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_2MB_PAGE_BIT |
2764 if (cpu_has_vmx_ept_execute_only())
2765 vmx->nested.nested_vmx_ept_caps |=
2766 VMX_EPT_EXECUTE_ONLY_BIT;
2767 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2768 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
2769 VMX_EPT_EXTENT_CONTEXT_BIT;
2771 vmx->nested.nested_vmx_ept_caps = 0;
2774 * Old versions of KVM use the single-context version without
2775 * checking for support, so declare that it is supported even
2776 * though it is treated as global context. The alternative is
2777 * not failing the single-context invvpid, and it is worse.
2780 vmx->nested.nested_vmx_secondary_ctls_high |=
2781 SECONDARY_EXEC_ENABLE_VPID;
2782 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2783 VMX_VPID_EXTENT_SUPPORTED_MASK;
2785 vmx->nested.nested_vmx_vpid_caps = 0;
2787 if (enable_unrestricted_guest)
2788 vmx->nested.nested_vmx_secondary_ctls_high |=
2789 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2791 /* miscellaneous data */
2792 rdmsr(MSR_IA32_VMX_MISC,
2793 vmx->nested.nested_vmx_misc_low,
2794 vmx->nested.nested_vmx_misc_high);
2795 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2796 vmx->nested.nested_vmx_misc_low |=
2797 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2798 VMX_MISC_ACTIVITY_HLT;
2799 vmx->nested.nested_vmx_misc_high = 0;
2802 * This MSR reports some information about VMX support. We
2803 * should return information about the VMX we emulate for the
2804 * guest, and the VMCS structure we give it - not about the
2805 * VMX support of the underlying hardware.
2807 vmx->nested.nested_vmx_basic =
2809 VMX_BASIC_TRUE_CTLS |
2810 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2811 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2813 if (cpu_has_vmx_basic_inout())
2814 vmx->nested.nested_vmx_basic |= VMX_BASIC_INOUT;
2817 * These MSRs specify bits which the guest must keep fixed on
2818 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2819 * We picked the standard core2 setting.
2821 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2822 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2823 vmx->nested.nested_vmx_cr0_fixed0 = VMXON_CR0_ALWAYSON;
2824 vmx->nested.nested_vmx_cr4_fixed0 = VMXON_CR4_ALWAYSON;
2826 /* These MSRs specify bits which the guest must keep fixed off. */
2827 rdmsrl(MSR_IA32_VMX_CR0_FIXED1, vmx->nested.nested_vmx_cr0_fixed1);
2828 rdmsrl(MSR_IA32_VMX_CR4_FIXED1, vmx->nested.nested_vmx_cr4_fixed1);
2830 /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2831 vmx->nested.nested_vmx_vmcs_enum = 0x2e;
2835 * if fixed0[i] == 1: val[i] must be 1
2836 * if fixed1[i] == 0: val[i] must be 0
2838 static inline bool fixed_bits_valid(u64 val, u64 fixed0, u64 fixed1)
2840 return ((val & fixed1) | fixed0) == val;
2843 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2845 return fixed_bits_valid(control, low, high);
2848 static inline u64 vmx_control_msr(u32 low, u32 high)
2850 return low | ((u64)high << 32);
2853 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
2858 return (superset | subset) == superset;
2861 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
2863 const u64 feature_and_reserved =
2864 /* feature (except bit 48; see below) */
2865 BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
2867 BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
2868 u64 vmx_basic = vmx->nested.nested_vmx_basic;
2870 if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
2874 * KVM does not emulate a version of VMX that constrains physical
2875 * addresses of VMX structures (e.g. VMCS) to 32-bits.
2877 if (data & BIT_ULL(48))
2880 if (vmx_basic_vmcs_revision_id(vmx_basic) !=
2881 vmx_basic_vmcs_revision_id(data))
2884 if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
2887 vmx->nested.nested_vmx_basic = data;
2892 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
2897 switch (msr_index) {
2898 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2899 lowp = &vmx->nested.nested_vmx_pinbased_ctls_low;
2900 highp = &vmx->nested.nested_vmx_pinbased_ctls_high;
2902 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2903 lowp = &vmx->nested.nested_vmx_procbased_ctls_low;
2904 highp = &vmx->nested.nested_vmx_procbased_ctls_high;
2906 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2907 lowp = &vmx->nested.nested_vmx_exit_ctls_low;
2908 highp = &vmx->nested.nested_vmx_exit_ctls_high;
2910 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2911 lowp = &vmx->nested.nested_vmx_entry_ctls_low;
2912 highp = &vmx->nested.nested_vmx_entry_ctls_high;
2914 case MSR_IA32_VMX_PROCBASED_CTLS2:
2915 lowp = &vmx->nested.nested_vmx_secondary_ctls_low;
2916 highp = &vmx->nested.nested_vmx_secondary_ctls_high;
2922 supported = vmx_control_msr(*lowp, *highp);
2924 /* Check must-be-1 bits are still 1. */
2925 if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
2928 /* Check must-be-0 bits are still 0. */
2929 if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
2933 *highp = data >> 32;
2937 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
2939 const u64 feature_and_reserved_bits =
2941 BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
2942 BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
2944 GENMASK_ULL(13, 9) | BIT_ULL(31);
2947 vmx_misc = vmx_control_msr(vmx->nested.nested_vmx_misc_low,
2948 vmx->nested.nested_vmx_misc_high);
2950 if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
2953 if ((vmx->nested.nested_vmx_pinbased_ctls_high &
2954 PIN_BASED_VMX_PREEMPTION_TIMER) &&
2955 vmx_misc_preemption_timer_rate(data) !=
2956 vmx_misc_preemption_timer_rate(vmx_misc))
2959 if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
2962 if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
2965 if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
2968 vmx->nested.nested_vmx_misc_low = data;
2969 vmx->nested.nested_vmx_misc_high = data >> 32;
2973 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
2975 u64 vmx_ept_vpid_cap;
2977 vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.nested_vmx_ept_caps,
2978 vmx->nested.nested_vmx_vpid_caps);
2980 /* Every bit is either reserved or a feature bit. */
2981 if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
2984 vmx->nested.nested_vmx_ept_caps = data;
2985 vmx->nested.nested_vmx_vpid_caps = data >> 32;
2989 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
2993 switch (msr_index) {
2994 case MSR_IA32_VMX_CR0_FIXED0:
2995 msr = &vmx->nested.nested_vmx_cr0_fixed0;
2997 case MSR_IA32_VMX_CR4_FIXED0:
2998 msr = &vmx->nested.nested_vmx_cr4_fixed0;
3005 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
3006 * must be 1 in the restored value.
3008 if (!is_bitwise_subset(data, *msr, -1ULL))
3016 * Called when userspace is restoring VMX MSRs.
3018 * Returns 0 on success, non-0 otherwise.
3020 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
3022 struct vcpu_vmx *vmx = to_vmx(vcpu);
3024 switch (msr_index) {
3025 case MSR_IA32_VMX_BASIC:
3026 return vmx_restore_vmx_basic(vmx, data);
3027 case MSR_IA32_VMX_PINBASED_CTLS:
3028 case MSR_IA32_VMX_PROCBASED_CTLS:
3029 case MSR_IA32_VMX_EXIT_CTLS:
3030 case MSR_IA32_VMX_ENTRY_CTLS:
3032 * The "non-true" VMX capability MSRs are generated from the
3033 * "true" MSRs, so we do not support restoring them directly.
3035 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
3036 * should restore the "true" MSRs with the must-be-1 bits
3037 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
3038 * DEFAULT SETTINGS".
3041 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3042 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3043 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3044 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3045 case MSR_IA32_VMX_PROCBASED_CTLS2:
3046 return vmx_restore_control_msr(vmx, msr_index, data);
3047 case MSR_IA32_VMX_MISC:
3048 return vmx_restore_vmx_misc(vmx, data);
3049 case MSR_IA32_VMX_CR0_FIXED0:
3050 case MSR_IA32_VMX_CR4_FIXED0:
3051 return vmx_restore_fixed0_msr(vmx, msr_index, data);
3052 case MSR_IA32_VMX_CR0_FIXED1:
3053 case MSR_IA32_VMX_CR4_FIXED1:
3055 * These MSRs are generated based on the vCPU's CPUID, so we
3056 * do not support restoring them directly.
3059 case MSR_IA32_VMX_EPT_VPID_CAP:
3060 return vmx_restore_vmx_ept_vpid_cap(vmx, data);
3061 case MSR_IA32_VMX_VMCS_ENUM:
3062 vmx->nested.nested_vmx_vmcs_enum = data;
3066 * The rest of the VMX capability MSRs do not support restore.
3072 /* Returns 0 on success, non-0 otherwise. */
3073 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
3075 struct vcpu_vmx *vmx = to_vmx(vcpu);
3077 switch (msr_index) {
3078 case MSR_IA32_VMX_BASIC:
3079 *pdata = vmx->nested.nested_vmx_basic;
3081 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3082 case MSR_IA32_VMX_PINBASED_CTLS:
3083 *pdata = vmx_control_msr(
3084 vmx->nested.nested_vmx_pinbased_ctls_low,
3085 vmx->nested.nested_vmx_pinbased_ctls_high);
3086 if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
3087 *pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3089 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3090 case MSR_IA32_VMX_PROCBASED_CTLS:
3091 *pdata = vmx_control_msr(
3092 vmx->nested.nested_vmx_procbased_ctls_low,
3093 vmx->nested.nested_vmx_procbased_ctls_high);
3094 if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
3095 *pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3097 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3098 case MSR_IA32_VMX_EXIT_CTLS:
3099 *pdata = vmx_control_msr(
3100 vmx->nested.nested_vmx_exit_ctls_low,
3101 vmx->nested.nested_vmx_exit_ctls_high);
3102 if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
3103 *pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
3105 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3106 case MSR_IA32_VMX_ENTRY_CTLS:
3107 *pdata = vmx_control_msr(
3108 vmx->nested.nested_vmx_entry_ctls_low,
3109 vmx->nested.nested_vmx_entry_ctls_high);
3110 if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
3111 *pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
3113 case MSR_IA32_VMX_MISC:
3114 *pdata = vmx_control_msr(
3115 vmx->nested.nested_vmx_misc_low,
3116 vmx->nested.nested_vmx_misc_high);
3118 case MSR_IA32_VMX_CR0_FIXED0:
3119 *pdata = vmx->nested.nested_vmx_cr0_fixed0;
3121 case MSR_IA32_VMX_CR0_FIXED1:
3122 *pdata = vmx->nested.nested_vmx_cr0_fixed1;
3124 case MSR_IA32_VMX_CR4_FIXED0:
3125 *pdata = vmx->nested.nested_vmx_cr4_fixed0;
3127 case MSR_IA32_VMX_CR4_FIXED1:
3128 *pdata = vmx->nested.nested_vmx_cr4_fixed1;
3130 case MSR_IA32_VMX_VMCS_ENUM:
3131 *pdata = vmx->nested.nested_vmx_vmcs_enum;
3133 case MSR_IA32_VMX_PROCBASED_CTLS2:
3134 *pdata = vmx_control_msr(
3135 vmx->nested.nested_vmx_secondary_ctls_low,
3136 vmx->nested.nested_vmx_secondary_ctls_high);
3138 case MSR_IA32_VMX_EPT_VPID_CAP:
3139 *pdata = vmx->nested.nested_vmx_ept_caps |
3140 ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
3149 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
3152 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
3154 return !(val & ~valid_bits);
3158 * Reads an msr value (of 'msr_index') into 'pdata'.
3159 * Returns 0 on success, non-0 otherwise.
3160 * Assumes vcpu_load() was already called.
3162 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3164 struct shared_msr_entry *msr;
3166 switch (msr_info->index) {
3167 #ifdef CONFIG_X86_64
3169 msr_info->data = vmcs_readl(GUEST_FS_BASE);
3172 msr_info->data = vmcs_readl(GUEST_GS_BASE);
3174 case MSR_KERNEL_GS_BASE:
3175 vmx_load_host_state(to_vmx(vcpu));
3176 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
3180 return kvm_get_msr_common(vcpu, msr_info);
3182 msr_info->data = guest_read_tsc(vcpu);
3184 case MSR_IA32_SYSENTER_CS:
3185 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
3187 case MSR_IA32_SYSENTER_EIP:
3188 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
3190 case MSR_IA32_SYSENTER_ESP:
3191 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
3193 case MSR_IA32_BNDCFGS:
3194 if (!kvm_mpx_supported())
3196 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
3198 case MSR_IA32_MCG_EXT_CTL:
3199 if (!msr_info->host_initiated &&
3200 !(to_vmx(vcpu)->msr_ia32_feature_control &
3201 FEATURE_CONTROL_LMCE))
3203 msr_info->data = vcpu->arch.mcg_ext_ctl;
3205 case MSR_IA32_FEATURE_CONTROL:
3206 msr_info->data = to_vmx(vcpu)->msr_ia32_feature_control;
3208 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3209 if (!nested_vmx_allowed(vcpu))
3211 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
3213 if (!vmx_xsaves_supported())
3215 msr_info->data = vcpu->arch.ia32_xss;
3218 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3220 /* Otherwise falls through */
3222 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
3224 msr_info->data = msr->data;
3227 return kvm_get_msr_common(vcpu, msr_info);
3233 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3236 * Writes msr value into into the appropriate "register".
3237 * Returns 0 on success, non-0 otherwise.
3238 * Assumes vcpu_load() was already called.
3240 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3242 struct vcpu_vmx *vmx = to_vmx(vcpu);
3243 struct shared_msr_entry *msr;
3245 u32 msr_index = msr_info->index;
3246 u64 data = msr_info->data;
3248 switch (msr_index) {
3250 ret = kvm_set_msr_common(vcpu, msr_info);
3252 #ifdef CONFIG_X86_64
3254 vmx_segment_cache_clear(vmx);
3255 vmcs_writel(GUEST_FS_BASE, data);
3258 vmx_segment_cache_clear(vmx);
3259 vmcs_writel(GUEST_GS_BASE, data);
3261 case MSR_KERNEL_GS_BASE:
3262 vmx_load_host_state(vmx);
3263 vmx->msr_guest_kernel_gs_base = data;
3266 case MSR_IA32_SYSENTER_CS:
3267 vmcs_write32(GUEST_SYSENTER_CS, data);
3269 case MSR_IA32_SYSENTER_EIP:
3270 vmcs_writel(GUEST_SYSENTER_EIP, data);
3272 case MSR_IA32_SYSENTER_ESP:
3273 vmcs_writel(GUEST_SYSENTER_ESP, data);
3275 case MSR_IA32_BNDCFGS:
3276 if (!kvm_mpx_supported())
3278 vmcs_write64(GUEST_BNDCFGS, data);
3281 kvm_write_tsc(vcpu, msr_info);
3283 case MSR_IA32_CR_PAT:
3284 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3285 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3287 vmcs_write64(GUEST_IA32_PAT, data);
3288 vcpu->arch.pat = data;
3291 ret = kvm_set_msr_common(vcpu, msr_info);
3293 case MSR_IA32_TSC_ADJUST:
3294 ret = kvm_set_msr_common(vcpu, msr_info);
3296 case MSR_IA32_MCG_EXT_CTL:
3297 if ((!msr_info->host_initiated &&
3298 !(to_vmx(vcpu)->msr_ia32_feature_control &
3299 FEATURE_CONTROL_LMCE)) ||
3300 (data & ~MCG_EXT_CTL_LMCE_EN))
3302 vcpu->arch.mcg_ext_ctl = data;
3304 case MSR_IA32_FEATURE_CONTROL:
3305 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3306 (to_vmx(vcpu)->msr_ia32_feature_control &
3307 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3309 vmx->msr_ia32_feature_control = data;
3310 if (msr_info->host_initiated && data == 0)
3311 vmx_leave_nested(vcpu);
3313 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3314 if (!msr_info->host_initiated)
3315 return 1; /* they are read-only */
3316 if (!nested_vmx_allowed(vcpu))
3318 return vmx_set_vmx_msr(vcpu, msr_index, data);
3320 if (!vmx_xsaves_supported())
3323 * The only supported bit as of Skylake is bit 8, but
3324 * it is not supported on KVM.
3328 vcpu->arch.ia32_xss = data;
3329 if (vcpu->arch.ia32_xss != host_xss)
3330 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3331 vcpu->arch.ia32_xss, host_xss);
3333 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3336 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3338 /* Check reserved bit, higher 32 bits should be zero */
3339 if ((data >> 32) != 0)
3341 /* Otherwise falls through */
3343 msr = find_msr_entry(vmx, msr_index);
3345 u64 old_msr_data = msr->data;
3347 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3349 ret = kvm_set_shared_msr(msr->index, msr->data,
3353 msr->data = old_msr_data;
3357 ret = kvm_set_msr_common(vcpu, msr_info);
3363 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3365 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3368 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3371 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3373 case VCPU_EXREG_PDPTR:
3375 ept_save_pdptrs(vcpu);
3382 static __init int cpu_has_kvm_support(void)
3384 return cpu_has_vmx();
3387 static __init int vmx_disabled_by_bios(void)
3391 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3392 if (msr & FEATURE_CONTROL_LOCKED) {
3393 /* launched w/ TXT and VMX disabled */
3394 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3397 /* launched w/o TXT and VMX only enabled w/ TXT */
3398 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3399 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3400 && !tboot_enabled()) {
3401 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3402 "activate TXT before enabling KVM\n");
3405 /* launched w/o TXT and VMX disabled */
3406 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3407 && !tboot_enabled())
3414 static void kvm_cpu_vmxon(u64 addr)
3416 intel_pt_handle_vmx(1);
3418 asm volatile (ASM_VMX_VMXON_RAX
3419 : : "a"(&addr), "m"(addr)
3423 static int hardware_enable(void)
3425 int cpu = raw_smp_processor_id();
3426 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3429 if (cr4_read_shadow() & X86_CR4_VMXE)
3432 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3433 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3434 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3437 * Now we can enable the vmclear operation in kdump
3438 * since the loaded_vmcss_on_cpu list on this cpu
3439 * has been initialized.
3441 * Though the cpu is not in VMX operation now, there
3442 * is no problem to enable the vmclear operation
3443 * for the loaded_vmcss_on_cpu list is empty!
3445 crash_enable_local_vmclear(cpu);
3447 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3449 test_bits = FEATURE_CONTROL_LOCKED;
3450 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3451 if (tboot_enabled())
3452 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3454 if ((old & test_bits) != test_bits) {
3455 /* enable and lock */
3456 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3458 cr4_set_bits(X86_CR4_VMXE);
3460 if (vmm_exclusive) {
3461 kvm_cpu_vmxon(phys_addr);
3465 native_store_gdt(this_cpu_ptr(&host_gdt));
3470 static void vmclear_local_loaded_vmcss(void)
3472 int cpu = raw_smp_processor_id();
3473 struct loaded_vmcs *v, *n;
3475 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3476 loaded_vmcss_on_cpu_link)
3477 __loaded_vmcs_clear(v);
3481 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3484 static void kvm_cpu_vmxoff(void)
3486 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3488 intel_pt_handle_vmx(0);
3491 static void hardware_disable(void)
3493 if (vmm_exclusive) {
3494 vmclear_local_loaded_vmcss();
3497 cr4_clear_bits(X86_CR4_VMXE);
3500 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3501 u32 msr, u32 *result)
3503 u32 vmx_msr_low, vmx_msr_high;
3504 u32 ctl = ctl_min | ctl_opt;
3506 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3508 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3509 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3511 /* Ensure minimum (required) set of control bits are supported. */
3519 static __init bool allow_1_setting(u32 msr, u32 ctl)
3521 u32 vmx_msr_low, vmx_msr_high;
3523 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3524 return vmx_msr_high & ctl;
3527 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3529 u32 vmx_msr_low, vmx_msr_high;
3530 u32 min, opt, min2, opt2;
3531 u32 _pin_based_exec_control = 0;
3532 u32 _cpu_based_exec_control = 0;
3533 u32 _cpu_based_2nd_exec_control = 0;
3534 u32 _vmexit_control = 0;
3535 u32 _vmentry_control = 0;
3537 min = CPU_BASED_HLT_EXITING |
3538 #ifdef CONFIG_X86_64
3539 CPU_BASED_CR8_LOAD_EXITING |
3540 CPU_BASED_CR8_STORE_EXITING |
3542 CPU_BASED_CR3_LOAD_EXITING |
3543 CPU_BASED_CR3_STORE_EXITING |
3544 CPU_BASED_USE_IO_BITMAPS |
3545 CPU_BASED_MOV_DR_EXITING |
3546 CPU_BASED_USE_TSC_OFFSETING |
3547 CPU_BASED_MWAIT_EXITING |
3548 CPU_BASED_MONITOR_EXITING |
3549 CPU_BASED_INVLPG_EXITING |
3550 CPU_BASED_RDPMC_EXITING;
3552 opt = CPU_BASED_TPR_SHADOW |
3553 CPU_BASED_USE_MSR_BITMAPS |
3554 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3555 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3556 &_cpu_based_exec_control) < 0)
3558 #ifdef CONFIG_X86_64
3559 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3560 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3561 ~CPU_BASED_CR8_STORE_EXITING;
3563 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3565 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3566 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3567 SECONDARY_EXEC_WBINVD_EXITING |
3568 SECONDARY_EXEC_ENABLE_VPID |
3569 SECONDARY_EXEC_ENABLE_EPT |
3570 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3571 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3572 SECONDARY_EXEC_RDTSCP |
3573 SECONDARY_EXEC_ENABLE_INVPCID |
3574 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3575 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3576 SECONDARY_EXEC_SHADOW_VMCS |
3577 SECONDARY_EXEC_XSAVES |
3578 SECONDARY_EXEC_ENABLE_PML |
3579 SECONDARY_EXEC_TSC_SCALING;
3580 if (adjust_vmx_controls(min2, opt2,
3581 MSR_IA32_VMX_PROCBASED_CTLS2,
3582 &_cpu_based_2nd_exec_control) < 0)
3585 #ifndef CONFIG_X86_64
3586 if (!(_cpu_based_2nd_exec_control &
3587 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3588 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3591 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3592 _cpu_based_2nd_exec_control &= ~(
3593 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3594 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3595 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3597 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3598 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3600 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3601 CPU_BASED_CR3_STORE_EXITING |
3602 CPU_BASED_INVLPG_EXITING);
3603 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3604 vmx_capability.ept, vmx_capability.vpid);
3607 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
3608 #ifdef CONFIG_X86_64
3609 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3611 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3612 VM_EXIT_CLEAR_BNDCFGS;
3613 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3614 &_vmexit_control) < 0)
3617 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
3618 PIN_BASED_VIRTUAL_NMIS;
3619 opt = PIN_BASED_POSTED_INTR | PIN_BASED_VMX_PREEMPTION_TIMER;
3620 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3621 &_pin_based_exec_control) < 0)
3624 if (cpu_has_broken_vmx_preemption_timer())
3625 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3626 if (!(_cpu_based_2nd_exec_control &
3627 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
3628 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3630 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3631 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3632 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3633 &_vmentry_control) < 0)
3636 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3638 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3639 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3642 #ifdef CONFIG_X86_64
3643 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3644 if (vmx_msr_high & (1u<<16))
3648 /* Require Write-Back (WB) memory type for VMCS accesses. */
3649 if (((vmx_msr_high >> 18) & 15) != 6)
3652 vmcs_conf->size = vmx_msr_high & 0x1fff;
3653 vmcs_conf->order = get_order(vmcs_conf->size);
3654 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
3655 vmcs_conf->revision_id = vmx_msr_low;
3657 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3658 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3659 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3660 vmcs_conf->vmexit_ctrl = _vmexit_control;
3661 vmcs_conf->vmentry_ctrl = _vmentry_control;
3663 cpu_has_load_ia32_efer =
3664 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3665 VM_ENTRY_LOAD_IA32_EFER)
3666 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3667 VM_EXIT_LOAD_IA32_EFER);
3669 cpu_has_load_perf_global_ctrl =
3670 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3671 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3672 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3673 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3676 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3677 * but due to errata below it can't be used. Workaround is to use
3678 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3680 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3685 * BC86,AAY89,BD102 (model 44)
3689 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3690 switch (boot_cpu_data.x86_model) {
3696 cpu_has_load_perf_global_ctrl = false;
3697 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3698 "does not work properly. Using workaround\n");
3705 if (boot_cpu_has(X86_FEATURE_XSAVES))
3706 rdmsrl(MSR_IA32_XSS, host_xss);
3711 static struct vmcs *alloc_vmcs_cpu(int cpu)
3713 int node = cpu_to_node(cpu);
3717 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3720 vmcs = page_address(pages);
3721 memset(vmcs, 0, vmcs_config.size);
3722 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3726 static struct vmcs *alloc_vmcs(void)
3728 return alloc_vmcs_cpu(raw_smp_processor_id());
3731 static void free_vmcs(struct vmcs *vmcs)
3733 free_pages((unsigned long)vmcs, vmcs_config.order);
3737 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3739 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3741 if (!loaded_vmcs->vmcs)
3743 loaded_vmcs_clear(loaded_vmcs);
3744 free_vmcs(loaded_vmcs->vmcs);
3745 loaded_vmcs->vmcs = NULL;
3746 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
3749 static void free_kvm_area(void)
3753 for_each_possible_cpu(cpu) {
3754 free_vmcs(per_cpu(vmxarea, cpu));
3755 per_cpu(vmxarea, cpu) = NULL;
3759 static void init_vmcs_shadow_fields(void)
3763 /* No checks for read only fields yet */
3765 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3766 switch (shadow_read_write_fields[i]) {
3768 if (!kvm_mpx_supported())
3776 shadow_read_write_fields[j] =
3777 shadow_read_write_fields[i];
3780 max_shadow_read_write_fields = j;
3782 /* shadowed fields guest access without vmexit */
3783 for (i = 0; i < max_shadow_read_write_fields; i++) {
3784 clear_bit(shadow_read_write_fields[i],
3785 vmx_vmwrite_bitmap);
3786 clear_bit(shadow_read_write_fields[i],
3789 for (i = 0; i < max_shadow_read_only_fields; i++)
3790 clear_bit(shadow_read_only_fields[i],
3794 static __init int alloc_kvm_area(void)
3798 for_each_possible_cpu(cpu) {
3801 vmcs = alloc_vmcs_cpu(cpu);
3807 per_cpu(vmxarea, cpu) = vmcs;
3812 static bool emulation_required(struct kvm_vcpu *vcpu)
3814 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3817 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3818 struct kvm_segment *save)
3820 if (!emulate_invalid_guest_state) {
3822 * CS and SS RPL should be equal during guest entry according
3823 * to VMX spec, but in reality it is not always so. Since vcpu
3824 * is in the middle of the transition from real mode to
3825 * protected mode it is safe to assume that RPL 0 is a good
3828 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3829 save->selector &= ~SEGMENT_RPL_MASK;
3830 save->dpl = save->selector & SEGMENT_RPL_MASK;
3833 vmx_set_segment(vcpu, save, seg);
3836 static void enter_pmode(struct kvm_vcpu *vcpu)
3838 unsigned long flags;
3839 struct vcpu_vmx *vmx = to_vmx(vcpu);
3842 * Update real mode segment cache. It may be not up-to-date if sement
3843 * register was written while vcpu was in a guest mode.
3845 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3846 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3847 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3848 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3849 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3850 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3852 vmx->rmode.vm86_active = 0;
3854 vmx_segment_cache_clear(vmx);
3856 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3858 flags = vmcs_readl(GUEST_RFLAGS);
3859 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3860 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3861 vmcs_writel(GUEST_RFLAGS, flags);
3863 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3864 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3866 update_exception_bitmap(vcpu);
3868 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3869 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3870 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3871 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3872 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3873 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3876 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3878 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3879 struct kvm_segment var = *save;
3882 if (seg == VCPU_SREG_CS)
3885 if (!emulate_invalid_guest_state) {
3886 var.selector = var.base >> 4;
3887 var.base = var.base & 0xffff0;
3897 if (save->base & 0xf)
3898 printk_once(KERN_WARNING "kvm: segment base is not "
3899 "paragraph aligned when entering "
3900 "protected mode (seg=%d)", seg);
3903 vmcs_write16(sf->selector, var.selector);
3904 vmcs_writel(sf->base, var.base);
3905 vmcs_write32(sf->limit, var.limit);
3906 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3909 static void enter_rmode(struct kvm_vcpu *vcpu)
3911 unsigned long flags;
3912 struct vcpu_vmx *vmx = to_vmx(vcpu);
3914 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3915 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3916 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3917 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3918 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3919 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3920 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3922 vmx->rmode.vm86_active = 1;
3925 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3926 * vcpu. Warn the user that an update is overdue.
3928 if (!vcpu->kvm->arch.tss_addr)
3929 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3930 "called before entering vcpu\n");
3932 vmx_segment_cache_clear(vmx);
3934 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3935 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3936 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3938 flags = vmcs_readl(GUEST_RFLAGS);
3939 vmx->rmode.save_rflags = flags;
3941 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3943 vmcs_writel(GUEST_RFLAGS, flags);
3944 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3945 update_exception_bitmap(vcpu);
3947 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3948 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3949 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3950 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3951 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3952 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3954 kvm_mmu_reset_context(vcpu);
3957 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3959 struct vcpu_vmx *vmx = to_vmx(vcpu);
3960 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3966 * Force kernel_gs_base reloading before EFER changes, as control
3967 * of this msr depends on is_long_mode().
3969 vmx_load_host_state(to_vmx(vcpu));
3970 vcpu->arch.efer = efer;
3971 if (efer & EFER_LMA) {
3972 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3975 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3977 msr->data = efer & ~EFER_LME;
3982 #ifdef CONFIG_X86_64
3984 static void enter_lmode(struct kvm_vcpu *vcpu)
3988 vmx_segment_cache_clear(to_vmx(vcpu));
3990 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3991 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
3992 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3994 vmcs_write32(GUEST_TR_AR_BYTES,
3995 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
3996 | VMX_AR_TYPE_BUSY_64_TSS);
3998 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
4001 static void exit_lmode(struct kvm_vcpu *vcpu)
4003 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4004 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
4009 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
4012 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
4014 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
4016 vpid_sync_context(vpid);
4020 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
4022 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
4025 static void vmx_flush_tlb_ept_only(struct kvm_vcpu *vcpu)
4028 vmx_flush_tlb(vcpu);
4031 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
4033 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
4035 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
4036 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
4039 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
4041 if (enable_ept && is_paging(vcpu))
4042 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4043 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
4046 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
4048 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
4050 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
4051 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
4054 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
4056 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4058 if (!test_bit(VCPU_EXREG_PDPTR,
4059 (unsigned long *)&vcpu->arch.regs_dirty))
4062 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4063 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
4064 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
4065 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
4066 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
4070 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
4072 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4074 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4075 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
4076 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
4077 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
4078 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
4081 __set_bit(VCPU_EXREG_PDPTR,
4082 (unsigned long *)&vcpu->arch.regs_avail);
4083 __set_bit(VCPU_EXREG_PDPTR,
4084 (unsigned long *)&vcpu->arch.regs_dirty);
4087 static bool nested_guest_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4089 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4090 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4091 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4093 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
4094 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4095 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4096 fixed0 &= ~(X86_CR0_PE | X86_CR0_PG);
4098 return fixed_bits_valid(val, fixed0, fixed1);
4101 static bool nested_host_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4103 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4104 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4106 return fixed_bits_valid(val, fixed0, fixed1);
4109 static bool nested_cr4_valid(struct kvm_vcpu *vcpu, unsigned long val)
4111 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed0;
4112 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed1;
4114 return fixed_bits_valid(val, fixed0, fixed1);
4117 /* No difference in the restrictions on guest and host CR4 in VMX operation. */
4118 #define nested_guest_cr4_valid nested_cr4_valid
4119 #define nested_host_cr4_valid nested_cr4_valid
4121 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
4123 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
4125 struct kvm_vcpu *vcpu)
4127 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
4128 vmx_decache_cr3(vcpu);
4129 if (!(cr0 & X86_CR0_PG)) {
4130 /* From paging/starting to nonpaging */
4131 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4132 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
4133 (CPU_BASED_CR3_LOAD_EXITING |
4134 CPU_BASED_CR3_STORE_EXITING));
4135 vcpu->arch.cr0 = cr0;
4136 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4137 } else if (!is_paging(vcpu)) {
4138 /* From nonpaging to paging */
4139 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4140 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
4141 ~(CPU_BASED_CR3_LOAD_EXITING |
4142 CPU_BASED_CR3_STORE_EXITING));
4143 vcpu->arch.cr0 = cr0;
4144 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4147 if (!(cr0 & X86_CR0_WP))
4148 *hw_cr0 &= ~X86_CR0_WP;
4151 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
4153 struct vcpu_vmx *vmx = to_vmx(vcpu);
4154 unsigned long hw_cr0;
4156 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
4157 if (enable_unrestricted_guest)
4158 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
4160 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
4162 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
4165 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
4169 #ifdef CONFIG_X86_64
4170 if (vcpu->arch.efer & EFER_LME) {
4171 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
4173 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
4179 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
4181 vmcs_writel(CR0_READ_SHADOW, cr0);
4182 vmcs_writel(GUEST_CR0, hw_cr0);
4183 vcpu->arch.cr0 = cr0;
4185 /* depends on vcpu->arch.cr0 to be set to a new value */
4186 vmx->emulation_required = emulation_required(vcpu);
4189 static u64 construct_eptp(unsigned long root_hpa)
4193 /* TODO write the value reading from MSR */
4194 eptp = VMX_EPT_DEFAULT_MT |
4195 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
4196 if (enable_ept_ad_bits)
4197 eptp |= VMX_EPT_AD_ENABLE_BIT;
4198 eptp |= (root_hpa & PAGE_MASK);
4203 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
4205 unsigned long guest_cr3;
4210 eptp = construct_eptp(cr3);
4211 vmcs_write64(EPT_POINTER, eptp);
4212 if (is_paging(vcpu) || is_guest_mode(vcpu))
4213 guest_cr3 = kvm_read_cr3(vcpu);
4215 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
4216 ept_load_pdptrs(vcpu);
4219 vmx_flush_tlb(vcpu);
4220 vmcs_writel(GUEST_CR3, guest_cr3);
4223 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
4226 * Pass through host's Machine Check Enable value to hw_cr4, which
4227 * is in force while we are in guest mode. Do not let guests control
4228 * this bit, even if host CR4.MCE == 0.
4230 unsigned long hw_cr4 =
4231 (cr4_read_shadow() & X86_CR4_MCE) |
4232 (cr4 & ~X86_CR4_MCE) |
4233 (to_vmx(vcpu)->rmode.vm86_active ?
4234 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
4236 if (cr4 & X86_CR4_VMXE) {
4238 * To use VMXON (and later other VMX instructions), a guest
4239 * must first be able to turn on cr4.VMXE (see handle_vmon()).
4240 * So basically the check on whether to allow nested VMX
4243 if (!nested_vmx_allowed(vcpu))
4247 if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
4250 vcpu->arch.cr4 = cr4;
4252 if (!is_paging(vcpu)) {
4253 hw_cr4 &= ~X86_CR4_PAE;
4254 hw_cr4 |= X86_CR4_PSE;
4255 } else if (!(cr4 & X86_CR4_PAE)) {
4256 hw_cr4 &= ~X86_CR4_PAE;
4260 if (!enable_unrestricted_guest && !is_paging(vcpu))
4262 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4263 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
4264 * to be manually disabled when guest switches to non-paging
4267 * If !enable_unrestricted_guest, the CPU is always running
4268 * with CR0.PG=1 and CR4 needs to be modified.
4269 * If enable_unrestricted_guest, the CPU automatically
4270 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4272 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
4274 vmcs_writel(CR4_READ_SHADOW, cr4);
4275 vmcs_writel(GUEST_CR4, hw_cr4);
4279 static void vmx_get_segment(struct kvm_vcpu *vcpu,
4280 struct kvm_segment *var, int seg)
4282 struct vcpu_vmx *vmx = to_vmx(vcpu);
4285 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4286 *var = vmx->rmode.segs[seg];
4287 if (seg == VCPU_SREG_TR
4288 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
4290 var->base = vmx_read_guest_seg_base(vmx, seg);
4291 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4294 var->base = vmx_read_guest_seg_base(vmx, seg);
4295 var->limit = vmx_read_guest_seg_limit(vmx, seg);
4296 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4297 ar = vmx_read_guest_seg_ar(vmx, seg);
4298 var->unusable = (ar >> 16) & 1;
4299 var->type = ar & 15;
4300 var->s = (ar >> 4) & 1;
4301 var->dpl = (ar >> 5) & 3;
4303 * Some userspaces do not preserve unusable property. Since usable
4304 * segment has to be present according to VMX spec we can use present
4305 * property to amend userspace bug by making unusable segment always
4306 * nonpresent. vmx_segment_access_rights() already marks nonpresent
4307 * segment as unusable.
4309 var->present = !var->unusable;
4310 var->avl = (ar >> 12) & 1;
4311 var->l = (ar >> 13) & 1;
4312 var->db = (ar >> 14) & 1;
4313 var->g = (ar >> 15) & 1;
4316 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
4318 struct kvm_segment s;
4320 if (to_vmx(vcpu)->rmode.vm86_active) {
4321 vmx_get_segment(vcpu, &s, seg);
4324 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
4327 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
4329 struct vcpu_vmx *vmx = to_vmx(vcpu);
4331 if (unlikely(vmx->rmode.vm86_active))
4334 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
4335 return VMX_AR_DPL(ar);
4339 static u32 vmx_segment_access_rights(struct kvm_segment *var)
4343 if (var->unusable || !var->present)
4346 ar = var->type & 15;
4347 ar |= (var->s & 1) << 4;
4348 ar |= (var->dpl & 3) << 5;
4349 ar |= (var->present & 1) << 7;
4350 ar |= (var->avl & 1) << 12;
4351 ar |= (var->l & 1) << 13;
4352 ar |= (var->db & 1) << 14;
4353 ar |= (var->g & 1) << 15;
4359 static void vmx_set_segment(struct kvm_vcpu *vcpu,
4360 struct kvm_segment *var, int seg)
4362 struct vcpu_vmx *vmx = to_vmx(vcpu);
4363 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4365 vmx_segment_cache_clear(vmx);
4367 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4368 vmx->rmode.segs[seg] = *var;
4369 if (seg == VCPU_SREG_TR)
4370 vmcs_write16(sf->selector, var->selector);
4372 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
4376 vmcs_writel(sf->base, var->base);
4377 vmcs_write32(sf->limit, var->limit);
4378 vmcs_write16(sf->selector, var->selector);
4381 * Fix the "Accessed" bit in AR field of segment registers for older
4383 * IA32 arch specifies that at the time of processor reset the
4384 * "Accessed" bit in the AR field of segment registers is 1. And qemu
4385 * is setting it to 0 in the userland code. This causes invalid guest
4386 * state vmexit when "unrestricted guest" mode is turned on.
4387 * Fix for this setup issue in cpu_reset is being pushed in the qemu
4388 * tree. Newer qemu binaries with that qemu fix would not need this
4391 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
4392 var->type |= 0x1; /* Accessed */
4394 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4397 vmx->emulation_required = emulation_required(vcpu);
4400 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4402 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4404 *db = (ar >> 14) & 1;
4405 *l = (ar >> 13) & 1;
4408 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4410 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4411 dt->address = vmcs_readl(GUEST_IDTR_BASE);
4414 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4416 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4417 vmcs_writel(GUEST_IDTR_BASE, dt->address);
4420 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4422 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4423 dt->address = vmcs_readl(GUEST_GDTR_BASE);
4426 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4428 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4429 vmcs_writel(GUEST_GDTR_BASE, dt->address);
4432 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4434 struct kvm_segment var;
4437 vmx_get_segment(vcpu, &var, seg);
4439 if (seg == VCPU_SREG_CS)
4441 ar = vmx_segment_access_rights(&var);
4443 if (var.base != (var.selector << 4))
4445 if (var.limit != 0xffff)
4453 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4455 struct kvm_segment cs;
4456 unsigned int cs_rpl;
4458 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4459 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4463 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4467 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4468 if (cs.dpl > cs_rpl)
4471 if (cs.dpl != cs_rpl)
4477 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4481 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4483 struct kvm_segment ss;
4484 unsigned int ss_rpl;
4486 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4487 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4491 if (ss.type != 3 && ss.type != 7)
4495 if (ss.dpl != ss_rpl) /* DPL != RPL */
4503 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4505 struct kvm_segment var;
4508 vmx_get_segment(vcpu, &var, seg);
4509 rpl = var.selector & SEGMENT_RPL_MASK;
4517 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4518 if (var.dpl < rpl) /* DPL < RPL */
4522 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4528 static bool tr_valid(struct kvm_vcpu *vcpu)
4530 struct kvm_segment tr;
4532 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4536 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4538 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4546 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4548 struct kvm_segment ldtr;
4550 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4554 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4564 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4566 struct kvm_segment cs, ss;
4568 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4569 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4571 return ((cs.selector & SEGMENT_RPL_MASK) ==
4572 (ss.selector & SEGMENT_RPL_MASK));
4576 * Check if guest state is valid. Returns true if valid, false if
4578 * We assume that registers are always usable
4580 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4582 if (enable_unrestricted_guest)
4585 /* real mode guest state checks */
4586 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4587 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4589 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4591 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4593 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4595 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4597 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4600 /* protected mode guest state checks */
4601 if (!cs_ss_rpl_check(vcpu))
4603 if (!code_segment_valid(vcpu))
4605 if (!stack_segment_valid(vcpu))
4607 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4609 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4611 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4613 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4615 if (!tr_valid(vcpu))
4617 if (!ldtr_valid(vcpu))
4621 * - Add checks on RIP
4622 * - Add checks on RFLAGS
4628 static int init_rmode_tss(struct kvm *kvm)
4634 idx = srcu_read_lock(&kvm->srcu);
4635 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4636 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4639 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4640 r = kvm_write_guest_page(kvm, fn++, &data,
4641 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4644 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4647 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4651 r = kvm_write_guest_page(kvm, fn, &data,
4652 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4655 srcu_read_unlock(&kvm->srcu, idx);
4659 static int init_rmode_identity_map(struct kvm *kvm)
4662 kvm_pfn_t identity_map_pfn;
4668 /* Protect kvm->arch.ept_identity_pagetable_done. */
4669 mutex_lock(&kvm->slots_lock);
4671 if (likely(kvm->arch.ept_identity_pagetable_done))
4674 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4676 r = alloc_identity_pagetable(kvm);
4680 idx = srcu_read_lock(&kvm->srcu);
4681 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4684 /* Set up identity-mapping pagetable for EPT in real mode */
4685 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4686 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4687 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4688 r = kvm_write_guest_page(kvm, identity_map_pfn,
4689 &tmp, i * sizeof(tmp), sizeof(tmp));
4693 kvm->arch.ept_identity_pagetable_done = true;
4696 srcu_read_unlock(&kvm->srcu, idx);
4699 mutex_unlock(&kvm->slots_lock);
4703 static void seg_setup(int seg)
4705 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4708 vmcs_write16(sf->selector, 0);
4709 vmcs_writel(sf->base, 0);
4710 vmcs_write32(sf->limit, 0xffff);
4712 if (seg == VCPU_SREG_CS)
4713 ar |= 0x08; /* code segment */
4715 vmcs_write32(sf->ar_bytes, ar);
4718 static int alloc_apic_access_page(struct kvm *kvm)
4723 mutex_lock(&kvm->slots_lock);
4724 if (kvm->arch.apic_access_page_done)
4726 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4727 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4731 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4732 if (is_error_page(page)) {
4738 * Do not pin the page in memory, so that memory hot-unplug
4739 * is able to migrate it.
4742 kvm->arch.apic_access_page_done = true;
4744 mutex_unlock(&kvm->slots_lock);
4748 static int alloc_identity_pagetable(struct kvm *kvm)
4750 /* Called with kvm->slots_lock held. */
4754 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4756 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4757 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4762 static int allocate_vpid(void)
4768 spin_lock(&vmx_vpid_lock);
4769 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4770 if (vpid < VMX_NR_VPIDS)
4771 __set_bit(vpid, vmx_vpid_bitmap);
4774 spin_unlock(&vmx_vpid_lock);
4778 static void free_vpid(int vpid)
4780 if (!enable_vpid || vpid == 0)
4782 spin_lock(&vmx_vpid_lock);
4783 __clear_bit(vpid, vmx_vpid_bitmap);
4784 spin_unlock(&vmx_vpid_lock);
4787 #define MSR_TYPE_R 1
4788 #define MSR_TYPE_W 2
4789 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4792 int f = sizeof(unsigned long);
4794 if (!cpu_has_vmx_msr_bitmap())
4798 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4799 * have the write-low and read-high bitmap offsets the wrong way round.
4800 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4802 if (msr <= 0x1fff) {
4803 if (type & MSR_TYPE_R)
4805 __clear_bit(msr, msr_bitmap + 0x000 / f);
4807 if (type & MSR_TYPE_W)
4809 __clear_bit(msr, msr_bitmap + 0x800 / f);
4811 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4813 if (type & MSR_TYPE_R)
4815 __clear_bit(msr, msr_bitmap + 0x400 / f);
4817 if (type & MSR_TYPE_W)
4819 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4825 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4826 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4828 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4829 unsigned long *msr_bitmap_nested,
4832 int f = sizeof(unsigned long);
4834 if (!cpu_has_vmx_msr_bitmap()) {
4840 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4841 * have the write-low and read-high bitmap offsets the wrong way round.
4842 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4844 if (msr <= 0x1fff) {
4845 if (type & MSR_TYPE_R &&
4846 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4848 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4850 if (type & MSR_TYPE_W &&
4851 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4853 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4855 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4857 if (type & MSR_TYPE_R &&
4858 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4860 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4862 if (type & MSR_TYPE_W &&
4863 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4865 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4870 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4873 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4874 msr, MSR_TYPE_R | MSR_TYPE_W);
4875 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4876 msr, MSR_TYPE_R | MSR_TYPE_W);
4879 static void vmx_disable_intercept_msr_x2apic(u32 msr, int type, bool apicv_active)
4882 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic_apicv,
4884 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic_apicv,
4887 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4889 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4894 static bool vmx_get_enable_apicv(void)
4896 return enable_apicv;
4899 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
4901 struct vcpu_vmx *vmx = to_vmx(vcpu);
4906 if (vmx->nested.pi_desc &&
4907 vmx->nested.pi_pending) {
4908 vmx->nested.pi_pending = false;
4909 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
4912 max_irr = find_last_bit(
4913 (unsigned long *)vmx->nested.pi_desc->pir, 256);
4918 vapic_page = kmap(vmx->nested.virtual_apic_page);
4919 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
4920 kunmap(vmx->nested.virtual_apic_page);
4922 status = vmcs_read16(GUEST_INTR_STATUS);
4923 if ((u8)max_irr > ((u8)status & 0xff)) {
4925 status |= (u8)max_irr;
4926 vmcs_write16(GUEST_INTR_STATUS, status);
4931 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu)
4934 if (vcpu->mode == IN_GUEST_MODE) {
4935 struct vcpu_vmx *vmx = to_vmx(vcpu);
4938 * Currently, we don't support urgent interrupt,
4939 * all interrupts are recognized as non-urgent
4940 * interrupt, so we cannot post interrupts when
4943 * If the vcpu is in guest mode, it means it is
4944 * running instead of being scheduled out and
4945 * waiting in the run queue, and that's the only
4946 * case when 'SN' is set currently, warning if
4949 WARN_ON_ONCE(pi_test_sn(&vmx->pi_desc));
4951 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4952 POSTED_INTR_VECTOR);
4959 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4962 struct vcpu_vmx *vmx = to_vmx(vcpu);
4964 if (is_guest_mode(vcpu) &&
4965 vector == vmx->nested.posted_intr_nv) {
4966 /* the PIR and ON have been set by L1. */
4967 kvm_vcpu_trigger_posted_interrupt(vcpu);
4969 * If a posted intr is not recognized by hardware,
4970 * we will accomplish it in the next vmentry.
4972 vmx->nested.pi_pending = true;
4973 kvm_make_request(KVM_REQ_EVENT, vcpu);
4979 * Send interrupt to vcpu via posted interrupt way.
4980 * 1. If target vcpu is running(non-root mode), send posted interrupt
4981 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4982 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4983 * interrupt from PIR in next vmentry.
4985 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4987 struct vcpu_vmx *vmx = to_vmx(vcpu);
4990 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4994 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4997 /* If a previous notification has sent the IPI, nothing to do. */
4998 if (pi_test_and_set_on(&vmx->pi_desc))
5001 if (!kvm_vcpu_trigger_posted_interrupt(vcpu))
5002 kvm_vcpu_kick(vcpu);
5006 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
5007 * will not change in the lifetime of the guest.
5008 * Note that host-state that does change is set elsewhere. E.g., host-state
5009 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
5011 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
5016 unsigned long cr0, cr4;
5019 WARN_ON(cr0 & X86_CR0_TS);
5020 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
5021 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
5023 /* Save the most likely value for this task's CR4 in the VMCS. */
5024 cr4 = cr4_read_shadow();
5025 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
5026 vmx->host_state.vmcs_host_cr4 = cr4;
5028 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
5029 #ifdef CONFIG_X86_64
5031 * Load null selectors, so we can avoid reloading them in
5032 * __vmx_load_host_state(), in case userspace uses the null selectors
5033 * too (the expected case).
5035 vmcs_write16(HOST_DS_SELECTOR, 0);
5036 vmcs_write16(HOST_ES_SELECTOR, 0);
5038 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5039 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5041 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5042 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
5044 native_store_idt(&dt);
5045 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
5046 vmx->host_idt_base = dt.address;
5048 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
5050 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
5051 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
5052 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
5053 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
5055 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
5056 rdmsr(MSR_IA32_CR_PAT, low32, high32);
5057 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
5061 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
5063 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
5065 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
5066 if (is_guest_mode(&vmx->vcpu))
5067 vmx->vcpu.arch.cr4_guest_owned_bits &=
5068 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
5069 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
5072 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
5074 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
5076 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5077 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
5078 /* Enable the preemption timer dynamically */
5079 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
5080 return pin_based_exec_ctrl;
5083 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5085 struct vcpu_vmx *vmx = to_vmx(vcpu);
5087 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5088 if (cpu_has_secondary_exec_ctrls()) {
5089 if (kvm_vcpu_apicv_active(vcpu))
5090 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
5091 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5092 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5094 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
5095 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5096 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5099 if (cpu_has_vmx_msr_bitmap())
5100 vmx_set_msr_bitmap(vcpu);
5103 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
5105 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
5107 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
5108 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5110 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
5111 exec_control &= ~CPU_BASED_TPR_SHADOW;
5112 #ifdef CONFIG_X86_64
5113 exec_control |= CPU_BASED_CR8_STORE_EXITING |
5114 CPU_BASED_CR8_LOAD_EXITING;
5118 exec_control |= CPU_BASED_CR3_STORE_EXITING |
5119 CPU_BASED_CR3_LOAD_EXITING |
5120 CPU_BASED_INVLPG_EXITING;
5121 return exec_control;
5124 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
5126 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
5127 if (!cpu_need_virtualize_apic_accesses(&vmx->vcpu))
5128 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5130 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
5132 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
5133 enable_unrestricted_guest = 0;
5134 /* Enable INVPCID for non-ept guests may cause performance regression. */
5135 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5137 if (!enable_unrestricted_guest)
5138 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
5140 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
5141 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5142 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
5143 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5144 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
5145 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
5147 We can NOT enable shadow_vmcs here because we don't have yet
5150 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
5153 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
5155 return exec_control;
5158 static void ept_set_mmio_spte_mask(void)
5161 * EPT Misconfigurations can be generated if the value of bits 2:0
5162 * of an EPT paging-structure entry is 110b (write/execute).
5164 kvm_mmu_set_mmio_spte_mask(VMX_EPT_MISCONFIG_WX_VALUE);
5167 #define VMX_XSS_EXIT_BITMAP 0
5169 * Sets up the vmcs for emulated real mode.
5171 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
5173 #ifdef CONFIG_X86_64
5179 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
5180 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
5182 if (enable_shadow_vmcs) {
5183 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5184 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5186 if (cpu_has_vmx_msr_bitmap())
5187 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
5189 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
5192 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5193 vmx->hv_deadline_tsc = -1;
5195 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
5197 if (cpu_has_secondary_exec_ctrls()) {
5198 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5199 vmx_secondary_exec_control(vmx));
5202 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
5203 vmcs_write64(EOI_EXIT_BITMAP0, 0);
5204 vmcs_write64(EOI_EXIT_BITMAP1, 0);
5205 vmcs_write64(EOI_EXIT_BITMAP2, 0);
5206 vmcs_write64(EOI_EXIT_BITMAP3, 0);
5208 vmcs_write16(GUEST_INTR_STATUS, 0);
5210 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
5211 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
5215 vmcs_write32(PLE_GAP, ple_gap);
5216 vmx->ple_window = ple_window;
5217 vmx->ple_window_dirty = true;
5220 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
5221 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
5222 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
5224 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
5225 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
5226 vmx_set_constant_host_state(vmx);
5227 #ifdef CONFIG_X86_64
5228 rdmsrl(MSR_FS_BASE, a);
5229 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
5230 rdmsrl(MSR_GS_BASE, a);
5231 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
5233 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
5234 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
5237 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
5238 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
5239 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
5240 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
5241 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
5243 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
5244 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
5246 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
5247 u32 index = vmx_msr_index[i];
5248 u32 data_low, data_high;
5251 if (rdmsr_safe(index, &data_low, &data_high) < 0)
5253 if (wrmsr_safe(index, data_low, data_high) < 0)
5255 vmx->guest_msrs[j].index = i;
5256 vmx->guest_msrs[j].data = 0;
5257 vmx->guest_msrs[j].mask = -1ull;
5262 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
5264 /* 22.2.1, 20.8.1 */
5265 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
5267 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
5268 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
5270 set_cr4_guest_host_mask(vmx);
5272 if (vmx_xsaves_supported())
5273 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
5276 ASSERT(vmx->pml_pg);
5277 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
5278 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5284 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
5286 struct vcpu_vmx *vmx = to_vmx(vcpu);
5287 struct msr_data apic_base_msr;
5290 vmx->rmode.vm86_active = 0;
5292 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
5293 kvm_set_cr8(vcpu, 0);
5296 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
5297 MSR_IA32_APICBASE_ENABLE;
5298 if (kvm_vcpu_is_reset_bsp(vcpu))
5299 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
5300 apic_base_msr.host_initiated = true;
5301 kvm_set_apic_base(vcpu, &apic_base_msr);
5304 vmx_segment_cache_clear(vmx);
5306 seg_setup(VCPU_SREG_CS);
5307 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
5308 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
5310 seg_setup(VCPU_SREG_DS);
5311 seg_setup(VCPU_SREG_ES);
5312 seg_setup(VCPU_SREG_FS);
5313 seg_setup(VCPU_SREG_GS);
5314 seg_setup(VCPU_SREG_SS);
5316 vmcs_write16(GUEST_TR_SELECTOR, 0);
5317 vmcs_writel(GUEST_TR_BASE, 0);
5318 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
5319 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
5321 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
5322 vmcs_writel(GUEST_LDTR_BASE, 0);
5323 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
5324 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
5327 vmcs_write32(GUEST_SYSENTER_CS, 0);
5328 vmcs_writel(GUEST_SYSENTER_ESP, 0);
5329 vmcs_writel(GUEST_SYSENTER_EIP, 0);
5330 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
5333 vmcs_writel(GUEST_RFLAGS, 0x02);
5334 kvm_rip_write(vcpu, 0xfff0);
5336 vmcs_writel(GUEST_GDTR_BASE, 0);
5337 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
5339 vmcs_writel(GUEST_IDTR_BASE, 0);
5340 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
5342 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
5343 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
5344 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
5348 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
5350 if (cpu_has_vmx_tpr_shadow() && !init_event) {
5351 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
5352 if (cpu_need_tpr_shadow(vcpu))
5353 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
5354 __pa(vcpu->arch.apic->regs));
5355 vmcs_write32(TPR_THRESHOLD, 0);
5358 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5360 if (kvm_vcpu_apicv_active(vcpu))
5361 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
5364 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5366 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5367 vmx->vcpu.arch.cr0 = cr0;
5368 vmx_set_cr0(vcpu, cr0); /* enter rmode */
5369 vmx_set_cr4(vcpu, 0);
5370 vmx_set_efer(vcpu, 0);
5372 update_exception_bitmap(vcpu);
5374 vpid_sync_context(vmx->vpid);
5378 * In nested virtualization, check if L1 asked to exit on external interrupts.
5379 * For most existing hypervisors, this will always return true.
5381 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5383 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5384 PIN_BASED_EXT_INTR_MASK;
5388 * In nested virtualization, check if L1 has set
5389 * VM_EXIT_ACK_INTR_ON_EXIT
5391 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5393 return get_vmcs12(vcpu)->vm_exit_controls &
5394 VM_EXIT_ACK_INTR_ON_EXIT;
5397 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5399 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5400 PIN_BASED_NMI_EXITING;
5403 static void enable_irq_window(struct kvm_vcpu *vcpu)
5405 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5406 CPU_BASED_VIRTUAL_INTR_PENDING);
5409 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5411 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5412 enable_irq_window(vcpu);
5416 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5417 CPU_BASED_VIRTUAL_NMI_PENDING);
5420 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5422 struct vcpu_vmx *vmx = to_vmx(vcpu);
5424 int irq = vcpu->arch.interrupt.nr;
5426 trace_kvm_inj_virq(irq);
5428 ++vcpu->stat.irq_injections;
5429 if (vmx->rmode.vm86_active) {
5431 if (vcpu->arch.interrupt.soft)
5432 inc_eip = vcpu->arch.event_exit_inst_len;
5433 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5434 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5437 intr = irq | INTR_INFO_VALID_MASK;
5438 if (vcpu->arch.interrupt.soft) {
5439 intr |= INTR_TYPE_SOFT_INTR;
5440 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5441 vmx->vcpu.arch.event_exit_inst_len);
5443 intr |= INTR_TYPE_EXT_INTR;
5444 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5447 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5449 struct vcpu_vmx *vmx = to_vmx(vcpu);
5451 if (!is_guest_mode(vcpu)) {
5452 ++vcpu->stat.nmi_injections;
5453 vmx->nmi_known_unmasked = false;
5456 if (vmx->rmode.vm86_active) {
5457 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5458 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5462 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5463 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5466 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5468 if (to_vmx(vcpu)->nmi_known_unmasked)
5470 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5473 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5475 struct vcpu_vmx *vmx = to_vmx(vcpu);
5477 vmx->nmi_known_unmasked = !masked;
5479 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5480 GUEST_INTR_STATE_NMI);
5482 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5483 GUEST_INTR_STATE_NMI);
5486 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5488 if (to_vmx(vcpu)->nested.nested_run_pending)
5491 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5492 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5493 | GUEST_INTR_STATE_NMI));
5496 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5498 return (!to_vmx(vcpu)->nested.nested_run_pending &&
5499 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5500 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5501 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5504 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5508 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5512 kvm->arch.tss_addr = addr;
5513 return init_rmode_tss(kvm);
5516 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5521 * Update instruction length as we may reinject the exception
5522 * from user space while in guest debugging mode.
5524 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5525 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5526 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5530 if (vcpu->guest_debug &
5531 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5548 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5549 int vec, u32 err_code)
5552 * Instruction with address size override prefix opcode 0x67
5553 * Cause the #SS fault with 0 error code in VM86 mode.
5555 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5556 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5557 if (vcpu->arch.halt_request) {
5558 vcpu->arch.halt_request = 0;
5559 return kvm_vcpu_halt(vcpu);
5567 * Forward all other exceptions that are valid in real mode.
5568 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5569 * the required debugging infrastructure rework.
5571 kvm_queue_exception(vcpu, vec);
5576 * Trigger machine check on the host. We assume all the MSRs are already set up
5577 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5578 * We pass a fake environment to the machine check handler because we want
5579 * the guest to be always treated like user space, no matter what context
5580 * it used internally.
5582 static void kvm_machine_check(void)
5584 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5585 struct pt_regs regs = {
5586 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5587 .flags = X86_EFLAGS_IF,
5590 do_machine_check(®s, 0);
5594 static int handle_machine_check(struct kvm_vcpu *vcpu)
5596 /* already handled by vcpu_run */
5600 static int handle_exception(struct kvm_vcpu *vcpu)
5602 struct vcpu_vmx *vmx = to_vmx(vcpu);
5603 struct kvm_run *kvm_run = vcpu->run;
5604 u32 intr_info, ex_no, error_code;
5605 unsigned long cr2, rip, dr6;
5607 enum emulation_result er;
5609 vect_info = vmx->idt_vectoring_info;
5610 intr_info = vmx->exit_intr_info;
5612 if (is_machine_check(intr_info))
5613 return handle_machine_check(vcpu);
5615 if (is_nmi(intr_info))
5616 return 1; /* already handled by vmx_vcpu_run() */
5618 if (is_invalid_opcode(intr_info)) {
5619 if (is_guest_mode(vcpu)) {
5620 kvm_queue_exception(vcpu, UD_VECTOR);
5623 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5624 if (er != EMULATE_DONE)
5625 kvm_queue_exception(vcpu, UD_VECTOR);
5630 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5631 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5634 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5635 * MMIO, it is better to report an internal error.
5636 * See the comments in vmx_handle_exit.
5638 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5639 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5640 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5641 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5642 vcpu->run->internal.ndata = 3;
5643 vcpu->run->internal.data[0] = vect_info;
5644 vcpu->run->internal.data[1] = intr_info;
5645 vcpu->run->internal.data[2] = error_code;
5649 if (is_page_fault(intr_info)) {
5650 /* EPT won't cause page fault directly */
5652 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5653 trace_kvm_page_fault(cr2, error_code);
5655 if (kvm_event_needs_reinjection(vcpu))
5656 kvm_mmu_unprotect_page_virt(vcpu, cr2);
5657 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
5660 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5662 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5663 return handle_rmode_exception(vcpu, ex_no, error_code);
5667 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5670 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5671 if (!(vcpu->guest_debug &
5672 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5673 vcpu->arch.dr6 &= ~15;
5674 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5675 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5676 skip_emulated_instruction(vcpu);
5678 kvm_queue_exception(vcpu, DB_VECTOR);
5681 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5682 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5686 * Update instruction length as we may reinject #BP from
5687 * user space while in guest debugging mode. Reading it for
5688 * #DB as well causes no harm, it is not used in that case.
5690 vmx->vcpu.arch.event_exit_inst_len =
5691 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5692 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5693 rip = kvm_rip_read(vcpu);
5694 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5695 kvm_run->debug.arch.exception = ex_no;
5698 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5699 kvm_run->ex.exception = ex_no;
5700 kvm_run->ex.error_code = error_code;
5706 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5708 ++vcpu->stat.irq_exits;
5712 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5714 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5718 static int handle_io(struct kvm_vcpu *vcpu)
5720 unsigned long exit_qualification;
5721 int size, in, string, ret;
5724 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5725 string = (exit_qualification & 16) != 0;
5726 in = (exit_qualification & 8) != 0;
5728 ++vcpu->stat.io_exits;
5731 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5733 port = exit_qualification >> 16;
5734 size = (exit_qualification & 7) + 1;
5736 ret = kvm_skip_emulated_instruction(vcpu);
5739 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
5740 * KVM_EXIT_DEBUG here.
5742 return kvm_fast_pio_out(vcpu, size, port) && ret;
5746 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5749 * Patch in the VMCALL instruction:
5751 hypercall[0] = 0x0f;
5752 hypercall[1] = 0x01;
5753 hypercall[2] = 0xc1;
5756 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5757 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5759 if (is_guest_mode(vcpu)) {
5760 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5761 unsigned long orig_val = val;
5764 * We get here when L2 changed cr0 in a way that did not change
5765 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5766 * but did change L0 shadowed bits. So we first calculate the
5767 * effective cr0 value that L1 would like to write into the
5768 * hardware. It consists of the L2-owned bits from the new
5769 * value combined with the L1-owned bits from L1's guest_cr0.
5771 val = (val & ~vmcs12->cr0_guest_host_mask) |
5772 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5774 if (!nested_guest_cr0_valid(vcpu, val))
5777 if (kvm_set_cr0(vcpu, val))
5779 vmcs_writel(CR0_READ_SHADOW, orig_val);
5782 if (to_vmx(vcpu)->nested.vmxon &&
5783 !nested_host_cr0_valid(vcpu, val))
5786 return kvm_set_cr0(vcpu, val);
5790 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5792 if (is_guest_mode(vcpu)) {
5793 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5794 unsigned long orig_val = val;
5796 /* analogously to handle_set_cr0 */
5797 val = (val & ~vmcs12->cr4_guest_host_mask) |
5798 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5799 if (kvm_set_cr4(vcpu, val))
5801 vmcs_writel(CR4_READ_SHADOW, orig_val);
5804 return kvm_set_cr4(vcpu, val);
5807 static int handle_cr(struct kvm_vcpu *vcpu)
5809 unsigned long exit_qualification, val;
5815 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5816 cr = exit_qualification & 15;
5817 reg = (exit_qualification >> 8) & 15;
5818 switch ((exit_qualification >> 4) & 3) {
5819 case 0: /* mov to cr */
5820 val = kvm_register_readl(vcpu, reg);
5821 trace_kvm_cr_write(cr, val);
5824 err = handle_set_cr0(vcpu, val);
5825 return kvm_complete_insn_gp(vcpu, err);
5827 err = kvm_set_cr3(vcpu, val);
5828 return kvm_complete_insn_gp(vcpu, err);
5830 err = handle_set_cr4(vcpu, val);
5831 return kvm_complete_insn_gp(vcpu, err);
5833 u8 cr8_prev = kvm_get_cr8(vcpu);
5835 err = kvm_set_cr8(vcpu, cr8);
5836 ret = kvm_complete_insn_gp(vcpu, err);
5837 if (lapic_in_kernel(vcpu))
5839 if (cr8_prev <= cr8)
5842 * TODO: we might be squashing a
5843 * KVM_GUESTDBG_SINGLESTEP-triggered
5844 * KVM_EXIT_DEBUG here.
5846 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5852 WARN_ONCE(1, "Guest should always own CR0.TS");
5853 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5854 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5855 return kvm_skip_emulated_instruction(vcpu);
5856 case 1: /*mov from cr*/
5859 val = kvm_read_cr3(vcpu);
5860 kvm_register_write(vcpu, reg, val);
5861 trace_kvm_cr_read(cr, val);
5862 return kvm_skip_emulated_instruction(vcpu);
5864 val = kvm_get_cr8(vcpu);
5865 kvm_register_write(vcpu, reg, val);
5866 trace_kvm_cr_read(cr, val);
5867 return kvm_skip_emulated_instruction(vcpu);
5871 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5872 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5873 kvm_lmsw(vcpu, val);
5875 return kvm_skip_emulated_instruction(vcpu);
5879 vcpu->run->exit_reason = 0;
5880 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5881 (int)(exit_qualification >> 4) & 3, cr);
5885 static int handle_dr(struct kvm_vcpu *vcpu)
5887 unsigned long exit_qualification;
5890 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5891 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5893 /* First, if DR does not exist, trigger UD */
5894 if (!kvm_require_dr(vcpu, dr))
5897 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5898 if (!kvm_require_cpl(vcpu, 0))
5900 dr7 = vmcs_readl(GUEST_DR7);
5903 * As the vm-exit takes precedence over the debug trap, we
5904 * need to emulate the latter, either for the host or the
5905 * guest debugging itself.
5907 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5908 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5909 vcpu->run->debug.arch.dr7 = dr7;
5910 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5911 vcpu->run->debug.arch.exception = DB_VECTOR;
5912 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5915 vcpu->arch.dr6 &= ~15;
5916 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5917 kvm_queue_exception(vcpu, DB_VECTOR);
5922 if (vcpu->guest_debug == 0) {
5923 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
5924 CPU_BASED_MOV_DR_EXITING);
5927 * No more DR vmexits; force a reload of the debug registers
5928 * and reenter on this instruction. The next vmexit will
5929 * retrieve the full state of the debug registers.
5931 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5935 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5936 if (exit_qualification & TYPE_MOV_FROM_DR) {
5939 if (kvm_get_dr(vcpu, dr, &val))
5941 kvm_register_write(vcpu, reg, val);
5943 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5946 return kvm_skip_emulated_instruction(vcpu);
5949 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5951 return vcpu->arch.dr6;
5954 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5958 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5960 get_debugreg(vcpu->arch.db[0], 0);
5961 get_debugreg(vcpu->arch.db[1], 1);
5962 get_debugreg(vcpu->arch.db[2], 2);
5963 get_debugreg(vcpu->arch.db[3], 3);
5964 get_debugreg(vcpu->arch.dr6, 6);
5965 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5967 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5968 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
5971 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5973 vmcs_writel(GUEST_DR7, val);
5976 static int handle_cpuid(struct kvm_vcpu *vcpu)
5978 return kvm_emulate_cpuid(vcpu);
5981 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5983 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5984 struct msr_data msr_info;
5986 msr_info.index = ecx;
5987 msr_info.host_initiated = false;
5988 if (vmx_get_msr(vcpu, &msr_info)) {
5989 trace_kvm_msr_read_ex(ecx);
5990 kvm_inject_gp(vcpu, 0);
5994 trace_kvm_msr_read(ecx, msr_info.data);
5996 /* FIXME: handling of bits 32:63 of rax, rdx */
5997 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
5998 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
5999 return kvm_skip_emulated_instruction(vcpu);
6002 static int handle_wrmsr(struct kvm_vcpu *vcpu)
6004 struct msr_data msr;
6005 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6006 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
6007 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6011 msr.host_initiated = false;
6012 if (kvm_set_msr(vcpu, &msr) != 0) {
6013 trace_kvm_msr_write_ex(ecx, data);
6014 kvm_inject_gp(vcpu, 0);
6018 trace_kvm_msr_write(ecx, data);
6019 return kvm_skip_emulated_instruction(vcpu);
6022 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
6024 kvm_apic_update_ppr(vcpu);
6028 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
6030 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6031 CPU_BASED_VIRTUAL_INTR_PENDING);
6033 kvm_make_request(KVM_REQ_EVENT, vcpu);
6035 ++vcpu->stat.irq_window_exits;
6039 static int handle_halt(struct kvm_vcpu *vcpu)
6041 return kvm_emulate_halt(vcpu);
6044 static int handle_vmcall(struct kvm_vcpu *vcpu)
6046 return kvm_emulate_hypercall(vcpu);
6049 static int handle_invd(struct kvm_vcpu *vcpu)
6051 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6054 static int handle_invlpg(struct kvm_vcpu *vcpu)
6056 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6058 kvm_mmu_invlpg(vcpu, exit_qualification);
6059 return kvm_skip_emulated_instruction(vcpu);
6062 static int handle_rdpmc(struct kvm_vcpu *vcpu)
6066 err = kvm_rdpmc(vcpu);
6067 return kvm_complete_insn_gp(vcpu, err);
6070 static int handle_wbinvd(struct kvm_vcpu *vcpu)
6072 return kvm_emulate_wbinvd(vcpu);
6075 static int handle_xsetbv(struct kvm_vcpu *vcpu)
6077 u64 new_bv = kvm_read_edx_eax(vcpu);
6078 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
6080 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
6081 return kvm_skip_emulated_instruction(vcpu);
6085 static int handle_xsaves(struct kvm_vcpu *vcpu)
6087 kvm_skip_emulated_instruction(vcpu);
6088 WARN(1, "this should never happen\n");
6092 static int handle_xrstors(struct kvm_vcpu *vcpu)
6094 kvm_skip_emulated_instruction(vcpu);
6095 WARN(1, "this should never happen\n");
6099 static int handle_apic_access(struct kvm_vcpu *vcpu)
6101 if (likely(fasteoi)) {
6102 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6103 int access_type, offset;
6105 access_type = exit_qualification & APIC_ACCESS_TYPE;
6106 offset = exit_qualification & APIC_ACCESS_OFFSET;
6108 * Sane guest uses MOV to write EOI, with written value
6109 * not cared. So make a short-circuit here by avoiding
6110 * heavy instruction emulation.
6112 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
6113 (offset == APIC_EOI)) {
6114 kvm_lapic_set_eoi(vcpu);
6115 return kvm_skip_emulated_instruction(vcpu);
6118 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6121 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
6123 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6124 int vector = exit_qualification & 0xff;
6126 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
6127 kvm_apic_set_eoi_accelerated(vcpu, vector);
6131 static int handle_apic_write(struct kvm_vcpu *vcpu)
6133 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6134 u32 offset = exit_qualification & 0xfff;
6136 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6137 kvm_apic_write_nodecode(vcpu, offset);
6141 static int handle_task_switch(struct kvm_vcpu *vcpu)
6143 struct vcpu_vmx *vmx = to_vmx(vcpu);
6144 unsigned long exit_qualification;
6145 bool has_error_code = false;
6148 int reason, type, idt_v, idt_index;
6150 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
6151 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
6152 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
6154 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6156 reason = (u32)exit_qualification >> 30;
6157 if (reason == TASK_SWITCH_GATE && idt_v) {
6159 case INTR_TYPE_NMI_INTR:
6160 vcpu->arch.nmi_injected = false;
6161 vmx_set_nmi_mask(vcpu, true);
6163 case INTR_TYPE_EXT_INTR:
6164 case INTR_TYPE_SOFT_INTR:
6165 kvm_clear_interrupt_queue(vcpu);
6167 case INTR_TYPE_HARD_EXCEPTION:
6168 if (vmx->idt_vectoring_info &
6169 VECTORING_INFO_DELIVER_CODE_MASK) {
6170 has_error_code = true;
6172 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6175 case INTR_TYPE_SOFT_EXCEPTION:
6176 kvm_clear_exception_queue(vcpu);
6182 tss_selector = exit_qualification;
6184 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
6185 type != INTR_TYPE_EXT_INTR &&
6186 type != INTR_TYPE_NMI_INTR))
6187 skip_emulated_instruction(vcpu);
6189 if (kvm_task_switch(vcpu, tss_selector,
6190 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
6191 has_error_code, error_code) == EMULATE_FAIL) {
6192 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6193 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6194 vcpu->run->internal.ndata = 0;
6199 * TODO: What about debug traps on tss switch?
6200 * Are we supposed to inject them and update dr6?
6206 static int handle_ept_violation(struct kvm_vcpu *vcpu)
6208 unsigned long exit_qualification;
6213 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6215 gla_validity = (exit_qualification >> 7) & 0x3;
6216 if (gla_validity == 0x2) {
6217 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
6218 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
6219 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
6220 vmcs_readl(GUEST_LINEAR_ADDRESS));
6221 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
6222 (long unsigned int)exit_qualification);
6223 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6224 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
6229 * EPT violation happened while executing iret from NMI,
6230 * "blocked by NMI" bit has to be set before next VM entry.
6231 * There are errata that may cause this bit to not be set:
6234 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
6235 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
6236 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
6238 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6239 trace_kvm_page_fault(gpa, exit_qualification);
6241 /* Is it a read fault? */
6242 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
6243 ? PFERR_USER_MASK : 0;
6244 /* Is it a write fault? */
6245 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
6246 ? PFERR_WRITE_MASK : 0;
6247 /* Is it a fetch fault? */
6248 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
6249 ? PFERR_FETCH_MASK : 0;
6250 /* ept page table entry is present? */
6251 error_code |= (exit_qualification &
6252 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
6253 EPT_VIOLATION_EXECUTABLE))
6254 ? PFERR_PRESENT_MASK : 0;
6256 vcpu->arch.gpa_available = true;
6257 vcpu->arch.exit_qualification = exit_qualification;
6259 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
6262 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
6267 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6268 if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
6269 trace_kvm_fast_mmio(gpa);
6270 return kvm_skip_emulated_instruction(vcpu);
6273 ret = handle_mmio_page_fault(vcpu, gpa, true);
6274 vcpu->arch.gpa_available = true;
6275 if (likely(ret == RET_MMIO_PF_EMULATE))
6276 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
6279 if (unlikely(ret == RET_MMIO_PF_INVALID))
6280 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
6282 if (unlikely(ret == RET_MMIO_PF_RETRY))
6285 /* It is the real ept misconfig */
6288 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6289 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6294 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6296 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6297 CPU_BASED_VIRTUAL_NMI_PENDING);
6298 ++vcpu->stat.nmi_window_exits;
6299 kvm_make_request(KVM_REQ_EVENT, vcpu);
6304 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6306 struct vcpu_vmx *vmx = to_vmx(vcpu);
6307 enum emulation_result err = EMULATE_DONE;
6310 bool intr_window_requested;
6311 unsigned count = 130;
6313 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6314 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6316 while (vmx->emulation_required && count-- != 0) {
6317 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6318 return handle_interrupt_window(&vmx->vcpu);
6320 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
6323 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
6325 if (err == EMULATE_USER_EXIT) {
6326 ++vcpu->stat.mmio_exits;
6331 if (err != EMULATE_DONE) {
6332 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6333 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6334 vcpu->run->internal.ndata = 0;
6338 if (vcpu->arch.halt_request) {
6339 vcpu->arch.halt_request = 0;
6340 ret = kvm_vcpu_halt(vcpu);
6344 if (signal_pending(current))
6354 static int __grow_ple_window(int val)
6356 if (ple_window_grow < 1)
6359 val = min(val, ple_window_actual_max);
6361 if (ple_window_grow < ple_window)
6362 val *= ple_window_grow;
6364 val += ple_window_grow;
6369 static int __shrink_ple_window(int val, int modifier, int minimum)
6374 if (modifier < ple_window)
6379 return max(val, minimum);
6382 static void grow_ple_window(struct kvm_vcpu *vcpu)
6384 struct vcpu_vmx *vmx = to_vmx(vcpu);
6385 int old = vmx->ple_window;
6387 vmx->ple_window = __grow_ple_window(old);
6389 if (vmx->ple_window != old)
6390 vmx->ple_window_dirty = true;
6392 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6395 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6397 struct vcpu_vmx *vmx = to_vmx(vcpu);
6398 int old = vmx->ple_window;
6400 vmx->ple_window = __shrink_ple_window(old,
6401 ple_window_shrink, ple_window);
6403 if (vmx->ple_window != old)
6404 vmx->ple_window_dirty = true;
6406 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6410 * ple_window_actual_max is computed to be one grow_ple_window() below
6411 * ple_window_max. (See __grow_ple_window for the reason.)
6412 * This prevents overflows, because ple_window_max is int.
6413 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6415 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6417 static void update_ple_window_actual_max(void)
6419 ple_window_actual_max =
6420 __shrink_ple_window(max(ple_window_max, ple_window),
6421 ple_window_grow, INT_MIN);
6425 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6427 static void wakeup_handler(void)
6429 struct kvm_vcpu *vcpu;
6430 int cpu = smp_processor_id();
6432 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6433 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6434 blocked_vcpu_list) {
6435 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6437 if (pi_test_on(pi_desc) == 1)
6438 kvm_vcpu_kick(vcpu);
6440 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6443 void vmx_enable_tdp(void)
6445 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
6446 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
6447 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
6448 0ull, VMX_EPT_EXECUTABLE_MASK,
6449 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
6450 enable_ept_ad_bits ? 0ull : VMX_EPT_RWX_MASK);
6452 ept_set_mmio_spte_mask();
6456 static __init int hardware_setup(void)
6458 int r = -ENOMEM, i, msr;
6460 rdmsrl_safe(MSR_EFER, &host_efer);
6462 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6463 kvm_define_shared_msr(i, vmx_msr_index[i]);
6465 for (i = 0; i < VMX_BITMAP_NR; i++) {
6466 vmx_bitmap[i] = (unsigned long *)__get_free_page(GFP_KERNEL);
6471 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6472 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6473 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6476 * Allow direct access to the PC debug port (it is often used for I/O
6477 * delays, but the vmexits simply slow things down).
6479 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6480 clear_bit(0x80, vmx_io_bitmap_a);
6482 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6484 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6485 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6487 if (setup_vmcs_config(&vmcs_config) < 0) {
6492 if (boot_cpu_has(X86_FEATURE_NX))
6493 kvm_enable_efer_bits(EFER_NX);
6495 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
6496 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
6499 if (!cpu_has_vmx_shadow_vmcs())
6500 enable_shadow_vmcs = 0;
6501 if (enable_shadow_vmcs)
6502 init_vmcs_shadow_fields();
6504 if (!cpu_has_vmx_ept() ||
6505 !cpu_has_vmx_ept_4levels()) {
6507 enable_unrestricted_guest = 0;
6508 enable_ept_ad_bits = 0;
6511 if (!cpu_has_vmx_ept_ad_bits())
6512 enable_ept_ad_bits = 0;
6514 if (!cpu_has_vmx_unrestricted_guest())
6515 enable_unrestricted_guest = 0;
6517 if (!cpu_has_vmx_flexpriority())
6518 flexpriority_enabled = 0;
6521 * set_apic_access_page_addr() is used to reload apic access
6522 * page upon invalidation. No need to do anything if not
6523 * using the APIC_ACCESS_ADDR VMCS field.
6525 if (!flexpriority_enabled)
6526 kvm_x86_ops->set_apic_access_page_addr = NULL;
6528 if (!cpu_has_vmx_tpr_shadow())
6529 kvm_x86_ops->update_cr8_intercept = NULL;
6531 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6532 kvm_disable_largepages();
6534 if (!cpu_has_vmx_ple())
6537 if (!cpu_has_vmx_apicv()) {
6539 kvm_x86_ops->sync_pir_to_irr = NULL;
6542 if (cpu_has_vmx_tsc_scaling()) {
6543 kvm_has_tsc_control = true;
6544 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6545 kvm_tsc_scaling_ratio_frac_bits = 48;
6548 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6549 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6550 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6551 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6552 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6553 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6554 vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
6556 memcpy(vmx_msr_bitmap_legacy_x2apic_apicv,
6557 vmx_msr_bitmap_legacy, PAGE_SIZE);
6558 memcpy(vmx_msr_bitmap_longmode_x2apic_apicv,
6559 vmx_msr_bitmap_longmode, PAGE_SIZE);
6560 memcpy(vmx_msr_bitmap_legacy_x2apic,
6561 vmx_msr_bitmap_legacy, PAGE_SIZE);
6562 memcpy(vmx_msr_bitmap_longmode_x2apic,
6563 vmx_msr_bitmap_longmode, PAGE_SIZE);
6565 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6567 for (msr = 0x800; msr <= 0x8ff; msr++) {
6568 if (msr == 0x839 /* TMCCT */)
6570 vmx_disable_intercept_msr_x2apic(msr, MSR_TYPE_R, true);
6574 * TPR reads and writes can be virtualized even if virtual interrupt
6575 * delivery is not in use.
6577 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_W, true);
6578 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_R | MSR_TYPE_W, false);
6581 vmx_disable_intercept_msr_x2apic(0x80b, MSR_TYPE_W, true);
6583 vmx_disable_intercept_msr_x2apic(0x83f, MSR_TYPE_W, true);
6590 update_ple_window_actual_max();
6593 * Only enable PML when hardware supports PML feature, and both EPT
6594 * and EPT A/D bit features are enabled -- PML depends on them to work.
6596 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6600 kvm_x86_ops->slot_enable_log_dirty = NULL;
6601 kvm_x86_ops->slot_disable_log_dirty = NULL;
6602 kvm_x86_ops->flush_log_dirty = NULL;
6603 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6606 if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
6609 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
6610 cpu_preemption_timer_multi =
6611 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
6613 kvm_x86_ops->set_hv_timer = NULL;
6614 kvm_x86_ops->cancel_hv_timer = NULL;
6617 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6619 kvm_mce_cap_supported |= MCG_LMCE_P;
6621 return alloc_kvm_area();
6624 for (i = 0; i < VMX_BITMAP_NR; i++)
6625 free_page((unsigned long)vmx_bitmap[i]);
6630 static __exit void hardware_unsetup(void)
6634 for (i = 0; i < VMX_BITMAP_NR; i++)
6635 free_page((unsigned long)vmx_bitmap[i]);
6641 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6642 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6644 static int handle_pause(struct kvm_vcpu *vcpu)
6647 grow_ple_window(vcpu);
6649 kvm_vcpu_on_spin(vcpu);
6650 return kvm_skip_emulated_instruction(vcpu);
6653 static int handle_nop(struct kvm_vcpu *vcpu)
6655 return kvm_skip_emulated_instruction(vcpu);
6658 static int handle_mwait(struct kvm_vcpu *vcpu)
6660 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6661 return handle_nop(vcpu);
6664 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6669 static int handle_monitor(struct kvm_vcpu *vcpu)
6671 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6672 return handle_nop(vcpu);
6676 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6677 * We could reuse a single VMCS for all the L2 guests, but we also want the
6678 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6679 * allows keeping them loaded on the processor, and in the future will allow
6680 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6681 * every entry if they never change.
6682 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6683 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6685 * The following functions allocate and free a vmcs02 in this pool.
6688 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6689 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6691 struct vmcs02_list *item;
6692 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6693 if (item->vmptr == vmx->nested.current_vmptr) {
6694 list_move(&item->list, &vmx->nested.vmcs02_pool);
6695 return &item->vmcs02;
6698 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6699 /* Recycle the least recently used VMCS. */
6700 item = list_last_entry(&vmx->nested.vmcs02_pool,
6701 struct vmcs02_list, list);
6702 item->vmptr = vmx->nested.current_vmptr;
6703 list_move(&item->list, &vmx->nested.vmcs02_pool);
6704 return &item->vmcs02;
6707 /* Create a new VMCS */
6708 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6711 item->vmcs02.vmcs = alloc_vmcs();
6712 item->vmcs02.shadow_vmcs = NULL;
6713 if (!item->vmcs02.vmcs) {
6717 loaded_vmcs_init(&item->vmcs02);
6718 item->vmptr = vmx->nested.current_vmptr;
6719 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6720 vmx->nested.vmcs02_num++;
6721 return &item->vmcs02;
6724 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6725 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6727 struct vmcs02_list *item;
6728 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6729 if (item->vmptr == vmptr) {
6730 free_loaded_vmcs(&item->vmcs02);
6731 list_del(&item->list);
6733 vmx->nested.vmcs02_num--;
6739 * Free all VMCSs saved for this vcpu, except the one pointed by
6740 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6741 * must be &vmx->vmcs01.
6743 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6745 struct vmcs02_list *item, *n;
6747 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6748 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6750 * Something will leak if the above WARN triggers. Better than
6753 if (vmx->loaded_vmcs == &item->vmcs02)
6756 free_loaded_vmcs(&item->vmcs02);
6757 list_del(&item->list);
6759 vmx->nested.vmcs02_num--;
6764 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6765 * set the success or error code of an emulated VMX instruction, as specified
6766 * by Vol 2B, VMX Instruction Reference, "Conventions".
6768 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6770 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6771 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6772 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6775 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6777 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6778 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6779 X86_EFLAGS_SF | X86_EFLAGS_OF))
6783 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6784 u32 vm_instruction_error)
6786 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6788 * failValid writes the error number to the current VMCS, which
6789 * can't be done there isn't a current VMCS.
6791 nested_vmx_failInvalid(vcpu);
6794 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6795 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6796 X86_EFLAGS_SF | X86_EFLAGS_OF))
6798 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6800 * We don't need to force a shadow sync because
6801 * VM_INSTRUCTION_ERROR is not shadowed
6805 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6807 /* TODO: not to reset guest simply here. */
6808 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6809 pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
6812 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6814 struct vcpu_vmx *vmx =
6815 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6817 vmx->nested.preemption_timer_expired = true;
6818 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6819 kvm_vcpu_kick(&vmx->vcpu);
6821 return HRTIMER_NORESTART;
6825 * Decode the memory-address operand of a vmx instruction, as recorded on an
6826 * exit caused by such an instruction (run by a guest hypervisor).
6827 * On success, returns 0. When the operand is invalid, returns 1 and throws
6830 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6831 unsigned long exit_qualification,
6832 u32 vmx_instruction_info, bool wr, gva_t *ret)
6836 struct kvm_segment s;
6839 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6840 * Execution", on an exit, vmx_instruction_info holds most of the
6841 * addressing components of the operand. Only the displacement part
6842 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6843 * For how an actual address is calculated from all these components,
6844 * refer to Vol. 1, "Operand Addressing".
6846 int scaling = vmx_instruction_info & 3;
6847 int addr_size = (vmx_instruction_info >> 7) & 7;
6848 bool is_reg = vmx_instruction_info & (1u << 10);
6849 int seg_reg = (vmx_instruction_info >> 15) & 7;
6850 int index_reg = (vmx_instruction_info >> 18) & 0xf;
6851 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6852 int base_reg = (vmx_instruction_info >> 23) & 0xf;
6853 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
6856 kvm_queue_exception(vcpu, UD_VECTOR);
6860 /* Addr = segment_base + offset */
6861 /* offset = base + [index * scale] + displacement */
6862 off = exit_qualification; /* holds the displacement */
6864 off += kvm_register_read(vcpu, base_reg);
6866 off += kvm_register_read(vcpu, index_reg)<<scaling;
6867 vmx_get_segment(vcpu, &s, seg_reg);
6868 *ret = s.base + off;
6870 if (addr_size == 1) /* 32 bit */
6873 /* Checks for #GP/#SS exceptions. */
6875 if (is_long_mode(vcpu)) {
6876 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6877 * non-canonical form. This is the only check on the memory
6878 * destination for long mode!
6880 exn = is_noncanonical_address(*ret);
6881 } else if (is_protmode(vcpu)) {
6882 /* Protected mode: apply checks for segment validity in the
6884 * - segment type check (#GP(0) may be thrown)
6885 * - usability check (#GP(0)/#SS(0))
6886 * - limit check (#GP(0)/#SS(0))
6889 /* #GP(0) if the destination operand is located in a
6890 * read-only data segment or any code segment.
6892 exn = ((s.type & 0xa) == 0 || (s.type & 8));
6894 /* #GP(0) if the source operand is located in an
6895 * execute-only code segment
6897 exn = ((s.type & 0xa) == 8);
6899 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6902 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
6904 exn = (s.unusable != 0);
6905 /* Protected mode: #GP(0)/#SS(0) if the memory
6906 * operand is outside the segment limit.
6908 exn = exn || (off + sizeof(u64) > s.limit);
6911 kvm_queue_exception_e(vcpu,
6912 seg_reg == VCPU_SREG_SS ?
6913 SS_VECTOR : GP_VECTOR,
6922 * This function performs the various checks including
6923 * - if it's 4KB aligned
6924 * - No bits beyond the physical address width are set
6925 * - Returns 0 on success or else 1
6926 * (Intel SDM Section 30.3)
6928 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6933 struct x86_exception e;
6935 struct vcpu_vmx *vmx = to_vmx(vcpu);
6936 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6938 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6939 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
6942 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6943 sizeof(vmptr), &e)) {
6944 kvm_inject_page_fault(vcpu, &e);
6948 switch (exit_reason) {
6949 case EXIT_REASON_VMON:
6952 * The first 4 bytes of VMXON region contain the supported
6953 * VMCS revision identifier
6955 * Note - IA32_VMX_BASIC[48] will never be 1
6956 * for the nested case;
6957 * which replaces physical address width with 32
6960 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6961 nested_vmx_failInvalid(vcpu);
6962 return kvm_skip_emulated_instruction(vcpu);
6965 page = nested_get_page(vcpu, vmptr);
6967 nested_vmx_failInvalid(vcpu);
6968 return kvm_skip_emulated_instruction(vcpu);
6970 if (*(u32 *)kmap(page) != VMCS12_REVISION) {
6972 nested_release_page_clean(page);
6973 nested_vmx_failInvalid(vcpu);
6974 return kvm_skip_emulated_instruction(vcpu);
6977 nested_release_page_clean(page);
6978 vmx->nested.vmxon_ptr = vmptr;
6980 case EXIT_REASON_VMCLEAR:
6981 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6982 nested_vmx_failValid(vcpu,
6983 VMXERR_VMCLEAR_INVALID_ADDRESS);
6984 return kvm_skip_emulated_instruction(vcpu);
6987 if (vmptr == vmx->nested.vmxon_ptr) {
6988 nested_vmx_failValid(vcpu,
6989 VMXERR_VMCLEAR_VMXON_POINTER);
6990 return kvm_skip_emulated_instruction(vcpu);
6993 case EXIT_REASON_VMPTRLD:
6994 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6995 nested_vmx_failValid(vcpu,
6996 VMXERR_VMPTRLD_INVALID_ADDRESS);
6997 return kvm_skip_emulated_instruction(vcpu);
7000 if (vmptr == vmx->nested.vmxon_ptr) {
7001 nested_vmx_failValid(vcpu,
7002 VMXERR_VMPTRLD_VMXON_POINTER);
7003 return kvm_skip_emulated_instruction(vcpu);
7007 return 1; /* shouldn't happen */
7015 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
7017 struct vcpu_vmx *vmx = to_vmx(vcpu);
7018 struct vmcs *shadow_vmcs;
7020 if (cpu_has_vmx_msr_bitmap()) {
7021 vmx->nested.msr_bitmap =
7022 (unsigned long *)__get_free_page(GFP_KERNEL);
7023 if (!vmx->nested.msr_bitmap)
7024 goto out_msr_bitmap;
7027 vmx->nested.cached_vmcs12 = kmalloc(VMCS12_SIZE, GFP_KERNEL);
7028 if (!vmx->nested.cached_vmcs12)
7029 goto out_cached_vmcs12;
7031 if (enable_shadow_vmcs) {
7032 shadow_vmcs = alloc_vmcs();
7034 goto out_shadow_vmcs;
7035 /* mark vmcs as shadow */
7036 shadow_vmcs->revision_id |= (1u << 31);
7037 /* init shadow vmcs */
7038 vmcs_clear(shadow_vmcs);
7039 vmx->vmcs01.shadow_vmcs = shadow_vmcs;
7042 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
7043 vmx->nested.vmcs02_num = 0;
7045 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
7046 HRTIMER_MODE_REL_PINNED);
7047 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
7049 vmx->nested.vmxon = true;
7053 kfree(vmx->nested.cached_vmcs12);
7056 free_page((unsigned long)vmx->nested.msr_bitmap);
7063 * Emulate the VMXON instruction.
7064 * Currently, we just remember that VMX is active, and do not save or even
7065 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
7066 * do not currently need to store anything in that guest-allocated memory
7067 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
7068 * argument is different from the VMXON pointer (which the spec says they do).
7070 static int handle_vmon(struct kvm_vcpu *vcpu)
7073 struct kvm_segment cs;
7074 struct vcpu_vmx *vmx = to_vmx(vcpu);
7075 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
7076 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7078 /* The Intel VMX Instruction Reference lists a bunch of bits that
7079 * are prerequisite to running VMXON, most notably cr4.VMXE must be
7080 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
7081 * Otherwise, we should fail with #UD. We test these now:
7083 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
7084 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
7085 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
7086 kvm_queue_exception(vcpu, UD_VECTOR);
7090 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
7091 if (is_long_mode(vcpu) && !cs.l) {
7092 kvm_queue_exception(vcpu, UD_VECTOR);
7096 if (vmx_get_cpl(vcpu)) {
7097 kvm_inject_gp(vcpu, 0);
7101 if (vmx->nested.vmxon) {
7102 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
7103 return kvm_skip_emulated_instruction(vcpu);
7106 if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
7107 != VMXON_NEEDED_FEATURES) {
7108 kvm_inject_gp(vcpu, 0);
7112 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
7115 ret = enter_vmx_operation(vcpu);
7119 nested_vmx_succeed(vcpu);
7120 return kvm_skip_emulated_instruction(vcpu);
7124 * Intel's VMX Instruction Reference specifies a common set of prerequisites
7125 * for running VMX instructions (except VMXON, whose prerequisites are
7126 * slightly different). It also specifies what exception to inject otherwise.
7128 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
7130 struct kvm_segment cs;
7131 struct vcpu_vmx *vmx = to_vmx(vcpu);
7133 if (!vmx->nested.vmxon) {
7134 kvm_queue_exception(vcpu, UD_VECTOR);
7138 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
7139 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
7140 (is_long_mode(vcpu) && !cs.l)) {
7141 kvm_queue_exception(vcpu, UD_VECTOR);
7145 if (vmx_get_cpl(vcpu)) {
7146 kvm_inject_gp(vcpu, 0);
7153 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
7155 if (vmx->nested.current_vmptr == -1ull)
7158 /* current_vmptr and current_vmcs12 are always set/reset together */
7159 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
7162 if (enable_shadow_vmcs) {
7163 /* copy to memory all shadowed fields in case
7164 they were modified */
7165 copy_shadow_to_vmcs12(vmx);
7166 vmx->nested.sync_shadow_vmcs = false;
7167 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
7168 SECONDARY_EXEC_SHADOW_VMCS);
7169 vmcs_write64(VMCS_LINK_POINTER, -1ull);
7171 vmx->nested.posted_intr_nv = -1;
7173 /* Flush VMCS12 to guest memory */
7174 memcpy(vmx->nested.current_vmcs12, vmx->nested.cached_vmcs12,
7177 kunmap(vmx->nested.current_vmcs12_page);
7178 nested_release_page(vmx->nested.current_vmcs12_page);
7179 vmx->nested.current_vmptr = -1ull;
7180 vmx->nested.current_vmcs12 = NULL;
7184 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7185 * just stops using VMX.
7187 static void free_nested(struct vcpu_vmx *vmx)
7189 if (!vmx->nested.vmxon)
7192 vmx->nested.vmxon = false;
7193 free_vpid(vmx->nested.vpid02);
7194 nested_release_vmcs12(vmx);
7195 if (vmx->nested.msr_bitmap) {
7196 free_page((unsigned long)vmx->nested.msr_bitmap);
7197 vmx->nested.msr_bitmap = NULL;
7199 if (enable_shadow_vmcs) {
7200 vmcs_clear(vmx->vmcs01.shadow_vmcs);
7201 free_vmcs(vmx->vmcs01.shadow_vmcs);
7202 vmx->vmcs01.shadow_vmcs = NULL;
7204 kfree(vmx->nested.cached_vmcs12);
7205 /* Unpin physical memory we referred to in current vmcs02 */
7206 if (vmx->nested.apic_access_page) {
7207 nested_release_page(vmx->nested.apic_access_page);
7208 vmx->nested.apic_access_page = NULL;
7210 if (vmx->nested.virtual_apic_page) {
7211 nested_release_page(vmx->nested.virtual_apic_page);
7212 vmx->nested.virtual_apic_page = NULL;
7214 if (vmx->nested.pi_desc_page) {
7215 kunmap(vmx->nested.pi_desc_page);
7216 nested_release_page(vmx->nested.pi_desc_page);
7217 vmx->nested.pi_desc_page = NULL;
7218 vmx->nested.pi_desc = NULL;
7221 nested_free_all_saved_vmcss(vmx);
7224 /* Emulate the VMXOFF instruction */
7225 static int handle_vmoff(struct kvm_vcpu *vcpu)
7227 if (!nested_vmx_check_permission(vcpu))
7229 free_nested(to_vmx(vcpu));
7230 nested_vmx_succeed(vcpu);
7231 return kvm_skip_emulated_instruction(vcpu);
7234 /* Emulate the VMCLEAR instruction */
7235 static int handle_vmclear(struct kvm_vcpu *vcpu)
7237 struct vcpu_vmx *vmx = to_vmx(vcpu);
7241 if (!nested_vmx_check_permission(vcpu))
7244 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
7247 if (vmptr == vmx->nested.current_vmptr)
7248 nested_release_vmcs12(vmx);
7250 kvm_vcpu_write_guest(vcpu,
7251 vmptr + offsetof(struct vmcs12, launch_state),
7252 &zero, sizeof(zero));
7254 nested_free_vmcs02(vmx, vmptr);
7256 nested_vmx_succeed(vcpu);
7257 return kvm_skip_emulated_instruction(vcpu);
7260 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
7262 /* Emulate the VMLAUNCH instruction */
7263 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
7265 return nested_vmx_run(vcpu, true);
7268 /* Emulate the VMRESUME instruction */
7269 static int handle_vmresume(struct kvm_vcpu *vcpu)
7272 return nested_vmx_run(vcpu, false);
7275 enum vmcs_field_type {
7276 VMCS_FIELD_TYPE_U16 = 0,
7277 VMCS_FIELD_TYPE_U64 = 1,
7278 VMCS_FIELD_TYPE_U32 = 2,
7279 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
7282 static inline int vmcs_field_type(unsigned long field)
7284 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
7285 return VMCS_FIELD_TYPE_U32;
7286 return (field >> 13) & 0x3 ;
7289 static inline int vmcs_field_readonly(unsigned long field)
7291 return (((field >> 10) & 0x3) == 1);
7295 * Read a vmcs12 field. Since these can have varying lengths and we return
7296 * one type, we chose the biggest type (u64) and zero-extend the return value
7297 * to that size. Note that the caller, handle_vmread, might need to use only
7298 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7299 * 64-bit fields are to be returned).
7301 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7302 unsigned long field, u64 *ret)
7304 short offset = vmcs_field_to_offset(field);
7310 p = ((char *)(get_vmcs12(vcpu))) + offset;
7312 switch (vmcs_field_type(field)) {
7313 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7314 *ret = *((natural_width *)p);
7316 case VMCS_FIELD_TYPE_U16:
7319 case VMCS_FIELD_TYPE_U32:
7322 case VMCS_FIELD_TYPE_U64:
7332 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7333 unsigned long field, u64 field_value){
7334 short offset = vmcs_field_to_offset(field);
7335 char *p = ((char *) get_vmcs12(vcpu)) + offset;
7339 switch (vmcs_field_type(field)) {
7340 case VMCS_FIELD_TYPE_U16:
7341 *(u16 *)p = field_value;
7343 case VMCS_FIELD_TYPE_U32:
7344 *(u32 *)p = field_value;
7346 case VMCS_FIELD_TYPE_U64:
7347 *(u64 *)p = field_value;
7349 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7350 *(natural_width *)p = field_value;
7359 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7362 unsigned long field;
7364 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7365 const unsigned long *fields = shadow_read_write_fields;
7366 const int num_fields = max_shadow_read_write_fields;
7370 vmcs_load(shadow_vmcs);
7372 for (i = 0; i < num_fields; i++) {
7374 switch (vmcs_field_type(field)) {
7375 case VMCS_FIELD_TYPE_U16:
7376 field_value = vmcs_read16(field);
7378 case VMCS_FIELD_TYPE_U32:
7379 field_value = vmcs_read32(field);
7381 case VMCS_FIELD_TYPE_U64:
7382 field_value = vmcs_read64(field);
7384 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7385 field_value = vmcs_readl(field);
7391 vmcs12_write_any(&vmx->vcpu, field, field_value);
7394 vmcs_clear(shadow_vmcs);
7395 vmcs_load(vmx->loaded_vmcs->vmcs);
7400 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7402 const unsigned long *fields[] = {
7403 shadow_read_write_fields,
7404 shadow_read_only_fields
7406 const int max_fields[] = {
7407 max_shadow_read_write_fields,
7408 max_shadow_read_only_fields
7411 unsigned long field;
7412 u64 field_value = 0;
7413 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7415 vmcs_load(shadow_vmcs);
7417 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7418 for (i = 0; i < max_fields[q]; i++) {
7419 field = fields[q][i];
7420 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7422 switch (vmcs_field_type(field)) {
7423 case VMCS_FIELD_TYPE_U16:
7424 vmcs_write16(field, (u16)field_value);
7426 case VMCS_FIELD_TYPE_U32:
7427 vmcs_write32(field, (u32)field_value);
7429 case VMCS_FIELD_TYPE_U64:
7430 vmcs_write64(field, (u64)field_value);
7432 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7433 vmcs_writel(field, (long)field_value);
7442 vmcs_clear(shadow_vmcs);
7443 vmcs_load(vmx->loaded_vmcs->vmcs);
7447 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7448 * used before) all generate the same failure when it is missing.
7450 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7452 struct vcpu_vmx *vmx = to_vmx(vcpu);
7453 if (vmx->nested.current_vmptr == -1ull) {
7454 nested_vmx_failInvalid(vcpu);
7460 static int handle_vmread(struct kvm_vcpu *vcpu)
7462 unsigned long field;
7464 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7465 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7468 if (!nested_vmx_check_permission(vcpu))
7471 if (!nested_vmx_check_vmcs12(vcpu))
7472 return kvm_skip_emulated_instruction(vcpu);
7474 /* Decode instruction info and find the field to read */
7475 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7476 /* Read the field, zero-extended to a u64 field_value */
7477 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7478 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7479 return kvm_skip_emulated_instruction(vcpu);
7482 * Now copy part of this value to register or memory, as requested.
7483 * Note that the number of bits actually copied is 32 or 64 depending
7484 * on the guest's mode (32 or 64 bit), not on the given field's length.
7486 if (vmx_instruction_info & (1u << 10)) {
7487 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7490 if (get_vmx_mem_address(vcpu, exit_qualification,
7491 vmx_instruction_info, true, &gva))
7493 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
7494 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7495 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7498 nested_vmx_succeed(vcpu);
7499 return kvm_skip_emulated_instruction(vcpu);
7503 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7505 unsigned long field;
7507 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7508 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7509 /* The value to write might be 32 or 64 bits, depending on L1's long
7510 * mode, and eventually we need to write that into a field of several
7511 * possible lengths. The code below first zero-extends the value to 64
7512 * bit (field_value), and then copies only the appropriate number of
7513 * bits into the vmcs12 field.
7515 u64 field_value = 0;
7516 struct x86_exception e;
7518 if (!nested_vmx_check_permission(vcpu))
7521 if (!nested_vmx_check_vmcs12(vcpu))
7522 return kvm_skip_emulated_instruction(vcpu);
7524 if (vmx_instruction_info & (1u << 10))
7525 field_value = kvm_register_readl(vcpu,
7526 (((vmx_instruction_info) >> 3) & 0xf));
7528 if (get_vmx_mem_address(vcpu, exit_qualification,
7529 vmx_instruction_info, false, &gva))
7531 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7532 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7533 kvm_inject_page_fault(vcpu, &e);
7539 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7540 if (vmcs_field_readonly(field)) {
7541 nested_vmx_failValid(vcpu,
7542 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7543 return kvm_skip_emulated_instruction(vcpu);
7546 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7547 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7548 return kvm_skip_emulated_instruction(vcpu);
7551 nested_vmx_succeed(vcpu);
7552 return kvm_skip_emulated_instruction(vcpu);
7555 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
7557 vmx->nested.current_vmptr = vmptr;
7558 if (enable_shadow_vmcs) {
7559 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7560 SECONDARY_EXEC_SHADOW_VMCS);
7561 vmcs_write64(VMCS_LINK_POINTER,
7562 __pa(vmx->vmcs01.shadow_vmcs));
7563 vmx->nested.sync_shadow_vmcs = true;
7567 /* Emulate the VMPTRLD instruction */
7568 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7570 struct vcpu_vmx *vmx = to_vmx(vcpu);
7573 if (!nested_vmx_check_permission(vcpu))
7576 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
7579 if (vmx->nested.current_vmptr != vmptr) {
7580 struct vmcs12 *new_vmcs12;
7582 page = nested_get_page(vcpu, vmptr);
7584 nested_vmx_failInvalid(vcpu);
7585 return kvm_skip_emulated_instruction(vcpu);
7587 new_vmcs12 = kmap(page);
7588 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7590 nested_release_page_clean(page);
7591 nested_vmx_failValid(vcpu,
7592 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7593 return kvm_skip_emulated_instruction(vcpu);
7596 nested_release_vmcs12(vmx);
7597 vmx->nested.current_vmcs12 = new_vmcs12;
7598 vmx->nested.current_vmcs12_page = page;
7600 * Load VMCS12 from guest memory since it is not already
7603 memcpy(vmx->nested.cached_vmcs12,
7604 vmx->nested.current_vmcs12, VMCS12_SIZE);
7605 set_current_vmptr(vmx, vmptr);
7608 nested_vmx_succeed(vcpu);
7609 return kvm_skip_emulated_instruction(vcpu);
7612 /* Emulate the VMPTRST instruction */
7613 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7615 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7616 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7618 struct x86_exception e;
7620 if (!nested_vmx_check_permission(vcpu))
7623 if (get_vmx_mem_address(vcpu, exit_qualification,
7624 vmx_instruction_info, true, &vmcs_gva))
7626 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
7627 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7628 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7630 kvm_inject_page_fault(vcpu, &e);
7633 nested_vmx_succeed(vcpu);
7634 return kvm_skip_emulated_instruction(vcpu);
7637 /* Emulate the INVEPT instruction */
7638 static int handle_invept(struct kvm_vcpu *vcpu)
7640 struct vcpu_vmx *vmx = to_vmx(vcpu);
7641 u32 vmx_instruction_info, types;
7644 struct x86_exception e;
7649 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7650 SECONDARY_EXEC_ENABLE_EPT) ||
7651 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7652 kvm_queue_exception(vcpu, UD_VECTOR);
7656 if (!nested_vmx_check_permission(vcpu))
7659 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
7660 kvm_queue_exception(vcpu, UD_VECTOR);
7664 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7665 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7667 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7669 if (type >= 32 || !(types & (1 << type))) {
7670 nested_vmx_failValid(vcpu,
7671 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7672 return kvm_skip_emulated_instruction(vcpu);
7675 /* According to the Intel VMX instruction reference, the memory
7676 * operand is read even if it isn't needed (e.g., for type==global)
7678 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7679 vmx_instruction_info, false, &gva))
7681 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7682 sizeof(operand), &e)) {
7683 kvm_inject_page_fault(vcpu, &e);
7688 case VMX_EPT_EXTENT_GLOBAL:
7690 * TODO: track mappings and invalidate
7691 * single context requests appropriately
7693 case VMX_EPT_EXTENT_CONTEXT:
7694 kvm_mmu_sync_roots(vcpu);
7695 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7696 nested_vmx_succeed(vcpu);
7703 return kvm_skip_emulated_instruction(vcpu);
7706 static int handle_invvpid(struct kvm_vcpu *vcpu)
7708 struct vcpu_vmx *vmx = to_vmx(vcpu);
7709 u32 vmx_instruction_info;
7710 unsigned long type, types;
7712 struct x86_exception e;
7715 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7716 SECONDARY_EXEC_ENABLE_VPID) ||
7717 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7718 kvm_queue_exception(vcpu, UD_VECTOR);
7722 if (!nested_vmx_check_permission(vcpu))
7725 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7726 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7728 types = (vmx->nested.nested_vmx_vpid_caps &
7729 VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
7731 if (type >= 32 || !(types & (1 << type))) {
7732 nested_vmx_failValid(vcpu,
7733 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7734 return kvm_skip_emulated_instruction(vcpu);
7737 /* according to the intel vmx instruction reference, the memory
7738 * operand is read even if it isn't needed (e.g., for type==global)
7740 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7741 vmx_instruction_info, false, &gva))
7743 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vpid,
7745 kvm_inject_page_fault(vcpu, &e);
7750 case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
7751 case VMX_VPID_EXTENT_SINGLE_CONTEXT:
7752 case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
7754 nested_vmx_failValid(vcpu,
7755 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7756 return kvm_skip_emulated_instruction(vcpu);
7759 case VMX_VPID_EXTENT_ALL_CONTEXT:
7763 return kvm_skip_emulated_instruction(vcpu);
7766 __vmx_flush_tlb(vcpu, vmx->nested.vpid02);
7767 nested_vmx_succeed(vcpu);
7769 return kvm_skip_emulated_instruction(vcpu);
7772 static int handle_pml_full(struct kvm_vcpu *vcpu)
7774 unsigned long exit_qualification;
7776 trace_kvm_pml_full(vcpu->vcpu_id);
7778 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7781 * PML buffer FULL happened while executing iret from NMI,
7782 * "blocked by NMI" bit has to be set before next VM entry.
7784 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7785 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7786 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7787 GUEST_INTR_STATE_NMI);
7790 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7791 * here.., and there's no userspace involvement needed for PML.
7796 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
7798 kvm_lapic_expired_hv_timer(vcpu);
7803 * The exit handlers return 1 if the exit was handled fully and guest execution
7804 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
7805 * to be done to userspace and return 0.
7807 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7808 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
7809 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
7810 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
7811 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
7812 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
7813 [EXIT_REASON_CR_ACCESS] = handle_cr,
7814 [EXIT_REASON_DR_ACCESS] = handle_dr,
7815 [EXIT_REASON_CPUID] = handle_cpuid,
7816 [EXIT_REASON_MSR_READ] = handle_rdmsr,
7817 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
7818 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
7819 [EXIT_REASON_HLT] = handle_halt,
7820 [EXIT_REASON_INVD] = handle_invd,
7821 [EXIT_REASON_INVLPG] = handle_invlpg,
7822 [EXIT_REASON_RDPMC] = handle_rdpmc,
7823 [EXIT_REASON_VMCALL] = handle_vmcall,
7824 [EXIT_REASON_VMCLEAR] = handle_vmclear,
7825 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
7826 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
7827 [EXIT_REASON_VMPTRST] = handle_vmptrst,
7828 [EXIT_REASON_VMREAD] = handle_vmread,
7829 [EXIT_REASON_VMRESUME] = handle_vmresume,
7830 [EXIT_REASON_VMWRITE] = handle_vmwrite,
7831 [EXIT_REASON_VMOFF] = handle_vmoff,
7832 [EXIT_REASON_VMON] = handle_vmon,
7833 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
7834 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
7835 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
7836 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
7837 [EXIT_REASON_WBINVD] = handle_wbinvd,
7838 [EXIT_REASON_XSETBV] = handle_xsetbv,
7839 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
7840 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
7841 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
7842 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
7843 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
7844 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
7845 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
7846 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
7847 [EXIT_REASON_INVEPT] = handle_invept,
7848 [EXIT_REASON_INVVPID] = handle_invvpid,
7849 [EXIT_REASON_XSAVES] = handle_xsaves,
7850 [EXIT_REASON_XRSTORS] = handle_xrstors,
7851 [EXIT_REASON_PML_FULL] = handle_pml_full,
7852 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
7855 static const int kvm_vmx_max_exit_handlers =
7856 ARRAY_SIZE(kvm_vmx_exit_handlers);
7858 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7859 struct vmcs12 *vmcs12)
7861 unsigned long exit_qualification;
7862 gpa_t bitmap, last_bitmap;
7867 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7868 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7870 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7872 port = exit_qualification >> 16;
7873 size = (exit_qualification & 7) + 1;
7875 last_bitmap = (gpa_t)-1;
7880 bitmap = vmcs12->io_bitmap_a;
7881 else if (port < 0x10000)
7882 bitmap = vmcs12->io_bitmap_b;
7885 bitmap += (port & 0x7fff) / 8;
7887 if (last_bitmap != bitmap)
7888 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
7890 if (b & (1 << (port & 7)))
7895 last_bitmap = bitmap;
7902 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7903 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7904 * disinterest in the current event (read or write a specific MSR) by using an
7905 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7907 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7908 struct vmcs12 *vmcs12, u32 exit_reason)
7910 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7913 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7917 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7918 * for the four combinations of read/write and low/high MSR numbers.
7919 * First we need to figure out which of the four to use:
7921 bitmap = vmcs12->msr_bitmap;
7922 if (exit_reason == EXIT_REASON_MSR_WRITE)
7924 if (msr_index >= 0xc0000000) {
7925 msr_index -= 0xc0000000;
7929 /* Then read the msr_index'th bit from this bitmap: */
7930 if (msr_index < 1024*8) {
7932 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
7934 return 1 & (b >> (msr_index & 7));
7936 return true; /* let L1 handle the wrong parameter */
7940 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7941 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7942 * intercept (via guest_host_mask etc.) the current event.
7944 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7945 struct vmcs12 *vmcs12)
7947 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7948 int cr = exit_qualification & 15;
7949 int reg = (exit_qualification >> 8) & 15;
7950 unsigned long val = kvm_register_readl(vcpu, reg);
7952 switch ((exit_qualification >> 4) & 3) {
7953 case 0: /* mov to cr */
7956 if (vmcs12->cr0_guest_host_mask &
7957 (val ^ vmcs12->cr0_read_shadow))
7961 if ((vmcs12->cr3_target_count >= 1 &&
7962 vmcs12->cr3_target_value0 == val) ||
7963 (vmcs12->cr3_target_count >= 2 &&
7964 vmcs12->cr3_target_value1 == val) ||
7965 (vmcs12->cr3_target_count >= 3 &&
7966 vmcs12->cr3_target_value2 == val) ||
7967 (vmcs12->cr3_target_count >= 4 &&
7968 vmcs12->cr3_target_value3 == val))
7970 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
7974 if (vmcs12->cr4_guest_host_mask &
7975 (vmcs12->cr4_read_shadow ^ val))
7979 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
7985 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
7986 (vmcs12->cr0_read_shadow & X86_CR0_TS))
7989 case 1: /* mov from cr */
7992 if (vmcs12->cpu_based_vm_exec_control &
7993 CPU_BASED_CR3_STORE_EXITING)
7997 if (vmcs12->cpu_based_vm_exec_control &
7998 CPU_BASED_CR8_STORE_EXITING)
8005 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
8006 * cr0. Other attempted changes are ignored, with no exit.
8008 if (vmcs12->cr0_guest_host_mask & 0xe &
8009 (val ^ vmcs12->cr0_read_shadow))
8011 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
8012 !(vmcs12->cr0_read_shadow & 0x1) &&
8021 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
8022 * should handle it ourselves in L0 (and then continue L2). Only call this
8023 * when in is_guest_mode (L2).
8025 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
8027 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8028 struct vcpu_vmx *vmx = to_vmx(vcpu);
8029 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8030 u32 exit_reason = vmx->exit_reason;
8032 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
8033 vmcs_readl(EXIT_QUALIFICATION),
8034 vmx->idt_vectoring_info,
8036 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8039 if (vmx->nested.nested_run_pending)
8042 if (unlikely(vmx->fail)) {
8043 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
8044 vmcs_read32(VM_INSTRUCTION_ERROR));
8048 switch (exit_reason) {
8049 case EXIT_REASON_EXCEPTION_NMI:
8050 if (is_nmi(intr_info))
8052 else if (is_page_fault(intr_info))
8054 else if (is_no_device(intr_info) &&
8055 !(vmcs12->guest_cr0 & X86_CR0_TS))
8057 else if (is_debug(intr_info) &&
8059 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
8061 else if (is_breakpoint(intr_info) &&
8062 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
8064 return vmcs12->exception_bitmap &
8065 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
8066 case EXIT_REASON_EXTERNAL_INTERRUPT:
8068 case EXIT_REASON_TRIPLE_FAULT:
8070 case EXIT_REASON_PENDING_INTERRUPT:
8071 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
8072 case EXIT_REASON_NMI_WINDOW:
8073 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
8074 case EXIT_REASON_TASK_SWITCH:
8076 case EXIT_REASON_CPUID:
8078 case EXIT_REASON_HLT:
8079 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
8080 case EXIT_REASON_INVD:
8082 case EXIT_REASON_INVLPG:
8083 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8084 case EXIT_REASON_RDPMC:
8085 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
8086 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
8087 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
8088 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
8089 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
8090 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
8091 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
8092 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
8093 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
8095 * VMX instructions trap unconditionally. This allows L1 to
8096 * emulate them for its L2 guest, i.e., allows 3-level nesting!
8099 case EXIT_REASON_CR_ACCESS:
8100 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
8101 case EXIT_REASON_DR_ACCESS:
8102 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
8103 case EXIT_REASON_IO_INSTRUCTION:
8104 return nested_vmx_exit_handled_io(vcpu, vmcs12);
8105 case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
8106 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
8107 case EXIT_REASON_MSR_READ:
8108 case EXIT_REASON_MSR_WRITE:
8109 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
8110 case EXIT_REASON_INVALID_STATE:
8112 case EXIT_REASON_MWAIT_INSTRUCTION:
8113 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
8114 case EXIT_REASON_MONITOR_TRAP_FLAG:
8115 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
8116 case EXIT_REASON_MONITOR_INSTRUCTION:
8117 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
8118 case EXIT_REASON_PAUSE_INSTRUCTION:
8119 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
8120 nested_cpu_has2(vmcs12,
8121 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
8122 case EXIT_REASON_MCE_DURING_VMENTRY:
8124 case EXIT_REASON_TPR_BELOW_THRESHOLD:
8125 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
8126 case EXIT_REASON_APIC_ACCESS:
8127 return nested_cpu_has2(vmcs12,
8128 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
8129 case EXIT_REASON_APIC_WRITE:
8130 case EXIT_REASON_EOI_INDUCED:
8131 /* apic_write and eoi_induced should exit unconditionally. */
8133 case EXIT_REASON_EPT_VIOLATION:
8135 * L0 always deals with the EPT violation. If nested EPT is
8136 * used, and the nested mmu code discovers that the address is
8137 * missing in the guest EPT table (EPT12), the EPT violation
8138 * will be injected with nested_ept_inject_page_fault()
8141 case EXIT_REASON_EPT_MISCONFIG:
8143 * L2 never uses directly L1's EPT, but rather L0's own EPT
8144 * table (shadow on EPT) or a merged EPT table that L0 built
8145 * (EPT on EPT). So any problems with the structure of the
8146 * table is L0's fault.
8149 case EXIT_REASON_WBINVD:
8150 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
8151 case EXIT_REASON_XSETBV:
8153 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
8155 * This should never happen, since it is not possible to
8156 * set XSS to a non-zero value---neither in L1 nor in L2.
8157 * If if it were, XSS would have to be checked against
8158 * the XSS exit bitmap in vmcs12.
8160 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
8161 case EXIT_REASON_PREEMPTION_TIMER:
8168 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
8170 *info1 = vmcs_readl(EXIT_QUALIFICATION);
8171 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
8174 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
8177 __free_page(vmx->pml_pg);
8182 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
8184 struct vcpu_vmx *vmx = to_vmx(vcpu);
8188 pml_idx = vmcs_read16(GUEST_PML_INDEX);
8190 /* Do nothing if PML buffer is empty */
8191 if (pml_idx == (PML_ENTITY_NUM - 1))
8194 /* PML index always points to next available PML buffer entity */
8195 if (pml_idx >= PML_ENTITY_NUM)
8200 pml_buf = page_address(vmx->pml_pg);
8201 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
8204 gpa = pml_buf[pml_idx];
8205 WARN_ON(gpa & (PAGE_SIZE - 1));
8206 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
8209 /* reset PML index */
8210 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8214 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8215 * Called before reporting dirty_bitmap to userspace.
8217 static void kvm_flush_pml_buffers(struct kvm *kvm)
8220 struct kvm_vcpu *vcpu;
8222 * We only need to kick vcpu out of guest mode here, as PML buffer
8223 * is flushed at beginning of all VMEXITs, and it's obvious that only
8224 * vcpus running in guest are possible to have unflushed GPAs in PML
8227 kvm_for_each_vcpu(i, vcpu, kvm)
8228 kvm_vcpu_kick(vcpu);
8231 static void vmx_dump_sel(char *name, uint32_t sel)
8233 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8234 name, vmcs_read16(sel),
8235 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
8236 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
8237 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
8240 static void vmx_dump_dtsel(char *name, uint32_t limit)
8242 pr_err("%s limit=0x%08x, base=0x%016lx\n",
8243 name, vmcs_read32(limit),
8244 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
8247 static void dump_vmcs(void)
8249 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
8250 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
8251 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
8252 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
8253 u32 secondary_exec_control = 0;
8254 unsigned long cr4 = vmcs_readl(GUEST_CR4);
8255 u64 efer = vmcs_read64(GUEST_IA32_EFER);
8258 if (cpu_has_secondary_exec_ctrls())
8259 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8261 pr_err("*** Guest State ***\n");
8262 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8263 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
8264 vmcs_readl(CR0_GUEST_HOST_MASK));
8265 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8266 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
8267 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
8268 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
8269 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
8271 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
8272 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
8273 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
8274 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
8276 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
8277 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
8278 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
8279 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
8280 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8281 vmcs_readl(GUEST_SYSENTER_ESP),
8282 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8283 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
8284 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
8285 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
8286 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
8287 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
8288 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
8289 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8290 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8291 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8292 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
8293 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8294 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8295 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8296 efer, vmcs_read64(GUEST_IA32_PAT));
8297 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
8298 vmcs_read64(GUEST_IA32_DEBUGCTL),
8299 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8300 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8301 pr_err("PerfGlobCtl = 0x%016llx\n",
8302 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8303 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8304 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8305 pr_err("Interruptibility = %08x ActivityState = %08x\n",
8306 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8307 vmcs_read32(GUEST_ACTIVITY_STATE));
8308 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8309 pr_err("InterruptStatus = %04x\n",
8310 vmcs_read16(GUEST_INTR_STATUS));
8312 pr_err("*** Host State ***\n");
8313 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
8314 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8315 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8316 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8317 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8318 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8319 vmcs_read16(HOST_TR_SELECTOR));
8320 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8321 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8322 vmcs_readl(HOST_TR_BASE));
8323 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8324 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8325 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8326 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8327 vmcs_readl(HOST_CR4));
8328 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8329 vmcs_readl(HOST_IA32_SYSENTER_ESP),
8330 vmcs_read32(HOST_IA32_SYSENTER_CS),
8331 vmcs_readl(HOST_IA32_SYSENTER_EIP));
8332 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8333 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8334 vmcs_read64(HOST_IA32_EFER),
8335 vmcs_read64(HOST_IA32_PAT));
8336 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8337 pr_err("PerfGlobCtl = 0x%016llx\n",
8338 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8340 pr_err("*** Control State ***\n");
8341 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8342 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8343 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8344 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8345 vmcs_read32(EXCEPTION_BITMAP),
8346 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8347 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8348 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8349 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8350 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8351 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8352 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8353 vmcs_read32(VM_EXIT_INTR_INFO),
8354 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8355 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8356 pr_err(" reason=%08x qualification=%016lx\n",
8357 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8358 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8359 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8360 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8361 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8362 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8363 pr_err("TSC Multiplier = 0x%016llx\n",
8364 vmcs_read64(TSC_MULTIPLIER));
8365 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8366 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8367 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8368 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8369 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8370 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8371 n = vmcs_read32(CR3_TARGET_COUNT);
8372 for (i = 0; i + 1 < n; i += 4)
8373 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8374 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8375 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8377 pr_err("CR3 target%u=%016lx\n",
8378 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8379 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8380 pr_err("PLE Gap=%08x Window=%08x\n",
8381 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8382 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8383 pr_err("Virtual processor ID = 0x%04x\n",
8384 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8388 * The guest has exited. See if we can fix it or if we need userspace
8391 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8393 struct vcpu_vmx *vmx = to_vmx(vcpu);
8394 u32 exit_reason = vmx->exit_reason;
8395 u32 vectoring_info = vmx->idt_vectoring_info;
8397 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8398 vcpu->arch.gpa_available = false;
8401 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8402 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8403 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8404 * mode as if vcpus is in root mode, the PML buffer must has been
8408 vmx_flush_pml_buffer(vcpu);
8410 /* If guest state is invalid, start emulating */
8411 if (vmx->emulation_required)
8412 return handle_invalid_guest_state(vcpu);
8414 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
8415 nested_vmx_vmexit(vcpu, exit_reason,
8416 vmcs_read32(VM_EXIT_INTR_INFO),
8417 vmcs_readl(EXIT_QUALIFICATION));
8421 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8423 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8424 vcpu->run->fail_entry.hardware_entry_failure_reason
8429 if (unlikely(vmx->fail)) {
8430 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8431 vcpu->run->fail_entry.hardware_entry_failure_reason
8432 = vmcs_read32(VM_INSTRUCTION_ERROR);
8438 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8439 * delivery event since it indicates guest is accessing MMIO.
8440 * The vm-exit can be triggered again after return to guest that
8441 * will cause infinite loop.
8443 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8444 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8445 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8446 exit_reason != EXIT_REASON_PML_FULL &&
8447 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8448 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8449 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8450 vcpu->run->internal.ndata = 2;
8451 vcpu->run->internal.data[0] = vectoring_info;
8452 vcpu->run->internal.data[1] = exit_reason;
8456 if (exit_reason < kvm_vmx_max_exit_handlers
8457 && kvm_vmx_exit_handlers[exit_reason])
8458 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8460 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
8462 kvm_queue_exception(vcpu, UD_VECTOR);
8467 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8469 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8471 if (is_guest_mode(vcpu) &&
8472 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8475 if (irr == -1 || tpr < irr) {
8476 vmcs_write32(TPR_THRESHOLD, 0);
8480 vmcs_write32(TPR_THRESHOLD, irr);
8483 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8485 u32 sec_exec_control;
8487 /* Postpone execution until vmcs01 is the current VMCS. */
8488 if (is_guest_mode(vcpu)) {
8489 to_vmx(vcpu)->nested.change_vmcs01_virtual_x2apic_mode = true;
8493 if (!cpu_has_vmx_virtualize_x2apic_mode())
8496 if (!cpu_need_tpr_shadow(vcpu))
8499 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8502 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8503 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8505 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8506 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8507 vmx_flush_tlb_ept_only(vcpu);
8509 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8511 vmx_set_msr_bitmap(vcpu);
8514 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8516 struct vcpu_vmx *vmx = to_vmx(vcpu);
8519 * Currently we do not handle the nested case where L2 has an
8520 * APIC access page of its own; that page is still pinned.
8521 * Hence, we skip the case where the VCPU is in guest mode _and_
8522 * L1 prepared an APIC access page for L2.
8524 * For the case where L1 and L2 share the same APIC access page
8525 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8526 * in the vmcs12), this function will only update either the vmcs01
8527 * or the vmcs02. If the former, the vmcs02 will be updated by
8528 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8529 * the next L2->L1 exit.
8531 if (!is_guest_mode(vcpu) ||
8532 !nested_cpu_has2(get_vmcs12(&vmx->vcpu),
8533 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
8534 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8535 vmx_flush_tlb_ept_only(vcpu);
8539 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
8547 status = vmcs_read16(GUEST_INTR_STATUS);
8549 if (max_isr != old) {
8551 status |= max_isr << 8;
8552 vmcs_write16(GUEST_INTR_STATUS, status);
8556 static void vmx_set_rvi(int vector)
8564 status = vmcs_read16(GUEST_INTR_STATUS);
8565 old = (u8)status & 0xff;
8566 if ((u8)vector != old) {
8568 status |= (u8)vector;
8569 vmcs_write16(GUEST_INTR_STATUS, status);
8573 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8575 if (!is_guest_mode(vcpu)) {
8576 vmx_set_rvi(max_irr);
8584 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8587 if (nested_exit_on_intr(vcpu))
8591 * Else, fall back to pre-APICv interrupt injection since L2
8592 * is run without virtual interrupt delivery.
8594 if (!kvm_event_needs_reinjection(vcpu) &&
8595 vmx_interrupt_allowed(vcpu)) {
8596 kvm_queue_interrupt(vcpu, max_irr, false);
8597 vmx_inject_irq(vcpu);
8601 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
8603 struct vcpu_vmx *vmx = to_vmx(vcpu);
8606 WARN_ON(!vcpu->arch.apicv_active);
8607 if (pi_test_on(&vmx->pi_desc)) {
8608 pi_clear_on(&vmx->pi_desc);
8610 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
8611 * But on x86 this is just a compiler barrier anyway.
8613 smp_mb__after_atomic();
8614 max_irr = kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
8616 max_irr = kvm_lapic_find_highest_irr(vcpu);
8618 vmx_hwapic_irr_update(vcpu, max_irr);
8622 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
8624 if (!kvm_vcpu_apicv_active(vcpu))
8627 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8628 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8629 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8630 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8633 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
8635 struct vcpu_vmx *vmx = to_vmx(vcpu);
8637 pi_clear_on(&vmx->pi_desc);
8638 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
8641 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8645 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8646 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
8649 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8650 exit_intr_info = vmx->exit_intr_info;
8652 /* Handle machine checks before interrupts are enabled */
8653 if (is_machine_check(exit_intr_info))
8654 kvm_machine_check();
8656 /* We need to handle NMIs before interrupts are enabled */
8657 if (is_nmi(exit_intr_info)) {
8658 kvm_before_handle_nmi(&vmx->vcpu);
8660 kvm_after_handle_nmi(&vmx->vcpu);
8664 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8666 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8667 register void *__sp asm(_ASM_SP);
8669 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8670 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8671 unsigned int vector;
8672 unsigned long entry;
8674 struct vcpu_vmx *vmx = to_vmx(vcpu);
8675 #ifdef CONFIG_X86_64
8679 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8680 desc = (gate_desc *)vmx->host_idt_base + vector;
8681 entry = gate_offset(*desc);
8683 #ifdef CONFIG_X86_64
8684 "mov %%" _ASM_SP ", %[sp]\n\t"
8685 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8690 __ASM_SIZE(push) " $%c[cs]\n\t"
8691 "call *%[entry]\n\t"
8693 #ifdef CONFIG_X86_64
8699 [ss]"i"(__KERNEL_DS),
8700 [cs]"i"(__KERNEL_CS)
8705 static bool vmx_has_high_real_mode_segbase(void)
8707 return enable_unrestricted_guest || emulate_invalid_guest_state;
8710 static bool vmx_mpx_supported(void)
8712 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8713 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8716 static bool vmx_xsaves_supported(void)
8718 return vmcs_config.cpu_based_2nd_exec_ctrl &
8719 SECONDARY_EXEC_XSAVES;
8722 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8727 bool idtv_info_valid;
8729 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8731 if (vmx->nmi_known_unmasked)
8734 * Can't use vmx->exit_intr_info since we're not sure what
8735 * the exit reason is.
8737 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8738 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8739 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8741 * SDM 3: 27.7.1.2 (September 2008)
8742 * Re-set bit "block by NMI" before VM entry if vmexit caused by
8743 * a guest IRET fault.
8744 * SDM 3: 23.2.2 (September 2008)
8745 * Bit 12 is undefined in any of the following cases:
8746 * If the VM exit sets the valid bit in the IDT-vectoring
8747 * information field.
8748 * If the VM exit is due to a double fault.
8750 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8751 vector != DF_VECTOR && !idtv_info_valid)
8752 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8753 GUEST_INTR_STATE_NMI);
8755 vmx->nmi_known_unmasked =
8756 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8757 & GUEST_INTR_STATE_NMI);
8760 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8761 u32 idt_vectoring_info,
8762 int instr_len_field,
8763 int error_code_field)
8767 bool idtv_info_valid;
8769 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8771 vcpu->arch.nmi_injected = false;
8772 kvm_clear_exception_queue(vcpu);
8773 kvm_clear_interrupt_queue(vcpu);
8775 if (!idtv_info_valid)
8778 kvm_make_request(KVM_REQ_EVENT, vcpu);
8780 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8781 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8784 case INTR_TYPE_NMI_INTR:
8785 vcpu->arch.nmi_injected = true;
8787 * SDM 3: 27.7.1.2 (September 2008)
8788 * Clear bit "block by NMI" before VM entry if a NMI
8791 vmx_set_nmi_mask(vcpu, false);
8793 case INTR_TYPE_SOFT_EXCEPTION:
8794 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8796 case INTR_TYPE_HARD_EXCEPTION:
8797 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8798 u32 err = vmcs_read32(error_code_field);
8799 kvm_requeue_exception_e(vcpu, vector, err);
8801 kvm_requeue_exception(vcpu, vector);
8803 case INTR_TYPE_SOFT_INTR:
8804 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8806 case INTR_TYPE_EXT_INTR:
8807 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8814 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8816 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8817 VM_EXIT_INSTRUCTION_LEN,
8818 IDT_VECTORING_ERROR_CODE);
8821 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8823 __vmx_complete_interrupts(vcpu,
8824 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8825 VM_ENTRY_INSTRUCTION_LEN,
8826 VM_ENTRY_EXCEPTION_ERROR_CODE);
8828 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8831 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8834 struct perf_guest_switch_msr *msrs;
8836 msrs = perf_guest_get_msrs(&nr_msrs);
8841 for (i = 0; i < nr_msrs; i++)
8842 if (msrs[i].host == msrs[i].guest)
8843 clear_atomic_switch_msr(vmx, msrs[i].msr);
8845 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8849 static void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
8851 struct vcpu_vmx *vmx = to_vmx(vcpu);
8855 if (vmx->hv_deadline_tsc == -1)
8859 if (vmx->hv_deadline_tsc > tscl)
8860 /* sure to be 32 bit only because checked on set_hv_timer */
8861 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
8862 cpu_preemption_timer_multi);
8866 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
8869 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8871 struct vcpu_vmx *vmx = to_vmx(vcpu);
8872 unsigned long debugctlmsr, cr4;
8874 /* Don't enter VMX if guest state is invalid, let the exit handler
8875 start emulation until we arrive back to a valid state */
8876 if (vmx->emulation_required)
8879 if (vmx->ple_window_dirty) {
8880 vmx->ple_window_dirty = false;
8881 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8884 if (vmx->nested.sync_shadow_vmcs) {
8885 copy_vmcs12_to_shadow(vmx);
8886 vmx->nested.sync_shadow_vmcs = false;
8889 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8890 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
8891 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
8892 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
8894 cr4 = cr4_read_shadow();
8895 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
8896 vmcs_writel(HOST_CR4, cr4);
8897 vmx->host_state.vmcs_host_cr4 = cr4;
8900 /* When single-stepping over STI and MOV SS, we must clear the
8901 * corresponding interruptibility bits in the guest state. Otherwise
8902 * vmentry fails as it then expects bit 14 (BS) in pending debug
8903 * exceptions being set, but that's not correct for the guest debugging
8905 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8906 vmx_set_interrupt_shadow(vcpu, 0);
8908 if (vmx->guest_pkru_valid)
8909 __write_pkru(vmx->guest_pkru);
8911 atomic_switch_perf_msrs(vmx);
8912 debugctlmsr = get_debugctlmsr();
8914 vmx_arm_hv_timer(vcpu);
8916 vmx->__launched = vmx->loaded_vmcs->launched;
8918 /* Store host registers */
8919 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
8920 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
8921 "push %%" _ASM_CX " \n\t"
8922 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8924 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8925 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
8927 /* Reload cr2 if changed */
8928 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
8929 "mov %%cr2, %%" _ASM_DX " \n\t"
8930 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
8932 "mov %%" _ASM_AX", %%cr2 \n\t"
8934 /* Check if vmlaunch of vmresume is needed */
8935 "cmpl $0, %c[launched](%0) \n\t"
8936 /* Load guest registers. Don't clobber flags. */
8937 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
8938 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
8939 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
8940 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
8941 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
8942 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
8943 #ifdef CONFIG_X86_64
8944 "mov %c[r8](%0), %%r8 \n\t"
8945 "mov %c[r9](%0), %%r9 \n\t"
8946 "mov %c[r10](%0), %%r10 \n\t"
8947 "mov %c[r11](%0), %%r11 \n\t"
8948 "mov %c[r12](%0), %%r12 \n\t"
8949 "mov %c[r13](%0), %%r13 \n\t"
8950 "mov %c[r14](%0), %%r14 \n\t"
8951 "mov %c[r15](%0), %%r15 \n\t"
8953 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
8955 /* Enter guest mode */
8957 __ex(ASM_VMX_VMLAUNCH) "\n\t"
8959 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
8961 /* Save guest registers, load host registers, keep flags */
8962 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
8964 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
8965 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
8966 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
8967 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
8968 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
8969 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
8970 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
8971 #ifdef CONFIG_X86_64
8972 "mov %%r8, %c[r8](%0) \n\t"
8973 "mov %%r9, %c[r9](%0) \n\t"
8974 "mov %%r10, %c[r10](%0) \n\t"
8975 "mov %%r11, %c[r11](%0) \n\t"
8976 "mov %%r12, %c[r12](%0) \n\t"
8977 "mov %%r13, %c[r13](%0) \n\t"
8978 "mov %%r14, %c[r14](%0) \n\t"
8979 "mov %%r15, %c[r15](%0) \n\t"
8981 "mov %%cr2, %%" _ASM_AX " \n\t"
8982 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
8984 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
8985 "setbe %c[fail](%0) \n\t"
8986 ".pushsection .rodata \n\t"
8987 ".global vmx_return \n\t"
8988 "vmx_return: " _ASM_PTR " 2b \n\t"
8990 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
8991 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
8992 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
8993 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
8994 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
8995 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
8996 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
8997 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
8998 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
8999 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
9000 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
9001 #ifdef CONFIG_X86_64
9002 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
9003 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
9004 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
9005 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
9006 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
9007 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
9008 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
9009 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
9011 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
9012 [wordsize]"i"(sizeof(ulong))
9014 #ifdef CONFIG_X86_64
9015 , "rax", "rbx", "rdi", "rsi"
9016 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
9018 , "eax", "ebx", "edi", "esi"
9022 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
9024 update_debugctlmsr(debugctlmsr);
9026 #ifndef CONFIG_X86_64
9028 * The sysexit path does not restore ds/es, so we must set them to
9029 * a reasonable value ourselves.
9031 * We can't defer this to vmx_load_host_state() since that function
9032 * may be executed in interrupt context, which saves and restore segments
9033 * around it, nullifying its effect.
9035 loadsegment(ds, __USER_DS);
9036 loadsegment(es, __USER_DS);
9039 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
9040 | (1 << VCPU_EXREG_RFLAGS)
9041 | (1 << VCPU_EXREG_PDPTR)
9042 | (1 << VCPU_EXREG_SEGMENTS)
9043 | (1 << VCPU_EXREG_CR3));
9044 vcpu->arch.regs_dirty = 0;
9046 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9048 vmx->loaded_vmcs->launched = 1;
9050 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
9053 * eager fpu is enabled if PKEY is supported and CR4 is switched
9054 * back on host, so it is safe to read guest PKRU from current
9057 if (boot_cpu_has(X86_FEATURE_OSPKE)) {
9058 vmx->guest_pkru = __read_pkru();
9059 if (vmx->guest_pkru != vmx->host_pkru) {
9060 vmx->guest_pkru_valid = true;
9061 __write_pkru(vmx->host_pkru);
9063 vmx->guest_pkru_valid = false;
9067 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
9068 * we did not inject a still-pending event to L1 now because of
9069 * nested_run_pending, we need to re-enable this bit.
9071 if (vmx->nested.nested_run_pending)
9072 kvm_make_request(KVM_REQ_EVENT, vcpu);
9074 vmx->nested.nested_run_pending = 0;
9076 vmx_complete_atomic_exit(vmx);
9077 vmx_recover_nmi_blocking(vmx);
9078 vmx_complete_interrupts(vmx);
9081 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
9083 struct vcpu_vmx *vmx = to_vmx(vcpu);
9086 if (vmx->loaded_vmcs == vmcs)
9090 vmx->loaded_vmcs = vmcs;
9092 vmx_vcpu_load(vcpu, cpu);
9098 * Ensure that the current vmcs of the logical processor is the
9099 * vmcs01 of the vcpu before calling free_nested().
9101 static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
9103 struct vcpu_vmx *vmx = to_vmx(vcpu);
9106 r = vcpu_load(vcpu);
9108 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
9113 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
9115 struct vcpu_vmx *vmx = to_vmx(vcpu);
9118 vmx_destroy_pml_buffer(vmx);
9119 free_vpid(vmx->vpid);
9120 leave_guest_mode(vcpu);
9121 vmx_free_vcpu_nested(vcpu);
9122 free_loaded_vmcs(vmx->loaded_vmcs);
9123 kfree(vmx->guest_msrs);
9124 kvm_vcpu_uninit(vcpu);
9125 kmem_cache_free(kvm_vcpu_cache, vmx);
9128 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
9131 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
9135 return ERR_PTR(-ENOMEM);
9137 vmx->vpid = allocate_vpid();
9139 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
9146 * If PML is turned on, failure on enabling PML just results in failure
9147 * of creating the vcpu, therefore we can simplify PML logic (by
9148 * avoiding dealing with cases, such as enabling PML partially on vcpus
9149 * for the guest, etc.
9152 vmx->pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
9157 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
9158 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
9161 if (!vmx->guest_msrs)
9164 vmx->loaded_vmcs = &vmx->vmcs01;
9165 vmx->loaded_vmcs->vmcs = alloc_vmcs();
9166 vmx->loaded_vmcs->shadow_vmcs = NULL;
9167 if (!vmx->loaded_vmcs->vmcs)
9170 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
9171 loaded_vmcs_init(vmx->loaded_vmcs);
9176 vmx_vcpu_load(&vmx->vcpu, cpu);
9177 vmx->vcpu.cpu = cpu;
9178 err = vmx_vcpu_setup(vmx);
9179 vmx_vcpu_put(&vmx->vcpu);
9183 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9184 err = alloc_apic_access_page(kvm);
9190 if (!kvm->arch.ept_identity_map_addr)
9191 kvm->arch.ept_identity_map_addr =
9192 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
9193 err = init_rmode_identity_map(kvm);
9199 nested_vmx_setup_ctls_msrs(vmx);
9200 vmx->nested.vpid02 = allocate_vpid();
9203 vmx->nested.posted_intr_nv = -1;
9204 vmx->nested.current_vmptr = -1ull;
9205 vmx->nested.current_vmcs12 = NULL;
9207 vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
9212 free_vpid(vmx->nested.vpid02);
9213 free_loaded_vmcs(vmx->loaded_vmcs);
9215 kfree(vmx->guest_msrs);
9217 vmx_destroy_pml_buffer(vmx);
9219 kvm_vcpu_uninit(&vmx->vcpu);
9221 free_vpid(vmx->vpid);
9222 kmem_cache_free(kvm_vcpu_cache, vmx);
9223 return ERR_PTR(err);
9226 static void __init vmx_check_processor_compat(void *rtn)
9228 struct vmcs_config vmcs_conf;
9231 if (setup_vmcs_config(&vmcs_conf) < 0)
9233 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
9234 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
9235 smp_processor_id());
9240 static int get_ept_level(void)
9242 return VMX_EPT_DEFAULT_GAW + 1;
9245 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
9250 /* For VT-d and EPT combination
9251 * 1. MMIO: always map as UC
9253 * a. VT-d without snooping control feature: can't guarantee the
9254 * result, try to trust guest.
9255 * b. VT-d with snooping control feature: snooping control feature of
9256 * VT-d engine can guarantee the cache correctness. Just set it
9257 * to WB to keep consistent with host. So the same as item 3.
9258 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9259 * consistent with host MTRR
9262 cache = MTRR_TYPE_UNCACHABLE;
9266 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
9267 ipat = VMX_EPT_IPAT_BIT;
9268 cache = MTRR_TYPE_WRBACK;
9272 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
9273 ipat = VMX_EPT_IPAT_BIT;
9274 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
9275 cache = MTRR_TYPE_WRBACK;
9277 cache = MTRR_TYPE_UNCACHABLE;
9281 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
9284 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
9287 static int vmx_get_lpage_level(void)
9289 if (enable_ept && !cpu_has_vmx_ept_1g_page())
9290 return PT_DIRECTORY_LEVEL;
9292 /* For shadow and EPT supported 1GB page */
9293 return PT_PDPE_LEVEL;
9296 static void vmcs_set_secondary_exec_control(u32 new_ctl)
9299 * These bits in the secondary execution controls field
9300 * are dynamic, the others are mostly based on the hypervisor
9301 * architecture and the guest's CPUID. Do not touch the
9305 SECONDARY_EXEC_SHADOW_VMCS |
9306 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
9307 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9309 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9311 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
9312 (new_ctl & ~mask) | (cur_ctl & mask));
9316 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
9317 * (indicating "allowed-1") if they are supported in the guest's CPUID.
9319 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
9321 struct vcpu_vmx *vmx = to_vmx(vcpu);
9322 struct kvm_cpuid_entry2 *entry;
9324 vmx->nested.nested_vmx_cr0_fixed1 = 0xffffffff;
9325 vmx->nested.nested_vmx_cr4_fixed1 = X86_CR4_PCE;
9327 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
9328 if (entry && (entry->_reg & (_cpuid_mask))) \
9329 vmx->nested.nested_vmx_cr4_fixed1 |= (_cr4_mask); \
9332 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
9333 cr4_fixed1_update(X86_CR4_VME, edx, bit(X86_FEATURE_VME));
9334 cr4_fixed1_update(X86_CR4_PVI, edx, bit(X86_FEATURE_VME));
9335 cr4_fixed1_update(X86_CR4_TSD, edx, bit(X86_FEATURE_TSC));
9336 cr4_fixed1_update(X86_CR4_DE, edx, bit(X86_FEATURE_DE));
9337 cr4_fixed1_update(X86_CR4_PSE, edx, bit(X86_FEATURE_PSE));
9338 cr4_fixed1_update(X86_CR4_PAE, edx, bit(X86_FEATURE_PAE));
9339 cr4_fixed1_update(X86_CR4_MCE, edx, bit(X86_FEATURE_MCE));
9340 cr4_fixed1_update(X86_CR4_PGE, edx, bit(X86_FEATURE_PGE));
9341 cr4_fixed1_update(X86_CR4_OSFXSR, edx, bit(X86_FEATURE_FXSR));
9342 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
9343 cr4_fixed1_update(X86_CR4_VMXE, ecx, bit(X86_FEATURE_VMX));
9344 cr4_fixed1_update(X86_CR4_SMXE, ecx, bit(X86_FEATURE_SMX));
9345 cr4_fixed1_update(X86_CR4_PCIDE, ecx, bit(X86_FEATURE_PCID));
9346 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, bit(X86_FEATURE_XSAVE));
9348 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9349 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, bit(X86_FEATURE_FSGSBASE));
9350 cr4_fixed1_update(X86_CR4_SMEP, ebx, bit(X86_FEATURE_SMEP));
9351 cr4_fixed1_update(X86_CR4_SMAP, ebx, bit(X86_FEATURE_SMAP));
9352 cr4_fixed1_update(X86_CR4_PKE, ecx, bit(X86_FEATURE_PKU));
9353 /* TODO: Use X86_CR4_UMIP and X86_FEATURE_UMIP macros */
9354 cr4_fixed1_update(bit(11), ecx, bit(2));
9356 #undef cr4_fixed1_update
9359 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
9361 struct kvm_cpuid_entry2 *best;
9362 struct vcpu_vmx *vmx = to_vmx(vcpu);
9363 u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx);
9365 if (vmx_rdtscp_supported()) {
9366 bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu);
9367 if (!rdtscp_enabled)
9368 secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP;
9372 vmx->nested.nested_vmx_secondary_ctls_high |=
9373 SECONDARY_EXEC_RDTSCP;
9375 vmx->nested.nested_vmx_secondary_ctls_high &=
9376 ~SECONDARY_EXEC_RDTSCP;
9380 /* Exposing INVPCID only when PCID is exposed */
9381 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9382 if (vmx_invpcid_supported() &&
9383 (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) ||
9384 !guest_cpuid_has_pcid(vcpu))) {
9385 secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID;
9388 best->ebx &= ~bit(X86_FEATURE_INVPCID);
9391 if (cpu_has_secondary_exec_ctrls())
9392 vmcs_set_secondary_exec_control(secondary_exec_ctl);
9394 if (nested_vmx_allowed(vcpu))
9395 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
9396 FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9398 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
9399 ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9401 if (nested_vmx_allowed(vcpu))
9402 nested_vmx_cr_fixed1_bits_update(vcpu);
9405 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9407 if (func == 1 && nested)
9408 entry->ecx |= bit(X86_FEATURE_VMX);
9411 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9412 struct x86_exception *fault)
9414 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9417 if (fault->error_code & PFERR_RSVD_MASK)
9418 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9420 exit_reason = EXIT_REASON_EPT_VIOLATION;
9421 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
9422 vmcs12->guest_physical_address = fault->address;
9425 /* Callbacks for nested_ept_init_mmu_context: */
9427 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9429 /* return the page table to be shadowed - in our case, EPT12 */
9430 return get_vmcs12(vcpu)->ept_pointer;
9433 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9435 WARN_ON(mmu_is_nested(vcpu));
9436 kvm_init_shadow_ept_mmu(vcpu,
9437 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
9438 VMX_EPT_EXECUTE_ONLY_BIT);
9439 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
9440 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
9441 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
9443 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
9446 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
9448 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
9451 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9454 bool inequality, bit;
9456 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9458 (error_code & vmcs12->page_fault_error_code_mask) !=
9459 vmcs12->page_fault_error_code_match;
9460 return inequality ^ bit;
9463 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9464 struct x86_exception *fault)
9466 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9468 WARN_ON(!is_guest_mode(vcpu));
9470 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code))
9471 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
9472 vmcs_read32(VM_EXIT_INTR_INFO),
9473 vmcs_readl(EXIT_QUALIFICATION));
9475 kvm_inject_page_fault(vcpu, fault);
9478 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9479 struct vmcs12 *vmcs12);
9481 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9482 struct vmcs12 *vmcs12)
9484 struct vcpu_vmx *vmx = to_vmx(vcpu);
9487 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9489 * Translate L1 physical address to host physical
9490 * address for vmcs02. Keep the page pinned, so this
9491 * physical address remains valid. We keep a reference
9492 * to it so we can release it later.
9494 if (vmx->nested.apic_access_page) /* shouldn't happen */
9495 nested_release_page(vmx->nested.apic_access_page);
9496 vmx->nested.apic_access_page =
9497 nested_get_page(vcpu, vmcs12->apic_access_addr);
9499 * If translation failed, no matter: This feature asks
9500 * to exit when accessing the given address, and if it
9501 * can never be accessed, this feature won't do
9504 if (vmx->nested.apic_access_page) {
9505 hpa = page_to_phys(vmx->nested.apic_access_page);
9506 vmcs_write64(APIC_ACCESS_ADDR, hpa);
9508 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
9509 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9511 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9512 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9513 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
9514 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9515 kvm_vcpu_reload_apic_access_page(vcpu);
9518 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9519 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
9520 nested_release_page(vmx->nested.virtual_apic_page);
9521 vmx->nested.virtual_apic_page =
9522 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
9525 * If translation failed, VM entry will fail because
9526 * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
9527 * Failing the vm entry is _not_ what the processor
9528 * does but it's basically the only possibility we
9529 * have. We could still enter the guest if CR8 load
9530 * exits are enabled, CR8 store exits are enabled, and
9531 * virtualize APIC access is disabled; in this case
9532 * the processor would never use the TPR shadow and we
9533 * could simply clear the bit from the execution
9534 * control. But such a configuration is useless, so
9535 * let's keep the code simple.
9537 if (vmx->nested.virtual_apic_page) {
9538 hpa = page_to_phys(vmx->nested.virtual_apic_page);
9539 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
9543 if (nested_cpu_has_posted_intr(vmcs12)) {
9544 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9545 kunmap(vmx->nested.pi_desc_page);
9546 nested_release_page(vmx->nested.pi_desc_page);
9548 vmx->nested.pi_desc_page =
9549 nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
9550 vmx->nested.pi_desc =
9551 (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
9552 if (!vmx->nested.pi_desc) {
9553 nested_release_page_clean(vmx->nested.pi_desc_page);
9556 vmx->nested.pi_desc =
9557 (struct pi_desc *)((void *)vmx->nested.pi_desc +
9558 (unsigned long)(vmcs12->posted_intr_desc_addr &
9560 vmcs_write64(POSTED_INTR_DESC_ADDR,
9561 page_to_phys(vmx->nested.pi_desc_page) +
9562 (unsigned long)(vmcs12->posted_intr_desc_addr &
9565 if (cpu_has_vmx_msr_bitmap() &&
9566 nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS) &&
9567 nested_vmx_merge_msr_bitmap(vcpu, vmcs12))
9570 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
9571 CPU_BASED_USE_MSR_BITMAPS);
9574 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9576 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9577 struct vcpu_vmx *vmx = to_vmx(vcpu);
9579 if (vcpu->arch.virtual_tsc_khz == 0)
9582 /* Make sure short timeouts reliably trigger an immediate vmexit.
9583 * hrtimer_start does not guarantee this. */
9584 if (preemption_timeout <= 1) {
9585 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9589 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9590 preemption_timeout *= 1000000;
9591 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9592 hrtimer_start(&vmx->nested.preemption_timer,
9593 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9596 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9597 struct vmcs12 *vmcs12)
9602 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9605 if (vmcs12_read_any(vcpu, MSR_BITMAP, &addr)) {
9609 maxphyaddr = cpuid_maxphyaddr(vcpu);
9611 if (!PAGE_ALIGNED(vmcs12->msr_bitmap) ||
9612 ((addr + PAGE_SIZE) >> maxphyaddr))
9619 * Merge L0's and L1's MSR bitmap, return false to indicate that
9620 * we do not use the hardware.
9622 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9623 struct vmcs12 *vmcs12)
9627 unsigned long *msr_bitmap_l1;
9628 unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.msr_bitmap;
9630 /* This shortcut is ok because we support only x2APIC MSRs so far. */
9631 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
9634 page = nested_get_page(vcpu, vmcs12->msr_bitmap);
9637 msr_bitmap_l1 = (unsigned long *)kmap(page);
9639 memset(msr_bitmap_l0, 0xff, PAGE_SIZE);
9641 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
9642 if (nested_cpu_has_apic_reg_virt(vmcs12))
9643 for (msr = 0x800; msr <= 0x8ff; msr++)
9644 nested_vmx_disable_intercept_for_msr(
9645 msr_bitmap_l1, msr_bitmap_l0,
9648 nested_vmx_disable_intercept_for_msr(
9649 msr_bitmap_l1, msr_bitmap_l0,
9650 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9651 MSR_TYPE_R | MSR_TYPE_W);
9653 if (nested_cpu_has_vid(vmcs12)) {
9654 nested_vmx_disable_intercept_for_msr(
9655 msr_bitmap_l1, msr_bitmap_l0,
9656 APIC_BASE_MSR + (APIC_EOI >> 4),
9658 nested_vmx_disable_intercept_for_msr(
9659 msr_bitmap_l1, msr_bitmap_l0,
9660 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9665 nested_release_page_clean(page);
9670 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
9671 struct vmcs12 *vmcs12)
9673 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9674 !nested_cpu_has_apic_reg_virt(vmcs12) &&
9675 !nested_cpu_has_vid(vmcs12) &&
9676 !nested_cpu_has_posted_intr(vmcs12))
9680 * If virtualize x2apic mode is enabled,
9681 * virtualize apic access must be disabled.
9683 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9684 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9688 * If virtual interrupt delivery is enabled,
9689 * we must exit on external interrupts.
9691 if (nested_cpu_has_vid(vmcs12) &&
9692 !nested_exit_on_intr(vcpu))
9696 * bits 15:8 should be zero in posted_intr_nv,
9697 * the descriptor address has been already checked
9698 * in nested_get_vmcs12_pages.
9700 if (nested_cpu_has_posted_intr(vmcs12) &&
9701 (!nested_cpu_has_vid(vmcs12) ||
9702 !nested_exit_intr_ack_set(vcpu) ||
9703 vmcs12->posted_intr_nv & 0xff00))
9706 /* tpr shadow is needed by all apicv features. */
9707 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9713 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9714 unsigned long count_field,
9715 unsigned long addr_field)
9720 if (vmcs12_read_any(vcpu, count_field, &count) ||
9721 vmcs12_read_any(vcpu, addr_field, &addr)) {
9727 maxphyaddr = cpuid_maxphyaddr(vcpu);
9728 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9729 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9730 pr_debug_ratelimited(
9731 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9732 addr_field, maxphyaddr, count, addr);
9738 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9739 struct vmcs12 *vmcs12)
9741 if (vmcs12->vm_exit_msr_load_count == 0 &&
9742 vmcs12->vm_exit_msr_store_count == 0 &&
9743 vmcs12->vm_entry_msr_load_count == 0)
9744 return 0; /* Fast path */
9745 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9746 VM_EXIT_MSR_LOAD_ADDR) ||
9747 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9748 VM_EXIT_MSR_STORE_ADDR) ||
9749 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9750 VM_ENTRY_MSR_LOAD_ADDR))
9755 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9756 struct vmx_msr_entry *e)
9758 /* x2APIC MSR accesses are not allowed */
9759 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9761 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9762 e->index == MSR_IA32_UCODE_REV)
9764 if (e->reserved != 0)
9769 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9770 struct vmx_msr_entry *e)
9772 if (e->index == MSR_FS_BASE ||
9773 e->index == MSR_GS_BASE ||
9774 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9775 nested_vmx_msr_check_common(vcpu, e))
9780 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9781 struct vmx_msr_entry *e)
9783 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9784 nested_vmx_msr_check_common(vcpu, e))
9790 * Load guest's/host's msr at nested entry/exit.
9791 * return 0 for success, entry index for failure.
9793 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9796 struct vmx_msr_entry e;
9797 struct msr_data msr;
9799 msr.host_initiated = false;
9800 for (i = 0; i < count; i++) {
9801 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
9803 pr_debug_ratelimited(
9804 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9805 __func__, i, gpa + i * sizeof(e));
9808 if (nested_vmx_load_msr_check(vcpu, &e)) {
9809 pr_debug_ratelimited(
9810 "%s check failed (%u, 0x%x, 0x%x)\n",
9811 __func__, i, e.index, e.reserved);
9814 msr.index = e.index;
9816 if (kvm_set_msr(vcpu, &msr)) {
9817 pr_debug_ratelimited(
9818 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9819 __func__, i, e.index, e.value);
9828 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9831 struct vmx_msr_entry e;
9833 for (i = 0; i < count; i++) {
9834 struct msr_data msr_info;
9835 if (kvm_vcpu_read_guest(vcpu,
9836 gpa + i * sizeof(e),
9837 &e, 2 * sizeof(u32))) {
9838 pr_debug_ratelimited(
9839 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9840 __func__, i, gpa + i * sizeof(e));
9843 if (nested_vmx_store_msr_check(vcpu, &e)) {
9844 pr_debug_ratelimited(
9845 "%s check failed (%u, 0x%x, 0x%x)\n",
9846 __func__, i, e.index, e.reserved);
9849 msr_info.host_initiated = false;
9850 msr_info.index = e.index;
9851 if (kvm_get_msr(vcpu, &msr_info)) {
9852 pr_debug_ratelimited(
9853 "%s cannot read MSR (%u, 0x%x)\n",
9854 __func__, i, e.index);
9857 if (kvm_vcpu_write_guest(vcpu,
9858 gpa + i * sizeof(e) +
9859 offsetof(struct vmx_msr_entry, value),
9860 &msr_info.data, sizeof(msr_info.data))) {
9861 pr_debug_ratelimited(
9862 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9863 __func__, i, e.index, msr_info.data);
9870 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
9872 unsigned long invalid_mask;
9874 invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
9875 return (val & invalid_mask) == 0;
9879 * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
9880 * emulating VM entry into a guest with EPT enabled.
9881 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
9882 * is assigned to entry_failure_code on failure.
9884 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
9885 u32 *entry_failure_code)
9887 if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
9888 if (!nested_cr3_valid(vcpu, cr3)) {
9889 *entry_failure_code = ENTRY_FAIL_DEFAULT;
9894 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
9895 * must not be dereferenced.
9897 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
9899 if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
9900 *entry_failure_code = ENTRY_FAIL_PDPTE;
9905 vcpu->arch.cr3 = cr3;
9906 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
9909 kvm_mmu_reset_context(vcpu);
9914 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
9915 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
9916 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
9917 * guest in a way that will both be appropriate to L1's requests, and our
9918 * needs. In addition to modifying the active vmcs (which is vmcs02), this
9919 * function also has additional necessary side-effects, like setting various
9920 * vcpu->arch fields.
9921 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
9922 * is assigned to entry_failure_code on failure.
9924 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
9925 bool from_vmentry, u32 *entry_failure_code)
9927 struct vcpu_vmx *vmx = to_vmx(vcpu);
9930 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
9931 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
9932 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
9933 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
9934 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
9935 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
9936 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
9937 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
9938 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
9939 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
9940 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
9941 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
9942 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
9943 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
9944 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
9945 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
9946 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
9947 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
9948 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
9949 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
9950 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
9951 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
9952 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
9953 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
9954 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
9955 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
9956 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
9957 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
9958 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
9959 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
9960 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
9961 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
9962 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
9963 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
9964 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
9965 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
9968 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
9969 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
9970 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
9972 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
9973 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
9976 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
9977 vmcs12->vm_entry_intr_info_field);
9978 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
9979 vmcs12->vm_entry_exception_error_code);
9980 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
9981 vmcs12->vm_entry_instruction_len);
9982 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
9983 vmcs12->guest_interruptibility_info);
9985 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
9987 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
9988 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
9989 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
9990 vmcs12->guest_pending_dbg_exceptions);
9991 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
9992 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
9994 if (nested_cpu_has_xsaves(vmcs12))
9995 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
9996 vmcs_write64(VMCS_LINK_POINTER, -1ull);
9998 exec_control = vmcs12->pin_based_vm_exec_control;
10000 /* Preemption timer setting is only taken from vmcs01. */
10001 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10002 exec_control |= vmcs_config.pin_based_exec_ctrl;
10003 if (vmx->hv_deadline_tsc == -1)
10004 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10006 /* Posted interrupts setting is only taken from vmcs12. */
10007 if (nested_cpu_has_posted_intr(vmcs12)) {
10009 * Note that we use L0's vector here and in
10010 * vmx_deliver_nested_posted_interrupt.
10012 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
10013 vmx->nested.pi_pending = false;
10014 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
10016 exec_control &= ~PIN_BASED_POSTED_INTR;
10019 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
10021 vmx->nested.preemption_timer_expired = false;
10022 if (nested_cpu_has_preemption_timer(vmcs12))
10023 vmx_start_preemption_timer(vcpu);
10026 * Whether page-faults are trapped is determined by a combination of
10027 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
10028 * If enable_ept, L0 doesn't care about page faults and we should
10029 * set all of these to L1's desires. However, if !enable_ept, L0 does
10030 * care about (at least some) page faults, and because it is not easy
10031 * (if at all possible?) to merge L0 and L1's desires, we simply ask
10032 * to exit on each and every L2 page fault. This is done by setting
10033 * MASK=MATCH=0 and (see below) EB.PF=1.
10034 * Note that below we don't need special code to set EB.PF beyond the
10035 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
10036 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
10037 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
10039 * A problem with this approach (when !enable_ept) is that L1 may be
10040 * injected with more page faults than it asked for. This could have
10041 * caused problems, but in practice existing hypervisors don't care.
10042 * To fix this, we will need to emulate the PFEC checking (on the L1
10043 * page tables), using walk_addr(), when injecting PFs to L1.
10045 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
10046 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
10047 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
10048 enable_ept ? vmcs12->page_fault_error_code_match : 0);
10050 if (cpu_has_secondary_exec_ctrls()) {
10051 exec_control = vmx_secondary_exec_control(vmx);
10053 /* Take the following fields only from vmcs12 */
10054 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
10055 SECONDARY_EXEC_RDTSCP |
10056 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
10057 SECONDARY_EXEC_APIC_REGISTER_VIRT);
10058 if (nested_cpu_has(vmcs12,
10059 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
10060 exec_control |= vmcs12->secondary_vm_exec_control;
10062 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
10063 vmcs_write64(EOI_EXIT_BITMAP0,
10064 vmcs12->eoi_exit_bitmap0);
10065 vmcs_write64(EOI_EXIT_BITMAP1,
10066 vmcs12->eoi_exit_bitmap1);
10067 vmcs_write64(EOI_EXIT_BITMAP2,
10068 vmcs12->eoi_exit_bitmap2);
10069 vmcs_write64(EOI_EXIT_BITMAP3,
10070 vmcs12->eoi_exit_bitmap3);
10071 vmcs_write16(GUEST_INTR_STATUS,
10072 vmcs12->guest_intr_status);
10076 * Write an illegal value to APIC_ACCESS_ADDR. Later,
10077 * nested_get_vmcs12_pages will either fix it up or
10078 * remove the VM execution control.
10080 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
10081 vmcs_write64(APIC_ACCESS_ADDR, -1ull);
10083 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
10088 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
10089 * Some constant fields are set here by vmx_set_constant_host_state().
10090 * Other fields are different per CPU, and will be set later when
10091 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
10093 vmx_set_constant_host_state(vmx);
10096 * Set the MSR load/store lists to match L0's settings.
10098 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
10099 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10100 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
10101 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10102 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
10105 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
10106 * entry, but only if the current (host) sp changed from the value
10107 * we wrote last (vmx->host_rsp). This cache is no longer relevant
10108 * if we switch vmcs, and rather than hold a separate cache per vmcs,
10109 * here we just force the write to happen on entry.
10113 exec_control = vmx_exec_control(vmx); /* L0's desires */
10114 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
10115 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
10116 exec_control &= ~CPU_BASED_TPR_SHADOW;
10117 exec_control |= vmcs12->cpu_based_vm_exec_control;
10120 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
10121 * nested_get_vmcs12_pages can't fix it up, the illegal value
10122 * will result in a VM entry failure.
10124 if (exec_control & CPU_BASED_TPR_SHADOW) {
10125 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
10126 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
10130 * Merging of IO bitmap not currently supported.
10131 * Rather, exit every time.
10133 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
10134 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
10136 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
10138 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
10139 * bitwise-or of what L1 wants to trap for L2, and what we want to
10140 * trap. Note that CR0.TS also needs updating - we do this later.
10142 update_exception_bitmap(vcpu);
10143 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
10144 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10146 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
10147 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
10148 * bits are further modified by vmx_set_efer() below.
10150 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
10152 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
10153 * emulated by vmx_set_efer(), below.
10155 vm_entry_controls_init(vmx,
10156 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
10157 ~VM_ENTRY_IA32E_MODE) |
10158 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
10160 if (from_vmentry &&
10161 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
10162 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
10163 vcpu->arch.pat = vmcs12->guest_ia32_pat;
10164 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
10165 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
10168 set_cr4_guest_host_mask(vmx);
10170 if (from_vmentry &&
10171 vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
10172 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
10174 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
10175 vmcs_write64(TSC_OFFSET,
10176 vcpu->arch.tsc_offset + vmcs12->tsc_offset);
10178 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
10179 if (kvm_has_tsc_control)
10180 decache_tsc_multiplier(vmx);
10184 * There is no direct mapping between vpid02 and vpid12, the
10185 * vpid02 is per-vCPU for L0 and reused while the value of
10186 * vpid12 is changed w/ one invvpid during nested vmentry.
10187 * The vpid12 is allocated by L1 for L2, so it will not
10188 * influence global bitmap(for vpid01 and vpid02 allocation)
10189 * even if spawn a lot of nested vCPUs.
10191 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
10192 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
10193 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
10194 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
10195 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
10198 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
10199 vmx_flush_tlb(vcpu);
10204 if (nested_cpu_has_ept(vmcs12)) {
10205 kvm_mmu_unload(vcpu);
10206 nested_ept_init_mmu_context(vcpu);
10207 } else if (nested_cpu_has2(vmcs12,
10208 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10209 vmx_flush_tlb_ept_only(vcpu);
10213 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
10214 * bits which we consider mandatory enabled.
10215 * The CR0_READ_SHADOW is what L2 should have expected to read given
10216 * the specifications by L1; It's not enough to take
10217 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10218 * have more bits than L1 expected.
10220 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
10221 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
10223 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
10224 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
10226 if (from_vmentry &&
10227 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
10228 vcpu->arch.efer = vmcs12->guest_ia32_efer;
10229 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
10230 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10232 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10233 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10234 vmx_set_efer(vcpu, vcpu->arch.efer);
10236 /* Shadow page tables on either EPT or shadow page tables. */
10237 if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
10238 entry_failure_code))
10242 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
10245 * L1 may access the L2's PDPTR, so save them to construct vmcs12
10248 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
10249 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
10250 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
10251 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
10254 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
10255 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
10259 static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10261 struct vcpu_vmx *vmx = to_vmx(vcpu);
10263 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
10264 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
10265 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10267 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12))
10268 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10270 if (nested_vmx_check_apicv_controls(vcpu, vmcs12))
10271 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10273 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12))
10274 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10276 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
10277 vmx->nested.nested_vmx_procbased_ctls_low,
10278 vmx->nested.nested_vmx_procbased_ctls_high) ||
10279 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
10280 vmx->nested.nested_vmx_secondary_ctls_low,
10281 vmx->nested.nested_vmx_secondary_ctls_high) ||
10282 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
10283 vmx->nested.nested_vmx_pinbased_ctls_low,
10284 vmx->nested.nested_vmx_pinbased_ctls_high) ||
10285 !vmx_control_verify(vmcs12->vm_exit_controls,
10286 vmx->nested.nested_vmx_exit_ctls_low,
10287 vmx->nested.nested_vmx_exit_ctls_high) ||
10288 !vmx_control_verify(vmcs12->vm_entry_controls,
10289 vmx->nested.nested_vmx_entry_ctls_low,
10290 vmx->nested.nested_vmx_entry_ctls_high))
10291 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10293 if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
10294 !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
10295 !nested_cr3_valid(vcpu, vmcs12->host_cr3))
10296 return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
10301 static int check_vmentry_postreqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10306 *exit_qual = ENTRY_FAIL_DEFAULT;
10308 if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
10309 !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
10312 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_SHADOW_VMCS) &&
10313 vmcs12->vmcs_link_pointer != -1ull) {
10314 *exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
10319 * If the load IA32_EFER VM-entry control is 1, the following checks
10320 * are performed on the field for the IA32_EFER MSR:
10321 * - Bits reserved in the IA32_EFER MSR must be 0.
10322 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
10323 * the IA-32e mode guest VM-exit control. It must also be identical
10324 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
10327 if (to_vmx(vcpu)->nested.nested_run_pending &&
10328 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
10329 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
10330 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
10331 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
10332 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
10333 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
10338 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
10339 * IA32_EFER MSR must be 0 in the field for that register. In addition,
10340 * the values of the LMA and LME bits in the field must each be that of
10341 * the host address-space size VM-exit control.
10343 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
10344 ia32e = (vmcs12->vm_exit_controls &
10345 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
10346 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
10347 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
10348 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
10355 static int enter_vmx_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
10357 struct vcpu_vmx *vmx = to_vmx(vcpu);
10358 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10359 struct loaded_vmcs *vmcs02;
10363 vmcs02 = nested_get_current_vmcs02(vmx);
10367 enter_guest_mode(vcpu);
10369 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
10370 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10372 vmx_switch_vmcs(vcpu, vmcs02);
10373 vmx_segment_cache_clear(vmx);
10375 if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &exit_qual)) {
10376 leave_guest_mode(vcpu);
10377 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10378 nested_vmx_entry_failure(vcpu, vmcs12,
10379 EXIT_REASON_INVALID_STATE, exit_qual);
10383 nested_get_vmcs12_pages(vcpu, vmcs12);
10385 msr_entry_idx = nested_vmx_load_msr(vcpu,
10386 vmcs12->vm_entry_msr_load_addr,
10387 vmcs12->vm_entry_msr_load_count);
10388 if (msr_entry_idx) {
10389 leave_guest_mode(vcpu);
10390 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10391 nested_vmx_entry_failure(vcpu, vmcs12,
10392 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
10396 vmcs12->launch_state = 1;
10399 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10400 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10401 * returned as far as L1 is concerned. It will only return (and set
10402 * the success flag) when L2 exits (see nested_vmx_vmexit()).
10408 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
10409 * for running an L2 nested guest.
10411 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
10413 struct vmcs12 *vmcs12;
10414 struct vcpu_vmx *vmx = to_vmx(vcpu);
10418 if (!nested_vmx_check_permission(vcpu))
10421 if (!nested_vmx_check_vmcs12(vcpu))
10424 vmcs12 = get_vmcs12(vcpu);
10426 if (enable_shadow_vmcs)
10427 copy_shadow_to_vmcs12(vmx);
10430 * The nested entry process starts with enforcing various prerequisites
10431 * on vmcs12 as required by the Intel SDM, and act appropriately when
10432 * they fail: As the SDM explains, some conditions should cause the
10433 * instruction to fail, while others will cause the instruction to seem
10434 * to succeed, but return an EXIT_REASON_INVALID_STATE.
10435 * To speed up the normal (success) code path, we should avoid checking
10436 * for misconfigurations which will anyway be caught by the processor
10437 * when using the merged vmcs02.
10439 if (vmcs12->launch_state == launch) {
10440 nested_vmx_failValid(vcpu,
10441 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
10442 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
10446 ret = check_vmentry_prereqs(vcpu, vmcs12);
10448 nested_vmx_failValid(vcpu, ret);
10453 * After this point, the trap flag no longer triggers a singlestep trap
10454 * on the vm entry instructions; don't call kvm_skip_emulated_instruction.
10455 * This is not 100% correct; for performance reasons, we delegate most
10456 * of the checks on host state to the processor. If those fail,
10457 * the singlestep trap is missed.
10459 skip_emulated_instruction(vcpu);
10461 ret = check_vmentry_postreqs(vcpu, vmcs12, &exit_qual);
10463 nested_vmx_entry_failure(vcpu, vmcs12,
10464 EXIT_REASON_INVALID_STATE, exit_qual);
10469 * We're finally done with prerequisite checking, and can start with
10470 * the nested entry.
10473 ret = enter_vmx_non_root_mode(vcpu, true);
10477 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
10478 return kvm_vcpu_halt(vcpu);
10480 vmx->nested.nested_run_pending = 1;
10485 return kvm_skip_emulated_instruction(vcpu);
10489 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
10490 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
10491 * This function returns the new value we should put in vmcs12.guest_cr0.
10492 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
10493 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
10494 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
10495 * didn't trap the bit, because if L1 did, so would L0).
10496 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
10497 * been modified by L2, and L1 knows it. So just leave the old value of
10498 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
10499 * isn't relevant, because if L0 traps this bit it can set it to anything.
10500 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
10501 * changed these bits, and therefore they need to be updated, but L0
10502 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
10503 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
10505 static inline unsigned long
10506 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10509 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
10510 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
10511 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
10512 vcpu->arch.cr0_guest_owned_bits));
10515 static inline unsigned long
10516 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10519 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
10520 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
10521 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
10522 vcpu->arch.cr4_guest_owned_bits));
10525 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
10526 struct vmcs12 *vmcs12)
10531 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
10532 nr = vcpu->arch.exception.nr;
10533 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10535 if (kvm_exception_is_soft(nr)) {
10536 vmcs12->vm_exit_instruction_len =
10537 vcpu->arch.event_exit_inst_len;
10538 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
10540 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
10542 if (vcpu->arch.exception.has_error_code) {
10543 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
10544 vmcs12->idt_vectoring_error_code =
10545 vcpu->arch.exception.error_code;
10548 vmcs12->idt_vectoring_info_field = idt_vectoring;
10549 } else if (vcpu->arch.nmi_injected) {
10550 vmcs12->idt_vectoring_info_field =
10551 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
10552 } else if (vcpu->arch.interrupt.pending) {
10553 nr = vcpu->arch.interrupt.nr;
10554 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10556 if (vcpu->arch.interrupt.soft) {
10557 idt_vectoring |= INTR_TYPE_SOFT_INTR;
10558 vmcs12->vm_entry_instruction_len =
10559 vcpu->arch.event_exit_inst_len;
10561 idt_vectoring |= INTR_TYPE_EXT_INTR;
10563 vmcs12->idt_vectoring_info_field = idt_vectoring;
10567 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
10569 struct vcpu_vmx *vmx = to_vmx(vcpu);
10571 if (vcpu->arch.exception.pending ||
10572 vcpu->arch.nmi_injected ||
10573 vcpu->arch.interrupt.pending)
10576 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
10577 vmx->nested.preemption_timer_expired) {
10578 if (vmx->nested.nested_run_pending)
10580 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
10584 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
10585 if (vmx->nested.nested_run_pending)
10587 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10588 NMI_VECTOR | INTR_TYPE_NMI_INTR |
10589 INTR_INFO_VALID_MASK, 0);
10591 * The NMI-triggered VM exit counts as injection:
10592 * clear this one and block further NMIs.
10594 vcpu->arch.nmi_pending = 0;
10595 vmx_set_nmi_mask(vcpu, true);
10599 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
10600 nested_exit_on_intr(vcpu)) {
10601 if (vmx->nested.nested_run_pending)
10603 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
10607 vmx_complete_nested_posted_interrupt(vcpu);
10611 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
10613 ktime_t remaining =
10614 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
10617 if (ktime_to_ns(remaining) <= 0)
10620 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
10621 do_div(value, 1000000);
10622 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10626 * Update the guest state fields of vmcs12 to reflect changes that
10627 * occurred while L2 was running. (The "IA-32e mode guest" bit of the
10628 * VM-entry controls is also updated, since this is really a guest
10631 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10633 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
10634 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
10636 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
10637 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
10638 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
10640 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
10641 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
10642 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
10643 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
10644 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
10645 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
10646 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
10647 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
10648 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
10649 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
10650 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
10651 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
10652 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
10653 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
10654 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
10655 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
10656 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
10657 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
10658 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
10659 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
10660 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
10661 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
10662 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
10663 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
10664 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
10665 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
10666 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
10667 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
10668 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
10669 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
10670 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
10671 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
10672 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
10673 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
10674 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
10675 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
10677 vmcs12->guest_interruptibility_info =
10678 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
10679 vmcs12->guest_pending_dbg_exceptions =
10680 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
10681 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10682 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
10684 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
10686 if (nested_cpu_has_preemption_timer(vmcs12)) {
10687 if (vmcs12->vm_exit_controls &
10688 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
10689 vmcs12->vmx_preemption_timer_value =
10690 vmx_get_preemption_timer_value(vcpu);
10691 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
10695 * In some cases (usually, nested EPT), L2 is allowed to change its
10696 * own CR3 without exiting. If it has changed it, we must keep it.
10697 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10698 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10700 * Additionally, restore L2's PDPTR to vmcs12.
10703 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
10704 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
10705 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
10706 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
10707 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
10710 if (nested_cpu_has_ept(vmcs12))
10711 vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
10713 if (nested_cpu_has_vid(vmcs12))
10714 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
10716 vmcs12->vm_entry_controls =
10717 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
10718 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
10720 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
10721 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
10722 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10725 /* TODO: These cannot have changed unless we have MSR bitmaps and
10726 * the relevant bit asks not to trap the change */
10727 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
10728 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
10729 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
10730 vmcs12->guest_ia32_efer = vcpu->arch.efer;
10731 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
10732 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
10733 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
10734 if (kvm_mpx_supported())
10735 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
10736 if (nested_cpu_has_xsaves(vmcs12))
10737 vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP);
10741 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10742 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10743 * and this function updates it to reflect the changes to the guest state while
10744 * L2 was running (and perhaps made some exits which were handled directly by L0
10745 * without going back to L1), and to reflect the exit reason.
10746 * Note that we do not have to copy here all VMCS fields, just those that
10747 * could have changed by the L2 guest or the exit - i.e., the guest-state and
10748 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10749 * which already writes to vmcs12 directly.
10751 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10752 u32 exit_reason, u32 exit_intr_info,
10753 unsigned long exit_qualification)
10755 /* update guest state fields: */
10756 sync_vmcs12(vcpu, vmcs12);
10758 /* update exit information fields: */
10760 vmcs12->vm_exit_reason = exit_reason;
10761 vmcs12->exit_qualification = exit_qualification;
10763 vmcs12->vm_exit_intr_info = exit_intr_info;
10764 if ((vmcs12->vm_exit_intr_info &
10765 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
10766 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
10767 vmcs12->vm_exit_intr_error_code =
10768 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
10769 vmcs12->idt_vectoring_info_field = 0;
10770 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
10771 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
10773 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
10774 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10775 * instead of reading the real value. */
10776 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
10779 * Transfer the event that L0 or L1 may wanted to inject into
10780 * L2 to IDT_VECTORING_INFO_FIELD.
10782 vmcs12_save_pending_event(vcpu, vmcs12);
10786 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10787 * preserved above and would only end up incorrectly in L1.
10789 vcpu->arch.nmi_injected = false;
10790 kvm_clear_exception_queue(vcpu);
10791 kvm_clear_interrupt_queue(vcpu);
10795 * A part of what we need to when the nested L2 guest exits and we want to
10796 * run its L1 parent, is to reset L1's guest state to the host state specified
10798 * This function is to be called not only on normal nested exit, but also on
10799 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10800 * Failures During or After Loading Guest State").
10801 * This function should be called when the active VMCS is L1's (vmcs01).
10803 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10804 struct vmcs12 *vmcs12)
10806 struct kvm_segment seg;
10807 u32 entry_failure_code;
10809 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10810 vcpu->arch.efer = vmcs12->host_ia32_efer;
10811 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10812 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10814 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10815 vmx_set_efer(vcpu, vcpu->arch.efer);
10817 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10818 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10819 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10821 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10822 * actually changed, because vmx_set_cr0 refers to efer set above.
10824 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
10825 * (KVM doesn't change it);
10827 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
10828 vmx_set_cr0(vcpu, vmcs12->host_cr0);
10830 /* Same as above - no reason to call set_cr4_guest_host_mask(). */
10831 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
10832 kvm_set_cr4(vcpu, vmcs12->host_cr4);
10834 nested_ept_uninit_mmu_context(vcpu);
10837 * Only PDPTE load can fail as the value of cr3 was checked on entry and
10838 * couldn't have changed.
10840 if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
10841 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
10844 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
10848 * Trivially support vpid by letting L2s share their parent
10849 * L1's vpid. TODO: move to a more elaborate solution, giving
10850 * each L2 its own vpid and exposing the vpid feature to L1.
10852 vmx_flush_tlb(vcpu);
10856 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
10857 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
10858 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
10859 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
10860 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
10862 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
10863 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
10864 vmcs_write64(GUEST_BNDCFGS, 0);
10866 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
10867 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
10868 vcpu->arch.pat = vmcs12->host_ia32_pat;
10870 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
10871 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
10872 vmcs12->host_ia32_perf_global_ctrl);
10874 /* Set L1 segment info according to Intel SDM
10875 27.5.2 Loading Host Segment and Descriptor-Table Registers */
10876 seg = (struct kvm_segment) {
10878 .limit = 0xFFFFFFFF,
10879 .selector = vmcs12->host_cs_selector,
10885 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10889 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
10890 seg = (struct kvm_segment) {
10892 .limit = 0xFFFFFFFF,
10899 seg.selector = vmcs12->host_ds_selector;
10900 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
10901 seg.selector = vmcs12->host_es_selector;
10902 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
10903 seg.selector = vmcs12->host_ss_selector;
10904 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
10905 seg.selector = vmcs12->host_fs_selector;
10906 seg.base = vmcs12->host_fs_base;
10907 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
10908 seg.selector = vmcs12->host_gs_selector;
10909 seg.base = vmcs12->host_gs_base;
10910 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
10911 seg = (struct kvm_segment) {
10912 .base = vmcs12->host_tr_base,
10914 .selector = vmcs12->host_tr_selector,
10918 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
10920 kvm_set_dr(vcpu, 7, 0x400);
10921 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
10923 if (cpu_has_vmx_msr_bitmap())
10924 vmx_set_msr_bitmap(vcpu);
10926 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
10927 vmcs12->vm_exit_msr_load_count))
10928 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
10932 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
10933 * and modify vmcs12 to make it see what it would expect to see there if
10934 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
10936 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
10937 u32 exit_intr_info,
10938 unsigned long exit_qualification)
10940 struct vcpu_vmx *vmx = to_vmx(vcpu);
10941 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10942 u32 vm_inst_error = 0;
10944 /* trying to cancel vmlaunch/vmresume is a bug */
10945 WARN_ON_ONCE(vmx->nested.nested_run_pending);
10947 leave_guest_mode(vcpu);
10948 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
10949 exit_qualification);
10951 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
10952 vmcs12->vm_exit_msr_store_count))
10953 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
10955 if (unlikely(vmx->fail))
10956 vm_inst_error = vmcs_read32(VM_INSTRUCTION_ERROR);
10958 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10960 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
10961 && nested_exit_intr_ack_set(vcpu)) {
10962 int irq = kvm_cpu_get_interrupt(vcpu);
10964 vmcs12->vm_exit_intr_info = irq |
10965 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
10968 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
10969 vmcs12->exit_qualification,
10970 vmcs12->idt_vectoring_info_field,
10971 vmcs12->vm_exit_intr_info,
10972 vmcs12->vm_exit_intr_error_code,
10975 vm_entry_controls_reset_shadow(vmx);
10976 vm_exit_controls_reset_shadow(vmx);
10977 vmx_segment_cache_clear(vmx);
10979 /* if no vmcs02 cache requested, remove the one we used */
10980 if (VMCS02_POOL_SIZE == 0)
10981 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
10983 load_vmcs12_host_state(vcpu, vmcs12);
10985 /* Update any VMCS fields that might have changed while L2 ran */
10986 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10987 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10988 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
10989 if (vmx->hv_deadline_tsc == -1)
10990 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
10991 PIN_BASED_VMX_PREEMPTION_TIMER);
10993 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
10994 PIN_BASED_VMX_PREEMPTION_TIMER);
10995 if (kvm_has_tsc_control)
10996 decache_tsc_multiplier(vmx);
10998 if (vmx->nested.change_vmcs01_virtual_x2apic_mode) {
10999 vmx->nested.change_vmcs01_virtual_x2apic_mode = false;
11000 vmx_set_virtual_x2apic_mode(vcpu,
11001 vcpu->arch.apic_base & X2APIC_ENABLE);
11002 } else if (!nested_cpu_has_ept(vmcs12) &&
11003 nested_cpu_has2(vmcs12,
11004 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
11005 vmx_flush_tlb_ept_only(vcpu);
11008 /* This is needed for same reason as it was needed in prepare_vmcs02 */
11011 /* Unpin physical memory we referred to in vmcs02 */
11012 if (vmx->nested.apic_access_page) {
11013 nested_release_page(vmx->nested.apic_access_page);
11014 vmx->nested.apic_access_page = NULL;
11016 if (vmx->nested.virtual_apic_page) {
11017 nested_release_page(vmx->nested.virtual_apic_page);
11018 vmx->nested.virtual_apic_page = NULL;
11020 if (vmx->nested.pi_desc_page) {
11021 kunmap(vmx->nested.pi_desc_page);
11022 nested_release_page(vmx->nested.pi_desc_page);
11023 vmx->nested.pi_desc_page = NULL;
11024 vmx->nested.pi_desc = NULL;
11028 * We are now running in L2, mmu_notifier will force to reload the
11029 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
11031 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
11034 * Exiting from L2 to L1, we're now back to L1 which thinks it just
11035 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
11036 * success or failure flag accordingly.
11038 if (unlikely(vmx->fail)) {
11040 nested_vmx_failValid(vcpu, vm_inst_error);
11042 nested_vmx_succeed(vcpu);
11043 if (enable_shadow_vmcs)
11044 vmx->nested.sync_shadow_vmcs = true;
11046 /* in case we halted in L2 */
11047 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11051 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
11053 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
11055 if (is_guest_mode(vcpu)) {
11056 to_vmx(vcpu)->nested.nested_run_pending = 0;
11057 nested_vmx_vmexit(vcpu, -1, 0, 0);
11059 free_nested(to_vmx(vcpu));
11063 * L1's failure to enter L2 is a subset of a normal exit, as explained in
11064 * 23.7 "VM-entry failures during or after loading guest state" (this also
11065 * lists the acceptable exit-reason and exit-qualification parameters).
11066 * It should only be called before L2 actually succeeded to run, and when
11067 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
11069 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
11070 struct vmcs12 *vmcs12,
11071 u32 reason, unsigned long qualification)
11073 load_vmcs12_host_state(vcpu, vmcs12);
11074 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
11075 vmcs12->exit_qualification = qualification;
11076 nested_vmx_succeed(vcpu);
11077 if (enable_shadow_vmcs)
11078 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
11081 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
11082 struct x86_instruction_info *info,
11083 enum x86_intercept_stage stage)
11085 return X86EMUL_CONTINUE;
11088 #ifdef CONFIG_X86_64
11089 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
11090 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
11091 u64 divisor, u64 *result)
11093 u64 low = a << shift, high = a >> (64 - shift);
11095 /* To avoid the overflow on divq */
11096 if (high >= divisor)
11099 /* Low hold the result, high hold rem which is discarded */
11100 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
11101 "rm" (divisor), "0" (low), "1" (high));
11107 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
11109 struct vcpu_vmx *vmx = to_vmx(vcpu);
11110 u64 tscl = rdtsc();
11111 u64 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
11112 u64 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
11114 /* Convert to host delta tsc if tsc scaling is enabled */
11115 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
11116 u64_shl_div_u64(delta_tsc,
11117 kvm_tsc_scaling_ratio_frac_bits,
11118 vcpu->arch.tsc_scaling_ratio,
11123 * If the delta tsc can't fit in the 32 bit after the multi shift,
11124 * we can't use the preemption timer.
11125 * It's possible that it fits on later vmentries, but checking
11126 * on every vmentry is costly so we just use an hrtimer.
11128 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
11131 vmx->hv_deadline_tsc = tscl + delta_tsc;
11132 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11133 PIN_BASED_VMX_PREEMPTION_TIMER);
11137 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
11139 struct vcpu_vmx *vmx = to_vmx(vcpu);
11140 vmx->hv_deadline_tsc = -1;
11141 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11142 PIN_BASED_VMX_PREEMPTION_TIMER);
11146 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
11149 shrink_ple_window(vcpu);
11152 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
11153 struct kvm_memory_slot *slot)
11155 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
11156 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
11159 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
11160 struct kvm_memory_slot *slot)
11162 kvm_mmu_slot_set_dirty(kvm, slot);
11165 static void vmx_flush_log_dirty(struct kvm *kvm)
11167 kvm_flush_pml_buffers(kvm);
11170 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
11171 struct kvm_memory_slot *memslot,
11172 gfn_t offset, unsigned long mask)
11174 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
11178 * This routine does the following things for vCPU which is going
11179 * to be blocked if VT-d PI is enabled.
11180 * - Store the vCPU to the wakeup list, so when interrupts happen
11181 * we can find the right vCPU to wake up.
11182 * - Change the Posted-interrupt descriptor as below:
11183 * 'NDST' <-- vcpu->pre_pcpu
11184 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
11185 * - If 'ON' is set during this process, which means at least one
11186 * interrupt is posted for this vCPU, we cannot block it, in
11187 * this case, return 1, otherwise, return 0.
11190 static int pi_pre_block(struct kvm_vcpu *vcpu)
11192 unsigned long flags;
11194 struct pi_desc old, new;
11195 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11197 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11198 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11199 !kvm_vcpu_apicv_active(vcpu))
11202 vcpu->pre_pcpu = vcpu->cpu;
11203 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
11204 vcpu->pre_pcpu), flags);
11205 list_add_tail(&vcpu->blocked_vcpu_list,
11206 &per_cpu(blocked_vcpu_on_cpu,
11208 spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
11209 vcpu->pre_pcpu), flags);
11212 old.control = new.control = pi_desc->control;
11215 * We should not block the vCPU if
11216 * an interrupt is posted for it.
11218 if (pi_test_on(pi_desc) == 1) {
11219 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
11220 vcpu->pre_pcpu), flags);
11221 list_del(&vcpu->blocked_vcpu_list);
11222 spin_unlock_irqrestore(
11223 &per_cpu(blocked_vcpu_on_cpu_lock,
11224 vcpu->pre_pcpu), flags);
11225 vcpu->pre_pcpu = -1;
11230 WARN((pi_desc->sn == 1),
11231 "Warning: SN field of posted-interrupts "
11232 "is set before blocking\n");
11235 * Since vCPU can be preempted during this process,
11236 * vcpu->cpu could be different with pre_pcpu, we
11237 * need to set pre_pcpu as the destination of wakeup
11238 * notification event, then we can find the right vCPU
11239 * to wakeup in wakeup handler if interrupts happen
11240 * when the vCPU is in blocked state.
11242 dest = cpu_physical_id(vcpu->pre_pcpu);
11244 if (x2apic_enabled())
11247 new.ndst = (dest << 8) & 0xFF00;
11249 /* set 'NV' to 'wakeup vector' */
11250 new.nv = POSTED_INTR_WAKEUP_VECTOR;
11251 } while (cmpxchg(&pi_desc->control, old.control,
11252 new.control) != old.control);
11257 static int vmx_pre_block(struct kvm_vcpu *vcpu)
11259 if (pi_pre_block(vcpu))
11262 if (kvm_lapic_hv_timer_in_use(vcpu))
11263 kvm_lapic_switch_to_sw_timer(vcpu);
11268 static void pi_post_block(struct kvm_vcpu *vcpu)
11270 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11271 struct pi_desc old, new;
11273 unsigned long flags;
11275 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11276 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11277 !kvm_vcpu_apicv_active(vcpu))
11281 old.control = new.control = pi_desc->control;
11283 dest = cpu_physical_id(vcpu->cpu);
11285 if (x2apic_enabled())
11288 new.ndst = (dest << 8) & 0xFF00;
11290 /* Allow posting non-urgent interrupts */
11293 /* set 'NV' to 'notification vector' */
11294 new.nv = POSTED_INTR_VECTOR;
11295 } while (cmpxchg(&pi_desc->control, old.control,
11296 new.control) != old.control);
11298 if(vcpu->pre_pcpu != -1) {
11300 &per_cpu(blocked_vcpu_on_cpu_lock,
11301 vcpu->pre_pcpu), flags);
11302 list_del(&vcpu->blocked_vcpu_list);
11303 spin_unlock_irqrestore(
11304 &per_cpu(blocked_vcpu_on_cpu_lock,
11305 vcpu->pre_pcpu), flags);
11306 vcpu->pre_pcpu = -1;
11310 static void vmx_post_block(struct kvm_vcpu *vcpu)
11312 if (kvm_x86_ops->set_hv_timer)
11313 kvm_lapic_switch_to_hv_timer(vcpu);
11315 pi_post_block(vcpu);
11319 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
11322 * @host_irq: host irq of the interrupt
11323 * @guest_irq: gsi of the interrupt
11324 * @set: set or unset PI
11325 * returns 0 on success, < 0 on failure
11327 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
11328 uint32_t guest_irq, bool set)
11330 struct kvm_kernel_irq_routing_entry *e;
11331 struct kvm_irq_routing_table *irq_rt;
11332 struct kvm_lapic_irq irq;
11333 struct kvm_vcpu *vcpu;
11334 struct vcpu_data vcpu_info;
11335 int idx, ret = -EINVAL;
11337 if (!kvm_arch_has_assigned_device(kvm) ||
11338 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11339 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
11342 idx = srcu_read_lock(&kvm->irq_srcu);
11343 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
11344 BUG_ON(guest_irq >= irq_rt->nr_rt_entries);
11346 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
11347 if (e->type != KVM_IRQ_ROUTING_MSI)
11350 * VT-d PI cannot support posting multicast/broadcast
11351 * interrupts to a vCPU, we still use interrupt remapping
11352 * for these kind of interrupts.
11354 * For lowest-priority interrupts, we only support
11355 * those with single CPU as the destination, e.g. user
11356 * configures the interrupts via /proc/irq or uses
11357 * irqbalance to make the interrupts single-CPU.
11359 * We will support full lowest-priority interrupt later.
11362 kvm_set_msi_irq(kvm, e, &irq);
11363 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
11365 * Make sure the IRTE is in remapped mode if
11366 * we don't handle it in posted mode.
11368 ret = irq_set_vcpu_affinity(host_irq, NULL);
11371 "failed to back to remapped mode, irq: %u\n",
11379 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
11380 vcpu_info.vector = irq.vector;
11382 trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
11383 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
11386 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
11388 /* suppress notification event before unposting */
11389 pi_set_sn(vcpu_to_pi_desc(vcpu));
11390 ret = irq_set_vcpu_affinity(host_irq, NULL);
11391 pi_clear_sn(vcpu_to_pi_desc(vcpu));
11395 printk(KERN_INFO "%s: failed to update PI IRTE\n",
11403 srcu_read_unlock(&kvm->irq_srcu, idx);
11407 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
11409 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
11410 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
11411 FEATURE_CONTROL_LMCE;
11413 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
11414 ~FEATURE_CONTROL_LMCE;
11417 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
11418 .cpu_has_kvm_support = cpu_has_kvm_support,
11419 .disabled_by_bios = vmx_disabled_by_bios,
11420 .hardware_setup = hardware_setup,
11421 .hardware_unsetup = hardware_unsetup,
11422 .check_processor_compatibility = vmx_check_processor_compat,
11423 .hardware_enable = hardware_enable,
11424 .hardware_disable = hardware_disable,
11425 .cpu_has_accelerated_tpr = report_flexpriority,
11426 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
11428 .vcpu_create = vmx_create_vcpu,
11429 .vcpu_free = vmx_free_vcpu,
11430 .vcpu_reset = vmx_vcpu_reset,
11432 .prepare_guest_switch = vmx_save_host_state,
11433 .vcpu_load = vmx_vcpu_load,
11434 .vcpu_put = vmx_vcpu_put,
11436 .update_bp_intercept = update_exception_bitmap,
11437 .get_msr = vmx_get_msr,
11438 .set_msr = vmx_set_msr,
11439 .get_segment_base = vmx_get_segment_base,
11440 .get_segment = vmx_get_segment,
11441 .set_segment = vmx_set_segment,
11442 .get_cpl = vmx_get_cpl,
11443 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
11444 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
11445 .decache_cr3 = vmx_decache_cr3,
11446 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
11447 .set_cr0 = vmx_set_cr0,
11448 .set_cr3 = vmx_set_cr3,
11449 .set_cr4 = vmx_set_cr4,
11450 .set_efer = vmx_set_efer,
11451 .get_idt = vmx_get_idt,
11452 .set_idt = vmx_set_idt,
11453 .get_gdt = vmx_get_gdt,
11454 .set_gdt = vmx_set_gdt,
11455 .get_dr6 = vmx_get_dr6,
11456 .set_dr6 = vmx_set_dr6,
11457 .set_dr7 = vmx_set_dr7,
11458 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
11459 .cache_reg = vmx_cache_reg,
11460 .get_rflags = vmx_get_rflags,
11461 .set_rflags = vmx_set_rflags,
11463 .get_pkru = vmx_get_pkru,
11465 .tlb_flush = vmx_flush_tlb,
11467 .run = vmx_vcpu_run,
11468 .handle_exit = vmx_handle_exit,
11469 .skip_emulated_instruction = skip_emulated_instruction,
11470 .set_interrupt_shadow = vmx_set_interrupt_shadow,
11471 .get_interrupt_shadow = vmx_get_interrupt_shadow,
11472 .patch_hypercall = vmx_patch_hypercall,
11473 .set_irq = vmx_inject_irq,
11474 .set_nmi = vmx_inject_nmi,
11475 .queue_exception = vmx_queue_exception,
11476 .cancel_injection = vmx_cancel_injection,
11477 .interrupt_allowed = vmx_interrupt_allowed,
11478 .nmi_allowed = vmx_nmi_allowed,
11479 .get_nmi_mask = vmx_get_nmi_mask,
11480 .set_nmi_mask = vmx_set_nmi_mask,
11481 .enable_nmi_window = enable_nmi_window,
11482 .enable_irq_window = enable_irq_window,
11483 .update_cr8_intercept = update_cr8_intercept,
11484 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
11485 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
11486 .get_enable_apicv = vmx_get_enable_apicv,
11487 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
11488 .load_eoi_exitmap = vmx_load_eoi_exitmap,
11489 .apicv_post_state_restore = vmx_apicv_post_state_restore,
11490 .hwapic_irr_update = vmx_hwapic_irr_update,
11491 .hwapic_isr_update = vmx_hwapic_isr_update,
11492 .sync_pir_to_irr = vmx_sync_pir_to_irr,
11493 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
11495 .set_tss_addr = vmx_set_tss_addr,
11496 .get_tdp_level = get_ept_level,
11497 .get_mt_mask = vmx_get_mt_mask,
11499 .get_exit_info = vmx_get_exit_info,
11501 .get_lpage_level = vmx_get_lpage_level,
11503 .cpuid_update = vmx_cpuid_update,
11505 .rdtscp_supported = vmx_rdtscp_supported,
11506 .invpcid_supported = vmx_invpcid_supported,
11508 .set_supported_cpuid = vmx_set_supported_cpuid,
11510 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
11512 .write_tsc_offset = vmx_write_tsc_offset,
11514 .set_tdp_cr3 = vmx_set_cr3,
11516 .check_intercept = vmx_check_intercept,
11517 .handle_external_intr = vmx_handle_external_intr,
11518 .mpx_supported = vmx_mpx_supported,
11519 .xsaves_supported = vmx_xsaves_supported,
11521 .check_nested_events = vmx_check_nested_events,
11523 .sched_in = vmx_sched_in,
11525 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
11526 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
11527 .flush_log_dirty = vmx_flush_log_dirty,
11528 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
11530 .pre_block = vmx_pre_block,
11531 .post_block = vmx_post_block,
11533 .pmu_ops = &intel_pmu_ops,
11535 .update_pi_irte = vmx_update_pi_irte,
11537 #ifdef CONFIG_X86_64
11538 .set_hv_timer = vmx_set_hv_timer,
11539 .cancel_hv_timer = vmx_cancel_hv_timer,
11542 .setup_mce = vmx_setup_mce,
11545 static int __init vmx_init(void)
11547 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
11548 __alignof__(struct vcpu_vmx), THIS_MODULE);
11552 #ifdef CONFIG_KEXEC_CORE
11553 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
11554 crash_vmclear_local_loaded_vmcss);
11560 static void __exit vmx_exit(void)
11562 #ifdef CONFIG_KEXEC_CORE
11563 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
11570 module_init(vmx_init)
11571 module_exit(vmx_exit)