
7929 if (msr_index < 1024*8) {
7931 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
7933 return 1 & (b >> (msr_index & 7));
7935 return true; /* let L1 handle the wrong parameter */
7939 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7940 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7941 * intercept (via guest_host_mask etc.) the current event.
7943 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7944 struct vmcs12 *vmcs12)
7946 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7947 int cr = exit_qualification & 15;
7948 int reg = (exit_qualification >> 8) & 15;
7949 unsigned long val = kvm_register_readl(vcpu, reg);
7951 switch ((exit_qualification >> 4) & 3) {
7952 case 0: /* mov to cr */
7955 if (vmcs12->cr0_guest_host_mask &
7956 (val ^ vmcs12->cr0_read_shadow))
7960 if ((vmcs12->cr3_target_count >= 1 &&
7961 vmcs12->cr3_target_value0 == val) ||
7962 (vmcs12->cr3_target_count >= 2 &&
7963 vmcs12->cr3_target_value1 == val) ||
7964 (vmcs12->cr3_target_count >= 3 &&
7965 vmcs12->cr3_target_value2 == val) ||
7966 (vmcs12->cr3_target_count >= 4 &&
7967 vmcs12->cr3_target_value3 == val))
7969 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
7973 if (vmcs12->cr4_guest_host_mask &
7974 (vmcs12->cr4_read_shadow ^ val))
7978 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
7984 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
7985 (vmcs12->cr0_read_shadow & X86_CR0_TS))
7988 case 1: /* mov from cr */
7991 if (vmcs12->cpu_based_vm_exec_control &
7992 CPU_BASED_CR3_STORE_EXITING)
7996 if (vmcs12->cpu_based_vm_exec_control &
7997 CPU_BASED_CR8_STORE_EXITING)
8004 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
8005 * cr0. Other attempted changes are ignored, with no exit.
8007 if (vmcs12->cr0_guest_host_mask & 0xe &
8008 (val ^ vmcs12->cr0_read_shadow))
8010 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
8011 !(vmcs12->cr0_read_shadow & 0x1) &&
8020 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
8021 * should handle it ourselves in L0 (and then continue L2). Only call this
8022 * when in is_guest_mode (L2).
8024 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
8026 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8027 struct vcpu_vmx *vmx = to_vmx(vcpu);
8028 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8029 u32 exit_reason = vmx->exit_reason;
8031 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
8032 vmcs_readl(EXIT_QUALIFICATION),
8033 vmx->idt_vectoring_info,
8035 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8038 if (vmx->nested.nested_run_pending)
8041 if (unlikely(vmx->fail)) {
8042 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
8043 vmcs_read32(VM_INSTRUCTION_ERROR));
8047 switch (exit_reason) {
8048 case EXIT_REASON_EXCEPTION_NMI:
8049 if (is_nmi(intr_info))
8051 else if (is_page_fault(intr_info))
8053 else if (is_no_device(intr_info) &&
8054 !(vmcs12->guest_cr0 & X86_CR0_TS))
8056 else if (is_debug(intr_info) &&
8058 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
8060 else if (is_breakpoint(intr_info) &&
8061 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
8063 return vmcs12->exception_bitmap &
8064 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
8065 case EXIT_REASON_EXTERNAL_INTERRUPT:
8067 case EXIT_REASON_TRIPLE_FAULT:
8069 case EXIT_REASON_PENDING_INTERRUPT:
8070 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
8071 case EXIT_REASON_NMI_WINDOW:
8072 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
8073 case EXIT_REASON_TASK_SWITCH:
8075 case EXIT_REASON_CPUID:
8077 case EXIT_REASON_HLT:
8078 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
8079 case EXIT_REASON_INVD:
8081 case EXIT_REASON_INVLPG:
8082 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8083 case EXIT_REASON_RDPMC:
8084 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
8085 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
8086 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
8087 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
8088 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
8089 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
8090 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
8091 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
8092 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
8094 * VMX instructions trap unconditionally. This allows L1 to
8095 * emulate them for its L2 guest, i.e., allows 3-level nesting!
8098 case EXIT_REASON_CR_ACCESS:
8099 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
8100 case EXIT_REASON_DR_ACCESS:
8101 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
8102 case EXIT_REASON_IO_INSTRUCTION:
8103 return nested_vmx_exit_handled_io(vcpu, vmcs12);
8104 case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
8105 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
8106 case EXIT_REASON_MSR_READ:
8107 case EXIT_REASON_MSR_WRITE:
8108 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
8109 case EXIT_REASON_INVALID_STATE:
8111 case EXIT_REASON_MWAIT_INSTRUCTION:
8112 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
8113 case EXIT_REASON_MONITOR_TRAP_FLAG:
8114 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
8115 case EXIT_REASON_MONITOR_INSTRUCTION:
8116 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
8117 case EXIT_REASON_PAUSE_INSTRUCTION:
8118 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
8119 nested_cpu_has2(vmcs12,
8120 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
8121 case EXIT_REASON_MCE_DURING_VMENTRY:
8123 case EXIT_REASON_TPR_BELOW_THRESHOLD:
8124 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
8125 case EXIT_REASON_APIC_ACCESS:
8126 return nested_cpu_has2(vmcs12,
8127 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
8128 case EXIT_REASON_APIC_WRITE:
8129 case EXIT_REASON_EOI_INDUCED:
8130 /* apic_write and eoi_induced should exit unconditionally. */
8132 case EXIT_REASON_EPT_VIOLATION:
8134 * L0 always deals with the EPT violation. If nested EPT is
8135 * used, and the nested mmu code discovers that the address is
8136 * missing in the guest EPT table (EPT12), the EPT violation
8137 * will be injected with nested_ept_inject_page_fault()
8140 case EXIT_REASON_EPT_MISCONFIG:
8142 * L2 never uses directly L1's EPT, but rather L0's own EPT
8143 * table (shadow on EPT) or a merged EPT table that L0 built
8144 * (EPT on EPT). So any problems with the structure of the
8145 * table is L0's fault.
8148 case EXIT_REASON_WBINVD:
8149 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
8150 case EXIT_REASON_XSETBV:
8152 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
8154 * This should never happen, since it is not possible to
8155 * set XSS to a non-zero value---neither in L1 nor in L2.
8156 * If if it were, XSS would have to be checked against
8157 * the XSS exit bitmap in vmcs12.
8159 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
8160 case EXIT_REASON_PREEMPTION_TIMER:
8167 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
8169 *info1 = vmcs_readl(EXIT_QUALIFICATION);
8170 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
8173 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
8176 __free_page(vmx->pml_pg);
8181 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
8183 struct vcpu_vmx *vmx = to_vmx(vcpu);
8187 pml_idx = vmcs_read16(GUEST_PML_INDEX);
8189 /* Do nothing if PML buffer is empty */
8190 if (pml_idx == (PML_ENTITY_NUM - 1))
8193 /* PML index always points to next available PML buffer entity */
8194 if (pml_idx >= PML_ENTITY_NUM)
8199 pml_buf = page_address(vmx->pml_pg);
8200 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
8203 gpa = pml_buf[pml_idx];
8204 WARN_ON(gpa & (PAGE_SIZE - 1));
8205 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
8208 /* reset PML index */
8209 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8213 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8214 * Called before reporting dirty_bitmap to userspace.
8216 static void kvm_flush_pml_buffers(struct kvm *kvm)
8219 struct kvm_vcpu *vcpu;
8221 * We only need to kick vcpu out of guest mode here, as PML buffer
8222 * is flushed at beginning of all VMEXITs, and it's obvious that only
8223 * vcpus running in guest are possible to have unflushed GPAs in PML
8226 kvm_for_each_vcpu(i, vcpu, kvm)
8227 kvm_vcpu_kick(vcpu);
8230 static void vmx_dump_sel(char *name, uint32_t sel)
8232 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8233 name, vmcs_read16(sel),
8234 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
8235 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
8236 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
8239 static void vmx_dump_dtsel(char *name, uint32_t limit)
8241 pr_err("%s limit=0x%08x, base=0x%016lx\n",
8242 name, vmcs_read32(limit),
8243 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
8246 static void dump_vmcs(void)
8248 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
8249 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
8250 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
8251 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
8252 u32 secondary_exec_control = 0;
8253 unsigned long cr4 = vmcs_readl(GUEST_CR4);
8254 u64 efer = vmcs_read64(GUEST_IA32_EFER);
8257 if (cpu_has_secondary_exec_ctrls())
8258 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8260 pr_err("*** Guest State ***\n");
8261 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8262 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
8263 vmcs_readl(CR0_GUEST_HOST_MASK));
8264 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8265 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
8266 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
8267 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
8268 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
8270 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
8271 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
8272 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
8273 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
8275 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
8276 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
8277 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
8278 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
8279 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8280 vmcs_readl(GUEST_SYSENTER_ESP),
8281 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8282 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
8283 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
8284 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
8285 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
8286 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
8287 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
8288 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8289 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8290 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8291 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
8292 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8293 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8294 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8295 efer, vmcs_read64(GUEST_IA32_PAT));
8296 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
8297 vmcs_read64(GUEST_IA32_DEBUGCTL),
8298 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8299 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8300 pr_err("PerfGlobCtl = 0x%016llx\n",
8301 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8302 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8303 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8304 pr_err("Interruptibility = %08x ActivityState = %08x\n",
8305 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8306 vmcs_read32(GUEST_ACTIVITY_STATE));
8307 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8308 pr_err("InterruptStatus = %04x\n",
8309 vmcs_read16(GUEST_INTR_STATUS));
8311 pr_err("*** Host State ***\n");
8312 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
8313 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8314 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8315 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8316 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8317 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8318 vmcs_read16(HOST_TR_SELECTOR));
8319 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8320 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8321 vmcs_readl(HOST_TR_BASE));
8322 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8323 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8324 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8325 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8326 vmcs_readl(HOST_CR4));
8327 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8328 vmcs_readl(HOST_IA32_SYSENTER_ESP),
8329 vmcs_read32(HOST_IA32_SYSENTER_CS),
8330 vmcs_readl(HOST_IA32_SYSENTER_EIP));
8331 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8332 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8333 vmcs_read64(HOST_IA32_EFER),
8334 vmcs_read64(HOST_IA32_PAT));
8335 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8336 pr_err("PerfGlobCtl = 0x%016llx\n",
8337 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8339 pr_err("*** Control State ***\n");
8340 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8341 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8342 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8343 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8344 vmcs_read32(EXCEPTION_BITMAP),
8345 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8346 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8347 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8348 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8349 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8350 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8351 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8352 vmcs_read32(VM_EXIT_INTR_INFO),
8353 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8354 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8355 pr_err(" reason=%08x qualification=%016lx\n",
8356 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8357 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8358 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8359 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8360 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8361 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8362 pr_err("TSC Multiplier = 0x%016llx\n",
8363 vmcs_read64(TSC_MULTIPLIER));
8364 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8365 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8366 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8367 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8368 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8369 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8370 n = vmcs_read32(CR3_TARGET_COUNT);
8371 for (i = 0; i + 1 < n; i += 4)
8372 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8373 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8374 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8376 pr_err("CR3 target%u=%016lx\n",
8377 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8378 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8379 pr_err("PLE Gap=%08x Window=%08x\n",
8380 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8381 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8382 pr_err("Virtual processor ID = 0x%04x\n",
8383 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8387 * The guest has exited. See if we can fix it or if we need userspace
8390 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8392 struct vcpu_vmx *vmx = to_vmx(vcpu);
8393 u32 exit_reason = vmx->exit_reason;
8394 u32 vectoring_info = vmx->idt_vectoring_info;
8396 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8397 vcpu->arch.gpa_available = false;
8400 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8401 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8402 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8403 * mode as if vcpus is in root mode, the PML buffer must has been
8407 vmx_flush_pml_buffer(vcpu);
8409 /* If guest state is invalid, start emulating */
8410 if (vmx->emulation_required)
8411 return handle_invalid_guest_state(vcpu);
8413 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
8414 nested_vmx_vmexit(vcpu, exit_reason,
8415 vmcs_read32(VM_EXIT_INTR_INFO),
8416 vmcs_readl(EXIT_QUALIFICATION));
8420 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8422 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8423 vcpu->run->fail_entry.hardware_entry_failure_reason
8428 if (unlikely(vmx->fail)) {
8429 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8430 vcpu->run->fail_entry.hardware_entry_failure_reason
8431 = vmcs_read32(VM_INSTRUCTION_ERROR);
8437 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8438 * delivery event since it indicates guest is accessing MMIO.
8439 * The vm-exit can be triggered again after return to guest that
8440 * will cause infinite loop.
8442 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8443 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8444 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8445 exit_reason != EXIT_REASON_PML_FULL &&
8446 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8447 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8448 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8449 vcpu->run->internal.ndata = 2;
8450 vcpu->run->internal.data[0] = vectoring_info;
8451 vcpu->run->internal.data[1] = exit_reason;
8455 if (exit_reason < kvm_vmx_max_exit_handlers
8456 && kvm_vmx_exit_handlers[exit_reason])
8457 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8459 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
8461 kvm_queue_exception(vcpu, UD_VECTOR);
8466 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8468 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8470 if (is_guest_mode(vcpu) &&
8471 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8474 if (irr == -1 || tpr < irr) {
8475 vmcs_write32(TPR_THRESHOLD, 0);
8479 vmcs_write32(TPR_THRESHOLD, irr);
8482 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8484 u32 sec_exec_control;
8486 /* Postpone execution until vmcs01 is the current VMCS. */
8487 if (is_guest_mode(vcpu)) {
8488 to_vmx(vcpu)->nested.change_vmcs01_virtual_x2apic_mode = true;
8492 if (!cpu_has_vmx_virtualize_x2apic_mode())
8495 if (!cpu_need_tpr_shadow(vcpu))
8498 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8501 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8502 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8504 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8505 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8506 vmx_flush_tlb_ept_only(vcpu);
8508 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8510 vmx_set_msr_bitmap(vcpu);
8513 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8515 struct vcpu_vmx *vmx = to_vmx(vcpu);
8518 * Currently we do not handle the nested case where L2 has an
8519 * APIC access page of its own; that page is still pinned.
8520 * Hence, we skip the case where the VCPU is in guest mode _and_
8521 * L1 prepared an APIC access page for L2.
8523 * For the case where L1 and L2 share the same APIC access page
8524 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8525 * in the vmcs12), this function will only update either the vmcs01
8526 * or the vmcs02. If the former, the vmcs02 will be updated by
8527 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8528 * the next L2->L1 exit.
8530 if (!is_guest_mode(vcpu) ||
8531 !nested_cpu_has2(get_vmcs12(&vmx->vcpu),
8532 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
8533 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8534 vmx_flush_tlb_ept_only(vcpu);
8538 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
8546 status = vmcs_read16(GUEST_INTR_STATUS);
8548 if (max_isr != old) {
8550 status |= max_isr << 8;
8551 vmcs_write16(GUEST_INTR_STATUS, status);
8555 static void vmx_set_rvi(int vector)
8563 status = vmcs_read16(GUEST_INTR_STATUS);
8564 old = (u8)status & 0xff;
8565 if ((u8)vector != old) {
8567 status |= (u8)vector;
8568 vmcs_write16(GUEST_INTR_STATUS, status);
8572 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8574 if (!is_guest_mode(vcpu)) {
8575 vmx_set_rvi(max_irr);
8583 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8586 if (nested_exit_on_intr(vcpu))
8590 * Else, fall back to pre-APICv interrupt injection since L2
8591 * is run without virtual interrupt delivery.
8593 if (!kvm_event_needs_reinjection(vcpu) &&
8594 vmx_interrupt_allowed(vcpu)) {
8595 kvm_queue_interrupt(vcpu, max_irr, false);
8596 vmx_inject_irq(vcpu);
8600 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
8602 struct vcpu_vmx *vmx = to_vmx(vcpu);
8605 WARN_ON(!vcpu->arch.apicv_active);
8606 if (pi_test_on(&vmx->pi_desc)) {
8607 pi_clear_on(&vmx->pi_desc);
8609 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
8610 * But on x86 this is just a compiler barrier anyway.
8612 smp_mb__after_atomic();
8613 max_irr = kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
8615 max_irr = kvm_lapic_find_highest_irr(vcpu);
8617 vmx_hwapic_irr_update(vcpu, max_irr);
8621 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
8623 if (!kvm_vcpu_apicv_active(vcpu))
8626 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8627 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8628 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8629 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8632 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
8634 struct vcpu_vmx *vmx = to_vmx(vcpu);
8636 pi_clear_on(&vmx->pi_desc);
8637 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
8640 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8644 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8645 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
8648 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8649 exit_intr_info = vmx->exit_intr_info;
8651 /* Handle machine checks before interrupts are enabled */
8652 if (is_machine_check(exit_intr_info))
8653 kvm_machine_check();
8655 /* We need to handle NMIs before interrupts are enabled */
8656 if (is_nmi(exit_intr_info)) {
8657 kvm_before_handle_nmi(&vmx->vcpu);
8659 kvm_after_handle_nmi(&vmx->vcpu);
8663 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8665 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8666 register void *__sp asm(_ASM_SP);
8668 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8669 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8670 unsigned int vector;
8671 unsigned long entry;
8673 struct vcpu_vmx *vmx = to_vmx(vcpu);
8674 #ifdef CONFIG_X86_64
8678 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8679 desc = (gate_desc *)vmx->host_idt_base + vector;
8680 entry = gate_offset(*desc);
8682 #ifdef CONFIG_X86_64
8683 "mov %%" _ASM_SP ", %[sp]\n\t"
8684 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8689 __ASM_SIZE(push) " $%c[cs]\n\t"
8690 "call *%[entry]\n\t"
8692 #ifdef CONFIG_X86_64
8698 [ss]"i"(__KERNEL_DS),
8699 [cs]"i"(__KERNEL_CS)
8704 static bool vmx_has_high_real_mode_segbase(void)
8706 return enable_unrestricted_guest || emulate_invalid_guest_state;
8709 static bool vmx_mpx_supported(void)
8711 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8712 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8715 static bool vmx_xsaves_supported(void)
8717 return vmcs_config.cpu_based_2nd_exec_ctrl &
8718 SECONDARY_EXEC_XSAVES;
8721 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8726 bool idtv_info_valid;
8728 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8730 if (vmx->nmi_known_unmasked)
8733 * Can't use vmx->exit_intr_info since we're not sure what
8734 * the exit reason is.
8736 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8737 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8738 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8740 * SDM 3: 27.7.1.2 (September 2008)
8741 * Re-set bit "block by NMI" before VM entry if vmexit caused by
8742 * a guest IRET fault.
8743 * SDM 3: 23.2.2 (September 2008)
8744 * Bit 12 is undefined in any of the following cases:
8745 * If the VM exit sets the valid bit in the IDT-vectoring
8746 * information field.
8747 * If the VM exit is due to a double fault.
8749 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8750 vector != DF_VECTOR && !idtv_info_valid)
8751 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8752 GUEST_INTR_STATE_NMI);
8754 vmx->nmi_known_unmasked =
8755 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8756 & GUEST_INTR_STATE_NMI);
8759 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8760 u32 idt_vectoring_info,
8761 int instr_len_field,
8762 int error_code_field)
8766 bool idtv_info_valid;
8768 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8770 vcpu->arch.nmi_injected = false;
8771 kvm_clear_exception_queue(vcpu);
8772 kvm_clear_interrupt_queue(vcpu);
8774 if (!idtv_info_valid)
8777 kvm_make_request(KVM_REQ_EVENT, vcpu);
8779 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8780 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8783 case INTR_TYPE_NMI_INTR:
8784 vcpu->arch.nmi_injected = true;
8786 * SDM 3: 27.7.1.2 (September 2008)
8787 * Clear bit "block by NMI" before VM entry if a NMI
8790 vmx_set_nmi_mask(vcpu, false);
8792 case INTR_TYPE_SOFT_EXCEPTION:
8793 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8795 case INTR_TYPE_HARD_EXCEPTION:
8796 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8797 u32 err = vmcs_read32(error_code_field);
8798 kvm_requeue_exception_e(vcpu, vector, err);
8800 kvm_requeue_exception(vcpu, vector);
8802 case INTR_TYPE_SOFT_INTR:
8803 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8805 case INTR_TYPE_EXT_INTR:
8806 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8813 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8815 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8816 VM_EXIT_INSTRUCTION_LEN,
8817 IDT_VECTORING_ERROR_CODE);
8820 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8822 __vmx_complete_interrupts(vcpu,
8823 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8824 VM_ENTRY_INSTRUCTION_LEN,
8825 VM_ENTRY_EXCEPTION_ERROR_CODE);
8827 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8830 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8833 struct perf_guest_switch_msr *msrs;
8835 msrs = perf_guest_get_msrs(&nr_msrs);
8840 for (i = 0; i < nr_msrs; i++)
8841 if (msrs[i].host == msrs[i].guest)
8842 clear_atomic_switch_msr(vmx, msrs[i].msr);
8844 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8848 static void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
8850 struct vcpu_vmx *vmx = to_vmx(vcpu);
8854 if (vmx->hv_deadline_tsc == -1)
8858 if (vmx->hv_deadline_tsc > tscl)
8859 /* sure to be 32 bit only because checked on set_hv_timer */
8860 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
8861 cpu_preemption_timer_multi);
8865 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
8868 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8870 struct vcpu_vmx *vmx = to_vmx(vcpu);
8871 unsigned long debugctlmsr, cr4;
8873 /* Don't enter VMX if guest state is invalid, let the exit handler
8874 start emulation until we arrive back to a valid state */
8875 if (vmx->emulation_required)
8878 if (vmx->ple_window_dirty) {
8879 vmx->ple_window_dirty = false;
8880 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8883 if (vmx->nested.sync_shadow_vmcs) {
8884 copy_vmcs12_to_shadow(vmx);
8885 vmx->nested.sync_shadow_vmcs = false;
8888 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8889 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
8890 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
8891 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
8893 cr4 = cr4_read_shadow();
8894 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
8895 vmcs_writel(HOST_CR4, cr4);
8896 vmx->host_state.vmcs_host_cr4 = cr4;
8899 /* When single-stepping over STI and MOV SS, we must clear the
8900 * corresponding interruptibility bits in the guest state. Otherwise
8901 * vmentry fails as it then expects bit 14 (BS) in pending debug
8902 * exceptions being set, but that's not correct for the guest debugging
8904 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8905 vmx_set_interrupt_shadow(vcpu, 0);
8907 if (vmx->guest_pkru_valid)
8908 __write_pkru(vmx->guest_pkru);
8910 atomic_switch_perf_msrs(vmx);
8911 debugctlmsr = get_debugctlmsr();
8913 vmx_arm_hv_timer(vcpu);
8915 vmx->__launched = vmx->loaded_vmcs->launched;
8917 /* Store host registers */
8918 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
8919 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
8920 "push %%" _ASM_CX " \n\t"
8921 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8923 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8924 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
8926 /* Reload cr2 if changed */
8927 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
8928 "mov %%cr2, %%" _ASM_DX " \n\t"
8929 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
8931 "mov %%" _ASM_AX", %%cr2 \n\t"
8933 /* Check if vmlaunch of vmresume is needed */
8934 "cmpl $0, %c[launched](%0) \n\t"
8935 /* Load guest registers. Don't clobber flags. */
8936 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
8937 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
8938 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
8939 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
8940 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
8941 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
8942 #ifdef CONFIG_X86_64
8943 "mov %c[r8](%0), %%r8 \n\t"
8944 "mov %c[r9](%0), %%r9 \n\t"
8945 "mov %c[r10](%0), %%r10 \n\t"
8946 "mov %c[r11](%0), %%r11 \n\t"
8947 "mov %c[r12](%0), %%r12 \n\t"
8948 "mov %c[r13](%0), %%r13 \n\t"
8949 "mov %c[r14](%0), %%r14 \n\t"
8950 "mov %c[r15](%0), %%r15 \n\t"
8952 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
8954 /* Enter guest mode */
8956 __ex(ASM_VMX_VMLAUNCH) "\n\t"
8958 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
8960 /* Save guest registers, load host registers, keep flags */
8961 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
8963 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
8964 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
8965 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
8966 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
8967 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
8968 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
8969 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
8970 #ifdef CONFIG_X86_64
8971 "mov %%r8, %c[r8](%0) \n\t"
8972 "mov %%r9, %c[r9](%0) \n\t"
8973 "mov %%r10, %c[r10](%0) \n\t"
8974 "mov %%r11, %c[r11](%0) \n\t"
8975 "mov %%r12, %c[r12](%0) \n\t"
8976 "mov %%r13, %c[r13](%0) \n\t"
8977 "mov %%r14, %c[r14](%0) \n\t"
8978 "mov %%r15, %c[r15](%0) \n\t"
8980 "mov %%cr2, %%" _ASM_AX " \n\t"
8981 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
8983 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
8984 "setbe %c[fail](%0) \n\t"
8985 ".pushsection .rodata \n\t"
8986 ".global vmx_return \n\t"
8987 "vmx_return: " _ASM_PTR " 2b \n\t"
8989 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
8990 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
8991 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
8992 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
8993 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
8994 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
8995 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
8996 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
8997 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
8998 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
8999 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
9000 #ifdef CONFIG_X86_64
9001 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
9002 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
9003 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
9004 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
9005 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
9006 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
9007 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
9008 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
9010 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
9011 [wordsize]"i"(sizeof(ulong))
9013 #ifdef CONFIG_X86_64
9014 , "rax", "rbx", "rdi", "rsi"
9015 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
9017 , "eax", "ebx", "edi", "esi"
9021 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
9023 update_debugctlmsr(debugctlmsr);
9025 #ifndef CONFIG_X86_64
9027 * The sysexit path does not restore ds/es, so we must set them to
9028 * a reasonable value ourselves.
9030 * We can't defer this to vmx_load_host_state() since that function
9031 * may be executed in interrupt context, which saves and restore segments
9032 * around it, nullifying its effect.
9034 loadsegment(ds, __USER_DS);
9035 loadsegment(es, __USER_DS);
9038 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
9039 | (1 << VCPU_EXREG_RFLAGS)
9040 | (1 << VCPU_EXREG_PDPTR)
9041 | (1 << VCPU_EXREG_SEGMENTS)
9042 | (1 << VCPU_EXREG_CR3));
9043 vcpu->arch.regs_dirty = 0;
9045 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9047 vmx->loaded_vmcs->launched = 1;
9049 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
9052 * eager fpu is enabled if PKEY is supported and CR4 is switched
9053 * back on host, so it is safe to read guest PKRU from current
9056 if (boot_cpu_has(X86_FEATURE_OSPKE)) {
9057 vmx->guest_pkru = __read_pkru();
9058 if (vmx->guest_pkru != vmx->host_pkru) {
9059 vmx->guest_pkru_valid = true;
9060 __write_pkru(vmx->host_pkru);
9062 vmx->guest_pkru_valid = false;
9066 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
9067 * we did not inject a still-pending event to L1 now because of
9068 * nested_run_pending, we need to re-enable this bit.
9070 if (vmx->nested.nested_run_pending)
9071 kvm_make_request(KVM_REQ_EVENT, vcpu);
9073 vmx->nested.nested_run_pending = 0;
9075 vmx_complete_atomic_exit(vmx);
9076 vmx_recover_nmi_blocking(vmx);
9077 vmx_complete_interrupts(vmx);
9080 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
9082 struct vcpu_vmx *vmx = to_vmx(vcpu);
9085 if (vmx->loaded_vmcs == vmcs)
9089 vmx->loaded_vmcs = vmcs;
9091 vmx_vcpu_load(vcpu, cpu);
9097 * Ensure that the current vmcs of the logical processor is the
9098 * vmcs01 of the vcpu before calling free_nested().
9100 static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
9102 struct vcpu_vmx *vmx = to_vmx(vcpu);
9105 r = vcpu_load(vcpu);
9107 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
9112 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
9114 struct vcpu_vmx *vmx = to_vmx(vcpu);
9117 vmx_destroy_pml_buffer(vmx);
9118 free_vpid(vmx->vpid);
9119 leave_guest_mode(vcpu);
9120 vmx_free_vcpu_nested(vcpu);
9121 free_loaded_vmcs(vmx->loaded_vmcs);
9122 kfree(vmx->guest_msrs);
9123 kvm_vcpu_uninit(vcpu);
9124 kmem_cache_free(kvm_vcpu_cache, vmx);
9127 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
9130 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
9134 return ERR_PTR(-ENOMEM);
9136 vmx->vpid = allocate_vpid();
9138 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
9145 * If PML is turned on, failure on enabling PML just results in failure
9146 * of creating the vcpu, therefore we can simplify PML logic (by
9147 * avoiding dealing with cases, such as enabling PML partially on vcpus
9148 * for the guest, etc.
9151 vmx->pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
9156 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
9157 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
9160 if (!vmx->guest_msrs)
9163 vmx->loaded_vmcs = &vmx->vmcs01;
9164 vmx->loaded_vmcs->vmcs = alloc_vmcs();
9165 vmx->loaded_vmcs->shadow_vmcs = NULL;
9166 if (!vmx->loaded_vmcs->vmcs)
9169 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
9170 loaded_vmcs_init(vmx->loaded_vmcs);
9175 vmx_vcpu_load(&vmx->vcpu, cpu);
9176 vmx->vcpu.cpu = cpu;
9177 err = vmx_vcpu_setup(vmx);
9178 vmx_vcpu_put(&vmx->vcpu);
9182 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9183 err = alloc_apic_access_page(kvm);
9189 if (!kvm->arch.ept_identity_map_addr)
9190 kvm->arch.ept_identity_map_addr =
9191 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
9192 err = init_rmode_identity_map(kvm);
9198 nested_vmx_setup_ctls_msrs(vmx);
9199 vmx->nested.vpid02 = allocate_vpid();
9202 vmx->nested.posted_intr_nv = -1;
9203 vmx->nested.current_vmptr = -1ull;
9204 vmx->nested.current_vmcs12 = NULL;
9206 vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
9211 free_vpid(vmx->nested.vpid02);
9212 free_loaded_vmcs(vmx->loaded_vmcs);
9214 kfree(vmx->guest_msrs);
9216 vmx_destroy_pml_buffer(vmx);
9218 kvm_vcpu_uninit(&vmx->vcpu);
9220 free_vpid(vmx->vpid);
9221 kmem_cache_free(kvm_vcpu_cache, vmx);
9222 return ERR_PTR(err);
9225 static void __init vmx_check_processor_compat(void *rtn)
9227 struct vmcs_config vmcs_conf;
9230 if (setup_vmcs_config(&vmcs_conf) < 0)
9232 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
9233 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
9234 smp_processor_id());
9239 static int get_ept_level(void)
9241 return VMX_EPT_DEFAULT_GAW + 1;
9244 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
9249 /* For VT-d and EPT combination
9250 * 1. MMIO: always map as UC
9252 * a. VT-d without snooping control feature: can't guarantee the
9253 * result, try to trust guest.
9254 * b. VT-d with snooping control feature: snooping control feature of
9255 * VT-d engine can guarantee the cache correctness. Just set it
9256 * to WB to keep consistent with host. So the same as item 3.
9257 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9258 * consistent with host MTRR
9261 cache = MTRR_TYPE_UNCACHABLE;
9265 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
9266 ipat = VMX_EPT_IPAT_BIT;
9267 cache = MTRR_TYPE_WRBACK;
9271 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
9272 ipat = VMX_EPT_IPAT_BIT;
9273 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
9274 cache = MTRR_TYPE_WRBACK;
9276 cache = MTRR_TYPE_UNCACHABLE;
9280 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
9283 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
9286 static int vmx_get_lpage_level(void)
9288 if (enable_ept && !cpu_has_vmx_ept_1g_page())
9289 return PT_DIRECTORY_LEVEL;
9291 /* For shadow and EPT supported 1GB page */
9292 return PT_PDPE_LEVEL;
9295 static void vmcs_set_secondary_exec_control(u32 new_ctl)
9298 * These bits in the secondary execution controls field
9299 * are dynamic, the others are mostly based on the hypervisor
9300 * architecture and the guest's CPUID. Do not touch the
9304 SECONDARY_EXEC_SHADOW_VMCS |
9305 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
9306 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9308 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9310 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
9311 (new_ctl & ~mask) | (cur_ctl & mask));
9315 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
9316 * (indicating "allowed-1") if they are supported in the guest's CPUID.
9318 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
9320 struct vcpu_vmx *vmx = to_vmx(vcpu);
9321 struct kvm_cpuid_entry2 *entry;
9323 vmx->nested.nested_vmx_cr0_fixed1 = 0xffffffff;
9324 vmx->nested.nested_vmx_cr4_fixed1 = X86_CR4_PCE;
9326 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
9327 if (entry && (entry->_reg & (_cpuid_mask))) \
9328 vmx->nested.nested_vmx_cr4_fixed1 |= (_cr4_mask); \
9331 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
9332 cr4_fixed1_update(X86_CR4_VME, edx, bit(X86_FEATURE_VME));
9333 cr4_fixed1_update(X86_CR4_PVI, edx, bit(X86_FEATURE_VME));
9334 cr4_fixed1_update(X86_CR4_TSD, edx, bit(X86_FEATURE_TSC));
9335 cr4_fixed1_update(X86_CR4_DE, edx, bit(X86_FEATURE_DE));
9336 cr4_fixed1_update(X86_CR4_PSE, edx, bit(X86_FEATURE_PSE));
9337 cr4_fixed1_update(X86_CR4_PAE, edx, bit(X86_FEATURE_PAE));
9338 cr4_fixed1_update(X86_CR4_MCE, edx, bit(X86_FEATURE_MCE));
9339 cr4_fixed1_update(X86_CR4_PGE, edx, bit(X86_FEATURE_PGE));
9340 cr4_fixed1_update(X86_CR4_OSFXSR, edx, bit(X86_FEATURE_FXSR));
9341 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
9342 cr4_fixed1_update(X86_CR4_VMXE, ecx, bit(X86_FEATURE_VMX));
9343 cr4_fixed1_update(X86_CR4_SMXE, ecx, bit(X86_FEATURE_SMX));
9344 cr4_fixed1_update(X86_CR4_PCIDE, ecx, bit(X86_FEATURE_PCID));
9345 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, bit(X86_FEATURE_XSAVE));
9347 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9348 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, bit(X86_FEATURE_FSGSBASE));
9349 cr4_fixed1_update(X86_CR4_SMEP, ebx, bit(X86_FEATURE_SMEP));
9350 cr4_fixed1_update(X86_CR4_SMAP, ebx, bit(X86_FEATURE_SMAP));
9351 cr4_fixed1_update(X86_CR4_PKE, ecx, bit(X86_FEATURE_PKU));
9352 /* TODO: Use X86_CR4_UMIP and X86_FEATURE_UMIP macros */
9353 cr4_fixed1_update(bit(11), ecx, bit(2));
9355 #undef cr4_fixed1_update
9358 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
9360 struct kvm_cpuid_entry2 *best;
9361 struct vcpu_vmx *vmx = to_vmx(vcpu);
9362 u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx);
9364 if (vmx_rdtscp_supported()) {
9365 bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu);
9366 if (!rdtscp_enabled)
9367 secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP;
9371 vmx->nested.nested_vmx_secondary_ctls_high |=
9372 SECONDARY_EXEC_RDTSCP;
9374 vmx->nested.nested_vmx_secondary_ctls_high &=
9375 ~SECONDARY_EXEC_RDTSCP;
9379 /* Exposing INVPCID only when PCID is exposed */
9380 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9381 if (vmx_invpcid_supported() &&
9382 (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) ||
9383 !guest_cpuid_has_pcid(vcpu))) {
9384 secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID;
9387 best->ebx &= ~bit(X86_FEATURE_INVPCID);
9390 if (cpu_has_secondary_exec_ctrls())
9391 vmcs_set_secondary_exec_control(secondary_exec_ctl);
9393 if (nested_vmx_allowed(vcpu))
9394 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
9395 FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9397 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
9398 ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9400 if (nested_vmx_allowed(vcpu))
9401 nested_vmx_cr_fixed1_bits_update(vcpu);
9404 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9406 if (func == 1 && nested)
9407 entry->ecx |= bit(X86_FEATURE_VMX);
9410 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9411 struct x86_exception *fault)
9413 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9416 if (fault->error_code & PFERR_RSVD_MASK)
9417 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9419 exit_reason = EXIT_REASON_EPT_VIOLATION;
9420 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
9421 vmcs12->guest_physical_address = fault->address;
9424 /* Callbacks for nested_ept_init_mmu_context: */
9426 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9428 /* return the page table to be shadowed - in our case, EPT12 */
9429 return get_vmcs12(vcpu)->ept_pointer;
9432 static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9436 WARN_ON(mmu_is_nested(vcpu));
9437 eptp = nested_ept_get_cr3(vcpu);
9438 if ((eptp & VMX_EPT_AD_ENABLE_BIT) && !enable_ept_ad_bits)
9441 kvm_mmu_unload(vcpu);
9442 kvm_init_shadow_ept_mmu(vcpu,
9443 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
9444 VMX_EPT_EXECUTE_ONLY_BIT,
9445 eptp & VMX_EPT_AD_ENABLE_BIT);
9446 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
9447 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
9448 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
9450 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
9454 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
9456 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
9459 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9462 bool inequality, bit;
9464 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9466 (error_code & vmcs12->page_fault_error_code_mask) !=
9467 vmcs12->page_fault_error_code_match;
9468 return inequality ^ bit;
9471 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9472 struct x86_exception *fault)
9474 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9476 WARN_ON(!is_guest_mode(vcpu));
9478 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code))
9479 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
9480 vmcs_read32(VM_EXIT_INTR_INFO),
9481 vmcs_readl(EXIT_QUALIFICATION));
9483 kvm_inject_page_fault(vcpu, fault);
9486 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9487 struct vmcs12 *vmcs12);
9489 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9490 struct vmcs12 *vmcs12)
9492 struct vcpu_vmx *vmx = to_vmx(vcpu);
9495 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9497 * Translate L1 physical address to host physical
9498 * address for vmcs02. Keep the page pinned, so this
9499 * physical address remains valid. We keep a reference
9500 * to it so we can release it later.
9502 if (vmx->nested.apic_access_page) /* shouldn't happen */
9503 nested_release_page(vmx->nested.apic_access_page);
9504 vmx->nested.apic_access_page =
9505 nested_get_page(vcpu, vmcs12->apic_access_addr);
9507 * If translation failed, no matter: This feature asks
9508 * to exit when accessing the given address, and if it
9509 * can never be accessed, this feature won't do
9512 if (vmx->nested.apic_access_page) {
9513 hpa = page_to_phys(vmx->nested.apic_access_page);
9514 vmcs_write64(APIC_ACCESS_ADDR, hpa);
9516 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
9517 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9519 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9520 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9521 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
9522 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9523 kvm_vcpu_reload_apic_access_page(vcpu);
9526 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9527 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
9528 nested_release_page(vmx->nested.virtual_apic_page);
9529 vmx->nested.virtual_apic_page =
9530 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
9533 * If translation failed, VM entry will fail because
9534 * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
9535 * Failing the vm entry is _not_ what the processor
9536 * does but it's basically the only possibility we
9537 * have. We could still enter the guest if CR8 load
9538 * exits are enabled, CR8 store exits are enabled, and
9539 * virtualize APIC access is disabled; in this case
9540 * the processor would never use the TPR shadow and we
9541 * could simply clear the bit from the execution
9542 * control. But such a configuration is useless, so
9543 * let's keep the code simple.
9545 if (vmx->nested.virtual_apic_page) {
9546 hpa = page_to_phys(vmx->nested.virtual_apic_page);
9547 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
9551 if (nested_cpu_has_posted_intr(vmcs12)) {
9552 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9553 kunmap(vmx->nested.pi_desc_page);
9554 nested_release_page(vmx->nested.pi_desc_page);
9556 vmx->nested.pi_desc_page =
9557 nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
9558 vmx->nested.pi_desc =
9559 (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
9560 if (!vmx->nested.pi_desc) {
9561 nested_release_page_clean(vmx->nested.pi_desc_page);
9564 vmx->nested.pi_desc =
9565 (struct pi_desc *)((void *)vmx->nested.pi_desc +
9566 (unsigned long)(vmcs12->posted_intr_desc_addr &
9568 vmcs_write64(POSTED_INTR_DESC_ADDR,
9569 page_to_phys(vmx->nested.pi_desc_page) +
9570 (unsigned long)(vmcs12->posted_intr_desc_addr &
9573 if (cpu_has_vmx_msr_bitmap() &&
9574 nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS) &&
9575 nested_vmx_merge_msr_bitmap(vcpu, vmcs12))
9578 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
9579 CPU_BASED_USE_MSR_BITMAPS);
9582 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9584 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9585 struct vcpu_vmx *vmx = to_vmx(vcpu);
9587 if (vcpu->arch.virtual_tsc_khz == 0)
9590 /* Make sure short timeouts reliably trigger an immediate vmexit.
9591 * hrtimer_start does not guarantee this. */
9592 if (preemption_timeout <= 1) {
9593 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9597 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9598 preemption_timeout *= 1000000;
9599 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9600 hrtimer_start(&vmx->nested.preemption_timer,
9601 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9604 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9605 struct vmcs12 *vmcs12)
9610 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9613 if (vmcs12_read_any(vcpu, MSR_BITMAP, &addr)) {
9617 maxphyaddr = cpuid_maxphyaddr(vcpu);
9619 if (!PAGE_ALIGNED(vmcs12->msr_bitmap) ||
9620 ((addr + PAGE_SIZE) >> maxphyaddr))
9627 * Merge L0's and L1's MSR bitmap, return false to indicate that
9628 * we do not use the hardware.
9630 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9631 struct vmcs12 *vmcs12)
9635 unsigned long *msr_bitmap_l1;
9636 unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.msr_bitmap;
9638 /* This shortcut is ok because we support only x2APIC MSRs so far. */
9639 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
9642 page = nested_get_page(vcpu, vmcs12->msr_bitmap);
9645 msr_bitmap_l1 = (unsigned long *)kmap(page);
9647 memset(msr_bitmap_l0, 0xff, PAGE_SIZE);
9649 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
9650 if (nested_cpu_has_apic_reg_virt(vmcs12))
9651 for (msr = 0x800; msr <= 0x8ff; msr++)
9652 nested_vmx_disable_intercept_for_msr(
9653 msr_bitmap_l1, msr_bitmap_l0,
9656 nested_vmx_disable_intercept_for_msr(
9657 msr_bitmap_l1, msr_bitmap_l0,
9658 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9659 MSR_TYPE_R | MSR_TYPE_W);
9661 if (nested_cpu_has_vid(vmcs12)) {
9662 nested_vmx_disable_intercept_for_msr(
9663 msr_bitmap_l1, msr_bitmap_l0,
9664 APIC_BASE_MSR + (APIC_EOI >> 4),
9666 nested_vmx_disable_intercept_for_msr(
9667 msr_bitmap_l1, msr_bitmap_l0,
9668 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9673 nested_release_page_clean(page);
9678 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
9679 struct vmcs12 *vmcs12)
9681 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9682 !nested_cpu_has_apic_reg_virt(vmcs12) &&
9683 !nested_cpu_has_vid(vmcs12) &&
9684 !nested_cpu_has_posted_intr(vmcs12))
9688 * If virtualize x2apic mode is enabled,
9689 * virtualize apic access must be disabled.
9691 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9692 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9696 * If virtual interrupt delivery is enabled,
9697 * we must exit on external interrupts.
9699 if (nested_cpu_has_vid(vmcs12) &&
9700 !nested_exit_on_intr(vcpu))
9704 * bits 15:8 should be zero in posted_intr_nv,
9705 * the descriptor address has been already checked
9706 * in nested_get_vmcs12_pages.
9708 if (nested_cpu_has_posted_intr(vmcs12) &&
9709 (!nested_cpu_has_vid(vmcs12) ||
9710 !nested_exit_intr_ack_set(vcpu) ||
9711 vmcs12->posted_intr_nv & 0xff00))
9714 /* tpr shadow is needed by all apicv features. */
9715 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9721 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9722 unsigned long count_field,
9723 unsigned long addr_field)
9728 if (vmcs12_read_any(vcpu, count_field, &count) ||
9729 vmcs12_read_any(vcpu, addr_field, &addr)) {
9735 maxphyaddr = cpuid_maxphyaddr(vcpu);
9736 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9737 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9738 pr_debug_ratelimited(
9739 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9740 addr_field, maxphyaddr, count, addr);
9746 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9747 struct vmcs12 *vmcs12)
9749 if (vmcs12->vm_exit_msr_load_count == 0 &&
9750 vmcs12->vm_exit_msr_store_count == 0 &&
9751 vmcs12->vm_entry_msr_load_count == 0)
9752 return 0; /* Fast path */
9753 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9754 VM_EXIT_MSR_LOAD_ADDR) ||
9755 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9756 VM_EXIT_MSR_STORE_ADDR) ||
9757 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9758 VM_ENTRY_MSR_LOAD_ADDR))
9763 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9764 struct vmx_msr_entry *e)
9766 /* x2APIC MSR accesses are not allowed */
9767 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9769 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9770 e->index == MSR_IA32_UCODE_REV)
9772 if (e->reserved != 0)
9777 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9778 struct vmx_msr_entry *e)
9780 if (e->index == MSR_FS_BASE ||
9781 e->index == MSR_GS_BASE ||
9782 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9783 nested_vmx_msr_check_common(vcpu, e))
9788 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9789 struct vmx_msr_entry *e)
9791 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9792 nested_vmx_msr_check_common(vcpu, e))
9798 * Load guest's/host's msr at nested entry/exit.
9799 * return 0 for success, entry index for failure.
9801 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9804 struct vmx_msr_entry e;
9805 struct msr_data msr;
9807 msr.host_initiated = false;
9808 for (i = 0; i < count; i++) {
9809 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
9811 pr_debug_ratelimited(
9812 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9813 __func__, i, gpa + i * sizeof(e));
9816 if (nested_vmx_load_msr_check(vcpu, &e)) {
9817 pr_debug_ratelimited(
9818 "%s check failed (%u, 0x%x, 0x%x)\n",
9819 __func__, i, e.index, e.reserved);
9822 msr.index = e.index;
9824 if (kvm_set_msr(vcpu, &msr)) {
9825 pr_debug_ratelimited(
9826 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9827 __func__, i, e.index, e.value);
9836 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9839 struct vmx_msr_entry e;
9841 for (i = 0; i < count; i++) {
9842 struct msr_data msr_info;
9843 if (kvm_vcpu_read_guest(vcpu,
9844 gpa + i * sizeof(e),
9845 &e, 2 * sizeof(u32))) {
9846 pr_debug_ratelimited(
9847 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9848 __func__, i, gpa + i * sizeof(e));
9851 if (nested_vmx_store_msr_check(vcpu, &e)) {
9852 pr_debug_ratelimited(
9853 "%s check failed (%u, 0x%x, 0x%x)\n",
9854 __func__, i, e.index, e.reserved);
9857 msr_info.host_initiated = false;
9858 msr_info.index = e.index;
9859 if (kvm_get_msr(vcpu, &msr_info)) {
9860 pr_debug_ratelimited(
9861 "%s cannot read MSR (%u, 0x%x)\n",
9862 __func__, i, e.index);
9865 if (kvm_vcpu_write_guest(vcpu,
9866 gpa + i * sizeof(e) +
9867 offsetof(struct vmx_msr_entry, value),
9868 &msr_info.data, sizeof(msr_info.data))) {
9869 pr_debug_ratelimited(
9870 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9871 __func__, i, e.index, msr_info.data);
9878 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
9880 unsigned long invalid_mask;
9882 invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
9883 return (val & invalid_mask) == 0;
9887 * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
9888 * emulating VM entry into a guest with EPT enabled.
9889 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
9890 * is assigned to entry_failure_code on failure.
9892 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
9893 u32 *entry_failure_code)
9895 if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
9896 if (!nested_cr3_valid(vcpu, cr3)) {
9897 *entry_failure_code = ENTRY_FAIL_DEFAULT;
9902 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
9903 * must not be dereferenced.
9905 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
9907 if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
9908 *entry_failure_code = ENTRY_FAIL_PDPTE;
9913 vcpu->arch.cr3 = cr3;
9914 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
9917 kvm_mmu_reset_context(vcpu);
9922 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
9923 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
9924 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
9925 * guest in a way that will both be appropriate to L1's requests, and our
9926 * needs. In addition to modifying the active vmcs (which is vmcs02), this
9927 * function also has additional necessary side-effects, like setting various
9928 * vcpu->arch fields.
9929 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
9930 * is assigned to entry_failure_code on failure.
9932 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
9933 bool from_vmentry, u32 *entry_failure_code)
9935 struct vcpu_vmx *vmx = to_vmx(vcpu);
9938 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
9939 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
9940 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
9941 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
9942 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
9943 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
9944 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
9945 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
9946 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
9947 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
9948 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
9949 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
9950 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
9951 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
9952 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
9953 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
9954 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
9955 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
9956 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
9957 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
9958 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
9959 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
9960 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
9961 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
9962 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
9963 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
9964 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
9965 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
9966 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
9967 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
9968 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
9969 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
9970 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
9971 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
9972 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
9973 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
9976 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
9977 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
9978 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
9980 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
9981 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
9984 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
9985 vmcs12->vm_entry_intr_info_field);
9986 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
9987 vmcs12->vm_entry_exception_error_code);
9988 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
9989 vmcs12->vm_entry_instruction_len);
9990 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
9991 vmcs12->guest_interruptibility_info);
9993 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
9995 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
9996 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
9997 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
9998 vmcs12->guest_pending_dbg_exceptions);
9999 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
10000 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
10002 if (nested_cpu_has_xsaves(vmcs12))
10003 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
10004 vmcs_write64(VMCS_LINK_POINTER, -1ull);
10006 exec_control = vmcs12->pin_based_vm_exec_control;
10008 /* Preemption timer setting is only taken from vmcs01. */
10009 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10010 exec_control |= vmcs_config.pin_based_exec_ctrl;
10011 if (vmx->hv_deadline_tsc == -1)
10012 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10014 /* Posted interrupts setting is only taken from vmcs12. */
10015 if (nested_cpu_has_posted_intr(vmcs12)) {
10017 * Note that we use L0's vector here and in
10018 * vmx_deliver_nested_posted_interrupt.
10020 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
10021 vmx->nested.pi_pending = false;
10022 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
10024 exec_control &= ~PIN_BASED_POSTED_INTR;
10027 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
10029 vmx->nested.preemption_timer_expired = false;
10030 if (nested_cpu_has_preemption_timer(vmcs12))
10031 vmx_start_preemption_timer(vcpu);
10034 * Whether page-faults are trapped is determined by a combination of
10035 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
10036 * If enable_ept, L0 doesn't care about page faults and we should
10037 * set all of these to L1's desires. However, if !enable_ept, L0 does
10038 * care about (at least some) page faults, and because it is not easy
10039 * (if at all possible?) to merge L0 and L1's desires, we simply ask
10040 * to exit on each and every L2 page fault. This is done by setting
10041 * MASK=MATCH=0 and (see below) EB.PF=1.
10042 * Note that below we don't need special code to set EB.PF beyond the
10043 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
10044 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
10045 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
10047 * A problem with this approach (when !enable_ept) is that L1 may be
10048 * injected with more page faults than it asked for. This could have
10049 * caused problems, but in practice existing hypervisors don't care.
10050 * To fix this, we will need to emulate the PFEC checking (on the L1
10051 * page tables), using walk_addr(), when injecting PFs to L1.
10053 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
10054 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
10055 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
10056 enable_ept ? vmcs12->page_fault_error_code_match : 0);
10058 if (cpu_has_secondary_exec_ctrls()) {
10059 exec_control = vmx_secondary_exec_control(vmx);
10061 /* Take the following fields only from vmcs12 */
10062 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
10063 SECONDARY_EXEC_RDTSCP |
10064 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
10065 SECONDARY_EXEC_APIC_REGISTER_VIRT);
10066 if (nested_cpu_has(vmcs12,
10067 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
10068 exec_control |= vmcs12->secondary_vm_exec_control;
10070 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
10071 vmcs_write64(EOI_EXIT_BITMAP0,
10072 vmcs12->eoi_exit_bitmap0);
10073 vmcs_write64(EOI_EXIT_BITMAP1,
10074 vmcs12->eoi_exit_bitmap1);
10075 vmcs_write64(EOI_EXIT_BITMAP2,
10076 vmcs12->eoi_exit_bitmap2);
10077 vmcs_write64(EOI_EXIT_BITMAP3,
10078 vmcs12->eoi_exit_bitmap3);
10079 vmcs_write16(GUEST_INTR_STATUS,
10080 vmcs12->guest_intr_status);
10084 * Write an illegal value to APIC_ACCESS_ADDR. Later,
10085 * nested_get_vmcs12_pages will either fix it up or
10086 * remove the VM execution control.
10088 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
10089 vmcs_write64(APIC_ACCESS_ADDR, -1ull);
10091 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
10096 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
10097 * Some constant fields are set here by vmx_set_constant_host_state().
10098 * Other fields are different per CPU, and will be set later when
10099 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
10101 vmx_set_constant_host_state(vmx);
10104 * Set the MSR load/store lists to match L0's settings.
10106 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
10107 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10108 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
10109 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10110 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
10113 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
10114 * entry, but only if the current (host) sp changed from the value
10115 * we wrote last (vmx->host_rsp). This cache is no longer relevant
10116 * if we switch vmcs, and rather than hold a separate cache per vmcs,
10117 * here we just force the write to happen on entry.
10121 exec_control = vmx_exec_control(vmx); /* L0's desires */
10122 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
10123 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
10124 exec_control &= ~CPU_BASED_TPR_SHADOW;
10125 exec_control |= vmcs12->cpu_based_vm_exec_control;
10128 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
10129 * nested_get_vmcs12_pages can't fix it up, the illegal value
10130 * will result in a VM entry failure.
10132 if (exec_control & CPU_BASED_TPR_SHADOW) {
10133 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
10134 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
10138 * Merging of IO bitmap not currently supported.
10139 * Rather, exit every time.
10141 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
10142 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
10144 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
10146 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
10147 * bitwise-or of what L1 wants to trap for L2, and what we want to
10148 * trap. Note that CR0.TS also needs updating - we do this later.
10150 update_exception_bitmap(vcpu);
10151 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
10152 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10154 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
10155 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
10156 * bits are further modified by vmx_set_efer() below.
10158 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
10160 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
10161 * emulated by vmx_set_efer(), below.
10163 vm_entry_controls_init(vmx,
10164 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
10165 ~VM_ENTRY_IA32E_MODE) |
10166 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
10168 if (from_vmentry &&
10169 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
10170 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
10171 vcpu->arch.pat = vmcs12->guest_ia32_pat;
10172 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
10173 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
10176 set_cr4_guest_host_mask(vmx);
10178 if (from_vmentry &&
10179 vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
10180 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
10182 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
10183 vmcs_write64(TSC_OFFSET,
10184 vcpu->arch.tsc_offset + vmcs12->tsc_offset);
10186 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
10187 if (kvm_has_tsc_control)
10188 decache_tsc_multiplier(vmx);
10192 * There is no direct mapping between vpid02 and vpid12, the
10193 * vpid02 is per-vCPU for L0 and reused while the value of
10194 * vpid12 is changed w/ one invvpid during nested vmentry.
10195 * The vpid12 is allocated by L1 for L2, so it will not
10196 * influence global bitmap(for vpid01 and vpid02 allocation)
10197 * even if spawn a lot of nested vCPUs.
10199 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
10200 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
10201 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
10202 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
10203 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
10206 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
10207 vmx_flush_tlb(vcpu);
10212 if (nested_cpu_has_ept(vmcs12)) {
10213 if (nested_ept_init_mmu_context(vcpu)) {
10214 *entry_failure_code = ENTRY_FAIL_DEFAULT;
10217 } else if (nested_cpu_has2(vmcs12,
10218 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10219 vmx_flush_tlb_ept_only(vcpu);
10223 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
10224 * bits which we consider mandatory enabled.
10225 * The CR0_READ_SHADOW is what L2 should have expected to read given
10226 * the specifications by L1; It's not enough to take
10227 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10228 * have more bits than L1 expected.
10230 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
10231 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
10233 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
10234 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
10236 if (from_vmentry &&
10237 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
10238 vcpu->arch.efer = vmcs12->guest_ia32_efer;
10239 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
10240 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10242 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10243 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10244 vmx_set_efer(vcpu, vcpu->arch.efer);
10246 /* Shadow page tables on either EPT or shadow page tables. */
10247 if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
10248 entry_failure_code))
10252 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
10255 * L1 may access the L2's PDPTR, so save them to construct vmcs12
10258 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
10259 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
10260 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
10261 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
10264 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
10265 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
10269 static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10271 struct vcpu_vmx *vmx = to_vmx(vcpu);
10273 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
10274 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
10275 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10277 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12))
10278 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10280 if (nested_vmx_check_apicv_controls(vcpu, vmcs12))
10281 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10283 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12))
10284 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10286 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
10287 vmx->nested.nested_vmx_procbased_ctls_low,
10288 vmx->nested.nested_vmx_procbased_ctls_high) ||
10289 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
10290 vmx->nested.nested_vmx_secondary_ctls_low,
10291 vmx->nested.nested_vmx_secondary_ctls_high) ||
10292 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
10293 vmx->nested.nested_vmx_pinbased_ctls_low,
10294 vmx->nested.nested_vmx_pinbased_ctls_high) ||
10295 !vmx_control_verify(vmcs12->vm_exit_controls,
10296 vmx->nested.nested_vmx_exit_ctls_low,
10297 vmx->nested.nested_vmx_exit_ctls_high) ||
10298 !vmx_control_verify(vmcs12->vm_entry_controls,
10299 vmx->nested.nested_vmx_entry_ctls_low,
10300 vmx->nested.nested_vmx_entry_ctls_high))
10301 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10303 if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
10304 !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
10305 !nested_cr3_valid(vcpu, vmcs12->host_cr3))
10306 return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
10311 static int check_vmentry_postreqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10316 *exit_qual = ENTRY_FAIL_DEFAULT;
10318 if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
10319 !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
10322 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_SHADOW_VMCS) &&
10323 vmcs12->vmcs_link_pointer != -1ull) {
10324 *exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
10329 * If the load IA32_EFER VM-entry control is 1, the following checks
10330 * are performed on the field for the IA32_EFER MSR:
10331 * - Bits reserved in the IA32_EFER MSR must be 0.
10332 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
10333 * the IA-32e mode guest VM-exit control. It must also be identical
10334 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
10337 if (to_vmx(vcpu)->nested.nested_run_pending &&
10338 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
10339 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
10340 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
10341 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
10342 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
10343 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
10348 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
10349 * IA32_EFER MSR must be 0 in the field for that register. In addition,
10350 * the values of the LMA and LME bits in the field must each be that of
10351 * the host address-space size VM-exit control.
10353 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
10354 ia32e = (vmcs12->vm_exit_controls &
10355 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
10356 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
10357 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
10358 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
10365 static int enter_vmx_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
10367 struct vcpu_vmx *vmx = to_vmx(vcpu);
10368 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10369 struct loaded_vmcs *vmcs02;
10373 vmcs02 = nested_get_current_vmcs02(vmx);
10377 enter_guest_mode(vcpu);
10379 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
10380 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10382 vmx_switch_vmcs(vcpu, vmcs02);
10383 vmx_segment_cache_clear(vmx);
10385 if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &exit_qual)) {
10386 leave_guest_mode(vcpu);
10387 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10388 nested_vmx_entry_failure(vcpu, vmcs12,
10389 EXIT_REASON_INVALID_STATE, exit_qual);
10393 nested_get_vmcs12_pages(vcpu, vmcs12);
10395 msr_entry_idx = nested_vmx_load_msr(vcpu,
10396 vmcs12->vm_entry_msr_load_addr,
10397 vmcs12->vm_entry_msr_load_count);
10398 if (msr_entry_idx) {
10399 leave_guest_mode(vcpu);
10400 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10401 nested_vmx_entry_failure(vcpu, vmcs12,
10402 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
10406 vmcs12->launch_state = 1;
10409 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10410 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10411 * returned as far as L1 is concerned. It will only return (and set
10412 * the success flag) when L2 exits (see nested_vmx_vmexit()).
10418 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
10419 * for running an L2 nested guest.
10421 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
10423 struct vmcs12 *vmcs12;
10424 struct vcpu_vmx *vmx = to_vmx(vcpu);
10428 if (!nested_vmx_check_permission(vcpu))
10431 if (!nested_vmx_check_vmcs12(vcpu))
10434 vmcs12 = get_vmcs12(vcpu);
10436 if (enable_shadow_vmcs)
10437 copy_shadow_to_vmcs12(vmx);
10440 * The nested entry process starts with enforcing various prerequisites
10441 * on vmcs12 as required by the Intel SDM, and act appropriately when
10442 * they fail: As the SDM explains, some conditions should cause the
10443 * instruction to fail, while others will cause the instruction to seem
10444 * to succeed, but return an EXIT_REASON_INVALID_STATE.
10445 * To speed up the normal (success) code path, we should avoid checking
10446 * for misconfigurations which will anyway be caught by the processor
10447 * when using the merged vmcs02.
10449 if (vmcs12->launch_state == launch) {
10450 nested_vmx_failValid(vcpu,
10451 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
10452 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
10456 ret = check_vmentry_prereqs(vcpu, vmcs12);
10458 nested_vmx_failValid(vcpu, ret);
10463 * After this point, the trap flag no longer triggers a singlestep trap
10464 * on the vm entry instructions; don't call kvm_skip_emulated_instruction.
10465 * This is not 100% correct; for performance reasons, we delegate most
10466 * of the checks on host state to the processor. If those fail,
10467 * the singlestep trap is missed.
10469 skip_emulated_instruction(vcpu);
10471 ret = check_vmentry_postreqs(vcpu, vmcs12, &exit_qual);
10473 nested_vmx_entry_failure(vcpu, vmcs12,
10474 EXIT_REASON_INVALID_STATE, exit_qual);
10479 * We're finally done with prerequisite checking, and can start with
10480 * the nested entry.
10483 ret = enter_vmx_non_root_mode(vcpu, true);
10487 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
10488 return kvm_vcpu_halt(vcpu);
10490 vmx->nested.nested_run_pending = 1;
10495 return kvm_skip_emulated_instruction(vcpu);
10499 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
10500 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
10501 * This function returns the new value we should put in vmcs12.guest_cr0.
10502 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
10503 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
10504 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
10505 * didn't trap the bit, because if L1 did, so would L0).
10506 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
10507 * been modified by L2, and L1 knows it. So just leave the old value of
10508 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
10509 * isn't relevant, because if L0 traps this bit it can set it to anything.
10510 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
10511 * changed these bits, and therefore they need to be updated, but L0
10512 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
10513 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
10515 static inline unsigned long
10516 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10519 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
10520 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
10521 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
10522 vcpu->arch.cr0_guest_owned_bits));
10525 static inline unsigned long
10526 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10529 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
10530 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
10531 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
10532 vcpu->arch.cr4_guest_owned_bits));
10535 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
10536 struct vmcs12 *vmcs12)
10541 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
10542 nr = vcpu->arch.exception.nr;
10543 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10545 if (kvm_exception_is_soft(nr)) {
10546 vmcs12->vm_exit_instruction_len =
10547 vcpu->arch.event_exit_inst_len;
10548 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
10550 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
10552 if (vcpu->arch.exception.has_error_code) {
10553 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
10554 vmcs12->idt_vectoring_error_code =
10555 vcpu->arch.exception.error_code;
10558 vmcs12->idt_vectoring_info_field = idt_vectoring;
10559 } else if (vcpu->arch.nmi_injected) {
10560 vmcs12->idt_vectoring_info_field =
10561 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
10562 } else if (vcpu->arch.interrupt.pending) {
10563 nr = vcpu->arch.interrupt.nr;
10564 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10566 if (vcpu->arch.interrupt.soft) {
10567 idt_vectoring |= INTR_TYPE_SOFT_INTR;
10568 vmcs12->vm_entry_instruction_len =
10569 vcpu->arch.event_exit_inst_len;
10571 idt_vectoring |= INTR_TYPE_EXT_INTR;
10573 vmcs12->idt_vectoring_info_field = idt_vectoring;
10577 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
10579 struct vcpu_vmx *vmx = to_vmx(vcpu);
10581 if (vcpu->arch.exception.pending ||
10582 vcpu->arch.nmi_injected ||
10583 vcpu->arch.interrupt.pending)
10586 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
10587 vmx->nested.preemption_timer_expired) {
10588 if (vmx->nested.nested_run_pending)
10590 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
10594 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
10595 if (vmx->nested.nested_run_pending)
10597 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10598 NMI_VECTOR | INTR_TYPE_NMI_INTR |
10599 INTR_INFO_VALID_MASK, 0);
10601 * The NMI-triggered VM exit counts as injection:
10602 * clear this one and block further NMIs.
10604 vcpu->arch.nmi_pending = 0;
10605 vmx_set_nmi_mask(vcpu, true);
10609 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
10610 nested_exit_on_intr(vcpu)) {
10611 if (vmx->nested.nested_run_pending)
10613 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
10617 vmx_complete_nested_posted_interrupt(vcpu);
10621 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
10623 ktime_t remaining =
10624 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
10627 if (ktime_to_ns(remaining) <= 0)
10630 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
10631 do_div(value, 1000000);
10632 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10636 * Update the guest state fields of vmcs12 to reflect changes that
10637 * occurred while L2 was running. (The "IA-32e mode guest" bit of the
10638 * VM-entry controls is also updated, since this is really a guest
10641 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10643 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
10644 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
10646 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
10647 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
10648 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
10650 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
10651 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
10652 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
10653 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
10654 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
10655 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
10656 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
10657 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
10658 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
10659 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
10660 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
10661 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
10662 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
10663 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
10664 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
10665 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
10666 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
10667 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
10668 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
10669 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
10670 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
10671 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
10672 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
10673 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
10674 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
10675 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
10676 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
10677 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
10678 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
10679 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
10680 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
10681 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
10682 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
10683 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
10684 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
10685 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
10687 vmcs12->guest_interruptibility_info =
10688 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
10689 vmcs12->guest_pending_dbg_exceptions =
10690 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
10691 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10692 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
10694 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
10696 if (nested_cpu_has_preemption_timer(vmcs12)) {
10697 if (vmcs12->vm_exit_controls &
10698 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
10699 vmcs12->vmx_preemption_timer_value =
10700 vmx_get_preemption_timer_value(vcpu);
10701 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
10705 * In some cases (usually, nested EPT), L2 is allowed to change its
10706 * own CR3 without exiting. If it has changed it, we must keep it.
10707 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10708 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10710 * Additionally, restore L2's PDPTR to vmcs12.
10713 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
10714 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
10715 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
10716 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
10717 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
10720 if (nested_cpu_has_ept(vmcs12))
10721 vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
10723 if (nested_cpu_has_vid(vmcs12))
10724 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
10726 vmcs12->vm_entry_controls =
10727 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
10728 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
10730 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
10731 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
10732 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10735 /* TODO: These cannot have changed unless we have MSR bitmaps and
10736 * the relevant bit asks not to trap the change */
10737 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
10738 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
10739 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
10740 vmcs12->guest_ia32_efer = vcpu->arch.efer;
10741 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
10742 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
10743 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
10744 if (kvm_mpx_supported())
10745 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
10746 if (nested_cpu_has_xsaves(vmcs12))
10747 vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP);
10751 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10752 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10753 * and this function updates it to reflect the changes to the guest state while
10754 * L2 was running (and perhaps made some exits which were handled directly by L0
10755 * without going back to L1), and to reflect the exit reason.
10756 * Note that we do not have to copy here all VMCS fields, just those that
10757 * could have changed by the L2 guest or the exit - i.e., the guest-state and
10758 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10759 * which already writes to vmcs12 directly.
10761 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10762 u32 exit_reason, u32 exit_intr_info,
10763 unsigned long exit_qualification)
10765 /* update guest state fields: */
10766 sync_vmcs12(vcpu, vmcs12);
10768 /* update exit information fields: */
10770 vmcs12->vm_exit_reason = exit_reason;
10771 vmcs12->exit_qualification = exit_qualification;
10773 vmcs12->vm_exit_intr_info = exit_intr_info;
10774 if ((vmcs12->vm_exit_intr_info &
10775 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
10776 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
10777 vmcs12->vm_exit_intr_error_code =
10778 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
10779 vmcs12->idt_vectoring_info_field = 0;
10780 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
10781 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
10783 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
10784 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10785 * instead of reading the real value. */
10786 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
10789 * Transfer the event that L0 or L1 may wanted to inject into
10790 * L2 to IDT_VECTORING_INFO_FIELD.
10792 vmcs12_save_pending_event(vcpu, vmcs12);
10796 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10797 * preserved above and would only end up incorrectly in L1.
10799 vcpu->arch.nmi_injected = false;
10800 kvm_clear_exception_queue(vcpu);
10801 kvm_clear_interrupt_queue(vcpu);
10805 * A part of what we need to when the nested L2 guest exits and we want to
10806 * run its L1 parent, is to reset L1's guest state to the host state specified
10808 * This function is to be called not only on normal nested exit, but also on
10809 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10810 * Failures During or After Loading Guest State").
10811 * This function should be called when the active VMCS is L1's (vmcs01).
10813 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10814 struct vmcs12 *vmcs12)
10816 struct kvm_segment seg;
10817 u32 entry_failure_code;
10819 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10820 vcpu->arch.efer = vmcs12->host_ia32_efer;
10821 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10822 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10824 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10825 vmx_set_efer(vcpu, vcpu->arch.efer);
10827 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10828 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10829 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10831 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10832 * actually changed, because vmx_set_cr0 refers to efer set above.
10834 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
10835 * (KVM doesn't change it);
10837 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
10838 vmx_set_cr0(vcpu, vmcs12->host_cr0);
10840 /* Same as above - no reason to call set_cr4_guest_host_mask(). */
10841 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
10842 kvm_set_cr4(vcpu, vmcs12->host_cr4);
10844 nested_ept_uninit_mmu_context(vcpu);
10847 * Only PDPTE load can fail as the value of cr3 was checked on entry and
10848 * couldn't have changed.
10850 if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
10851 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
10854 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
10858 * Trivially support vpid by letting L2s share their parent
10859 * L1's vpid. TODO: move to a more elaborate solution, giving
10860 * each L2 its own vpid and exposing the vpid feature to L1.
10862 vmx_flush_tlb(vcpu);
10866 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
10867 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
10868 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
10869 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
10870 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
10872 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
10873 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
10874 vmcs_write64(GUEST_BNDCFGS, 0);
10876 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
10877 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
10878 vcpu->arch.pat = vmcs12->host_ia32_pat;
10880 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
10881 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
10882 vmcs12->host_ia32_perf_global_ctrl);
10884 /* Set L1 segment info according to Intel SDM
10885 27.5.2 Loading Host Segment and Descriptor-Table Registers */
10886 seg = (struct kvm_segment) {
10888 .limit = 0xFFFFFFFF,
10889 .selector = vmcs12->host_cs_selector,
10895 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10899 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
10900 seg = (struct kvm_segment) {
10902 .limit = 0xFFFFFFFF,
10909 seg.selector = vmcs12->host_ds_selector;
10910 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
10911 seg.selector = vmcs12->host_es_selector;
10912 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
10913 seg.selector = vmcs12->host_ss_selector;
10914 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
10915 seg.selector = vmcs12->host_fs_selector;
10916 seg.base = vmcs12->host_fs_base;
10917 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
10918 seg.selector = vmcs12->host_gs_selector;
10919 seg.base = vmcs12->host_gs_base;
10920 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
10921 seg = (struct kvm_segment) {
10922 .base = vmcs12->host_tr_base,
10924 .selector = vmcs12->host_tr_selector,
10928 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
10930 kvm_set_dr(vcpu, 7, 0x400);
10931 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
10933 if (cpu_has_vmx_msr_bitmap())
10934 vmx_set_msr_bitmap(vcpu);
10936 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
10937 vmcs12->vm_exit_msr_load_count))
10938 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
10942 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
10943 * and modify vmcs12 to make it see what it would expect to see there if
10944 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
10946 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
10947 u32 exit_intr_info,
10948 unsigned long exit_qualification)
10950 struct vcpu_vmx *vmx = to_vmx(vcpu);
10951 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10952 u32 vm_inst_error = 0;
10954 /* trying to cancel vmlaunch/vmresume is a bug */
10955 WARN_ON_ONCE(vmx->nested.nested_run_pending);
10957 leave_guest_mode(vcpu);
10958 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
10959 exit_qualification);
10961 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
10962 vmcs12->vm_exit_msr_store_count))
10963 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
10965 if (unlikely(vmx->fail))
10966 vm_inst_error = vmcs_read32(VM_INSTRUCTION_ERROR);
10968 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10970 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
10971 && nested_exit_intr_ack_set(vcpu)) {
10972 int irq = kvm_cpu_get_interrupt(vcpu);
10974 vmcs12->vm_exit_intr_info = irq |
10975 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
10978 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
10979 vmcs12->exit_qualification,
10980 vmcs12->idt_vectoring_info_field,
10981 vmcs12->vm_exit_intr_info,
10982 vmcs12->vm_exit_intr_error_code,
10985 vm_entry_controls_reset_shadow(vmx);
10986 vm_exit_controls_reset_shadow(vmx);
10987 vmx_segment_cache_clear(vmx);
10989 /* if no vmcs02 cache requested, remove the one we used */
10990 if (VMCS02_POOL_SIZE == 0)
10991 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
10993 load_vmcs12_host_state(vcpu, vmcs12);
10995 /* Update any VMCS fields that might have changed while L2 ran */
10996 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10997 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10998 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
10999 if (vmx->hv_deadline_tsc == -1)
11000 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11001 PIN_BASED_VMX_PREEMPTION_TIMER);
11003 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11004 PIN_BASED_VMX_PREEMPTION_TIMER);
11005 if (kvm_has_tsc_control)
11006 decache_tsc_multiplier(vmx);
11008 if (vmx->nested.change_vmcs01_virtual_x2apic_mode) {
11009 vmx->nested.change_vmcs01_virtual_x2apic_mode = false;
11010 vmx_set_virtual_x2apic_mode(vcpu,
11011 vcpu->arch.apic_base & X2APIC_ENABLE);
11012 } else if (!nested_cpu_has_ept(vmcs12) &&
11013 nested_cpu_has2(vmcs12,
11014 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
11015 vmx_flush_tlb_ept_only(vcpu);
11018 /* This is needed for same reason as it was needed in prepare_vmcs02 */
11021 /* Unpin physical memory we referred to in vmcs02 */
11022 if (vmx->nested.apic_access_page) {
11023 nested_release_page(vmx->nested.apic_access_page);
11024 vmx->nested.apic_access_page = NULL;
11026 if (vmx->nested.virtual_apic_page) {
11027 nested_release_page(vmx->nested.virtual_apic_page);
11028 vmx->nested.virtual_apic_page = NULL;
11030 if (vmx->nested.pi_desc_page) {
11031 kunmap(vmx->nested.pi_desc_page);
11032 nested_release_page(vmx->nested.pi_desc_page);
11033 vmx->nested.pi_desc_page = NULL;
11034 vmx->nested.pi_desc = NULL;
11038 * We are now running in L2, mmu_notifier will force to reload the
11039 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
11041 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
11044 * Exiting from L2 to L1, we're now back to L1 which thinks it just
11045 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
11046 * success or failure flag accordingly.
11048 if (unlikely(vmx->fail)) {
11050 nested_vmx_failValid(vcpu, vm_inst_error);
11052 nested_vmx_succeed(vcpu);
11053 if (enable_shadow_vmcs)
11054 vmx->nested.sync_shadow_vmcs = true;
11056 /* in case we halted in L2 */
11057 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11061 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
11063 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
11065 if (is_guest_mode(vcpu)) {
11066 to_vmx(vcpu)->nested.nested_run_pending = 0;
11067 nested_vmx_vmexit(vcpu, -1, 0, 0);
11069 free_nested(to_vmx(vcpu));
11073 * L1's failure to enter L2 is a subset of a normal exit, as explained in
11074 * 23.7 "VM-entry failures during or after loading guest state" (this also
11075 * lists the acceptable exit-reason and exit-qualification parameters).
11076 * It should only be called before L2 actually succeeded to run, and when
11077 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
11079 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
11080 struct vmcs12 *vmcs12,
11081 u32 reason, unsigned long qualification)
11083 load_vmcs12_host_state(vcpu, vmcs12);
11084 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
11085 vmcs12->exit_qualification = qualification;
11086 nested_vmx_succeed(vcpu);
11087 if (enable_shadow_vmcs)
11088 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
11091 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
11092 struct x86_instruction_info *info,
11093 enum x86_intercept_stage stage)
11095 return X86EMUL_CONTINUE;
11098 #ifdef CONFIG_X86_64
11099 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
11100 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
11101 u64 divisor, u64 *result)
11103 u64 low = a << shift, high = a >> (64 - shift);
11105 /* To avoid the overflow on divq */
11106 if (high >= divisor)
11109 /* Low hold the result, high hold rem which is discarded */
11110 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
11111 "rm" (divisor), "0" (low), "1" (high));
11117 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
11119 struct vcpu_vmx *vmx = to_vmx(vcpu);
11120 u64 tscl = rdtsc();
11121 u64 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
11122 u64 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
11124 /* Convert to host delta tsc if tsc scaling is enabled */
11125 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
11126 u64_shl_div_u64(delta_tsc,
11127 kvm_tsc_scaling_ratio_frac_bits,
11128 vcpu->arch.tsc_scaling_ratio,
11133 * If the delta tsc can't fit in the 32 bit after the multi shift,
11134 * we can't use the preemption timer.
11135 * It's possible that it fits on later vmentries, but checking
11136 * on every vmentry is costly so we just use an hrtimer.
11138 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
11141 vmx->hv_deadline_tsc = tscl + delta_tsc;
11142 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11143 PIN_BASED_VMX_PREEMPTION_TIMER);
11147 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
11149 struct vcpu_vmx *vmx = to_vmx(vcpu);
11150 vmx->hv_deadline_tsc = -1;
11151 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11152 PIN_BASED_VMX_PREEMPTION_TIMER);
11156 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
11159 shrink_ple_window(vcpu);
11162 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
11163 struct kvm_memory_slot *slot)
11165 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
11166 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
11169 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
11170 struct kvm_memory_slot *slot)
11172 kvm_mmu_slot_set_dirty(kvm, slot);
11175 static void vmx_flush_log_dirty(struct kvm *kvm)
11177 kvm_flush_pml_buffers(kvm);
11180 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
11181 struct kvm_memory_slot *memslot,
11182 gfn_t offset, unsigned long mask)
11184 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
11188 * This routine does the following things for vCPU which is going
11189 * to be blocked if VT-d PI is enabled.
11190 * - Store the vCPU to the wakeup list, so when interrupts happen
11191 * we can find the right vCPU to wake up.
11192 * - Change the Posted-interrupt descriptor as below:
11193 * 'NDST' <-- vcpu->pre_pcpu
11194 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
11195 * - If 'ON' is set during this process, which means at least one
11196 * interrupt is posted for this vCPU, we cannot block it, in
11197 * this case, return 1, otherwise, return 0.
11200 static int pi_pre_block(struct kvm_vcpu *vcpu)
11202 unsigned long flags;
11204 struct pi_desc old, new;
11205 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11207 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11208 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11209 !kvm_vcpu_apicv_active(vcpu))
11212 vcpu->pre_pcpu = vcpu->cpu;
11213 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
11214 vcpu->pre_pcpu), flags);
11215 list_add_tail(&vcpu->blocked_vcpu_list,
11216 &per_cpu(blocked_vcpu_on_cpu,
11218 spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
11219 vcpu->pre_pcpu), flags);
11222 old.control = new.control = pi_desc->control;
11225 * We should not block the vCPU if
11226 * an interrupt is posted for it.
11228 if (pi_test_on(pi_desc) == 1) {
11229 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
11230 vcpu->pre_pcpu), flags);
11231 list_del(&vcpu->blocked_vcpu_list);
11232 spin_unlock_irqrestore(
11233 &per_cpu(blocked_vcpu_on_cpu_lock,
11234 vcpu->pre_pcpu), flags);
11235 vcpu->pre_pcpu = -1;
11240 WARN((pi_desc->sn == 1),
11241 "Warning: SN field of posted-interrupts "
11242 "is set before blocking\n");
11245 * Since vCPU can be preempted during this process,
11246 * vcpu->cpu could be different with pre_pcpu, we
11247 * need to set pre_pcpu as the destination of wakeup
11248 * notification event, then we can find the right vCPU
11249 * to wakeup in wakeup handler if interrupts happen
11250 * when the vCPU is in blocked state.
11252 dest = cpu_physical_id(vcpu->pre_pcpu);
11254 if (x2apic_enabled())
11257 new.ndst = (dest << 8) & 0xFF00;
11259 /* set 'NV' to 'wakeup vector' */
11260 new.nv = POSTED_INTR_WAKEUP_VECTOR;
11261 } while (cmpxchg(&pi_desc->control, old.control,
11262 new.control) != old.control);
11267 static int vmx_pre_block(struct kvm_vcpu *vcpu)
11269 if (pi_pre_block(vcpu))
11272 if (kvm_lapic_hv_timer_in_use(vcpu))
11273 kvm_lapic_switch_to_sw_timer(vcpu);
11278 static void pi_post_block(struct kvm_vcpu *vcpu)
11280 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11281 struct pi_desc old, new;
11283 unsigned long flags;
11285 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11286 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11287 !kvm_vcpu_apicv_active(vcpu))
11291 old.control = new.control = pi_desc->control;
11293 dest = cpu_physical_id(vcpu->cpu);
11295 if (x2apic_enabled())
11298 new.ndst = (dest << 8) & 0xFF00;
11300 /* Allow posting non-urgent interrupts */
11303 /* set 'NV' to 'notification vector' */
11304 new.nv = POSTED_INTR_VECTOR;
11305 } while (cmpxchg(&pi_desc->control, old.control,
11306 new.control) != old.control);
11308 if(vcpu->pre_pcpu != -1) {
11310 &per_cpu(blocked_vcpu_on_cpu_lock,
11311 vcpu->pre_pcpu), flags);
11312 list_del(&vcpu->blocked_vcpu_list);
11313 spin_unlock_irqrestore(
11314 &per_cpu(blocked_vcpu_on_cpu_lock,
11315 vcpu->pre_pcpu), flags);
11316 vcpu->pre_pcpu = -1;
11320 static void vmx_post_block(struct kvm_vcpu *vcpu)
11322 if (kvm_x86_ops->set_hv_timer)
11323 kvm_lapic_switch_to_hv_timer(vcpu);
11325 pi_post_block(vcpu);
11329 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
11332 * @host_irq: host irq of the interrupt
11333 * @guest_irq: gsi of the interrupt
11334 * @set: set or unset PI
11335 * returns 0 on success, < 0 on failure
11337 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
11338 uint32_t guest_irq, bool set)
11340 struct kvm_kernel_irq_routing_entry *e;
11341 struct kvm_irq_routing_table *irq_rt;
11342 struct kvm_lapic_irq irq;
11343 struct kvm_vcpu *vcpu;
11344 struct vcpu_data vcpu_info;
11345 int idx, ret = -EINVAL;
11347 if (!kvm_arch_has_assigned_device(kvm) ||
11348 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11349 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
11352 idx = srcu_read_lock(&kvm->irq_srcu);
11353 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
11354 BUG_ON(guest_irq >= irq_rt->nr_rt_entries);
11356 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
11357 if (e->type != KVM_IRQ_ROUTING_MSI)
11360 * VT-d PI cannot support posting multicast/broadcast
11361 * interrupts to a vCPU, we still use interrupt remapping
11362 * for these kind of interrupts.
11364 * For lowest-priority interrupts, we only support
11365 * those with single CPU as the destination, e.g. user
11366 * configures the interrupts via /proc/irq or uses
11367 * irqbalance to make the interrupts single-CPU.
11369 * We will support full lowest-priority interrupt later.
11372 kvm_set_msi_irq(kvm, e, &irq);
11373 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
11375 * Make sure the IRTE is in remapped mode if
11376 * we don't handle it in posted mode.
11378 ret = irq_set_vcpu_affinity(host_irq, NULL);
11381 "failed to back to remapped mode, irq: %u\n",
11389 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
11390 vcpu_info.vector = irq.vector;
11392 trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
11393 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
11396 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
11398 /* suppress notification event before unposting */
11399 pi_set_sn(vcpu_to_pi_desc(vcpu));
11400 ret = irq_set_vcpu_affinity(host_irq, NULL);
11401 pi_clear_sn(vcpu_to_pi_desc(vcpu));
11405 printk(KERN_INFO "%s: failed to update PI IRTE\n",
11413 srcu_read_unlock(&kvm->irq_srcu, idx);
11417 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
11419 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
11420 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
11421 FEATURE_CONTROL_LMCE;
11423 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
11424 ~FEATURE_CONTROL_LMCE;
11427 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
11428 .cpu_has_kvm_support = cpu_has_kvm_support,
11429 .disabled_by_bios = vmx_disabled_by_bios,
11430 .hardware_setup = hardware_setup,
11431 .hardware_unsetup = hardware_unsetup,
11432 .check_processor_compatibility = vmx_check_processor_compat,
11433 .hardware_enable = hardware_enable,
11434 .hardware_disable = hardware_disable,
11435 .cpu_has_accelerated_tpr = report_flexpriority,
11436 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
11438 .vcpu_create = vmx_create_vcpu,
11439 .vcpu_free = vmx_free_vcpu,
11440 .vcpu_reset = vmx_vcpu_reset,
11442 .prepare_guest_switch = vmx_save_host_state,
11443 .vcpu_load = vmx_vcpu_load,
11444 .vcpu_put = vmx_vcpu_put,
11446 .update_bp_intercept = update_exception_bitmap,
11447 .get_msr = vmx_get_msr,
11448 .set_msr = vmx_set_msr,
11449 .get_segment_base = vmx_get_segment_base,
11450 .get_segment = vmx_get_segment,
11451 .set_segment = vmx_set_segment,
11452 .get_cpl = vmx_get_cpl,
11453 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
11454 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
11455 .decache_cr3 = vmx_decache_cr3,
11456 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
11457 .set_cr0 = vmx_set_cr0,
11458 .set_cr3 = vmx_set_cr3,
11459 .set_cr4 = vmx_set_cr4,
11460 .set_efer = vmx_set_efer,
11461 .get_idt = vmx_get_idt,
11462 .set_idt = vmx_set_idt,
11463 .get_gdt = vmx_get_gdt,
11464 .set_gdt = vmx_set_gdt,
11465 .get_dr6 = vmx_get_dr6,
11466 .set_dr6 = vmx_set_dr6,
11467 .set_dr7 = vmx_set_dr7,
11468 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
11469 .cache_reg = vmx_cache_reg,
11470 .get_rflags = vmx_get_rflags,
11471 .set_rflags = vmx_set_rflags,
11473 .get_pkru = vmx_get_pkru,
11475 .tlb_flush = vmx_flush_tlb,
11477 .run = vmx_vcpu_run,
11478 .handle_exit = vmx_handle_exit,
11479 .skip_emulated_instruction = skip_emulated_instruction,
11480 .set_interrupt_shadow = vmx_set_interrupt_shadow,
11481 .get_interrupt_shadow = vmx_get_interrupt_shadow,
11482 .patch_hypercall = vmx_patch_hypercall,
11483 .set_irq = vmx_inject_irq,
11484 .set_nmi = vmx_inject_nmi,
11485 .queue_exception = vmx_queue_exception,
11486 .cancel_injection = vmx_cancel_injection,
11487 .interrupt_allowed = vmx_interrupt_allowed,
11488 .nmi_allowed = vmx_nmi_allowed,
11489 .get_nmi_mask = vmx_get_nmi_mask,
11490 .set_nmi_mask = vmx_set_nmi_mask,
11491 .enable_nmi_window = enable_nmi_window,
11492 .enable_irq_window = enable_irq_window,
11493 .update_cr8_intercept = update_cr8_intercept,
11494 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
11495 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
11496 .get_enable_apicv = vmx_get_enable_apicv,
11497 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
11498 .load_eoi_exitmap = vmx_load_eoi_exitmap,
11499 .apicv_post_state_restore = vmx_apicv_post_state_restore,
11500 .hwapic_irr_update = vmx_hwapic_irr_update,
11501 .hwapic_isr_update = vmx_hwapic_isr_update,
11502 .sync_pir_to_irr = vmx_sync_pir_to_irr,
11503 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
11505 .set_tss_addr = vmx_set_tss_addr,
11506 .get_tdp_level = get_ept_level,
11507 .get_mt_mask = vmx_get_mt_mask,
11509 .get_exit_info = vmx_get_exit_info,
11511 .get_lpage_level = vmx_get_lpage_level,
11513 .cpuid_update = vmx_cpuid_update,
11515 .rdtscp_supported = vmx_rdtscp_supported,
11516 .invpcid_supported = vmx_invpcid_supported,
11518 .set_supported_cpuid = vmx_set_supported_cpuid,
11520 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
11522 .write_tsc_offset = vmx_write_tsc_offset,
11524 .set_tdp_cr3 = vmx_set_cr3,
11526 .check_intercept = vmx_check_intercept,
11527 .handle_external_intr = vmx_handle_external_intr,
11528 .mpx_supported = vmx_mpx_supported,
11529 .xsaves_supported = vmx_xsaves_supported,
11531 .check_nested_events = vmx_check_nested_events,
11533 .sched_in = vmx_sched_in,
11535 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
11536 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
11537 .flush_log_dirty = vmx_flush_log_dirty,
11538 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
11540 .pre_block = vmx_pre_block,
11541 .post_block = vmx_post_block,
11543 .pmu_ops = &intel_pmu_ops,
11545 .update_pi_irte = vmx_update_pi_irte,
11547 #ifdef CONFIG_X86_64
11548 .set_hv_timer = vmx_set_hv_timer,
11549 .cancel_hv_timer = vmx_cancel_hv_timer,
11552 .setup_mce = vmx_setup_mce,
11555 static int __init vmx_init(void)
11557 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
11558 __alignof__(struct vcpu_vmx), THIS_MODULE);
11562 #ifdef CONFIG_KEXEC_CORE
11563 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
11564 crash_vmclear_local_loaded_vmcss);
11570 static void __exit vmx_exit(void)
11572 #ifdef CONFIG_KEXEC_CORE
11573 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
11580 module_init(vmx_init)
11581 module_exit(vmx_exit)