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;
618 /* Support for vnmi-less CPUs */
619 int soft_vnmi_blocked;
621 s64 vnmi_blocked_time;
624 /* Posted interrupt descriptor */
625 struct pi_desc pi_desc;
627 /* Support for a guest hypervisor (nested VMX) */
628 struct nested_vmx nested;
630 /* Dynamic PLE window. */
632 bool ple_window_dirty;
634 /* Support for PML */
635 #define PML_ENTITY_NUM 512
638 /* apic deadline value in host tsc */
641 u64 current_tsc_ratio;
643 bool guest_pkru_valid;
648 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
649 * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
650 * in msr_ia32_feature_control_valid_bits.
652 u64 msr_ia32_feature_control;
653 u64 msr_ia32_feature_control_valid_bits;
656 enum segment_cache_field {
665 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
667 return container_of(vcpu, struct vcpu_vmx, vcpu);
670 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
672 return &(to_vmx(vcpu)->pi_desc);
675 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
676 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
677 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
678 [number##_HIGH] = VMCS12_OFFSET(name)+4
681 static unsigned long shadow_read_only_fields[] = {
683 * We do NOT shadow fields that are modified when L0
684 * traps and emulates any vmx instruction (e.g. VMPTRLD,
685 * VMXON...) executed by L1.
686 * For example, VM_INSTRUCTION_ERROR is read
687 * by L1 if a vmx instruction fails (part of the error path).
688 * Note the code assumes this logic. If for some reason
689 * we start shadowing these fields then we need to
690 * force a shadow sync when L0 emulates vmx instructions
691 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
692 * by nested_vmx_failValid)
696 VM_EXIT_INSTRUCTION_LEN,
697 IDT_VECTORING_INFO_FIELD,
698 IDT_VECTORING_ERROR_CODE,
699 VM_EXIT_INTR_ERROR_CODE,
701 GUEST_LINEAR_ADDRESS,
702 GUEST_PHYSICAL_ADDRESS
704 static int max_shadow_read_only_fields =
705 ARRAY_SIZE(shadow_read_only_fields);
707 static unsigned long shadow_read_write_fields[] = {
714 GUEST_INTERRUPTIBILITY_INFO,
727 CPU_BASED_VM_EXEC_CONTROL,
728 VM_ENTRY_EXCEPTION_ERROR_CODE,
729 VM_ENTRY_INTR_INFO_FIELD,
730 VM_ENTRY_INSTRUCTION_LEN,
731 VM_ENTRY_EXCEPTION_ERROR_CODE,
737 static int max_shadow_read_write_fields =
738 ARRAY_SIZE(shadow_read_write_fields);
740 static const unsigned short vmcs_field_to_offset_table[] = {
741 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
742 FIELD(POSTED_INTR_NV, posted_intr_nv),
743 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
744 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
745 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
746 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
747 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
748 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
749 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
750 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
751 FIELD(GUEST_INTR_STATUS, guest_intr_status),
752 FIELD(HOST_ES_SELECTOR, host_es_selector),
753 FIELD(HOST_CS_SELECTOR, host_cs_selector),
754 FIELD(HOST_SS_SELECTOR, host_ss_selector),
755 FIELD(HOST_DS_SELECTOR, host_ds_selector),
756 FIELD(HOST_FS_SELECTOR, host_fs_selector),
757 FIELD(HOST_GS_SELECTOR, host_gs_selector),
758 FIELD(HOST_TR_SELECTOR, host_tr_selector),
759 FIELD64(IO_BITMAP_A, io_bitmap_a),
760 FIELD64(IO_BITMAP_B, io_bitmap_b),
761 FIELD64(MSR_BITMAP, msr_bitmap),
762 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
763 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
764 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
765 FIELD64(TSC_OFFSET, tsc_offset),
766 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
767 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
768 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
769 FIELD64(EPT_POINTER, ept_pointer),
770 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
771 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
772 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
773 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
774 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
775 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
776 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
777 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
778 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
779 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
780 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
781 FIELD64(GUEST_PDPTR0, guest_pdptr0),
782 FIELD64(GUEST_PDPTR1, guest_pdptr1),
783 FIELD64(GUEST_PDPTR2, guest_pdptr2),
784 FIELD64(GUEST_PDPTR3, guest_pdptr3),
785 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
786 FIELD64(HOST_IA32_PAT, host_ia32_pat),
787 FIELD64(HOST_IA32_EFER, host_ia32_efer),
788 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
789 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
790 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
791 FIELD(EXCEPTION_BITMAP, exception_bitmap),
792 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
793 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
794 FIELD(CR3_TARGET_COUNT, cr3_target_count),
795 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
796 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
797 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
798 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
799 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
800 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
801 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
802 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
803 FIELD(TPR_THRESHOLD, tpr_threshold),
804 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
805 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
806 FIELD(VM_EXIT_REASON, vm_exit_reason),
807 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
808 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
809 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
810 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
811 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
812 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
813 FIELD(GUEST_ES_LIMIT, guest_es_limit),
814 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
815 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
816 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
817 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
818 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
819 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
820 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
821 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
822 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
823 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
824 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
825 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
826 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
827 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
828 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
829 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
830 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
831 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
832 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
833 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
834 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
835 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
836 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
837 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
838 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
839 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
840 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
841 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
842 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
843 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
844 FIELD(EXIT_QUALIFICATION, exit_qualification),
845 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
846 FIELD(GUEST_CR0, guest_cr0),
847 FIELD(GUEST_CR3, guest_cr3),
848 FIELD(GUEST_CR4, guest_cr4),
849 FIELD(GUEST_ES_BASE, guest_es_base),
850 FIELD(GUEST_CS_BASE, guest_cs_base),
851 FIELD(GUEST_SS_BASE, guest_ss_base),
852 FIELD(GUEST_DS_BASE, guest_ds_base),
853 FIELD(GUEST_FS_BASE, guest_fs_base),
854 FIELD(GUEST_GS_BASE, guest_gs_base),
855 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
856 FIELD(GUEST_TR_BASE, guest_tr_base),
857 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
858 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
859 FIELD(GUEST_DR7, guest_dr7),
860 FIELD(GUEST_RSP, guest_rsp),
861 FIELD(GUEST_RIP, guest_rip),
862 FIELD(GUEST_RFLAGS, guest_rflags),
863 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
864 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
865 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
866 FIELD(HOST_CR0, host_cr0),
867 FIELD(HOST_CR3, host_cr3),
868 FIELD(HOST_CR4, host_cr4),
869 FIELD(HOST_FS_BASE, host_fs_base),
870 FIELD(HOST_GS_BASE, host_gs_base),
871 FIELD(HOST_TR_BASE, host_tr_base),
872 FIELD(HOST_GDTR_BASE, host_gdtr_base),
873 FIELD(HOST_IDTR_BASE, host_idtr_base),
874 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
875 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
876 FIELD(HOST_RSP, host_rsp),
877 FIELD(HOST_RIP, host_rip),
880 static inline short vmcs_field_to_offset(unsigned long field)
882 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
884 if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
885 vmcs_field_to_offset_table[field] == 0)
888 return vmcs_field_to_offset_table[field];
891 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
893 return to_vmx(vcpu)->nested.cached_vmcs12;
896 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
898 struct page *page = kvm_vcpu_gfn_to_page(vcpu, addr >> PAGE_SHIFT);
899 if (is_error_page(page))
905 static void nested_release_page(struct page *page)
907 kvm_release_page_dirty(page);
910 static void nested_release_page_clean(struct page *page)
912 kvm_release_page_clean(page);
915 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
916 static u64 construct_eptp(unsigned long root_hpa);
917 static void kvm_cpu_vmxon(u64 addr);
918 static void kvm_cpu_vmxoff(void);
919 static bool vmx_xsaves_supported(void);
920 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
921 static void vmx_set_segment(struct kvm_vcpu *vcpu,
922 struct kvm_segment *var, int seg);
923 static void vmx_get_segment(struct kvm_vcpu *vcpu,
924 struct kvm_segment *var, int seg);
925 static bool guest_state_valid(struct kvm_vcpu *vcpu);
926 static u32 vmx_segment_access_rights(struct kvm_segment *var);
927 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
928 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
929 static int alloc_identity_pagetable(struct kvm *kvm);
931 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
932 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
934 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
935 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
937 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
938 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
941 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
942 * can find which vCPU should be waken up.
944 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
945 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
950 VMX_MSR_BITMAP_LEGACY,
951 VMX_MSR_BITMAP_LONGMODE,
952 VMX_MSR_BITMAP_LEGACY_X2APIC_APICV,
953 VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV,
954 VMX_MSR_BITMAP_LEGACY_X2APIC,
955 VMX_MSR_BITMAP_LONGMODE_X2APIC,
961 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
963 #define vmx_io_bitmap_a (vmx_bitmap[VMX_IO_BITMAP_A])
964 #define vmx_io_bitmap_b (vmx_bitmap[VMX_IO_BITMAP_B])
965 #define vmx_msr_bitmap_legacy (vmx_bitmap[VMX_MSR_BITMAP_LEGACY])
966 #define vmx_msr_bitmap_longmode (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE])
967 #define vmx_msr_bitmap_legacy_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC_APICV])
968 #define vmx_msr_bitmap_longmode_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV])
969 #define vmx_msr_bitmap_legacy_x2apic (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC])
970 #define vmx_msr_bitmap_longmode_x2apic (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC])
971 #define vmx_vmread_bitmap (vmx_bitmap[VMX_VMREAD_BITMAP])
972 #define vmx_vmwrite_bitmap (vmx_bitmap[VMX_VMWRITE_BITMAP])
974 static bool cpu_has_load_ia32_efer;
975 static bool cpu_has_load_perf_global_ctrl;
977 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
978 static DEFINE_SPINLOCK(vmx_vpid_lock);
980 static struct vmcs_config {
985 u32 pin_based_exec_ctrl;
986 u32 cpu_based_exec_ctrl;
987 u32 cpu_based_2nd_exec_ctrl;
992 static struct vmx_capability {
997 #define VMX_SEGMENT_FIELD(seg) \
998 [VCPU_SREG_##seg] = { \
999 .selector = GUEST_##seg##_SELECTOR, \
1000 .base = GUEST_##seg##_BASE, \
1001 .limit = GUEST_##seg##_LIMIT, \
1002 .ar_bytes = GUEST_##seg##_AR_BYTES, \
1005 static const struct kvm_vmx_segment_field {
1010 } kvm_vmx_segment_fields[] = {
1011 VMX_SEGMENT_FIELD(CS),
1012 VMX_SEGMENT_FIELD(DS),
1013 VMX_SEGMENT_FIELD(ES),
1014 VMX_SEGMENT_FIELD(FS),
1015 VMX_SEGMENT_FIELD(GS),
1016 VMX_SEGMENT_FIELD(SS),
1017 VMX_SEGMENT_FIELD(TR),
1018 VMX_SEGMENT_FIELD(LDTR),
1021 static u64 host_efer;
1023 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
1026 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
1027 * away by decrementing the array size.
1029 static const u32 vmx_msr_index[] = {
1030 #ifdef CONFIG_X86_64
1031 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
1033 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
1036 static inline bool is_exception_n(u32 intr_info, u8 vector)
1038 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1039 INTR_INFO_VALID_MASK)) ==
1040 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
1043 static inline bool is_debug(u32 intr_info)
1045 return is_exception_n(intr_info, DB_VECTOR);
1048 static inline bool is_breakpoint(u32 intr_info)
1050 return is_exception_n(intr_info, BP_VECTOR);
1053 static inline bool is_page_fault(u32 intr_info)
1055 return is_exception_n(intr_info, PF_VECTOR);
1058 static inline bool is_no_device(u32 intr_info)
1060 return is_exception_n(intr_info, NM_VECTOR);
1063 static inline bool is_invalid_opcode(u32 intr_info)
1065 return is_exception_n(intr_info, UD_VECTOR);
1068 static inline bool is_external_interrupt(u32 intr_info)
1070 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1071 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1074 static inline bool is_machine_check(u32 intr_info)
1076 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1077 INTR_INFO_VALID_MASK)) ==
1078 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1081 static inline bool cpu_has_vmx_msr_bitmap(void)
1083 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1086 static inline bool cpu_has_vmx_tpr_shadow(void)
1088 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1091 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1093 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1096 static inline bool cpu_has_secondary_exec_ctrls(void)
1098 return vmcs_config.cpu_based_exec_ctrl &
1099 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1102 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1104 return vmcs_config.cpu_based_2nd_exec_ctrl &
1105 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1108 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1110 return vmcs_config.cpu_based_2nd_exec_ctrl &
1111 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1114 static inline bool cpu_has_vmx_apic_register_virt(void)
1116 return vmcs_config.cpu_based_2nd_exec_ctrl &
1117 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1120 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1122 return vmcs_config.cpu_based_2nd_exec_ctrl &
1123 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1127 * Comment's format: document - errata name - stepping - processor name.
1129 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1131 static u32 vmx_preemption_cpu_tfms[] = {
1132 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
1134 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
1135 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1136 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1138 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1140 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
1141 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
1143 * 320767.pdf - AAP86 - B1 -
1144 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1147 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1149 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1151 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1153 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1154 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1155 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1159 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1161 u32 eax = cpuid_eax(0x00000001), i;
1163 /* Clear the reserved bits */
1164 eax &= ~(0x3U << 14 | 0xfU << 28);
1165 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
1166 if (eax == vmx_preemption_cpu_tfms[i])
1172 static inline bool cpu_has_vmx_preemption_timer(void)
1174 return vmcs_config.pin_based_exec_ctrl &
1175 PIN_BASED_VMX_PREEMPTION_TIMER;
1178 static inline bool cpu_has_vmx_posted_intr(void)
1180 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1181 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1184 static inline bool cpu_has_vmx_apicv(void)
1186 return cpu_has_vmx_apic_register_virt() &&
1187 cpu_has_vmx_virtual_intr_delivery() &&
1188 cpu_has_vmx_posted_intr();
1191 static inline bool cpu_has_vmx_flexpriority(void)
1193 return cpu_has_vmx_tpr_shadow() &&
1194 cpu_has_vmx_virtualize_apic_accesses();
1197 static inline bool cpu_has_vmx_ept_execute_only(void)
1199 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1202 static inline bool cpu_has_vmx_ept_2m_page(void)
1204 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1207 static inline bool cpu_has_vmx_ept_1g_page(void)
1209 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1212 static inline bool cpu_has_vmx_ept_4levels(void)
1214 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1217 static inline bool cpu_has_vmx_ept_ad_bits(void)
1219 return vmx_capability.ept & VMX_EPT_AD_BIT;
1222 static inline bool cpu_has_vmx_invept_context(void)
1224 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1227 static inline bool cpu_has_vmx_invept_global(void)
1229 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1232 static inline bool cpu_has_vmx_invvpid_single(void)
1234 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1237 static inline bool cpu_has_vmx_invvpid_global(void)
1239 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1242 static inline bool cpu_has_vmx_ept(void)
1244 return vmcs_config.cpu_based_2nd_exec_ctrl &
1245 SECONDARY_EXEC_ENABLE_EPT;
1248 static inline bool cpu_has_vmx_unrestricted_guest(void)
1250 return vmcs_config.cpu_based_2nd_exec_ctrl &
1251 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1254 static inline bool cpu_has_vmx_ple(void)
1256 return vmcs_config.cpu_based_2nd_exec_ctrl &
1257 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1260 static inline bool cpu_has_vmx_basic_inout(void)
1262 return (((u64)vmcs_config.basic_cap << 32) & VMX_BASIC_INOUT);
1265 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1267 return flexpriority_enabled && lapic_in_kernel(vcpu);
1270 static inline bool cpu_has_vmx_vpid(void)
1272 return vmcs_config.cpu_based_2nd_exec_ctrl &
1273 SECONDARY_EXEC_ENABLE_VPID;
1276 static inline bool cpu_has_vmx_rdtscp(void)
1278 return vmcs_config.cpu_based_2nd_exec_ctrl &
1279 SECONDARY_EXEC_RDTSCP;
1282 static inline bool cpu_has_vmx_invpcid(void)
1284 return vmcs_config.cpu_based_2nd_exec_ctrl &
1285 SECONDARY_EXEC_ENABLE_INVPCID;
1288 static inline bool cpu_has_virtual_nmis(void)
1290 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1293 static inline bool cpu_has_vmx_wbinvd_exit(void)
1295 return vmcs_config.cpu_based_2nd_exec_ctrl &
1296 SECONDARY_EXEC_WBINVD_EXITING;
1299 static inline bool cpu_has_vmx_shadow_vmcs(void)
1302 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1303 /* check if the cpu supports writing r/o exit information fields */
1304 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1307 return vmcs_config.cpu_based_2nd_exec_ctrl &
1308 SECONDARY_EXEC_SHADOW_VMCS;
1311 static inline bool cpu_has_vmx_pml(void)
1313 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1316 static inline bool cpu_has_vmx_tsc_scaling(void)
1318 return vmcs_config.cpu_based_2nd_exec_ctrl &
1319 SECONDARY_EXEC_TSC_SCALING;
1322 static inline bool report_flexpriority(void)
1324 return flexpriority_enabled;
1327 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1329 return vmcs12->cpu_based_vm_exec_control & bit;
1332 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1334 return (vmcs12->cpu_based_vm_exec_control &
1335 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1336 (vmcs12->secondary_vm_exec_control & bit);
1339 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1341 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1344 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1346 return vmcs12->pin_based_vm_exec_control &
1347 PIN_BASED_VMX_PREEMPTION_TIMER;
1350 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1352 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1355 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1357 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) &&
1358 vmx_xsaves_supported();
1361 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1363 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1366 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1368 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1371 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1373 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1376 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1378 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1381 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1383 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1386 static inline bool is_nmi(u32 intr_info)
1388 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1389 == (INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK);
1392 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1394 unsigned long exit_qualification);
1395 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1396 struct vmcs12 *vmcs12,
1397 u32 reason, unsigned long qualification);
1399 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1403 for (i = 0; i < vmx->nmsrs; ++i)
1404 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1409 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1415 } operand = { vpid, 0, gva };
1417 asm volatile (__ex(ASM_VMX_INVVPID)
1418 /* CF==1 or ZF==1 --> rc = -1 */
1419 "; ja 1f ; ud2 ; 1:"
1420 : : "a"(&operand), "c"(ext) : "cc", "memory");
1423 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1427 } operand = {eptp, gpa};
1429 asm volatile (__ex(ASM_VMX_INVEPT)
1430 /* CF==1 or ZF==1 --> rc = -1 */
1431 "; ja 1f ; ud2 ; 1:\n"
1432 : : "a" (&operand), "c" (ext) : "cc", "memory");
1435 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1439 i = __find_msr_index(vmx, msr);
1441 return &vmx->guest_msrs[i];
1445 static void vmcs_clear(struct vmcs *vmcs)
1447 u64 phys_addr = __pa(vmcs);
1450 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1451 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1454 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1458 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1460 vmcs_clear(loaded_vmcs->vmcs);
1461 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
1462 vmcs_clear(loaded_vmcs->shadow_vmcs);
1463 loaded_vmcs->cpu = -1;
1464 loaded_vmcs->launched = 0;
1467 static void vmcs_load(struct vmcs *vmcs)
1469 u64 phys_addr = __pa(vmcs);
1472 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1473 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1476 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1480 #ifdef CONFIG_KEXEC_CORE
1482 * This bitmap is used to indicate whether the vmclear
1483 * operation is enabled on all cpus. All disabled by
1486 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1488 static inline void crash_enable_local_vmclear(int cpu)
1490 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1493 static inline void crash_disable_local_vmclear(int cpu)
1495 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1498 static inline int crash_local_vmclear_enabled(int cpu)
1500 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1503 static void crash_vmclear_local_loaded_vmcss(void)
1505 int cpu = raw_smp_processor_id();
1506 struct loaded_vmcs *v;
1508 if (!crash_local_vmclear_enabled(cpu))
1511 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1512 loaded_vmcss_on_cpu_link)
1513 vmcs_clear(v->vmcs);
1516 static inline void crash_enable_local_vmclear(int cpu) { }
1517 static inline void crash_disable_local_vmclear(int cpu) { }
1518 #endif /* CONFIG_KEXEC_CORE */
1520 static void __loaded_vmcs_clear(void *arg)
1522 struct loaded_vmcs *loaded_vmcs = arg;
1523 int cpu = raw_smp_processor_id();
1525 if (loaded_vmcs->cpu != cpu)
1526 return; /* vcpu migration can race with cpu offline */
1527 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1528 per_cpu(current_vmcs, cpu) = NULL;
1529 crash_disable_local_vmclear(cpu);
1530 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1533 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1534 * is before setting loaded_vmcs->vcpu to -1 which is done in
1535 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1536 * then adds the vmcs into percpu list before it is deleted.
1540 loaded_vmcs_init(loaded_vmcs);
1541 crash_enable_local_vmclear(cpu);
1544 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1546 int cpu = loaded_vmcs->cpu;
1549 smp_call_function_single(cpu,
1550 __loaded_vmcs_clear, loaded_vmcs, 1);
1553 static inline void vpid_sync_vcpu_single(int vpid)
1558 if (cpu_has_vmx_invvpid_single())
1559 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1562 static inline void vpid_sync_vcpu_global(void)
1564 if (cpu_has_vmx_invvpid_global())
1565 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1568 static inline void vpid_sync_context(int vpid)
1570 if (cpu_has_vmx_invvpid_single())
1571 vpid_sync_vcpu_single(vpid);
1573 vpid_sync_vcpu_global();
1576 static inline void ept_sync_global(void)
1578 if (cpu_has_vmx_invept_global())
1579 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1582 static inline void ept_sync_context(u64 eptp)
1585 if (cpu_has_vmx_invept_context())
1586 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1592 static __always_inline void vmcs_check16(unsigned long field)
1594 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1595 "16-bit accessor invalid for 64-bit field");
1596 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1597 "16-bit accessor invalid for 64-bit high field");
1598 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1599 "16-bit accessor invalid for 32-bit high field");
1600 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1601 "16-bit accessor invalid for natural width field");
1604 static __always_inline void vmcs_check32(unsigned long field)
1606 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1607 "32-bit accessor invalid for 16-bit field");
1608 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1609 "32-bit accessor invalid for natural width field");
1612 static __always_inline void vmcs_check64(unsigned long field)
1614 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1615 "64-bit accessor invalid for 16-bit field");
1616 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1617 "64-bit accessor invalid for 64-bit high field");
1618 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1619 "64-bit accessor invalid for 32-bit field");
1620 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1621 "64-bit accessor invalid for natural width field");
1624 static __always_inline void vmcs_checkl(unsigned long field)
1626 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1627 "Natural width accessor invalid for 16-bit field");
1628 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1629 "Natural width accessor invalid for 64-bit field");
1630 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1631 "Natural width accessor invalid for 64-bit high field");
1632 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1633 "Natural width accessor invalid for 32-bit field");
1636 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1638 unsigned long value;
1640 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1641 : "=a"(value) : "d"(field) : "cc");
1645 static __always_inline u16 vmcs_read16(unsigned long field)
1647 vmcs_check16(field);
1648 return __vmcs_readl(field);
1651 static __always_inline u32 vmcs_read32(unsigned long field)
1653 vmcs_check32(field);
1654 return __vmcs_readl(field);
1657 static __always_inline u64 vmcs_read64(unsigned long field)
1659 vmcs_check64(field);
1660 #ifdef CONFIG_X86_64
1661 return __vmcs_readl(field);
1663 return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1667 static __always_inline unsigned long vmcs_readl(unsigned long field)
1670 return __vmcs_readl(field);
1673 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1675 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1676 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1680 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1684 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1685 : "=q"(error) : "a"(value), "d"(field) : "cc");
1686 if (unlikely(error))
1687 vmwrite_error(field, value);
1690 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1692 vmcs_check16(field);
1693 __vmcs_writel(field, value);
1696 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1698 vmcs_check32(field);
1699 __vmcs_writel(field, value);
1702 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1704 vmcs_check64(field);
1705 __vmcs_writel(field, value);
1706 #ifndef CONFIG_X86_64
1708 __vmcs_writel(field+1, value >> 32);
1712 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1715 __vmcs_writel(field, value);
1718 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1720 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1721 "vmcs_clear_bits does not support 64-bit fields");
1722 __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1725 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1727 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1728 "vmcs_set_bits does not support 64-bit fields");
1729 __vmcs_writel(field, __vmcs_readl(field) | mask);
1732 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
1734 vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
1737 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1739 vmcs_write32(VM_ENTRY_CONTROLS, val);
1740 vmx->vm_entry_controls_shadow = val;
1743 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1745 if (vmx->vm_entry_controls_shadow != val)
1746 vm_entry_controls_init(vmx, val);
1749 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1751 return vmx->vm_entry_controls_shadow;
1755 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1757 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1760 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1762 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1765 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
1767 vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
1770 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1772 vmcs_write32(VM_EXIT_CONTROLS, val);
1773 vmx->vm_exit_controls_shadow = val;
1776 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1778 if (vmx->vm_exit_controls_shadow != val)
1779 vm_exit_controls_init(vmx, val);
1782 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1784 return vmx->vm_exit_controls_shadow;
1788 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1790 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1793 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1795 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1798 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1800 vmx->segment_cache.bitmask = 0;
1803 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1807 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1809 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1810 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1811 vmx->segment_cache.bitmask = 0;
1813 ret = vmx->segment_cache.bitmask & mask;
1814 vmx->segment_cache.bitmask |= mask;
1818 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1820 u16 *p = &vmx->segment_cache.seg[seg].selector;
1822 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1823 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1827 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1829 ulong *p = &vmx->segment_cache.seg[seg].base;
1831 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1832 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1836 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1838 u32 *p = &vmx->segment_cache.seg[seg].limit;
1840 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1841 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1845 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1847 u32 *p = &vmx->segment_cache.seg[seg].ar;
1849 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1850 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1854 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1858 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1859 (1u << DB_VECTOR) | (1u << AC_VECTOR);
1860 if ((vcpu->guest_debug &
1861 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1862 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1863 eb |= 1u << BP_VECTOR;
1864 if (to_vmx(vcpu)->rmode.vm86_active)
1867 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1869 /* When we are running a nested L2 guest and L1 specified for it a
1870 * certain exception bitmap, we must trap the same exceptions and pass
1871 * them to L1. When running L2, we will only handle the exceptions
1872 * specified above if L1 did not want them.
1874 if (is_guest_mode(vcpu))
1875 eb |= get_vmcs12(vcpu)->exception_bitmap;
1877 vmcs_write32(EXCEPTION_BITMAP, eb);
1880 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1881 unsigned long entry, unsigned long exit)
1883 vm_entry_controls_clearbit(vmx, entry);
1884 vm_exit_controls_clearbit(vmx, exit);
1887 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1890 struct msr_autoload *m = &vmx->msr_autoload;
1894 if (cpu_has_load_ia32_efer) {
1895 clear_atomic_switch_msr_special(vmx,
1896 VM_ENTRY_LOAD_IA32_EFER,
1897 VM_EXIT_LOAD_IA32_EFER);
1901 case MSR_CORE_PERF_GLOBAL_CTRL:
1902 if (cpu_has_load_perf_global_ctrl) {
1903 clear_atomic_switch_msr_special(vmx,
1904 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1905 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1911 for (i = 0; i < m->nr; ++i)
1912 if (m->guest[i].index == msr)
1918 m->guest[i] = m->guest[m->nr];
1919 m->host[i] = m->host[m->nr];
1920 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1921 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1924 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1925 unsigned long entry, unsigned long exit,
1926 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1927 u64 guest_val, u64 host_val)
1929 vmcs_write64(guest_val_vmcs, guest_val);
1930 vmcs_write64(host_val_vmcs, host_val);
1931 vm_entry_controls_setbit(vmx, entry);
1932 vm_exit_controls_setbit(vmx, exit);
1935 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1936 u64 guest_val, u64 host_val)
1939 struct msr_autoload *m = &vmx->msr_autoload;
1943 if (cpu_has_load_ia32_efer) {
1944 add_atomic_switch_msr_special(vmx,
1945 VM_ENTRY_LOAD_IA32_EFER,
1946 VM_EXIT_LOAD_IA32_EFER,
1949 guest_val, host_val);
1953 case MSR_CORE_PERF_GLOBAL_CTRL:
1954 if (cpu_has_load_perf_global_ctrl) {
1955 add_atomic_switch_msr_special(vmx,
1956 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1957 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1958 GUEST_IA32_PERF_GLOBAL_CTRL,
1959 HOST_IA32_PERF_GLOBAL_CTRL,
1960 guest_val, host_val);
1964 case MSR_IA32_PEBS_ENABLE:
1965 /* PEBS needs a quiescent period after being disabled (to write
1966 * a record). Disabling PEBS through VMX MSR swapping doesn't
1967 * provide that period, so a CPU could write host's record into
1970 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1973 for (i = 0; i < m->nr; ++i)
1974 if (m->guest[i].index == msr)
1977 if (i == NR_AUTOLOAD_MSRS) {
1978 printk_once(KERN_WARNING "Not enough msr switch entries. "
1979 "Can't add msr %x\n", msr);
1981 } else if (i == m->nr) {
1983 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1984 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1987 m->guest[i].index = msr;
1988 m->guest[i].value = guest_val;
1989 m->host[i].index = msr;
1990 m->host[i].value = host_val;
1993 static void reload_tss(void)
1996 * VT restores TR but not its size. Useless.
1998 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1999 struct desc_struct *descs;
2001 descs = (void *)gdt->address;
2002 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
2006 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
2008 u64 guest_efer = vmx->vcpu.arch.efer;
2009 u64 ignore_bits = 0;
2013 * NX is needed to handle CR0.WP=1, CR4.SMEP=1. Testing
2014 * host CPUID is more efficient than testing guest CPUID
2015 * or CR4. Host SMEP is anyway a requirement for guest SMEP.
2017 if (boot_cpu_has(X86_FEATURE_SMEP))
2018 guest_efer |= EFER_NX;
2019 else if (!(guest_efer & EFER_NX))
2020 ignore_bits |= EFER_NX;
2024 * LMA and LME handled by hardware; SCE meaningless outside long mode.
2026 ignore_bits |= EFER_SCE;
2027 #ifdef CONFIG_X86_64
2028 ignore_bits |= EFER_LMA | EFER_LME;
2029 /* SCE is meaningful only in long mode on Intel */
2030 if (guest_efer & EFER_LMA)
2031 ignore_bits &= ~(u64)EFER_SCE;
2034 clear_atomic_switch_msr(vmx, MSR_EFER);
2037 * On EPT, we can't emulate NX, so we must switch EFER atomically.
2038 * On CPUs that support "load IA32_EFER", always switch EFER
2039 * atomically, since it's faster than switching it manually.
2041 if (cpu_has_load_ia32_efer ||
2042 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
2043 if (!(guest_efer & EFER_LMA))
2044 guest_efer &= ~EFER_LME;
2045 if (guest_efer != host_efer)
2046 add_atomic_switch_msr(vmx, MSR_EFER,
2047 guest_efer, host_efer);
2050 guest_efer &= ~ignore_bits;
2051 guest_efer |= host_efer & ignore_bits;
2053 vmx->guest_msrs[efer_offset].data = guest_efer;
2054 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
2060 static unsigned long segment_base(u16 selector)
2062 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2063 struct desc_struct *d;
2064 unsigned long table_base;
2067 if (!(selector & ~3))
2070 table_base = gdt->address;
2072 if (selector & 4) { /* from ldt */
2073 u16 ldt_selector = kvm_read_ldt();
2075 if (!(ldt_selector & ~3))
2078 table_base = segment_base(ldt_selector);
2080 d = (struct desc_struct *)(table_base + (selector & ~7));
2081 v = get_desc_base(d);
2082 #ifdef CONFIG_X86_64
2083 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
2084 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
2089 static inline unsigned long kvm_read_tr_base(void)
2092 asm("str %0" : "=g"(tr));
2093 return segment_base(tr);
2096 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
2098 struct vcpu_vmx *vmx = to_vmx(vcpu);
2101 if (vmx->host_state.loaded)
2104 vmx->host_state.loaded = 1;
2106 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
2107 * allow segment selectors with cpl > 0 or ti == 1.
2109 vmx->host_state.ldt_sel = kvm_read_ldt();
2110 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
2111 savesegment(fs, vmx->host_state.fs_sel);
2112 if (!(vmx->host_state.fs_sel & 7)) {
2113 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
2114 vmx->host_state.fs_reload_needed = 0;
2116 vmcs_write16(HOST_FS_SELECTOR, 0);
2117 vmx->host_state.fs_reload_needed = 1;
2119 savesegment(gs, vmx->host_state.gs_sel);
2120 if (!(vmx->host_state.gs_sel & 7))
2121 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
2123 vmcs_write16(HOST_GS_SELECTOR, 0);
2124 vmx->host_state.gs_ldt_reload_needed = 1;
2127 #ifdef CONFIG_X86_64
2128 savesegment(ds, vmx->host_state.ds_sel);
2129 savesegment(es, vmx->host_state.es_sel);
2132 #ifdef CONFIG_X86_64
2133 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
2134 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
2136 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
2137 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
2140 #ifdef CONFIG_X86_64
2141 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2142 if (is_long_mode(&vmx->vcpu))
2143 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2145 if (boot_cpu_has(X86_FEATURE_MPX))
2146 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2147 for (i = 0; i < vmx->save_nmsrs; ++i)
2148 kvm_set_shared_msr(vmx->guest_msrs[i].index,
2149 vmx->guest_msrs[i].data,
2150 vmx->guest_msrs[i].mask);
2153 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
2155 if (!vmx->host_state.loaded)
2158 ++vmx->vcpu.stat.host_state_reload;
2159 vmx->host_state.loaded = 0;
2160 #ifdef CONFIG_X86_64
2161 if (is_long_mode(&vmx->vcpu))
2162 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2164 if (vmx->host_state.gs_ldt_reload_needed) {
2165 kvm_load_ldt(vmx->host_state.ldt_sel);
2166 #ifdef CONFIG_X86_64
2167 load_gs_index(vmx->host_state.gs_sel);
2169 loadsegment(gs, vmx->host_state.gs_sel);
2172 if (vmx->host_state.fs_reload_needed)
2173 loadsegment(fs, vmx->host_state.fs_sel);
2174 #ifdef CONFIG_X86_64
2175 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2176 loadsegment(ds, vmx->host_state.ds_sel);
2177 loadsegment(es, vmx->host_state.es_sel);
2181 #ifdef CONFIG_X86_64
2182 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2184 if (vmx->host_state.msr_host_bndcfgs)
2185 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2186 load_gdt(this_cpu_ptr(&host_gdt));
2189 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2192 __vmx_load_host_state(vmx);
2196 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2198 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2199 struct pi_desc old, new;
2202 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2203 !irq_remapping_cap(IRQ_POSTING_CAP) ||
2204 !kvm_vcpu_apicv_active(vcpu))
2208 old.control = new.control = pi_desc->control;
2211 * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
2212 * are two possible cases:
2213 * 1. After running 'pre_block', context switch
2214 * happened. For this case, 'sn' was set in
2215 * vmx_vcpu_put(), so we need to clear it here.
2216 * 2. After running 'pre_block', we were blocked,
2217 * and woken up by some other guy. For this case,
2218 * we don't need to do anything, 'pi_post_block'
2219 * will do everything for us. However, we cannot
2220 * check whether it is case #1 or case #2 here
2221 * (maybe, not needed), so we also clear sn here,
2222 * I think it is not a big deal.
2224 if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) {
2225 if (vcpu->cpu != cpu) {
2226 dest = cpu_physical_id(cpu);
2228 if (x2apic_enabled())
2231 new.ndst = (dest << 8) & 0xFF00;
2234 /* set 'NV' to 'notification vector' */
2235 new.nv = POSTED_INTR_VECTOR;
2238 /* Allow posting non-urgent interrupts */
2240 } while (cmpxchg(&pi_desc->control, old.control,
2241 new.control) != old.control);
2244 static void decache_tsc_multiplier(struct vcpu_vmx *vmx)
2246 vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
2247 vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2251 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2252 * vcpu mutex is already taken.
2254 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2256 struct vcpu_vmx *vmx = to_vmx(vcpu);
2257 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2258 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
2261 kvm_cpu_vmxon(phys_addr);
2262 else if (!already_loaded)
2263 loaded_vmcs_clear(vmx->loaded_vmcs);
2265 if (!already_loaded) {
2266 local_irq_disable();
2267 crash_disable_local_vmclear(cpu);
2270 * Read loaded_vmcs->cpu should be before fetching
2271 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2272 * See the comments in __loaded_vmcs_clear().
2276 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2277 &per_cpu(loaded_vmcss_on_cpu, cpu));
2278 crash_enable_local_vmclear(cpu);
2282 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2283 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2284 vmcs_load(vmx->loaded_vmcs->vmcs);
2287 if (!already_loaded) {
2288 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2289 unsigned long sysenter_esp;
2291 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2294 * Linux uses per-cpu TSS and GDT, so set these when switching
2297 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
2298 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
2300 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2301 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2303 vmx->loaded_vmcs->cpu = cpu;
2306 /* Setup TSC multiplier */
2307 if (kvm_has_tsc_control &&
2308 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
2309 decache_tsc_multiplier(vmx);
2311 vmx_vcpu_pi_load(vcpu, cpu);
2312 vmx->host_pkru = read_pkru();
2315 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2317 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2319 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2320 !irq_remapping_cap(IRQ_POSTING_CAP) ||
2321 !kvm_vcpu_apicv_active(vcpu))
2324 /* Set SN when the vCPU is preempted */
2325 if (vcpu->preempted)
2329 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2331 vmx_vcpu_pi_put(vcpu);
2333 __vmx_load_host_state(to_vmx(vcpu));
2334 if (!vmm_exclusive) {
2335 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
2341 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2344 * Return the cr0 value that a nested guest would read. This is a combination
2345 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2346 * its hypervisor (cr0_read_shadow).
2348 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2350 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2351 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2353 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2355 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2356 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2359 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2361 unsigned long rflags, save_rflags;
2363 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2364 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2365 rflags = vmcs_readl(GUEST_RFLAGS);
2366 if (to_vmx(vcpu)->rmode.vm86_active) {
2367 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2368 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2369 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2371 to_vmx(vcpu)->rflags = rflags;
2373 return to_vmx(vcpu)->rflags;
2376 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2378 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2379 to_vmx(vcpu)->rflags = rflags;
2380 if (to_vmx(vcpu)->rmode.vm86_active) {
2381 to_vmx(vcpu)->rmode.save_rflags = rflags;
2382 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2384 vmcs_writel(GUEST_RFLAGS, rflags);
2387 static u32 vmx_get_pkru(struct kvm_vcpu *vcpu)
2389 return to_vmx(vcpu)->guest_pkru;
2392 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2394 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2397 if (interruptibility & GUEST_INTR_STATE_STI)
2398 ret |= KVM_X86_SHADOW_INT_STI;
2399 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2400 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2405 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2407 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2408 u32 interruptibility = interruptibility_old;
2410 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2412 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2413 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2414 else if (mask & KVM_X86_SHADOW_INT_STI)
2415 interruptibility |= GUEST_INTR_STATE_STI;
2417 if ((interruptibility != interruptibility_old))
2418 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2421 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2425 rip = kvm_rip_read(vcpu);
2426 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2427 kvm_rip_write(vcpu, rip);
2429 /* skipping an emulated instruction also counts */
2430 vmx_set_interrupt_shadow(vcpu, 0);
2434 * KVM wants to inject page-faults which it got to the guest. This function
2435 * checks whether in a nested guest, we need to inject them to L1 or L2.
2437 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
2439 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2441 if (!(vmcs12->exception_bitmap & (1u << nr)))
2444 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
2445 vmcs_read32(VM_EXIT_INTR_INFO),
2446 vmcs_readl(EXIT_QUALIFICATION));
2450 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2451 bool has_error_code, u32 error_code,
2454 struct vcpu_vmx *vmx = to_vmx(vcpu);
2455 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2457 if (!reinject && is_guest_mode(vcpu) &&
2458 nested_vmx_check_exception(vcpu, nr))
2461 if (has_error_code) {
2462 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2463 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2466 if (vmx->rmode.vm86_active) {
2468 if (kvm_exception_is_soft(nr))
2469 inc_eip = vcpu->arch.event_exit_inst_len;
2470 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2471 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2475 if (kvm_exception_is_soft(nr)) {
2476 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2477 vmx->vcpu.arch.event_exit_inst_len);
2478 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2480 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2482 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2485 static bool vmx_rdtscp_supported(void)
2487 return cpu_has_vmx_rdtscp();
2490 static bool vmx_invpcid_supported(void)
2492 return cpu_has_vmx_invpcid() && enable_ept;
2496 * Swap MSR entry in host/guest MSR entry array.
2498 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2500 struct shared_msr_entry tmp;
2502 tmp = vmx->guest_msrs[to];
2503 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2504 vmx->guest_msrs[from] = tmp;
2507 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2509 unsigned long *msr_bitmap;
2511 if (is_guest_mode(vcpu))
2512 msr_bitmap = to_vmx(vcpu)->nested.msr_bitmap;
2513 else if (cpu_has_secondary_exec_ctrls() &&
2514 (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
2515 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
2516 if (enable_apicv && kvm_vcpu_apicv_active(vcpu)) {
2517 if (is_long_mode(vcpu))
2518 msr_bitmap = vmx_msr_bitmap_longmode_x2apic_apicv;
2520 msr_bitmap = vmx_msr_bitmap_legacy_x2apic_apicv;
2522 if (is_long_mode(vcpu))
2523 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2525 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2528 if (is_long_mode(vcpu))
2529 msr_bitmap = vmx_msr_bitmap_longmode;
2531 msr_bitmap = vmx_msr_bitmap_legacy;
2534 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2538 * Set up the vmcs to automatically save and restore system
2539 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2540 * mode, as fiddling with msrs is very expensive.
2542 static void setup_msrs(struct vcpu_vmx *vmx)
2544 int save_nmsrs, index;
2547 #ifdef CONFIG_X86_64
2548 if (is_long_mode(&vmx->vcpu)) {
2549 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2551 move_msr_up(vmx, index, save_nmsrs++);
2552 index = __find_msr_index(vmx, MSR_LSTAR);
2554 move_msr_up(vmx, index, save_nmsrs++);
2555 index = __find_msr_index(vmx, MSR_CSTAR);
2557 move_msr_up(vmx, index, save_nmsrs++);
2558 index = __find_msr_index(vmx, MSR_TSC_AUX);
2559 if (index >= 0 && guest_cpuid_has_rdtscp(&vmx->vcpu))
2560 move_msr_up(vmx, index, save_nmsrs++);
2562 * MSR_STAR is only needed on long mode guests, and only
2563 * if efer.sce is enabled.
2565 index = __find_msr_index(vmx, MSR_STAR);
2566 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2567 move_msr_up(vmx, index, save_nmsrs++);
2570 index = __find_msr_index(vmx, MSR_EFER);
2571 if (index >= 0 && update_transition_efer(vmx, index))
2572 move_msr_up(vmx, index, save_nmsrs++);
2574 vmx->save_nmsrs = save_nmsrs;
2576 if (cpu_has_vmx_msr_bitmap())
2577 vmx_set_msr_bitmap(&vmx->vcpu);
2581 * reads and returns guest's timestamp counter "register"
2582 * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2583 * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2585 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2587 u64 host_tsc, tsc_offset;
2590 tsc_offset = vmcs_read64(TSC_OFFSET);
2591 return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2595 * writes 'offset' into guest's timestamp counter offset register
2597 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2599 if (is_guest_mode(vcpu)) {
2601 * We're here if L1 chose not to trap WRMSR to TSC. According
2602 * to the spec, this should set L1's TSC; The offset that L1
2603 * set for L2 remains unchanged, and still needs to be added
2604 * to the newly set TSC to get L2's TSC.
2606 struct vmcs12 *vmcs12;
2607 /* recalculate vmcs02.TSC_OFFSET: */
2608 vmcs12 = get_vmcs12(vcpu);
2609 vmcs_write64(TSC_OFFSET, offset +
2610 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2611 vmcs12->tsc_offset : 0));
2613 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2614 vmcs_read64(TSC_OFFSET), offset);
2615 vmcs_write64(TSC_OFFSET, offset);
2619 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2621 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2622 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2626 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2627 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2628 * all guests if the "nested" module option is off, and can also be disabled
2629 * for a single guest by disabling its VMX cpuid bit.
2631 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2633 return nested && guest_cpuid_has_vmx(vcpu);
2637 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2638 * returned for the various VMX controls MSRs when nested VMX is enabled.
2639 * The same values should also be used to verify that vmcs12 control fields are
2640 * valid during nested entry from L1 to L2.
2641 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2642 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2643 * bit in the high half is on if the corresponding bit in the control field
2644 * may be on. See also vmx_control_verify().
2646 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2649 * Note that as a general rule, the high half of the MSRs (bits in
2650 * the control fields which may be 1) should be initialized by the
2651 * intersection of the underlying hardware's MSR (i.e., features which
2652 * can be supported) and the list of features we want to expose -
2653 * because they are known to be properly supported in our code.
2654 * Also, usually, the low half of the MSRs (bits which must be 1) can
2655 * be set to 0, meaning that L1 may turn off any of these bits. The
2656 * reason is that if one of these bits is necessary, it will appear
2657 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2658 * fields of vmcs01 and vmcs02, will turn these bits off - and
2659 * nested_vmx_exit_handled() will not pass related exits to L1.
2660 * These rules have exceptions below.
2663 /* pin-based controls */
2664 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2665 vmx->nested.nested_vmx_pinbased_ctls_low,
2666 vmx->nested.nested_vmx_pinbased_ctls_high);
2667 vmx->nested.nested_vmx_pinbased_ctls_low |=
2668 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2669 vmx->nested.nested_vmx_pinbased_ctls_high &=
2670 PIN_BASED_EXT_INTR_MASK |
2671 PIN_BASED_NMI_EXITING |
2672 PIN_BASED_VIRTUAL_NMIS;
2673 vmx->nested.nested_vmx_pinbased_ctls_high |=
2674 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2675 PIN_BASED_VMX_PREEMPTION_TIMER;
2676 if (kvm_vcpu_apicv_active(&vmx->vcpu))
2677 vmx->nested.nested_vmx_pinbased_ctls_high |=
2678 PIN_BASED_POSTED_INTR;
2681 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2682 vmx->nested.nested_vmx_exit_ctls_low,
2683 vmx->nested.nested_vmx_exit_ctls_high);
2684 vmx->nested.nested_vmx_exit_ctls_low =
2685 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2687 vmx->nested.nested_vmx_exit_ctls_high &=
2688 #ifdef CONFIG_X86_64
2689 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2691 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2692 vmx->nested.nested_vmx_exit_ctls_high |=
2693 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2694 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2695 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2697 if (kvm_mpx_supported())
2698 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2700 /* We support free control of debug control saving. */
2701 vmx->nested.nested_vmx_exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2703 /* entry controls */
2704 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2705 vmx->nested.nested_vmx_entry_ctls_low,
2706 vmx->nested.nested_vmx_entry_ctls_high);
2707 vmx->nested.nested_vmx_entry_ctls_low =
2708 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2709 vmx->nested.nested_vmx_entry_ctls_high &=
2710 #ifdef CONFIG_X86_64
2711 VM_ENTRY_IA32E_MODE |
2713 VM_ENTRY_LOAD_IA32_PAT;
2714 vmx->nested.nested_vmx_entry_ctls_high |=
2715 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2716 if (kvm_mpx_supported())
2717 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2719 /* We support free control of debug control loading. */
2720 vmx->nested.nested_vmx_entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2722 /* cpu-based controls */
2723 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2724 vmx->nested.nested_vmx_procbased_ctls_low,
2725 vmx->nested.nested_vmx_procbased_ctls_high);
2726 vmx->nested.nested_vmx_procbased_ctls_low =
2727 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2728 vmx->nested.nested_vmx_procbased_ctls_high &=
2729 CPU_BASED_VIRTUAL_INTR_PENDING |
2730 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2731 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2732 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2733 CPU_BASED_CR3_STORE_EXITING |
2734 #ifdef CONFIG_X86_64
2735 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2737 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2738 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2739 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2740 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2741 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2743 * We can allow some features even when not supported by the
2744 * hardware. For example, L1 can specify an MSR bitmap - and we
2745 * can use it to avoid exits to L1 - even when L0 runs L2
2746 * without MSR bitmaps.
2748 vmx->nested.nested_vmx_procbased_ctls_high |=
2749 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2750 CPU_BASED_USE_MSR_BITMAPS;
2752 /* We support free control of CR3 access interception. */
2753 vmx->nested.nested_vmx_procbased_ctls_low &=
2754 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2756 /* secondary cpu-based controls */
2757 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2758 vmx->nested.nested_vmx_secondary_ctls_low,
2759 vmx->nested.nested_vmx_secondary_ctls_high);
2760 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2761 vmx->nested.nested_vmx_secondary_ctls_high &=
2762 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2763 SECONDARY_EXEC_RDTSCP |
2764 SECONDARY_EXEC_DESC |
2765 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2766 SECONDARY_EXEC_ENABLE_VPID |
2767 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2768 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2769 SECONDARY_EXEC_WBINVD_EXITING |
2770 SECONDARY_EXEC_XSAVES;
2773 /* nested EPT: emulate EPT also to L1 */
2774 vmx->nested.nested_vmx_secondary_ctls_high |=
2775 SECONDARY_EXEC_ENABLE_EPT;
2776 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2777 VMX_EPTP_WB_BIT | VMX_EPT_2MB_PAGE_BIT |
2779 if (cpu_has_vmx_ept_execute_only())
2780 vmx->nested.nested_vmx_ept_caps |=
2781 VMX_EPT_EXECUTE_ONLY_BIT;
2782 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2783 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
2784 VMX_EPT_EXTENT_CONTEXT_BIT;
2786 vmx->nested.nested_vmx_ept_caps = 0;
2789 * Old versions of KVM use the single-context version without
2790 * checking for support, so declare that it is supported even
2791 * though it is treated as global context. The alternative is
2792 * not failing the single-context invvpid, and it is worse.
2795 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2796 VMX_VPID_EXTENT_SUPPORTED_MASK;
2798 vmx->nested.nested_vmx_vpid_caps = 0;
2800 if (enable_unrestricted_guest)
2801 vmx->nested.nested_vmx_secondary_ctls_high |=
2802 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2804 /* miscellaneous data */
2805 rdmsr(MSR_IA32_VMX_MISC,
2806 vmx->nested.nested_vmx_misc_low,
2807 vmx->nested.nested_vmx_misc_high);
2808 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2809 vmx->nested.nested_vmx_misc_low |=
2810 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2811 VMX_MISC_ACTIVITY_HLT;
2812 vmx->nested.nested_vmx_misc_high = 0;
2815 * This MSR reports some information about VMX support. We
2816 * should return information about the VMX we emulate for the
2817 * guest, and the VMCS structure we give it - not about the
2818 * VMX support of the underlying hardware.
2820 vmx->nested.nested_vmx_basic =
2822 VMX_BASIC_TRUE_CTLS |
2823 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2824 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2826 if (cpu_has_vmx_basic_inout())
2827 vmx->nested.nested_vmx_basic |= VMX_BASIC_INOUT;
2830 * These MSRs specify bits which the guest must keep fixed on
2831 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2832 * We picked the standard core2 setting.
2834 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2835 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2836 vmx->nested.nested_vmx_cr0_fixed0 = VMXON_CR0_ALWAYSON;
2837 vmx->nested.nested_vmx_cr4_fixed0 = VMXON_CR4_ALWAYSON;
2839 /* These MSRs specify bits which the guest must keep fixed off. */
2840 rdmsrl(MSR_IA32_VMX_CR0_FIXED1, vmx->nested.nested_vmx_cr0_fixed1);
2841 rdmsrl(MSR_IA32_VMX_CR4_FIXED1, vmx->nested.nested_vmx_cr4_fixed1);
2843 /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2844 vmx->nested.nested_vmx_vmcs_enum = 0x2e;
2848 * if fixed0[i] == 1: val[i] must be 1
2849 * if fixed1[i] == 0: val[i] must be 0
2851 static inline bool fixed_bits_valid(u64 val, u64 fixed0, u64 fixed1)
2853 return ((val & fixed1) | fixed0) == val;
2856 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2858 return fixed_bits_valid(control, low, high);
2861 static inline u64 vmx_control_msr(u32 low, u32 high)
2863 return low | ((u64)high << 32);
2866 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
2871 return (superset | subset) == superset;
2874 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
2876 const u64 feature_and_reserved =
2877 /* feature (except bit 48; see below) */
2878 BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
2880 BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
2881 u64 vmx_basic = vmx->nested.nested_vmx_basic;
2883 if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
2887 * KVM does not emulate a version of VMX that constrains physical
2888 * addresses of VMX structures (e.g. VMCS) to 32-bits.
2890 if (data & BIT_ULL(48))
2893 if (vmx_basic_vmcs_revision_id(vmx_basic) !=
2894 vmx_basic_vmcs_revision_id(data))
2897 if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
2900 vmx->nested.nested_vmx_basic = data;
2905 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
2910 switch (msr_index) {
2911 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2912 lowp = &vmx->nested.nested_vmx_pinbased_ctls_low;
2913 highp = &vmx->nested.nested_vmx_pinbased_ctls_high;
2915 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2916 lowp = &vmx->nested.nested_vmx_procbased_ctls_low;
2917 highp = &vmx->nested.nested_vmx_procbased_ctls_high;
2919 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2920 lowp = &vmx->nested.nested_vmx_exit_ctls_low;
2921 highp = &vmx->nested.nested_vmx_exit_ctls_high;
2923 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2924 lowp = &vmx->nested.nested_vmx_entry_ctls_low;
2925 highp = &vmx->nested.nested_vmx_entry_ctls_high;
2927 case MSR_IA32_VMX_PROCBASED_CTLS2:
2928 lowp = &vmx->nested.nested_vmx_secondary_ctls_low;
2929 highp = &vmx->nested.nested_vmx_secondary_ctls_high;
2935 supported = vmx_control_msr(*lowp, *highp);
2937 /* Check must-be-1 bits are still 1. */
2938 if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
2941 /* Check must-be-0 bits are still 0. */
2942 if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
2946 *highp = data >> 32;
2950 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
2952 const u64 feature_and_reserved_bits =
2954 BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
2955 BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
2957 GENMASK_ULL(13, 9) | BIT_ULL(31);
2960 vmx_misc = vmx_control_msr(vmx->nested.nested_vmx_misc_low,
2961 vmx->nested.nested_vmx_misc_high);
2963 if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
2966 if ((vmx->nested.nested_vmx_pinbased_ctls_high &
2967 PIN_BASED_VMX_PREEMPTION_TIMER) &&
2968 vmx_misc_preemption_timer_rate(data) !=
2969 vmx_misc_preemption_timer_rate(vmx_misc))
2972 if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
2975 if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
2978 if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
2981 vmx->nested.nested_vmx_misc_low = data;
2982 vmx->nested.nested_vmx_misc_high = data >> 32;
2986 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
2988 u64 vmx_ept_vpid_cap;
2990 vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.nested_vmx_ept_caps,
2991 vmx->nested.nested_vmx_vpid_caps);
2993 /* Every bit is either reserved or a feature bit. */
2994 if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
2997 vmx->nested.nested_vmx_ept_caps = data;
2998 vmx->nested.nested_vmx_vpid_caps = data >> 32;
3002 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
3006 switch (msr_index) {
3007 case MSR_IA32_VMX_CR0_FIXED0:
3008 msr = &vmx->nested.nested_vmx_cr0_fixed0;
3010 case MSR_IA32_VMX_CR4_FIXED0:
3011 msr = &vmx->nested.nested_vmx_cr4_fixed0;
3018 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
3019 * must be 1 in the restored value.
3021 if (!is_bitwise_subset(data, *msr, -1ULL))
3029 * Called when userspace is restoring VMX MSRs.
3031 * Returns 0 on success, non-0 otherwise.
3033 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
3035 struct vcpu_vmx *vmx = to_vmx(vcpu);
3037 switch (msr_index) {
3038 case MSR_IA32_VMX_BASIC:
3039 return vmx_restore_vmx_basic(vmx, data);
3040 case MSR_IA32_VMX_PINBASED_CTLS:
3041 case MSR_IA32_VMX_PROCBASED_CTLS:
3042 case MSR_IA32_VMX_EXIT_CTLS:
3043 case MSR_IA32_VMX_ENTRY_CTLS:
3045 * The "non-true" VMX capability MSRs are generated from the
3046 * "true" MSRs, so we do not support restoring them directly.
3048 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
3049 * should restore the "true" MSRs with the must-be-1 bits
3050 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
3051 * DEFAULT SETTINGS".
3054 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3055 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3056 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3057 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3058 case MSR_IA32_VMX_PROCBASED_CTLS2:
3059 return vmx_restore_control_msr(vmx, msr_index, data);
3060 case MSR_IA32_VMX_MISC:
3061 return vmx_restore_vmx_misc(vmx, data);
3062 case MSR_IA32_VMX_CR0_FIXED0:
3063 case MSR_IA32_VMX_CR4_FIXED0:
3064 return vmx_restore_fixed0_msr(vmx, msr_index, data);
3065 case MSR_IA32_VMX_CR0_FIXED1:
3066 case MSR_IA32_VMX_CR4_FIXED1:
3068 * These MSRs are generated based on the vCPU's CPUID, so we
3069 * do not support restoring them directly.
3072 case MSR_IA32_VMX_EPT_VPID_CAP:
3073 return vmx_restore_vmx_ept_vpid_cap(vmx, data);
3074 case MSR_IA32_VMX_VMCS_ENUM:
3075 vmx->nested.nested_vmx_vmcs_enum = data;
3079 * The rest of the VMX capability MSRs do not support restore.
3085 /* Returns 0 on success, non-0 otherwise. */
3086 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
3088 struct vcpu_vmx *vmx = to_vmx(vcpu);
3090 switch (msr_index) {
3091 case MSR_IA32_VMX_BASIC:
3092 *pdata = vmx->nested.nested_vmx_basic;
3094 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3095 case MSR_IA32_VMX_PINBASED_CTLS:
3096 *pdata = vmx_control_msr(
3097 vmx->nested.nested_vmx_pinbased_ctls_low,
3098 vmx->nested.nested_vmx_pinbased_ctls_high);
3099 if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
3100 *pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3102 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3103 case MSR_IA32_VMX_PROCBASED_CTLS:
3104 *pdata = vmx_control_msr(
3105 vmx->nested.nested_vmx_procbased_ctls_low,
3106 vmx->nested.nested_vmx_procbased_ctls_high);
3107 if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
3108 *pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3110 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3111 case MSR_IA32_VMX_EXIT_CTLS:
3112 *pdata = vmx_control_msr(
3113 vmx->nested.nested_vmx_exit_ctls_low,
3114 vmx->nested.nested_vmx_exit_ctls_high);
3115 if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
3116 *pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
3118 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3119 case MSR_IA32_VMX_ENTRY_CTLS:
3120 *pdata = vmx_control_msr(
3121 vmx->nested.nested_vmx_entry_ctls_low,
3122 vmx->nested.nested_vmx_entry_ctls_high);
3123 if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
3124 *pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
3126 case MSR_IA32_VMX_MISC:
3127 *pdata = vmx_control_msr(
3128 vmx->nested.nested_vmx_misc_low,
3129 vmx->nested.nested_vmx_misc_high);
3131 case MSR_IA32_VMX_CR0_FIXED0:
3132 *pdata = vmx->nested.nested_vmx_cr0_fixed0;
3134 case MSR_IA32_VMX_CR0_FIXED1:
3135 *pdata = vmx->nested.nested_vmx_cr0_fixed1;
3137 case MSR_IA32_VMX_CR4_FIXED0:
3138 *pdata = vmx->nested.nested_vmx_cr4_fixed0;
3140 case MSR_IA32_VMX_CR4_FIXED1:
3141 *pdata = vmx->nested.nested_vmx_cr4_fixed1;
3143 case MSR_IA32_VMX_VMCS_ENUM:
3144 *pdata = vmx->nested.nested_vmx_vmcs_enum;
3146 case MSR_IA32_VMX_PROCBASED_CTLS2:
3147 *pdata = vmx_control_msr(
3148 vmx->nested.nested_vmx_secondary_ctls_low,
3149 vmx->nested.nested_vmx_secondary_ctls_high);
3151 case MSR_IA32_VMX_EPT_VPID_CAP:
3152 *pdata = vmx->nested.nested_vmx_ept_caps |
3153 ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
3162 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
3165 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
3167 return !(val & ~valid_bits);
3171 * Reads an msr value (of 'msr_index') into 'pdata'.
3172 * Returns 0 on success, non-0 otherwise.
3173 * Assumes vcpu_load() was already called.
3175 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3177 struct shared_msr_entry *msr;
3179 switch (msr_info->index) {
3180 #ifdef CONFIG_X86_64
3182 msr_info->data = vmcs_readl(GUEST_FS_BASE);
3185 msr_info->data = vmcs_readl(GUEST_GS_BASE);
3187 case MSR_KERNEL_GS_BASE:
3188 vmx_load_host_state(to_vmx(vcpu));
3189 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
3193 return kvm_get_msr_common(vcpu, msr_info);
3195 msr_info->data = guest_read_tsc(vcpu);
3197 case MSR_IA32_SYSENTER_CS:
3198 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
3200 case MSR_IA32_SYSENTER_EIP:
3201 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
3203 case MSR_IA32_SYSENTER_ESP:
3204 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
3206 case MSR_IA32_BNDCFGS:
3207 if (!kvm_mpx_supported())
3209 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
3211 case MSR_IA32_MCG_EXT_CTL:
3212 if (!msr_info->host_initiated &&
3213 !(to_vmx(vcpu)->msr_ia32_feature_control &
3214 FEATURE_CONTROL_LMCE))
3216 msr_info->data = vcpu->arch.mcg_ext_ctl;
3218 case MSR_IA32_FEATURE_CONTROL:
3219 msr_info->data = to_vmx(vcpu)->msr_ia32_feature_control;
3221 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3222 if (!nested_vmx_allowed(vcpu))
3224 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
3226 if (!vmx_xsaves_supported())
3228 msr_info->data = vcpu->arch.ia32_xss;
3231 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3233 /* Otherwise falls through */
3235 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
3237 msr_info->data = msr->data;
3240 return kvm_get_msr_common(vcpu, msr_info);
3246 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3249 * Writes msr value into into the appropriate "register".
3250 * Returns 0 on success, non-0 otherwise.
3251 * Assumes vcpu_load() was already called.
3253 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3255 struct vcpu_vmx *vmx = to_vmx(vcpu);
3256 struct shared_msr_entry *msr;
3258 u32 msr_index = msr_info->index;
3259 u64 data = msr_info->data;
3261 switch (msr_index) {
3263 ret = kvm_set_msr_common(vcpu, msr_info);
3265 #ifdef CONFIG_X86_64
3267 vmx_segment_cache_clear(vmx);
3268 vmcs_writel(GUEST_FS_BASE, data);
3271 vmx_segment_cache_clear(vmx);
3272 vmcs_writel(GUEST_GS_BASE, data);
3274 case MSR_KERNEL_GS_BASE:
3275 vmx_load_host_state(vmx);
3276 vmx->msr_guest_kernel_gs_base = data;
3279 case MSR_IA32_SYSENTER_CS:
3280 vmcs_write32(GUEST_SYSENTER_CS, data);
3282 case MSR_IA32_SYSENTER_EIP:
3283 vmcs_writel(GUEST_SYSENTER_EIP, data);
3285 case MSR_IA32_SYSENTER_ESP:
3286 vmcs_writel(GUEST_SYSENTER_ESP, data);
3288 case MSR_IA32_BNDCFGS:
3289 if (!kvm_mpx_supported())
3291 vmcs_write64(GUEST_BNDCFGS, data);
3294 kvm_write_tsc(vcpu, msr_info);
3296 case MSR_IA32_CR_PAT:
3297 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3298 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3300 vmcs_write64(GUEST_IA32_PAT, data);
3301 vcpu->arch.pat = data;
3304 ret = kvm_set_msr_common(vcpu, msr_info);
3306 case MSR_IA32_TSC_ADJUST:
3307 ret = kvm_set_msr_common(vcpu, msr_info);
3309 case MSR_IA32_MCG_EXT_CTL:
3310 if ((!msr_info->host_initiated &&
3311 !(to_vmx(vcpu)->msr_ia32_feature_control &
3312 FEATURE_CONTROL_LMCE)) ||
3313 (data & ~MCG_EXT_CTL_LMCE_EN))
3315 vcpu->arch.mcg_ext_ctl = data;
3317 case MSR_IA32_FEATURE_CONTROL:
3318 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3319 (to_vmx(vcpu)->msr_ia32_feature_control &
3320 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3322 vmx->msr_ia32_feature_control = data;
3323 if (msr_info->host_initiated && data == 0)
3324 vmx_leave_nested(vcpu);
3326 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3327 if (!msr_info->host_initiated)
3328 return 1; /* they are read-only */
3329 if (!nested_vmx_allowed(vcpu))
3331 return vmx_set_vmx_msr(vcpu, msr_index, data);
3333 if (!vmx_xsaves_supported())
3336 * The only supported bit as of Skylake is bit 8, but
3337 * it is not supported on KVM.
3341 vcpu->arch.ia32_xss = data;
3342 if (vcpu->arch.ia32_xss != host_xss)
3343 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3344 vcpu->arch.ia32_xss, host_xss);
3346 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3349 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
3351 /* Check reserved bit, higher 32 bits should be zero */
3352 if ((data >> 32) != 0)
3354 /* Otherwise falls through */
3356 msr = find_msr_entry(vmx, msr_index);
3358 u64 old_msr_data = msr->data;
3360 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3362 ret = kvm_set_shared_msr(msr->index, msr->data,
3366 msr->data = old_msr_data;
3370 ret = kvm_set_msr_common(vcpu, msr_info);
3376 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3378 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3381 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3384 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3386 case VCPU_EXREG_PDPTR:
3388 ept_save_pdptrs(vcpu);
3395 static __init int cpu_has_kvm_support(void)
3397 return cpu_has_vmx();
3400 static __init int vmx_disabled_by_bios(void)
3404 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3405 if (msr & FEATURE_CONTROL_LOCKED) {
3406 /* launched w/ TXT and VMX disabled */
3407 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3410 /* launched w/o TXT and VMX only enabled w/ TXT */
3411 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3412 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3413 && !tboot_enabled()) {
3414 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3415 "activate TXT before enabling KVM\n");
3418 /* launched w/o TXT and VMX disabled */
3419 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3420 && !tboot_enabled())
3427 static void kvm_cpu_vmxon(u64 addr)
3429 intel_pt_handle_vmx(1);
3431 asm volatile (ASM_VMX_VMXON_RAX
3432 : : "a"(&addr), "m"(addr)
3436 static int hardware_enable(void)
3438 int cpu = raw_smp_processor_id();
3439 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3442 if (cr4_read_shadow() & X86_CR4_VMXE)
3445 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3446 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3447 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3450 * Now we can enable the vmclear operation in kdump
3451 * since the loaded_vmcss_on_cpu list on this cpu
3452 * has been initialized.
3454 * Though the cpu is not in VMX operation now, there
3455 * is no problem to enable the vmclear operation
3456 * for the loaded_vmcss_on_cpu list is empty!
3458 crash_enable_local_vmclear(cpu);
3460 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3462 test_bits = FEATURE_CONTROL_LOCKED;
3463 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3464 if (tboot_enabled())
3465 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3467 if ((old & test_bits) != test_bits) {
3468 /* enable and lock */
3469 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3471 cr4_set_bits(X86_CR4_VMXE);
3473 if (vmm_exclusive) {
3474 kvm_cpu_vmxon(phys_addr);
3478 native_store_gdt(this_cpu_ptr(&host_gdt));
3483 static void vmclear_local_loaded_vmcss(void)
3485 int cpu = raw_smp_processor_id();
3486 struct loaded_vmcs *v, *n;
3488 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3489 loaded_vmcss_on_cpu_link)
3490 __loaded_vmcs_clear(v);
3494 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3497 static void kvm_cpu_vmxoff(void)
3499 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3501 intel_pt_handle_vmx(0);
3504 static void hardware_disable(void)
3506 if (vmm_exclusive) {
3507 vmclear_local_loaded_vmcss();
3510 cr4_clear_bits(X86_CR4_VMXE);
3513 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3514 u32 msr, u32 *result)
3516 u32 vmx_msr_low, vmx_msr_high;
3517 u32 ctl = ctl_min | ctl_opt;
3519 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3521 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3522 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3524 /* Ensure minimum (required) set of control bits are supported. */
3532 static __init bool allow_1_setting(u32 msr, u32 ctl)
3534 u32 vmx_msr_low, vmx_msr_high;
3536 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3537 return vmx_msr_high & ctl;
3540 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3542 u32 vmx_msr_low, vmx_msr_high;
3543 u32 min, opt, min2, opt2;
3544 u32 _pin_based_exec_control = 0;
3545 u32 _cpu_based_exec_control = 0;
3546 u32 _cpu_based_2nd_exec_control = 0;
3547 u32 _vmexit_control = 0;
3548 u32 _vmentry_control = 0;
3550 min = CPU_BASED_HLT_EXITING |
3551 #ifdef CONFIG_X86_64
3552 CPU_BASED_CR8_LOAD_EXITING |
3553 CPU_BASED_CR8_STORE_EXITING |
3555 CPU_BASED_CR3_LOAD_EXITING |
3556 CPU_BASED_CR3_STORE_EXITING |
3557 CPU_BASED_USE_IO_BITMAPS |
3558 CPU_BASED_MOV_DR_EXITING |
3559 CPU_BASED_USE_TSC_OFFSETING |
3560 CPU_BASED_MWAIT_EXITING |
3561 CPU_BASED_MONITOR_EXITING |
3562 CPU_BASED_INVLPG_EXITING |
3563 CPU_BASED_RDPMC_EXITING;
3565 opt = CPU_BASED_TPR_SHADOW |
3566 CPU_BASED_USE_MSR_BITMAPS |
3567 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3568 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3569 &_cpu_based_exec_control) < 0)
3571 #ifdef CONFIG_X86_64
3572 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3573 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3574 ~CPU_BASED_CR8_STORE_EXITING;
3576 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3578 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3579 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3580 SECONDARY_EXEC_WBINVD_EXITING |
3581 SECONDARY_EXEC_ENABLE_VPID |
3582 SECONDARY_EXEC_ENABLE_EPT |
3583 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3584 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3585 SECONDARY_EXEC_RDTSCP |
3586 SECONDARY_EXEC_ENABLE_INVPCID |
3587 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3588 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3589 SECONDARY_EXEC_SHADOW_VMCS |
3590 SECONDARY_EXEC_XSAVES |
3591 SECONDARY_EXEC_ENABLE_PML |
3592 SECONDARY_EXEC_TSC_SCALING;
3593 if (adjust_vmx_controls(min2, opt2,
3594 MSR_IA32_VMX_PROCBASED_CTLS2,
3595 &_cpu_based_2nd_exec_control) < 0)
3598 #ifndef CONFIG_X86_64
3599 if (!(_cpu_based_2nd_exec_control &
3600 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3601 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3604 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3605 _cpu_based_2nd_exec_control &= ~(
3606 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3607 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3608 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3610 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3611 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3613 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3614 CPU_BASED_CR3_STORE_EXITING |
3615 CPU_BASED_INVLPG_EXITING);
3616 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3617 vmx_capability.ept, vmx_capability.vpid);
3620 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
3621 #ifdef CONFIG_X86_64
3622 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3624 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3625 VM_EXIT_CLEAR_BNDCFGS;
3626 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3627 &_vmexit_control) < 0)
3630 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
3631 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR |
3632 PIN_BASED_VMX_PREEMPTION_TIMER;
3633 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3634 &_pin_based_exec_control) < 0)
3637 if (cpu_has_broken_vmx_preemption_timer())
3638 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3639 if (!(_cpu_based_2nd_exec_control &
3640 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
3641 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3643 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3644 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3645 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3646 &_vmentry_control) < 0)
3649 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3651 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3652 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3655 #ifdef CONFIG_X86_64
3656 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3657 if (vmx_msr_high & (1u<<16))
3661 /* Require Write-Back (WB) memory type for VMCS accesses. */
3662 if (((vmx_msr_high >> 18) & 15) != 6)
3665 vmcs_conf->size = vmx_msr_high & 0x1fff;
3666 vmcs_conf->order = get_order(vmcs_conf->size);
3667 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
3668 vmcs_conf->revision_id = vmx_msr_low;
3670 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3671 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3672 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3673 vmcs_conf->vmexit_ctrl = _vmexit_control;
3674 vmcs_conf->vmentry_ctrl = _vmentry_control;
3676 cpu_has_load_ia32_efer =
3677 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3678 VM_ENTRY_LOAD_IA32_EFER)
3679 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3680 VM_EXIT_LOAD_IA32_EFER);
3682 cpu_has_load_perf_global_ctrl =
3683 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3684 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3685 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3686 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3689 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3690 * but due to errata below it can't be used. Workaround is to use
3691 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3693 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3698 * BC86,AAY89,BD102 (model 44)
3702 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3703 switch (boot_cpu_data.x86_model) {
3709 cpu_has_load_perf_global_ctrl = false;
3710 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3711 "does not work properly. Using workaround\n");
3718 if (boot_cpu_has(X86_FEATURE_XSAVES))
3719 rdmsrl(MSR_IA32_XSS, host_xss);
3724 static struct vmcs *alloc_vmcs_cpu(int cpu)
3726 int node = cpu_to_node(cpu);
3730 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3733 vmcs = page_address(pages);
3734 memset(vmcs, 0, vmcs_config.size);
3735 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3739 static struct vmcs *alloc_vmcs(void)
3741 return alloc_vmcs_cpu(raw_smp_processor_id());
3744 static void free_vmcs(struct vmcs *vmcs)
3746 free_pages((unsigned long)vmcs, vmcs_config.order);
3750 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3752 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3754 if (!loaded_vmcs->vmcs)
3756 loaded_vmcs_clear(loaded_vmcs);
3757 free_vmcs(loaded_vmcs->vmcs);
3758 loaded_vmcs->vmcs = NULL;
3759 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
3762 static void free_kvm_area(void)
3766 for_each_possible_cpu(cpu) {
3767 free_vmcs(per_cpu(vmxarea, cpu));
3768 per_cpu(vmxarea, cpu) = NULL;
3772 static void init_vmcs_shadow_fields(void)
3776 /* No checks for read only fields yet */
3778 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3779 switch (shadow_read_write_fields[i]) {
3781 if (!kvm_mpx_supported())
3789 shadow_read_write_fields[j] =
3790 shadow_read_write_fields[i];
3793 max_shadow_read_write_fields = j;
3795 /* shadowed fields guest access without vmexit */
3796 for (i = 0; i < max_shadow_read_write_fields; i++) {
3797 clear_bit(shadow_read_write_fields[i],
3798 vmx_vmwrite_bitmap);
3799 clear_bit(shadow_read_write_fields[i],
3802 for (i = 0; i < max_shadow_read_only_fields; i++)
3803 clear_bit(shadow_read_only_fields[i],
3807 static __init int alloc_kvm_area(void)
3811 for_each_possible_cpu(cpu) {
3814 vmcs = alloc_vmcs_cpu(cpu);
3820 per_cpu(vmxarea, cpu) = vmcs;
3825 static bool emulation_required(struct kvm_vcpu *vcpu)
3827 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3830 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3831 struct kvm_segment *save)
3833 if (!emulate_invalid_guest_state) {
3835 * CS and SS RPL should be equal during guest entry according
3836 * to VMX spec, but in reality it is not always so. Since vcpu
3837 * is in the middle of the transition from real mode to
3838 * protected mode it is safe to assume that RPL 0 is a good
3841 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3842 save->selector &= ~SEGMENT_RPL_MASK;
3843 save->dpl = save->selector & SEGMENT_RPL_MASK;
3846 vmx_set_segment(vcpu, save, seg);
3849 static void enter_pmode(struct kvm_vcpu *vcpu)
3851 unsigned long flags;
3852 struct vcpu_vmx *vmx = to_vmx(vcpu);
3855 * Update real mode segment cache. It may be not up-to-date if sement
3856 * register was written while vcpu was in a guest mode.
3858 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3859 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3860 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3861 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3862 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3863 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3865 vmx->rmode.vm86_active = 0;
3867 vmx_segment_cache_clear(vmx);
3869 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3871 flags = vmcs_readl(GUEST_RFLAGS);
3872 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3873 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3874 vmcs_writel(GUEST_RFLAGS, flags);
3876 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3877 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3879 update_exception_bitmap(vcpu);
3881 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3882 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3883 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3884 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3885 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3886 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3889 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3891 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3892 struct kvm_segment var = *save;
3895 if (seg == VCPU_SREG_CS)
3898 if (!emulate_invalid_guest_state) {
3899 var.selector = var.base >> 4;
3900 var.base = var.base & 0xffff0;
3910 if (save->base & 0xf)
3911 printk_once(KERN_WARNING "kvm: segment base is not "
3912 "paragraph aligned when entering "
3913 "protected mode (seg=%d)", seg);
3916 vmcs_write16(sf->selector, var.selector);
3917 vmcs_write32(sf->base, var.base);
3918 vmcs_write32(sf->limit, var.limit);
3919 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3922 static void enter_rmode(struct kvm_vcpu *vcpu)
3924 unsigned long flags;
3925 struct vcpu_vmx *vmx = to_vmx(vcpu);
3927 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3928 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3929 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3930 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3931 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3932 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3933 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3935 vmx->rmode.vm86_active = 1;
3938 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3939 * vcpu. Warn the user that an update is overdue.
3941 if (!vcpu->kvm->arch.tss_addr)
3942 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3943 "called before entering vcpu\n");
3945 vmx_segment_cache_clear(vmx);
3947 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3948 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3949 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3951 flags = vmcs_readl(GUEST_RFLAGS);
3952 vmx->rmode.save_rflags = flags;
3954 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3956 vmcs_writel(GUEST_RFLAGS, flags);
3957 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3958 update_exception_bitmap(vcpu);
3960 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3961 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3962 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3963 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3964 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3965 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3967 kvm_mmu_reset_context(vcpu);
3970 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3972 struct vcpu_vmx *vmx = to_vmx(vcpu);
3973 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3979 * Force kernel_gs_base reloading before EFER changes, as control
3980 * of this msr depends on is_long_mode().
3982 vmx_load_host_state(to_vmx(vcpu));
3983 vcpu->arch.efer = efer;
3984 if (efer & EFER_LMA) {
3985 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3988 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3990 msr->data = efer & ~EFER_LME;
3995 #ifdef CONFIG_X86_64
3997 static void enter_lmode(struct kvm_vcpu *vcpu)
4001 vmx_segment_cache_clear(to_vmx(vcpu));
4003 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
4004 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
4005 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
4007 vmcs_write32(GUEST_TR_AR_BYTES,
4008 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
4009 | VMX_AR_TYPE_BUSY_64_TSS);
4011 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
4014 static void exit_lmode(struct kvm_vcpu *vcpu)
4016 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4017 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
4022 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
4024 vpid_sync_context(vpid);
4026 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
4028 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
4032 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
4034 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
4037 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
4039 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
4041 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
4042 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
4045 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
4047 if (enable_ept && is_paging(vcpu))
4048 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4049 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
4052 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
4054 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
4056 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
4057 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
4060 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
4062 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4064 if (!test_bit(VCPU_EXREG_PDPTR,
4065 (unsigned long *)&vcpu->arch.regs_dirty))
4068 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4069 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
4070 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
4071 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
4072 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
4076 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
4078 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4080 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4081 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
4082 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
4083 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
4084 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
4087 __set_bit(VCPU_EXREG_PDPTR,
4088 (unsigned long *)&vcpu->arch.regs_avail);
4089 __set_bit(VCPU_EXREG_PDPTR,
4090 (unsigned long *)&vcpu->arch.regs_dirty);
4093 static bool nested_guest_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4095 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4096 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4097 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4099 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
4100 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4101 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4102 fixed0 &= ~(X86_CR0_PE | X86_CR0_PG);
4104 return fixed_bits_valid(val, fixed0, fixed1);
4107 static bool nested_host_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4109 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4110 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4112 return fixed_bits_valid(val, fixed0, fixed1);
4115 static bool nested_cr4_valid(struct kvm_vcpu *vcpu, unsigned long val)
4117 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed0;
4118 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed1;
4120 return fixed_bits_valid(val, fixed0, fixed1);
4123 /* No difference in the restrictions on guest and host CR4 in VMX operation. */
4124 #define nested_guest_cr4_valid nested_cr4_valid
4125 #define nested_host_cr4_valid nested_cr4_valid
4127 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
4129 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
4131 struct kvm_vcpu *vcpu)
4133 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
4134 vmx_decache_cr3(vcpu);
4135 if (!(cr0 & X86_CR0_PG)) {
4136 /* From paging/starting to nonpaging */
4137 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4138 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
4139 (CPU_BASED_CR3_LOAD_EXITING |
4140 CPU_BASED_CR3_STORE_EXITING));
4141 vcpu->arch.cr0 = cr0;
4142 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4143 } else if (!is_paging(vcpu)) {
4144 /* From nonpaging to paging */
4145 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4146 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
4147 ~(CPU_BASED_CR3_LOAD_EXITING |
4148 CPU_BASED_CR3_STORE_EXITING));
4149 vcpu->arch.cr0 = cr0;
4150 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4153 if (!(cr0 & X86_CR0_WP))
4154 *hw_cr0 &= ~X86_CR0_WP;
4157 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
4159 struct vcpu_vmx *vmx = to_vmx(vcpu);
4160 unsigned long hw_cr0;
4162 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
4163 if (enable_unrestricted_guest)
4164 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
4166 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
4168 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
4171 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
4175 #ifdef CONFIG_X86_64
4176 if (vcpu->arch.efer & EFER_LME) {
4177 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
4179 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
4185 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
4187 vmcs_writel(CR0_READ_SHADOW, cr0);
4188 vmcs_writel(GUEST_CR0, hw_cr0);
4189 vcpu->arch.cr0 = cr0;
4191 /* depends on vcpu->arch.cr0 to be set to a new value */
4192 vmx->emulation_required = emulation_required(vcpu);
4195 static u64 construct_eptp(unsigned long root_hpa)
4199 /* TODO write the value reading from MSR */
4200 eptp = VMX_EPT_DEFAULT_MT |
4201 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
4202 if (enable_ept_ad_bits)
4203 eptp |= VMX_EPT_AD_ENABLE_BIT;
4204 eptp |= (root_hpa & PAGE_MASK);
4209 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
4211 unsigned long guest_cr3;
4216 eptp = construct_eptp(cr3);
4217 vmcs_write64(EPT_POINTER, eptp);
4218 if (is_paging(vcpu) || is_guest_mode(vcpu))
4219 guest_cr3 = kvm_read_cr3(vcpu);
4221 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
4222 ept_load_pdptrs(vcpu);
4225 vmx_flush_tlb(vcpu);
4226 vmcs_writel(GUEST_CR3, guest_cr3);
4229 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
4232 * Pass through host's Machine Check Enable value to hw_cr4, which
4233 * is in force while we are in guest mode. Do not let guests control
4234 * this bit, even if host CR4.MCE == 0.
4236 unsigned long hw_cr4 =
4237 (cr4_read_shadow() & X86_CR4_MCE) |
4238 (cr4 & ~X86_CR4_MCE) |
4239 (to_vmx(vcpu)->rmode.vm86_active ?
4240 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
4242 if (cr4 & X86_CR4_VMXE) {
4244 * To use VMXON (and later other VMX instructions), a guest
4245 * must first be able to turn on cr4.VMXE (see handle_vmon()).
4246 * So basically the check on whether to allow nested VMX
4249 if (!nested_vmx_allowed(vcpu))
4253 if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
4256 vcpu->arch.cr4 = cr4;
4258 if (!is_paging(vcpu)) {
4259 hw_cr4 &= ~X86_CR4_PAE;
4260 hw_cr4 |= X86_CR4_PSE;
4261 } else if (!(cr4 & X86_CR4_PAE)) {
4262 hw_cr4 &= ~X86_CR4_PAE;
4266 if (!enable_unrestricted_guest && !is_paging(vcpu))
4268 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4269 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
4270 * to be manually disabled when guest switches to non-paging
4273 * If !enable_unrestricted_guest, the CPU is always running
4274 * with CR0.PG=1 and CR4 needs to be modified.
4275 * If enable_unrestricted_guest, the CPU automatically
4276 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4278 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
4280 vmcs_writel(CR4_READ_SHADOW, cr4);
4281 vmcs_writel(GUEST_CR4, hw_cr4);
4285 static void vmx_get_segment(struct kvm_vcpu *vcpu,
4286 struct kvm_segment *var, int seg)
4288 struct vcpu_vmx *vmx = to_vmx(vcpu);
4291 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4292 *var = vmx->rmode.segs[seg];
4293 if (seg == VCPU_SREG_TR
4294 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
4296 var->base = vmx_read_guest_seg_base(vmx, seg);
4297 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4300 var->base = vmx_read_guest_seg_base(vmx, seg);
4301 var->limit = vmx_read_guest_seg_limit(vmx, seg);
4302 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4303 ar = vmx_read_guest_seg_ar(vmx, seg);
4304 var->unusable = (ar >> 16) & 1;
4305 var->type = ar & 15;
4306 var->s = (ar >> 4) & 1;
4307 var->dpl = (ar >> 5) & 3;
4309 * Some userspaces do not preserve unusable property. Since usable
4310 * segment has to be present according to VMX spec we can use present
4311 * property to amend userspace bug by making unusable segment always
4312 * nonpresent. vmx_segment_access_rights() already marks nonpresent
4313 * segment as unusable.
4315 var->present = !var->unusable;
4316 var->avl = (ar >> 12) & 1;
4317 var->l = (ar >> 13) & 1;
4318 var->db = (ar >> 14) & 1;
4319 var->g = (ar >> 15) & 1;
4322 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
4324 struct kvm_segment s;
4326 if (to_vmx(vcpu)->rmode.vm86_active) {
4327 vmx_get_segment(vcpu, &s, seg);
4330 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
4333 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
4335 struct vcpu_vmx *vmx = to_vmx(vcpu);
4337 if (unlikely(vmx->rmode.vm86_active))
4340 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
4341 return VMX_AR_DPL(ar);
4345 static u32 vmx_segment_access_rights(struct kvm_segment *var)
4349 if (var->unusable || !var->present)
4352 ar = var->type & 15;
4353 ar |= (var->s & 1) << 4;
4354 ar |= (var->dpl & 3) << 5;
4355 ar |= (var->present & 1) << 7;
4356 ar |= (var->avl & 1) << 12;
4357 ar |= (var->l & 1) << 13;
4358 ar |= (var->db & 1) << 14;
4359 ar |= (var->g & 1) << 15;
4365 static void vmx_set_segment(struct kvm_vcpu *vcpu,
4366 struct kvm_segment *var, int seg)
4368 struct vcpu_vmx *vmx = to_vmx(vcpu);
4369 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4371 vmx_segment_cache_clear(vmx);
4373 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4374 vmx->rmode.segs[seg] = *var;
4375 if (seg == VCPU_SREG_TR)
4376 vmcs_write16(sf->selector, var->selector);
4378 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
4382 vmcs_writel(sf->base, var->base);
4383 vmcs_write32(sf->limit, var->limit);
4384 vmcs_write16(sf->selector, var->selector);
4387 * Fix the "Accessed" bit in AR field of segment registers for older
4389 * IA32 arch specifies that at the time of processor reset the
4390 * "Accessed" bit in the AR field of segment registers is 1. And qemu
4391 * is setting it to 0 in the userland code. This causes invalid guest
4392 * state vmexit when "unrestricted guest" mode is turned on.
4393 * Fix for this setup issue in cpu_reset is being pushed in the qemu
4394 * tree. Newer qemu binaries with that qemu fix would not need this
4397 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
4398 var->type |= 0x1; /* Accessed */
4400 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4403 vmx->emulation_required = emulation_required(vcpu);
4406 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4408 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4410 *db = (ar >> 14) & 1;
4411 *l = (ar >> 13) & 1;
4414 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4416 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4417 dt->address = vmcs_readl(GUEST_IDTR_BASE);
4420 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4422 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4423 vmcs_writel(GUEST_IDTR_BASE, dt->address);
4426 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4428 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4429 dt->address = vmcs_readl(GUEST_GDTR_BASE);
4432 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4434 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4435 vmcs_writel(GUEST_GDTR_BASE, dt->address);
4438 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4440 struct kvm_segment var;
4443 vmx_get_segment(vcpu, &var, seg);
4445 if (seg == VCPU_SREG_CS)
4447 ar = vmx_segment_access_rights(&var);
4449 if (var.base != (var.selector << 4))
4451 if (var.limit != 0xffff)
4459 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4461 struct kvm_segment cs;
4462 unsigned int cs_rpl;
4464 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4465 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4469 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4473 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4474 if (cs.dpl > cs_rpl)
4477 if (cs.dpl != cs_rpl)
4483 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4487 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4489 struct kvm_segment ss;
4490 unsigned int ss_rpl;
4492 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4493 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4497 if (ss.type != 3 && ss.type != 7)
4501 if (ss.dpl != ss_rpl) /* DPL != RPL */
4509 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4511 struct kvm_segment var;
4514 vmx_get_segment(vcpu, &var, seg);
4515 rpl = var.selector & SEGMENT_RPL_MASK;
4523 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4524 if (var.dpl < rpl) /* DPL < RPL */
4528 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4534 static bool tr_valid(struct kvm_vcpu *vcpu)
4536 struct kvm_segment tr;
4538 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4542 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4544 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4552 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4554 struct kvm_segment ldtr;
4556 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4560 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4570 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4572 struct kvm_segment cs, ss;
4574 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4575 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4577 return ((cs.selector & SEGMENT_RPL_MASK) ==
4578 (ss.selector & SEGMENT_RPL_MASK));
4582 * Check if guest state is valid. Returns true if valid, false if
4584 * We assume that registers are always usable
4586 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4588 if (enable_unrestricted_guest)
4591 /* real mode guest state checks */
4592 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4593 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4595 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4597 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4599 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4601 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4603 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4606 /* protected mode guest state checks */
4607 if (!cs_ss_rpl_check(vcpu))
4609 if (!code_segment_valid(vcpu))
4611 if (!stack_segment_valid(vcpu))
4613 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4615 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4617 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4619 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4621 if (!tr_valid(vcpu))
4623 if (!ldtr_valid(vcpu))
4627 * - Add checks on RIP
4628 * - Add checks on RFLAGS
4634 static int init_rmode_tss(struct kvm *kvm)
4640 idx = srcu_read_lock(&kvm->srcu);
4641 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4642 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4645 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4646 r = kvm_write_guest_page(kvm, fn++, &data,
4647 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4650 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4653 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4657 r = kvm_write_guest_page(kvm, fn, &data,
4658 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4661 srcu_read_unlock(&kvm->srcu, idx);
4665 static int init_rmode_identity_map(struct kvm *kvm)
4668 kvm_pfn_t identity_map_pfn;
4674 /* Protect kvm->arch.ept_identity_pagetable_done. */
4675 mutex_lock(&kvm->slots_lock);
4677 if (likely(kvm->arch.ept_identity_pagetable_done))
4680 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4682 r = alloc_identity_pagetable(kvm);
4686 idx = srcu_read_lock(&kvm->srcu);
4687 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4690 /* Set up identity-mapping pagetable for EPT in real mode */
4691 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4692 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4693 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4694 r = kvm_write_guest_page(kvm, identity_map_pfn,
4695 &tmp, i * sizeof(tmp), sizeof(tmp));
4699 kvm->arch.ept_identity_pagetable_done = true;
4702 srcu_read_unlock(&kvm->srcu, idx);
4705 mutex_unlock(&kvm->slots_lock);
4709 static void seg_setup(int seg)
4711 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4714 vmcs_write16(sf->selector, 0);
4715 vmcs_writel(sf->base, 0);
4716 vmcs_write32(sf->limit, 0xffff);
4718 if (seg == VCPU_SREG_CS)
4719 ar |= 0x08; /* code segment */
4721 vmcs_write32(sf->ar_bytes, ar);
4724 static int alloc_apic_access_page(struct kvm *kvm)
4729 mutex_lock(&kvm->slots_lock);
4730 if (kvm->arch.apic_access_page_done)
4732 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4733 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4737 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4738 if (is_error_page(page)) {
4744 * Do not pin the page in memory, so that memory hot-unplug
4745 * is able to migrate it.
4748 kvm->arch.apic_access_page_done = true;
4750 mutex_unlock(&kvm->slots_lock);
4754 static int alloc_identity_pagetable(struct kvm *kvm)
4756 /* Called with kvm->slots_lock held. */
4760 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4762 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4763 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4768 static int allocate_vpid(void)
4774 spin_lock(&vmx_vpid_lock);
4775 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4776 if (vpid < VMX_NR_VPIDS)
4777 __set_bit(vpid, vmx_vpid_bitmap);
4780 spin_unlock(&vmx_vpid_lock);
4784 static void free_vpid(int vpid)
4786 if (!enable_vpid || vpid == 0)
4788 spin_lock(&vmx_vpid_lock);
4789 __clear_bit(vpid, vmx_vpid_bitmap);
4790 spin_unlock(&vmx_vpid_lock);
4793 #define MSR_TYPE_R 1
4794 #define MSR_TYPE_W 2
4795 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4798 int f = sizeof(unsigned long);
4800 if (!cpu_has_vmx_msr_bitmap())
4804 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4805 * have the write-low and read-high bitmap offsets the wrong way round.
4806 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4808 if (msr <= 0x1fff) {
4809 if (type & MSR_TYPE_R)
4811 __clear_bit(msr, msr_bitmap + 0x000 / f);
4813 if (type & MSR_TYPE_W)
4815 __clear_bit(msr, msr_bitmap + 0x800 / f);
4817 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4819 if (type & MSR_TYPE_R)
4821 __clear_bit(msr, msr_bitmap + 0x400 / f);
4823 if (type & MSR_TYPE_W)
4825 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4831 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4832 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4834 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4835 unsigned long *msr_bitmap_nested,
4838 int f = sizeof(unsigned long);
4840 if (!cpu_has_vmx_msr_bitmap()) {
4846 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4847 * have the write-low and read-high bitmap offsets the wrong way round.
4848 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4850 if (msr <= 0x1fff) {
4851 if (type & MSR_TYPE_R &&
4852 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4854 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4856 if (type & MSR_TYPE_W &&
4857 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4859 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4861 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4863 if (type & MSR_TYPE_R &&
4864 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4866 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4868 if (type & MSR_TYPE_W &&
4869 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4871 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4876 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4879 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4880 msr, MSR_TYPE_R | MSR_TYPE_W);
4881 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4882 msr, MSR_TYPE_R | MSR_TYPE_W);
4885 static void vmx_disable_intercept_msr_x2apic(u32 msr, int type, bool apicv_active)
4888 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic_apicv,
4890 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic_apicv,
4893 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4895 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4900 static bool vmx_get_enable_apicv(void)
4902 return enable_apicv;
4905 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
4907 struct vcpu_vmx *vmx = to_vmx(vcpu);
4912 if (vmx->nested.pi_desc &&
4913 vmx->nested.pi_pending) {
4914 vmx->nested.pi_pending = false;
4915 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
4918 max_irr = find_last_bit(
4919 (unsigned long *)vmx->nested.pi_desc->pir, 256);
4924 vapic_page = kmap(vmx->nested.virtual_apic_page);
4925 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
4926 kunmap(vmx->nested.virtual_apic_page);
4928 status = vmcs_read16(GUEST_INTR_STATUS);
4929 if ((u8)max_irr > ((u8)status & 0xff)) {
4931 status |= (u8)max_irr;
4932 vmcs_write16(GUEST_INTR_STATUS, status);
4937 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu)
4940 if (vcpu->mode == IN_GUEST_MODE) {
4941 struct vcpu_vmx *vmx = to_vmx(vcpu);
4944 * Currently, we don't support urgent interrupt,
4945 * all interrupts are recognized as non-urgent
4946 * interrupt, so we cannot post interrupts when
4949 * If the vcpu is in guest mode, it means it is
4950 * running instead of being scheduled out and
4951 * waiting in the run queue, and that's the only
4952 * case when 'SN' is set currently, warning if
4955 WARN_ON_ONCE(pi_test_sn(&vmx->pi_desc));
4957 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4958 POSTED_INTR_VECTOR);
4965 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4968 struct vcpu_vmx *vmx = to_vmx(vcpu);
4970 if (is_guest_mode(vcpu) &&
4971 vector == vmx->nested.posted_intr_nv) {
4972 /* the PIR and ON have been set by L1. */
4973 kvm_vcpu_trigger_posted_interrupt(vcpu);
4975 * If a posted intr is not recognized by hardware,
4976 * we will accomplish it in the next vmentry.
4978 vmx->nested.pi_pending = true;
4979 kvm_make_request(KVM_REQ_EVENT, vcpu);
4985 * Send interrupt to vcpu via posted interrupt way.
4986 * 1. If target vcpu is running(non-root mode), send posted interrupt
4987 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4988 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4989 * interrupt from PIR in next vmentry.
4991 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4993 struct vcpu_vmx *vmx = to_vmx(vcpu);
4996 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
5000 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
5003 /* If a previous notification has sent the IPI, nothing to do. */
5004 if (pi_test_and_set_on(&vmx->pi_desc))
5007 if (!kvm_vcpu_trigger_posted_interrupt(vcpu))
5008 kvm_vcpu_kick(vcpu);
5012 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
5013 * will not change in the lifetime of the guest.
5014 * Note that host-state that does change is set elsewhere. E.g., host-state
5015 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
5017 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
5022 unsigned long cr0, cr4;
5025 WARN_ON(cr0 & X86_CR0_TS);
5026 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
5027 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
5029 /* Save the most likely value for this task's CR4 in the VMCS. */
5030 cr4 = cr4_read_shadow();
5031 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
5032 vmx->host_state.vmcs_host_cr4 = cr4;
5034 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
5035 #ifdef CONFIG_X86_64
5037 * Load null selectors, so we can avoid reloading them in
5038 * __vmx_load_host_state(), in case userspace uses the null selectors
5039 * too (the expected case).
5041 vmcs_write16(HOST_DS_SELECTOR, 0);
5042 vmcs_write16(HOST_ES_SELECTOR, 0);
5044 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5045 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5047 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5048 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
5050 native_store_idt(&dt);
5051 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
5052 vmx->host_idt_base = dt.address;
5054 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
5056 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
5057 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
5058 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
5059 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
5061 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
5062 rdmsr(MSR_IA32_CR_PAT, low32, high32);
5063 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
5067 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
5069 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
5071 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
5072 if (is_guest_mode(&vmx->vcpu))
5073 vmx->vcpu.arch.cr4_guest_owned_bits &=
5074 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
5075 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
5078 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
5080 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
5082 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5083 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
5084 /* Enable the preemption timer dynamically */
5085 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
5086 return pin_based_exec_ctrl;
5089 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5091 struct vcpu_vmx *vmx = to_vmx(vcpu);
5093 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5094 if (cpu_has_secondary_exec_ctrls()) {
5095 if (kvm_vcpu_apicv_active(vcpu))
5096 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
5097 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5098 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5100 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
5101 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5102 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5105 if (cpu_has_vmx_msr_bitmap())
5106 vmx_set_msr_bitmap(vcpu);
5109 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
5111 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
5113 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
5114 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5116 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
5117 exec_control &= ~CPU_BASED_TPR_SHADOW;
5118 #ifdef CONFIG_X86_64
5119 exec_control |= CPU_BASED_CR8_STORE_EXITING |
5120 CPU_BASED_CR8_LOAD_EXITING;
5124 exec_control |= CPU_BASED_CR3_STORE_EXITING |
5125 CPU_BASED_CR3_LOAD_EXITING |
5126 CPU_BASED_INVLPG_EXITING;
5127 return exec_control;
5130 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
5132 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
5133 if (!cpu_need_virtualize_apic_accesses(&vmx->vcpu))
5134 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5136 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
5138 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
5139 enable_unrestricted_guest = 0;
5140 /* Enable INVPCID for non-ept guests may cause performance regression. */
5141 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5143 if (!enable_unrestricted_guest)
5144 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
5146 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
5147 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5148 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
5149 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5150 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
5151 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
5153 We can NOT enable shadow_vmcs here because we don't have yet
5156 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
5159 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
5161 return exec_control;
5164 static void ept_set_mmio_spte_mask(void)
5167 * EPT Misconfigurations can be generated if the value of bits 2:0
5168 * of an EPT paging-structure entry is 110b (write/execute).
5170 kvm_mmu_set_mmio_spte_mask(VMX_EPT_MISCONFIG_WX_VALUE);
5173 #define VMX_XSS_EXIT_BITMAP 0
5175 * Sets up the vmcs for emulated real mode.
5177 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
5179 #ifdef CONFIG_X86_64
5185 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
5186 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
5188 if (enable_shadow_vmcs) {
5189 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5190 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5192 if (cpu_has_vmx_msr_bitmap())
5193 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
5195 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
5198 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5199 vmx->hv_deadline_tsc = -1;
5201 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
5203 if (cpu_has_secondary_exec_ctrls()) {
5204 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5205 vmx_secondary_exec_control(vmx));
5208 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
5209 vmcs_write64(EOI_EXIT_BITMAP0, 0);
5210 vmcs_write64(EOI_EXIT_BITMAP1, 0);
5211 vmcs_write64(EOI_EXIT_BITMAP2, 0);
5212 vmcs_write64(EOI_EXIT_BITMAP3, 0);
5214 vmcs_write16(GUEST_INTR_STATUS, 0);
5216 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
5217 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
5221 vmcs_write32(PLE_GAP, ple_gap);
5222 vmx->ple_window = ple_window;
5223 vmx->ple_window_dirty = true;
5226 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
5227 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
5228 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
5230 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
5231 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
5232 vmx_set_constant_host_state(vmx);
5233 #ifdef CONFIG_X86_64
5234 rdmsrl(MSR_FS_BASE, a);
5235 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
5236 rdmsrl(MSR_GS_BASE, a);
5237 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
5239 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
5240 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
5243 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
5244 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
5245 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
5246 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
5247 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
5249 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
5250 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
5252 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
5253 u32 index = vmx_msr_index[i];
5254 u32 data_low, data_high;
5257 if (rdmsr_safe(index, &data_low, &data_high) < 0)
5259 if (wrmsr_safe(index, data_low, data_high) < 0)
5261 vmx->guest_msrs[j].index = i;
5262 vmx->guest_msrs[j].data = 0;
5263 vmx->guest_msrs[j].mask = -1ull;
5268 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
5270 /* 22.2.1, 20.8.1 */
5271 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
5273 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
5274 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
5276 set_cr4_guest_host_mask(vmx);
5278 if (vmx_xsaves_supported())
5279 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
5282 ASSERT(vmx->pml_pg);
5283 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
5284 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5290 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
5292 struct vcpu_vmx *vmx = to_vmx(vcpu);
5293 struct msr_data apic_base_msr;
5296 vmx->rmode.vm86_active = 0;
5298 vmx->soft_vnmi_blocked = 0;
5300 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
5301 kvm_set_cr8(vcpu, 0);
5304 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
5305 MSR_IA32_APICBASE_ENABLE;
5306 if (kvm_vcpu_is_reset_bsp(vcpu))
5307 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
5308 apic_base_msr.host_initiated = true;
5309 kvm_set_apic_base(vcpu, &apic_base_msr);
5312 vmx_segment_cache_clear(vmx);
5314 seg_setup(VCPU_SREG_CS);
5315 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
5316 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
5318 seg_setup(VCPU_SREG_DS);
5319 seg_setup(VCPU_SREG_ES);
5320 seg_setup(VCPU_SREG_FS);
5321 seg_setup(VCPU_SREG_GS);
5322 seg_setup(VCPU_SREG_SS);
5324 vmcs_write16(GUEST_TR_SELECTOR, 0);
5325 vmcs_writel(GUEST_TR_BASE, 0);
5326 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
5327 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
5329 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
5330 vmcs_writel(GUEST_LDTR_BASE, 0);
5331 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
5332 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
5335 vmcs_write32(GUEST_SYSENTER_CS, 0);
5336 vmcs_writel(GUEST_SYSENTER_ESP, 0);
5337 vmcs_writel(GUEST_SYSENTER_EIP, 0);
5338 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
5341 vmcs_writel(GUEST_RFLAGS, 0x02);
5342 kvm_rip_write(vcpu, 0xfff0);
5344 vmcs_writel(GUEST_GDTR_BASE, 0);
5345 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
5347 vmcs_writel(GUEST_IDTR_BASE, 0);
5348 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
5350 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
5351 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
5352 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
5356 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
5358 if (cpu_has_vmx_tpr_shadow() && !init_event) {
5359 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
5360 if (cpu_need_tpr_shadow(vcpu))
5361 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
5362 __pa(vcpu->arch.apic->regs));
5363 vmcs_write32(TPR_THRESHOLD, 0);
5366 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5368 if (kvm_vcpu_apicv_active(vcpu))
5369 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
5372 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5374 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5375 vmx->vcpu.arch.cr0 = cr0;
5376 vmx_set_cr0(vcpu, cr0); /* enter rmode */
5377 vmx_set_cr4(vcpu, 0);
5378 vmx_set_efer(vcpu, 0);
5380 update_exception_bitmap(vcpu);
5382 vpid_sync_context(vmx->vpid);
5386 * In nested virtualization, check if L1 asked to exit on external interrupts.
5387 * For most existing hypervisors, this will always return true.
5389 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5391 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5392 PIN_BASED_EXT_INTR_MASK;
5396 * In nested virtualization, check if L1 has set
5397 * VM_EXIT_ACK_INTR_ON_EXIT
5399 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5401 return get_vmcs12(vcpu)->vm_exit_controls &
5402 VM_EXIT_ACK_INTR_ON_EXIT;
5405 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5407 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5408 PIN_BASED_NMI_EXITING;
5411 static void enable_irq_window(struct kvm_vcpu *vcpu)
5413 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5414 CPU_BASED_VIRTUAL_INTR_PENDING);
5417 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5419 if (!cpu_has_virtual_nmis() ||
5420 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5421 enable_irq_window(vcpu);
5425 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5426 CPU_BASED_VIRTUAL_NMI_PENDING);
5429 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5431 struct vcpu_vmx *vmx = to_vmx(vcpu);
5433 int irq = vcpu->arch.interrupt.nr;
5435 trace_kvm_inj_virq(irq);
5437 ++vcpu->stat.irq_injections;
5438 if (vmx->rmode.vm86_active) {
5440 if (vcpu->arch.interrupt.soft)
5441 inc_eip = vcpu->arch.event_exit_inst_len;
5442 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5443 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5446 intr = irq | INTR_INFO_VALID_MASK;
5447 if (vcpu->arch.interrupt.soft) {
5448 intr |= INTR_TYPE_SOFT_INTR;
5449 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5450 vmx->vcpu.arch.event_exit_inst_len);
5452 intr |= INTR_TYPE_EXT_INTR;
5453 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5456 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5458 struct vcpu_vmx *vmx = to_vmx(vcpu);
5460 if (!is_guest_mode(vcpu)) {
5461 if (!cpu_has_virtual_nmis()) {
5463 * Tracking the NMI-blocked state in software is built upon
5464 * finding the next open IRQ window. This, in turn, depends on
5465 * well-behaving guests: They have to keep IRQs disabled at
5466 * least as long as the NMI handler runs. Otherwise we may
5467 * cause NMI nesting, maybe breaking the guest. But as this is
5468 * highly unlikely, we can live with the residual risk.
5470 vmx->soft_vnmi_blocked = 1;
5471 vmx->vnmi_blocked_time = 0;
5474 ++vcpu->stat.nmi_injections;
5475 vmx->nmi_known_unmasked = false;
5478 if (vmx->rmode.vm86_active) {
5479 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5480 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5484 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5485 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5488 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5490 if (!cpu_has_virtual_nmis())
5491 return to_vmx(vcpu)->soft_vnmi_blocked;
5492 if (to_vmx(vcpu)->nmi_known_unmasked)
5494 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5497 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5499 struct vcpu_vmx *vmx = to_vmx(vcpu);
5501 if (!cpu_has_virtual_nmis()) {
5502 if (vmx->soft_vnmi_blocked != masked) {
5503 vmx->soft_vnmi_blocked = masked;
5504 vmx->vnmi_blocked_time = 0;
5507 vmx->nmi_known_unmasked = !masked;
5509 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5510 GUEST_INTR_STATE_NMI);
5512 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5513 GUEST_INTR_STATE_NMI);
5517 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5519 if (to_vmx(vcpu)->nested.nested_run_pending)
5522 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
5525 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5526 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5527 | GUEST_INTR_STATE_NMI));
5530 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5532 return (!to_vmx(vcpu)->nested.nested_run_pending &&
5533 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5534 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5535 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5538 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5542 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5546 kvm->arch.tss_addr = addr;
5547 return init_rmode_tss(kvm);
5550 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5555 * Update instruction length as we may reinject the exception
5556 * from user space while in guest debugging mode.
5558 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5559 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5560 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5564 if (vcpu->guest_debug &
5565 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5582 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5583 int vec, u32 err_code)
5586 * Instruction with address size override prefix opcode 0x67
5587 * Cause the #SS fault with 0 error code in VM86 mode.
5589 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5590 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5591 if (vcpu->arch.halt_request) {
5592 vcpu->arch.halt_request = 0;
5593 return kvm_vcpu_halt(vcpu);
5601 * Forward all other exceptions that are valid in real mode.
5602 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5603 * the required debugging infrastructure rework.
5605 kvm_queue_exception(vcpu, vec);
5610 * Trigger machine check on the host. We assume all the MSRs are already set up
5611 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5612 * We pass a fake environment to the machine check handler because we want
5613 * the guest to be always treated like user space, no matter what context
5614 * it used internally.
5616 static void kvm_machine_check(void)
5618 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5619 struct pt_regs regs = {
5620 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5621 .flags = X86_EFLAGS_IF,
5624 do_machine_check(®s, 0);
5628 static int handle_machine_check(struct kvm_vcpu *vcpu)
5630 /* already handled by vcpu_run */
5634 static int handle_exception(struct kvm_vcpu *vcpu)
5636 struct vcpu_vmx *vmx = to_vmx(vcpu);
5637 struct kvm_run *kvm_run = vcpu->run;
5638 u32 intr_info, ex_no, error_code;
5639 unsigned long cr2, rip, dr6;
5641 enum emulation_result er;
5643 vect_info = vmx->idt_vectoring_info;
5644 intr_info = vmx->exit_intr_info;
5646 if (is_machine_check(intr_info))
5647 return handle_machine_check(vcpu);
5649 if (is_nmi(intr_info))
5650 return 1; /* already handled by vmx_vcpu_run() */
5652 if (is_invalid_opcode(intr_info)) {
5653 if (is_guest_mode(vcpu)) {
5654 kvm_queue_exception(vcpu, UD_VECTOR);
5657 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5658 if (er != EMULATE_DONE)
5659 kvm_queue_exception(vcpu, UD_VECTOR);
5664 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5665 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5668 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5669 * MMIO, it is better to report an internal error.
5670 * See the comments in vmx_handle_exit.
5672 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5673 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5674 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5675 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5676 vcpu->run->internal.ndata = 3;
5677 vcpu->run->internal.data[0] = vect_info;
5678 vcpu->run->internal.data[1] = intr_info;
5679 vcpu->run->internal.data[2] = error_code;
5683 if (is_page_fault(intr_info)) {
5684 /* EPT won't cause page fault directly */
5686 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5687 trace_kvm_page_fault(cr2, error_code);
5689 if (kvm_event_needs_reinjection(vcpu))
5690 kvm_mmu_unprotect_page_virt(vcpu, cr2);
5691 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
5694 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5696 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5697 return handle_rmode_exception(vcpu, ex_no, error_code);
5701 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5704 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5705 if (!(vcpu->guest_debug &
5706 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5707 vcpu->arch.dr6 &= ~15;
5708 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5709 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5710 skip_emulated_instruction(vcpu);
5712 kvm_queue_exception(vcpu, DB_VECTOR);
5715 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5716 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5720 * Update instruction length as we may reinject #BP from
5721 * user space while in guest debugging mode. Reading it for
5722 * #DB as well causes no harm, it is not used in that case.
5724 vmx->vcpu.arch.event_exit_inst_len =
5725 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5726 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5727 rip = kvm_rip_read(vcpu);
5728 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5729 kvm_run->debug.arch.exception = ex_no;
5732 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5733 kvm_run->ex.exception = ex_no;
5734 kvm_run->ex.error_code = error_code;
5740 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5742 ++vcpu->stat.irq_exits;
5746 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5748 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5752 static int handle_io(struct kvm_vcpu *vcpu)
5754 unsigned long exit_qualification;
5755 int size, in, string, ret;
5758 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5759 string = (exit_qualification & 16) != 0;
5760 in = (exit_qualification & 8) != 0;
5762 ++vcpu->stat.io_exits;
5765 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5767 port = exit_qualification >> 16;
5768 size = (exit_qualification & 7) + 1;
5770 ret = kvm_skip_emulated_instruction(vcpu);
5773 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
5774 * KVM_EXIT_DEBUG here.
5776 return kvm_fast_pio_out(vcpu, size, port) && ret;
5780 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5783 * Patch in the VMCALL instruction:
5785 hypercall[0] = 0x0f;
5786 hypercall[1] = 0x01;
5787 hypercall[2] = 0xc1;
5790 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5791 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5793 if (is_guest_mode(vcpu)) {
5794 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5795 unsigned long orig_val = val;
5798 * We get here when L2 changed cr0 in a way that did not change
5799 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5800 * but did change L0 shadowed bits. So we first calculate the
5801 * effective cr0 value that L1 would like to write into the
5802 * hardware. It consists of the L2-owned bits from the new
5803 * value combined with the L1-owned bits from L1's guest_cr0.
5805 val = (val & ~vmcs12->cr0_guest_host_mask) |
5806 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5808 if (!nested_guest_cr0_valid(vcpu, val))
5811 if (kvm_set_cr0(vcpu, val))
5813 vmcs_writel(CR0_READ_SHADOW, orig_val);
5816 if (to_vmx(vcpu)->nested.vmxon &&
5817 !nested_host_cr0_valid(vcpu, val))
5820 return kvm_set_cr0(vcpu, val);
5824 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5826 if (is_guest_mode(vcpu)) {
5827 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5828 unsigned long orig_val = val;
5830 /* analogously to handle_set_cr0 */
5831 val = (val & ~vmcs12->cr4_guest_host_mask) |
5832 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5833 if (kvm_set_cr4(vcpu, val))
5835 vmcs_writel(CR4_READ_SHADOW, orig_val);
5838 return kvm_set_cr4(vcpu, val);
5841 static int handle_cr(struct kvm_vcpu *vcpu)
5843 unsigned long exit_qualification, val;
5849 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5850 cr = exit_qualification & 15;
5851 reg = (exit_qualification >> 8) & 15;
5852 switch ((exit_qualification >> 4) & 3) {
5853 case 0: /* mov to cr */
5854 val = kvm_register_readl(vcpu, reg);
5855 trace_kvm_cr_write(cr, val);
5858 err = handle_set_cr0(vcpu, val);
5859 return kvm_complete_insn_gp(vcpu, err);
5861 err = kvm_set_cr3(vcpu, val);
5862 return kvm_complete_insn_gp(vcpu, err);
5864 err = handle_set_cr4(vcpu, val);
5865 return kvm_complete_insn_gp(vcpu, err);
5867 u8 cr8_prev = kvm_get_cr8(vcpu);
5869 err = kvm_set_cr8(vcpu, cr8);
5870 ret = kvm_complete_insn_gp(vcpu, err);
5871 if (lapic_in_kernel(vcpu))
5873 if (cr8_prev <= cr8)
5876 * TODO: we might be squashing a
5877 * KVM_GUESTDBG_SINGLESTEP-triggered
5878 * KVM_EXIT_DEBUG here.
5880 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5886 WARN_ONCE(1, "Guest should always own CR0.TS");
5887 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5888 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5889 return kvm_skip_emulated_instruction(vcpu);
5890 case 1: /*mov from cr*/
5893 val = kvm_read_cr3(vcpu);
5894 kvm_register_write(vcpu, reg, val);
5895 trace_kvm_cr_read(cr, val);
5896 return kvm_skip_emulated_instruction(vcpu);
5898 val = kvm_get_cr8(vcpu);
5899 kvm_register_write(vcpu, reg, val);
5900 trace_kvm_cr_read(cr, val);
5901 return kvm_skip_emulated_instruction(vcpu);
5905 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5906 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5907 kvm_lmsw(vcpu, val);
5909 return kvm_skip_emulated_instruction(vcpu);
5913 vcpu->run->exit_reason = 0;
5914 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5915 (int)(exit_qualification >> 4) & 3, cr);
5919 static int handle_dr(struct kvm_vcpu *vcpu)
5921 unsigned long exit_qualification;
5924 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5925 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5927 /* First, if DR does not exist, trigger UD */
5928 if (!kvm_require_dr(vcpu, dr))
5931 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5932 if (!kvm_require_cpl(vcpu, 0))
5934 dr7 = vmcs_readl(GUEST_DR7);
5937 * As the vm-exit takes precedence over the debug trap, we
5938 * need to emulate the latter, either for the host or the
5939 * guest debugging itself.
5941 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5942 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5943 vcpu->run->debug.arch.dr7 = dr7;
5944 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5945 vcpu->run->debug.arch.exception = DB_VECTOR;
5946 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5949 vcpu->arch.dr6 &= ~15;
5950 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5951 kvm_queue_exception(vcpu, DB_VECTOR);
5956 if (vcpu->guest_debug == 0) {
5957 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
5958 CPU_BASED_MOV_DR_EXITING);
5961 * No more DR vmexits; force a reload of the debug registers
5962 * and reenter on this instruction. The next vmexit will
5963 * retrieve the full state of the debug registers.
5965 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5969 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5970 if (exit_qualification & TYPE_MOV_FROM_DR) {
5973 if (kvm_get_dr(vcpu, dr, &val))
5975 kvm_register_write(vcpu, reg, val);
5977 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5980 return kvm_skip_emulated_instruction(vcpu);
5983 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5985 return vcpu->arch.dr6;
5988 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5992 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5994 get_debugreg(vcpu->arch.db[0], 0);
5995 get_debugreg(vcpu->arch.db[1], 1);
5996 get_debugreg(vcpu->arch.db[2], 2);
5997 get_debugreg(vcpu->arch.db[3], 3);
5998 get_debugreg(vcpu->arch.dr6, 6);
5999 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
6001 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
6002 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
6005 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
6007 vmcs_writel(GUEST_DR7, val);
6010 static int handle_cpuid(struct kvm_vcpu *vcpu)
6012 return kvm_emulate_cpuid(vcpu);
6015 static int handle_rdmsr(struct kvm_vcpu *vcpu)
6017 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6018 struct msr_data msr_info;
6020 msr_info.index = ecx;
6021 msr_info.host_initiated = false;
6022 if (vmx_get_msr(vcpu, &msr_info)) {
6023 trace_kvm_msr_read_ex(ecx);
6024 kvm_inject_gp(vcpu, 0);
6028 trace_kvm_msr_read(ecx, msr_info.data);
6030 /* FIXME: handling of bits 32:63 of rax, rdx */
6031 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
6032 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
6033 return kvm_skip_emulated_instruction(vcpu);
6036 static int handle_wrmsr(struct kvm_vcpu *vcpu)
6038 struct msr_data msr;
6039 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6040 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
6041 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6045 msr.host_initiated = false;
6046 if (kvm_set_msr(vcpu, &msr) != 0) {
6047 trace_kvm_msr_write_ex(ecx, data);
6048 kvm_inject_gp(vcpu, 0);
6052 trace_kvm_msr_write(ecx, data);
6053 return kvm_skip_emulated_instruction(vcpu);
6056 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
6058 kvm_apic_update_ppr(vcpu);
6062 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
6064 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6065 CPU_BASED_VIRTUAL_INTR_PENDING);
6067 kvm_make_request(KVM_REQ_EVENT, vcpu);
6069 ++vcpu->stat.irq_window_exits;
6073 static int handle_halt(struct kvm_vcpu *vcpu)
6075 return kvm_emulate_halt(vcpu);
6078 static int handle_vmcall(struct kvm_vcpu *vcpu)
6080 return kvm_emulate_hypercall(vcpu);
6083 static int handle_invd(struct kvm_vcpu *vcpu)
6085 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6088 static int handle_invlpg(struct kvm_vcpu *vcpu)
6090 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6092 kvm_mmu_invlpg(vcpu, exit_qualification);
6093 return kvm_skip_emulated_instruction(vcpu);
6096 static int handle_rdpmc(struct kvm_vcpu *vcpu)
6100 err = kvm_rdpmc(vcpu);
6101 return kvm_complete_insn_gp(vcpu, err);
6104 static int handle_wbinvd(struct kvm_vcpu *vcpu)
6106 return kvm_emulate_wbinvd(vcpu);
6109 static int handle_xsetbv(struct kvm_vcpu *vcpu)
6111 u64 new_bv = kvm_read_edx_eax(vcpu);
6112 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
6114 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
6115 return kvm_skip_emulated_instruction(vcpu);
6119 static int handle_xsaves(struct kvm_vcpu *vcpu)
6121 kvm_skip_emulated_instruction(vcpu);
6122 WARN(1, "this should never happen\n");
6126 static int handle_xrstors(struct kvm_vcpu *vcpu)
6128 kvm_skip_emulated_instruction(vcpu);
6129 WARN(1, "this should never happen\n");
6133 static int handle_apic_access(struct kvm_vcpu *vcpu)
6135 if (likely(fasteoi)) {
6136 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6137 int access_type, offset;
6139 access_type = exit_qualification & APIC_ACCESS_TYPE;
6140 offset = exit_qualification & APIC_ACCESS_OFFSET;
6142 * Sane guest uses MOV to write EOI, with written value
6143 * not cared. So make a short-circuit here by avoiding
6144 * heavy instruction emulation.
6146 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
6147 (offset == APIC_EOI)) {
6148 kvm_lapic_set_eoi(vcpu);
6149 return kvm_skip_emulated_instruction(vcpu);
6152 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6155 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
6157 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6158 int vector = exit_qualification & 0xff;
6160 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
6161 kvm_apic_set_eoi_accelerated(vcpu, vector);
6165 static int handle_apic_write(struct kvm_vcpu *vcpu)
6167 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6168 u32 offset = exit_qualification & 0xfff;
6170 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6171 kvm_apic_write_nodecode(vcpu, offset);
6175 static int handle_task_switch(struct kvm_vcpu *vcpu)
6177 struct vcpu_vmx *vmx = to_vmx(vcpu);
6178 unsigned long exit_qualification;
6179 bool has_error_code = false;
6182 int reason, type, idt_v, idt_index;
6184 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
6185 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
6186 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
6188 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6190 reason = (u32)exit_qualification >> 30;
6191 if (reason == TASK_SWITCH_GATE && idt_v) {
6193 case INTR_TYPE_NMI_INTR:
6194 vcpu->arch.nmi_injected = false;
6195 vmx_set_nmi_mask(vcpu, true);
6197 case INTR_TYPE_EXT_INTR:
6198 case INTR_TYPE_SOFT_INTR:
6199 kvm_clear_interrupt_queue(vcpu);
6201 case INTR_TYPE_HARD_EXCEPTION:
6202 if (vmx->idt_vectoring_info &
6203 VECTORING_INFO_DELIVER_CODE_MASK) {
6204 has_error_code = true;
6206 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6209 case INTR_TYPE_SOFT_EXCEPTION:
6210 kvm_clear_exception_queue(vcpu);
6216 tss_selector = exit_qualification;
6218 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
6219 type != INTR_TYPE_EXT_INTR &&
6220 type != INTR_TYPE_NMI_INTR))
6221 skip_emulated_instruction(vcpu);
6223 if (kvm_task_switch(vcpu, tss_selector,
6224 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
6225 has_error_code, error_code) == EMULATE_FAIL) {
6226 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6227 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6228 vcpu->run->internal.ndata = 0;
6233 * TODO: What about debug traps on tss switch?
6234 * Are we supposed to inject them and update dr6?
6240 static int handle_ept_violation(struct kvm_vcpu *vcpu)
6242 unsigned long exit_qualification;
6247 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6249 gla_validity = (exit_qualification >> 7) & 0x3;
6250 if (gla_validity == 0x2) {
6251 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
6252 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
6253 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
6254 vmcs_readl(GUEST_LINEAR_ADDRESS));
6255 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
6256 (long unsigned int)exit_qualification);
6257 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6258 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
6263 * EPT violation happened while executing iret from NMI,
6264 * "blocked by NMI" bit has to be set before next VM entry.
6265 * There are errata that may cause this bit to not be set:
6268 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
6269 cpu_has_virtual_nmis() &&
6270 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
6271 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
6273 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6274 trace_kvm_page_fault(gpa, exit_qualification);
6276 /* Is it a read fault? */
6277 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
6278 ? PFERR_USER_MASK : 0;
6279 /* Is it a write fault? */
6280 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
6281 ? PFERR_WRITE_MASK : 0;
6282 /* Is it a fetch fault? */
6283 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
6284 ? PFERR_FETCH_MASK : 0;
6285 /* ept page table entry is present? */
6286 error_code |= (exit_qualification &
6287 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
6288 EPT_VIOLATION_EXECUTABLE))
6289 ? PFERR_PRESENT_MASK : 0;
6291 vcpu->arch.gpa_available = true;
6292 vcpu->arch.exit_qualification = exit_qualification;
6294 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
6297 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
6302 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6303 if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
6304 trace_kvm_fast_mmio(gpa);
6305 return kvm_skip_emulated_instruction(vcpu);
6308 ret = handle_mmio_page_fault(vcpu, gpa, true);
6309 vcpu->arch.gpa_available = true;
6310 if (likely(ret == RET_MMIO_PF_EMULATE))
6311 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
6314 if (unlikely(ret == RET_MMIO_PF_INVALID))
6315 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
6317 if (unlikely(ret == RET_MMIO_PF_RETRY))
6320 /* It is the real ept misconfig */
6323 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6324 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6329 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6331 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6332 CPU_BASED_VIRTUAL_NMI_PENDING);
6333 ++vcpu->stat.nmi_window_exits;
6334 kvm_make_request(KVM_REQ_EVENT, vcpu);
6339 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6341 struct vcpu_vmx *vmx = to_vmx(vcpu);
6342 enum emulation_result err = EMULATE_DONE;
6345 bool intr_window_requested;
6346 unsigned count = 130;
6348 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6349 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6351 while (vmx->emulation_required && count-- != 0) {
6352 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6353 return handle_interrupt_window(&vmx->vcpu);
6355 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
6358 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
6360 if (err == EMULATE_USER_EXIT) {
6361 ++vcpu->stat.mmio_exits;
6366 if (err != EMULATE_DONE) {
6367 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6368 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6369 vcpu->run->internal.ndata = 0;
6373 if (vcpu->arch.halt_request) {
6374 vcpu->arch.halt_request = 0;
6375 ret = kvm_vcpu_halt(vcpu);
6379 if (signal_pending(current))
6389 static int __grow_ple_window(int val)
6391 if (ple_window_grow < 1)
6394 val = min(val, ple_window_actual_max);
6396 if (ple_window_grow < ple_window)
6397 val *= ple_window_grow;
6399 val += ple_window_grow;
6404 static int __shrink_ple_window(int val, int modifier, int minimum)
6409 if (modifier < ple_window)
6414 return max(val, minimum);
6417 static void grow_ple_window(struct kvm_vcpu *vcpu)
6419 struct vcpu_vmx *vmx = to_vmx(vcpu);
6420 int old = vmx->ple_window;
6422 vmx->ple_window = __grow_ple_window(old);
6424 if (vmx->ple_window != old)
6425 vmx->ple_window_dirty = true;
6427 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6430 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6432 struct vcpu_vmx *vmx = to_vmx(vcpu);
6433 int old = vmx->ple_window;
6435 vmx->ple_window = __shrink_ple_window(old,
6436 ple_window_shrink, ple_window);
6438 if (vmx->ple_window != old)
6439 vmx->ple_window_dirty = true;
6441 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6445 * ple_window_actual_max is computed to be one grow_ple_window() below
6446 * ple_window_max. (See __grow_ple_window for the reason.)
6447 * This prevents overflows, because ple_window_max is int.
6448 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6450 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6452 static void update_ple_window_actual_max(void)
6454 ple_window_actual_max =
6455 __shrink_ple_window(max(ple_window_max, ple_window),
6456 ple_window_grow, INT_MIN);
6460 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6462 static void wakeup_handler(void)
6464 struct kvm_vcpu *vcpu;
6465 int cpu = smp_processor_id();
6467 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6468 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6469 blocked_vcpu_list) {
6470 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6472 if (pi_test_on(pi_desc) == 1)
6473 kvm_vcpu_kick(vcpu);
6475 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6478 void vmx_enable_tdp(void)
6480 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
6481 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
6482 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
6483 0ull, VMX_EPT_EXECUTABLE_MASK,
6484 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
6485 enable_ept_ad_bits ? 0ull : VMX_EPT_RWX_MASK);
6487 ept_set_mmio_spte_mask();
6491 static __init int hardware_setup(void)
6493 int r = -ENOMEM, i, msr;
6495 rdmsrl_safe(MSR_EFER, &host_efer);
6497 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6498 kvm_define_shared_msr(i, vmx_msr_index[i]);
6500 for (i = 0; i < VMX_BITMAP_NR; i++) {
6501 vmx_bitmap[i] = (unsigned long *)__get_free_page(GFP_KERNEL);
6506 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6507 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6508 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6511 * Allow direct access to the PC debug port (it is often used for I/O
6512 * delays, but the vmexits simply slow things down).
6514 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6515 clear_bit(0x80, vmx_io_bitmap_a);
6517 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6519 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6520 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6522 if (setup_vmcs_config(&vmcs_config) < 0) {
6527 if (boot_cpu_has(X86_FEATURE_NX))
6528 kvm_enable_efer_bits(EFER_NX);
6530 if (!cpu_has_vmx_vpid())
6532 if (!cpu_has_vmx_shadow_vmcs())
6533 enable_shadow_vmcs = 0;
6534 if (enable_shadow_vmcs)
6535 init_vmcs_shadow_fields();
6537 if (!cpu_has_vmx_ept() ||
6538 !cpu_has_vmx_ept_4levels()) {
6540 enable_unrestricted_guest = 0;
6541 enable_ept_ad_bits = 0;
6544 if (!cpu_has_vmx_ept_ad_bits())
6545 enable_ept_ad_bits = 0;
6547 if (!cpu_has_vmx_unrestricted_guest())
6548 enable_unrestricted_guest = 0;
6550 if (!cpu_has_vmx_flexpriority())
6551 flexpriority_enabled = 0;
6554 * set_apic_access_page_addr() is used to reload apic access
6555 * page upon invalidation. No need to do anything if not
6556 * using the APIC_ACCESS_ADDR VMCS field.
6558 if (!flexpriority_enabled)
6559 kvm_x86_ops->set_apic_access_page_addr = NULL;
6561 if (!cpu_has_vmx_tpr_shadow())
6562 kvm_x86_ops->update_cr8_intercept = NULL;
6564 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6565 kvm_disable_largepages();
6567 if (!cpu_has_vmx_ple())
6570 if (!cpu_has_vmx_apicv()) {
6572 kvm_x86_ops->sync_pir_to_irr = NULL;
6575 if (cpu_has_vmx_tsc_scaling()) {
6576 kvm_has_tsc_control = true;
6577 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6578 kvm_tsc_scaling_ratio_frac_bits = 48;
6581 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6582 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6583 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6584 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6585 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6586 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6587 vmx_disable_intercept_for_msr(MSR_IA32_BNDCFGS, true);
6589 memcpy(vmx_msr_bitmap_legacy_x2apic_apicv,
6590 vmx_msr_bitmap_legacy, PAGE_SIZE);
6591 memcpy(vmx_msr_bitmap_longmode_x2apic_apicv,
6592 vmx_msr_bitmap_longmode, PAGE_SIZE);
6593 memcpy(vmx_msr_bitmap_legacy_x2apic,
6594 vmx_msr_bitmap_legacy, PAGE_SIZE);
6595 memcpy(vmx_msr_bitmap_longmode_x2apic,
6596 vmx_msr_bitmap_longmode, PAGE_SIZE);
6598 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6600 for (msr = 0x800; msr <= 0x8ff; msr++) {
6601 if (msr == 0x839 /* TMCCT */)
6603 vmx_disable_intercept_msr_x2apic(msr, MSR_TYPE_R, true);
6607 * TPR reads and writes can be virtualized even if virtual interrupt
6608 * delivery is not in use.
6610 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_W, true);
6611 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_R | MSR_TYPE_W, false);
6614 vmx_disable_intercept_msr_x2apic(0x80b, MSR_TYPE_W, true);
6616 vmx_disable_intercept_msr_x2apic(0x83f, MSR_TYPE_W, true);
6623 update_ple_window_actual_max();
6626 * Only enable PML when hardware supports PML feature, and both EPT
6627 * and EPT A/D bit features are enabled -- PML depends on them to work.
6629 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6633 kvm_x86_ops->slot_enable_log_dirty = NULL;
6634 kvm_x86_ops->slot_disable_log_dirty = NULL;
6635 kvm_x86_ops->flush_log_dirty = NULL;
6636 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6639 if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
6642 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
6643 cpu_preemption_timer_multi =
6644 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
6646 kvm_x86_ops->set_hv_timer = NULL;
6647 kvm_x86_ops->cancel_hv_timer = NULL;
6650 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6652 kvm_mce_cap_supported |= MCG_LMCE_P;
6654 return alloc_kvm_area();
6657 for (i = 0; i < VMX_BITMAP_NR; i++)
6658 free_page((unsigned long)vmx_bitmap[i]);
6663 static __exit void hardware_unsetup(void)
6667 for (i = 0; i < VMX_BITMAP_NR; i++)
6668 free_page((unsigned long)vmx_bitmap[i]);
6674 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6675 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6677 static int handle_pause(struct kvm_vcpu *vcpu)
6680 grow_ple_window(vcpu);
6682 kvm_vcpu_on_spin(vcpu);
6683 return kvm_skip_emulated_instruction(vcpu);
6686 static int handle_nop(struct kvm_vcpu *vcpu)
6688 return kvm_skip_emulated_instruction(vcpu);
6691 static int handle_mwait(struct kvm_vcpu *vcpu)
6693 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6694 return handle_nop(vcpu);
6697 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6702 static int handle_monitor(struct kvm_vcpu *vcpu)
6704 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6705 return handle_nop(vcpu);
6709 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6710 * We could reuse a single VMCS for all the L2 guests, but we also want the
6711 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6712 * allows keeping them loaded on the processor, and in the future will allow
6713 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6714 * every entry if they never change.
6715 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6716 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6718 * The following functions allocate and free a vmcs02 in this pool.
6721 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6722 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6724 struct vmcs02_list *item;
6725 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6726 if (item->vmptr == vmx->nested.current_vmptr) {
6727 list_move(&item->list, &vmx->nested.vmcs02_pool);
6728 return &item->vmcs02;
6731 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6732 /* Recycle the least recently used VMCS. */
6733 item = list_last_entry(&vmx->nested.vmcs02_pool,
6734 struct vmcs02_list, list);
6735 item->vmptr = vmx->nested.current_vmptr;
6736 list_move(&item->list, &vmx->nested.vmcs02_pool);
6737 return &item->vmcs02;
6740 /* Create a new VMCS */
6741 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6744 item->vmcs02.vmcs = alloc_vmcs();
6745 item->vmcs02.shadow_vmcs = NULL;
6746 if (!item->vmcs02.vmcs) {
6750 loaded_vmcs_init(&item->vmcs02);
6751 item->vmptr = vmx->nested.current_vmptr;
6752 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6753 vmx->nested.vmcs02_num++;
6754 return &item->vmcs02;
6757 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6758 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6760 struct vmcs02_list *item;
6761 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6762 if (item->vmptr == vmptr) {
6763 free_loaded_vmcs(&item->vmcs02);
6764 list_del(&item->list);
6766 vmx->nested.vmcs02_num--;
6772 * Free all VMCSs saved for this vcpu, except the one pointed by
6773 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6774 * must be &vmx->vmcs01.
6776 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6778 struct vmcs02_list *item, *n;
6780 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6781 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6783 * Something will leak if the above WARN triggers. Better than
6786 if (vmx->loaded_vmcs == &item->vmcs02)
6789 free_loaded_vmcs(&item->vmcs02);
6790 list_del(&item->list);
6792 vmx->nested.vmcs02_num--;
6797 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6798 * set the success or error code of an emulated VMX instruction, as specified
6799 * by Vol 2B, VMX Instruction Reference, "Conventions".
6801 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6803 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6804 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6805 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6808 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6810 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6811 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6812 X86_EFLAGS_SF | X86_EFLAGS_OF))
6816 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6817 u32 vm_instruction_error)
6819 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6821 * failValid writes the error number to the current VMCS, which
6822 * can't be done there isn't a current VMCS.
6824 nested_vmx_failInvalid(vcpu);
6827 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6828 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6829 X86_EFLAGS_SF | X86_EFLAGS_OF))
6831 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6833 * We don't need to force a shadow sync because
6834 * VM_INSTRUCTION_ERROR is not shadowed
6838 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6840 /* TODO: not to reset guest simply here. */
6841 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6842 pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
6845 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6847 struct vcpu_vmx *vmx =
6848 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6850 vmx->nested.preemption_timer_expired = true;
6851 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6852 kvm_vcpu_kick(&vmx->vcpu);
6854 return HRTIMER_NORESTART;
6858 * Decode the memory-address operand of a vmx instruction, as recorded on an
6859 * exit caused by such an instruction (run by a guest hypervisor).
6860 * On success, returns 0. When the operand is invalid, returns 1 and throws
6863 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6864 unsigned long exit_qualification,
6865 u32 vmx_instruction_info, bool wr, gva_t *ret)
6869 struct kvm_segment s;
6872 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6873 * Execution", on an exit, vmx_instruction_info holds most of the
6874 * addressing components of the operand. Only the displacement part
6875 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6876 * For how an actual address is calculated from all these components,
6877 * refer to Vol. 1, "Operand Addressing".
6879 int scaling = vmx_instruction_info & 3;
6880 int addr_size = (vmx_instruction_info >> 7) & 7;
6881 bool is_reg = vmx_instruction_info & (1u << 10);
6882 int seg_reg = (vmx_instruction_info >> 15) & 7;
6883 int index_reg = (vmx_instruction_info >> 18) & 0xf;
6884 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6885 int base_reg = (vmx_instruction_info >> 23) & 0xf;
6886 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
6889 kvm_queue_exception(vcpu, UD_VECTOR);
6893 /* Addr = segment_base + offset */
6894 /* offset = base + [index * scale] + displacement */
6895 off = exit_qualification; /* holds the displacement */
6897 off += kvm_register_read(vcpu, base_reg);
6899 off += kvm_register_read(vcpu, index_reg)<<scaling;
6900 vmx_get_segment(vcpu, &s, seg_reg);
6901 *ret = s.base + off;
6903 if (addr_size == 1) /* 32 bit */
6906 /* Checks for #GP/#SS exceptions. */
6908 if (is_long_mode(vcpu)) {
6909 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6910 * non-canonical form. This is the only check on the memory
6911 * destination for long mode!
6913 exn = is_noncanonical_address(*ret);
6914 } else if (is_protmode(vcpu)) {
6915 /* Protected mode: apply checks for segment validity in the
6917 * - segment type check (#GP(0) may be thrown)
6918 * - usability check (#GP(0)/#SS(0))
6919 * - limit check (#GP(0)/#SS(0))
6922 /* #GP(0) if the destination operand is located in a
6923 * read-only data segment or any code segment.
6925 exn = ((s.type & 0xa) == 0 || (s.type & 8));
6927 /* #GP(0) if the source operand is located in an
6928 * execute-only code segment
6930 exn = ((s.type & 0xa) == 8);
6932 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6935 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
6937 exn = (s.unusable != 0);
6938 /* Protected mode: #GP(0)/#SS(0) if the memory
6939 * operand is outside the segment limit.
6941 exn = exn || (off + sizeof(u64) > s.limit);
6944 kvm_queue_exception_e(vcpu,
6945 seg_reg == VCPU_SREG_SS ?
6946 SS_VECTOR : GP_VECTOR,
6955 * This function performs the various checks including
6956 * - if it's 4KB aligned
6957 * - No bits beyond the physical address width are set
6958 * - Returns 0 on success or else 1
6959 * (Intel SDM Section 30.3)
6961 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6966 struct x86_exception e;
6968 struct vcpu_vmx *vmx = to_vmx(vcpu);
6969 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6971 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6972 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
6975 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6976 sizeof(vmptr), &e)) {
6977 kvm_inject_page_fault(vcpu, &e);
6981 switch (exit_reason) {
6982 case EXIT_REASON_VMON:
6985 * The first 4 bytes of VMXON region contain the supported
6986 * VMCS revision identifier
6988 * Note - IA32_VMX_BASIC[48] will never be 1
6989 * for the nested case;
6990 * which replaces physical address width with 32
6993 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6994 nested_vmx_failInvalid(vcpu);
6995 return kvm_skip_emulated_instruction(vcpu);
6998 page = nested_get_page(vcpu, vmptr);
7000 nested_vmx_failInvalid(vcpu);
7001 return kvm_skip_emulated_instruction(vcpu);
7003 if (*(u32 *)kmap(page) != VMCS12_REVISION) {
7005 nested_release_page_clean(page);
7006 nested_vmx_failInvalid(vcpu);
7007 return kvm_skip_emulated_instruction(vcpu);
7010 nested_release_page_clean(page);
7011 vmx->nested.vmxon_ptr = vmptr;
7013 case EXIT_REASON_VMCLEAR:
7014 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
7015 nested_vmx_failValid(vcpu,
7016 VMXERR_VMCLEAR_INVALID_ADDRESS);
7017 return kvm_skip_emulated_instruction(vcpu);
7020 if (vmptr == vmx->nested.vmxon_ptr) {
7021 nested_vmx_failValid(vcpu,
7022 VMXERR_VMCLEAR_VMXON_POINTER);
7023 return kvm_skip_emulated_instruction(vcpu);
7026 case EXIT_REASON_VMPTRLD:
7027 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
7028 nested_vmx_failValid(vcpu,
7029 VMXERR_VMPTRLD_INVALID_ADDRESS);
7030 return kvm_skip_emulated_instruction(vcpu);
7033 if (vmptr == vmx->nested.vmxon_ptr) {
7034 nested_vmx_failValid(vcpu,
7035 VMXERR_VMPTRLD_VMXON_POINTER);
7036 return kvm_skip_emulated_instruction(vcpu);
7040 return 1; /* shouldn't happen */
7048 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
7050 struct vcpu_vmx *vmx = to_vmx(vcpu);
7051 struct vmcs *shadow_vmcs;
7053 if (cpu_has_vmx_msr_bitmap()) {
7054 vmx->nested.msr_bitmap =
7055 (unsigned long *)__get_free_page(GFP_KERNEL);
7056 if (!vmx->nested.msr_bitmap)
7057 goto out_msr_bitmap;
7060 vmx->nested.cached_vmcs12 = kmalloc(VMCS12_SIZE, GFP_KERNEL);
7061 if (!vmx->nested.cached_vmcs12)
7062 goto out_cached_vmcs12;
7064 if (enable_shadow_vmcs) {
7065 shadow_vmcs = alloc_vmcs();
7067 goto out_shadow_vmcs;
7068 /* mark vmcs as shadow */
7069 shadow_vmcs->revision_id |= (1u << 31);
7070 /* init shadow vmcs */
7071 vmcs_clear(shadow_vmcs);
7072 vmx->vmcs01.shadow_vmcs = shadow_vmcs;
7075 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
7076 vmx->nested.vmcs02_num = 0;
7078 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
7079 HRTIMER_MODE_REL_PINNED);
7080 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
7082 vmx->nested.vmxon = true;
7086 kfree(vmx->nested.cached_vmcs12);
7089 free_page((unsigned long)vmx->nested.msr_bitmap);
7096 * Emulate the VMXON instruction.
7097 * Currently, we just remember that VMX is active, and do not save or even
7098 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
7099 * do not currently need to store anything in that guest-allocated memory
7100 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
7101 * argument is different from the VMXON pointer (which the spec says they do).
7103 static int handle_vmon(struct kvm_vcpu *vcpu)
7106 struct kvm_segment cs;
7107 struct vcpu_vmx *vmx = to_vmx(vcpu);
7108 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
7109 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7111 /* The Intel VMX Instruction Reference lists a bunch of bits that
7112 * are prerequisite to running VMXON, most notably cr4.VMXE must be
7113 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
7114 * Otherwise, we should fail with #UD. We test these now:
7116 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
7117 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
7118 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
7119 kvm_queue_exception(vcpu, UD_VECTOR);
7123 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
7124 if (is_long_mode(vcpu) && !cs.l) {
7125 kvm_queue_exception(vcpu, UD_VECTOR);
7129 if (vmx_get_cpl(vcpu)) {
7130 kvm_inject_gp(vcpu, 0);
7134 if (vmx->nested.vmxon) {
7135 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
7136 return kvm_skip_emulated_instruction(vcpu);
7139 if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
7140 != VMXON_NEEDED_FEATURES) {
7141 kvm_inject_gp(vcpu, 0);
7145 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
7148 ret = enter_vmx_operation(vcpu);
7152 nested_vmx_succeed(vcpu);
7153 return kvm_skip_emulated_instruction(vcpu);
7157 * Intel's VMX Instruction Reference specifies a common set of prerequisites
7158 * for running VMX instructions (except VMXON, whose prerequisites are
7159 * slightly different). It also specifies what exception to inject otherwise.
7161 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
7163 struct kvm_segment cs;
7164 struct vcpu_vmx *vmx = to_vmx(vcpu);
7166 if (!vmx->nested.vmxon) {
7167 kvm_queue_exception(vcpu, UD_VECTOR);
7171 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
7172 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
7173 (is_long_mode(vcpu) && !cs.l)) {
7174 kvm_queue_exception(vcpu, UD_VECTOR);
7178 if (vmx_get_cpl(vcpu)) {
7179 kvm_inject_gp(vcpu, 0);
7186 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
7188 if (vmx->nested.current_vmptr == -1ull)
7191 /* current_vmptr and current_vmcs12 are always set/reset together */
7192 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
7195 if (enable_shadow_vmcs) {
7196 /* copy to memory all shadowed fields in case
7197 they were modified */
7198 copy_shadow_to_vmcs12(vmx);
7199 vmx->nested.sync_shadow_vmcs = false;
7200 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
7201 SECONDARY_EXEC_SHADOW_VMCS);
7202 vmcs_write64(VMCS_LINK_POINTER, -1ull);
7204 vmx->nested.posted_intr_nv = -1;
7206 /* Flush VMCS12 to guest memory */
7207 memcpy(vmx->nested.current_vmcs12, vmx->nested.cached_vmcs12,
7210 kunmap(vmx->nested.current_vmcs12_page);
7211 nested_release_page(vmx->nested.current_vmcs12_page);
7212 vmx->nested.current_vmptr = -1ull;
7213 vmx->nested.current_vmcs12 = NULL;
7217 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7218 * just stops using VMX.
7220 static void free_nested(struct vcpu_vmx *vmx)
7222 if (!vmx->nested.vmxon)
7225 vmx->nested.vmxon = false;
7226 free_vpid(vmx->nested.vpid02);
7227 nested_release_vmcs12(vmx);
7228 if (vmx->nested.msr_bitmap) {
7229 free_page((unsigned long)vmx->nested.msr_bitmap);
7230 vmx->nested.msr_bitmap = NULL;
7232 if (enable_shadow_vmcs) {
7233 vmcs_clear(vmx->vmcs01.shadow_vmcs);
7234 free_vmcs(vmx->vmcs01.shadow_vmcs);
7235 vmx->vmcs01.shadow_vmcs = NULL;
7237 kfree(vmx->nested.cached_vmcs12);
7238 /* Unpin physical memory we referred to in current vmcs02 */
7239 if (vmx->nested.apic_access_page) {
7240 nested_release_page(vmx->nested.apic_access_page);
7241 vmx->nested.apic_access_page = NULL;
7243 if (vmx->nested.virtual_apic_page) {
7244 nested_release_page(vmx->nested.virtual_apic_page);
7245 vmx->nested.virtual_apic_page = NULL;
7247 if (vmx->nested.pi_desc_page) {
7248 kunmap(vmx->nested.pi_desc_page);
7249 nested_release_page(vmx->nested.pi_desc_page);
7250 vmx->nested.pi_desc_page = NULL;
7251 vmx->nested.pi_desc = NULL;
7254 nested_free_all_saved_vmcss(vmx);
7257 /* Emulate the VMXOFF instruction */
7258 static int handle_vmoff(struct kvm_vcpu *vcpu)
7260 if (!nested_vmx_check_permission(vcpu))
7262 free_nested(to_vmx(vcpu));
7263 nested_vmx_succeed(vcpu);
7264 return kvm_skip_emulated_instruction(vcpu);
7267 /* Emulate the VMCLEAR instruction */
7268 static int handle_vmclear(struct kvm_vcpu *vcpu)
7270 struct vcpu_vmx *vmx = to_vmx(vcpu);
7272 struct vmcs12 *vmcs12;
7275 if (!nested_vmx_check_permission(vcpu))
7278 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
7281 if (vmptr == vmx->nested.current_vmptr)
7282 nested_release_vmcs12(vmx);
7284 page = nested_get_page(vcpu, vmptr);
7287 * For accurate processor emulation, VMCLEAR beyond available
7288 * physical memory should do nothing at all. However, it is
7289 * possible that a nested vmx bug, not a guest hypervisor bug,
7290 * resulted in this case, so let's shut down before doing any
7293 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7296 vmcs12 = kmap(page);
7297 vmcs12->launch_state = 0;
7299 nested_release_page(page);
7301 nested_free_vmcs02(vmx, vmptr);
7303 nested_vmx_succeed(vcpu);
7304 return kvm_skip_emulated_instruction(vcpu);
7307 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
7309 /* Emulate the VMLAUNCH instruction */
7310 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
7312 return nested_vmx_run(vcpu, true);
7315 /* Emulate the VMRESUME instruction */
7316 static int handle_vmresume(struct kvm_vcpu *vcpu)
7319 return nested_vmx_run(vcpu, false);
7322 enum vmcs_field_type {
7323 VMCS_FIELD_TYPE_U16 = 0,
7324 VMCS_FIELD_TYPE_U64 = 1,
7325 VMCS_FIELD_TYPE_U32 = 2,
7326 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
7329 static inline int vmcs_field_type(unsigned long field)
7331 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
7332 return VMCS_FIELD_TYPE_U32;
7333 return (field >> 13) & 0x3 ;
7336 static inline int vmcs_field_readonly(unsigned long field)
7338 return (((field >> 10) & 0x3) == 1);
7342 * Read a vmcs12 field. Since these can have varying lengths and we return
7343 * one type, we chose the biggest type (u64) and zero-extend the return value
7344 * to that size. Note that the caller, handle_vmread, might need to use only
7345 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7346 * 64-bit fields are to be returned).
7348 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7349 unsigned long field, u64 *ret)
7351 short offset = vmcs_field_to_offset(field);
7357 p = ((char *)(get_vmcs12(vcpu))) + offset;
7359 switch (vmcs_field_type(field)) {
7360 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7361 *ret = *((natural_width *)p);
7363 case VMCS_FIELD_TYPE_U16:
7366 case VMCS_FIELD_TYPE_U32:
7369 case VMCS_FIELD_TYPE_U64:
7379 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7380 unsigned long field, u64 field_value){
7381 short offset = vmcs_field_to_offset(field);
7382 char *p = ((char *) get_vmcs12(vcpu)) + offset;
7386 switch (vmcs_field_type(field)) {
7387 case VMCS_FIELD_TYPE_U16:
7388 *(u16 *)p = field_value;
7390 case VMCS_FIELD_TYPE_U32:
7391 *(u32 *)p = field_value;
7393 case VMCS_FIELD_TYPE_U64:
7394 *(u64 *)p = field_value;
7396 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7397 *(natural_width *)p = field_value;
7406 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7409 unsigned long field;
7411 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7412 const unsigned long *fields = shadow_read_write_fields;
7413 const int num_fields = max_shadow_read_write_fields;
7417 vmcs_load(shadow_vmcs);
7419 for (i = 0; i < num_fields; i++) {
7421 switch (vmcs_field_type(field)) {
7422 case VMCS_FIELD_TYPE_U16:
7423 field_value = vmcs_read16(field);
7425 case VMCS_FIELD_TYPE_U32:
7426 field_value = vmcs_read32(field);
7428 case VMCS_FIELD_TYPE_U64:
7429 field_value = vmcs_read64(field);
7431 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7432 field_value = vmcs_readl(field);
7438 vmcs12_write_any(&vmx->vcpu, field, field_value);
7441 vmcs_clear(shadow_vmcs);
7442 vmcs_load(vmx->loaded_vmcs->vmcs);
7447 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7449 const unsigned long *fields[] = {
7450 shadow_read_write_fields,
7451 shadow_read_only_fields
7453 const int max_fields[] = {
7454 max_shadow_read_write_fields,
7455 max_shadow_read_only_fields
7458 unsigned long field;
7459 u64 field_value = 0;
7460 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7462 vmcs_load(shadow_vmcs);
7464 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7465 for (i = 0; i < max_fields[q]; i++) {
7466 field = fields[q][i];
7467 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7469 switch (vmcs_field_type(field)) {
7470 case VMCS_FIELD_TYPE_U16:
7471 vmcs_write16(field, (u16)field_value);
7473 case VMCS_FIELD_TYPE_U32:
7474 vmcs_write32(field, (u32)field_value);
7476 case VMCS_FIELD_TYPE_U64:
7477 vmcs_write64(field, (u64)field_value);
7479 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7480 vmcs_writel(field, (long)field_value);
7489 vmcs_clear(shadow_vmcs);
7490 vmcs_load(vmx->loaded_vmcs->vmcs);
7494 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7495 * used before) all generate the same failure when it is missing.
7497 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7499 struct vcpu_vmx *vmx = to_vmx(vcpu);
7500 if (vmx->nested.current_vmptr == -1ull) {
7501 nested_vmx_failInvalid(vcpu);
7507 static int handle_vmread(struct kvm_vcpu *vcpu)
7509 unsigned long field;
7511 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7512 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7515 if (!nested_vmx_check_permission(vcpu))
7518 if (!nested_vmx_check_vmcs12(vcpu))
7519 return kvm_skip_emulated_instruction(vcpu);
7521 /* Decode instruction info and find the field to read */
7522 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7523 /* Read the field, zero-extended to a u64 field_value */
7524 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7525 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7526 return kvm_skip_emulated_instruction(vcpu);
7529 * Now copy part of this value to register or memory, as requested.
7530 * Note that the number of bits actually copied is 32 or 64 depending
7531 * on the guest's mode (32 or 64 bit), not on the given field's length.
7533 if (vmx_instruction_info & (1u << 10)) {
7534 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7537 if (get_vmx_mem_address(vcpu, exit_qualification,
7538 vmx_instruction_info, true, &gva))
7540 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
7541 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7542 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7545 nested_vmx_succeed(vcpu);
7546 return kvm_skip_emulated_instruction(vcpu);
7550 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7552 unsigned long field;
7554 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7555 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7556 /* The value to write might be 32 or 64 bits, depending on L1's long
7557 * mode, and eventually we need to write that into a field of several
7558 * possible lengths. The code below first zero-extends the value to 64
7559 * bit (field_value), and then copies only the appropriate number of
7560 * bits into the vmcs12 field.
7562 u64 field_value = 0;
7563 struct x86_exception e;
7565 if (!nested_vmx_check_permission(vcpu))
7568 if (!nested_vmx_check_vmcs12(vcpu))
7569 return kvm_skip_emulated_instruction(vcpu);
7571 if (vmx_instruction_info & (1u << 10))
7572 field_value = kvm_register_readl(vcpu,
7573 (((vmx_instruction_info) >> 3) & 0xf));
7575 if (get_vmx_mem_address(vcpu, exit_qualification,
7576 vmx_instruction_info, false, &gva))
7578 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7579 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7580 kvm_inject_page_fault(vcpu, &e);
7586 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7587 if (vmcs_field_readonly(field)) {
7588 nested_vmx_failValid(vcpu,
7589 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7590 return kvm_skip_emulated_instruction(vcpu);
7593 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7594 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7595 return kvm_skip_emulated_instruction(vcpu);
7598 nested_vmx_succeed(vcpu);
7599 return kvm_skip_emulated_instruction(vcpu);
7602 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
7604 vmx->nested.current_vmptr = vmptr;
7605 if (enable_shadow_vmcs) {
7606 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7607 SECONDARY_EXEC_SHADOW_VMCS);
7608 vmcs_write64(VMCS_LINK_POINTER,
7609 __pa(vmx->vmcs01.shadow_vmcs));
7610 vmx->nested.sync_shadow_vmcs = true;
7614 /* Emulate the VMPTRLD instruction */
7615 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7617 struct vcpu_vmx *vmx = to_vmx(vcpu);
7620 if (!nested_vmx_check_permission(vcpu))
7623 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
7626 if (vmx->nested.current_vmptr != vmptr) {
7627 struct vmcs12 *new_vmcs12;
7629 page = nested_get_page(vcpu, vmptr);
7631 nested_vmx_failInvalid(vcpu);
7632 return kvm_skip_emulated_instruction(vcpu);
7634 new_vmcs12 = kmap(page);
7635 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7637 nested_release_page_clean(page);
7638 nested_vmx_failValid(vcpu,
7639 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7640 return kvm_skip_emulated_instruction(vcpu);
7643 nested_release_vmcs12(vmx);
7644 vmx->nested.current_vmcs12 = new_vmcs12;
7645 vmx->nested.current_vmcs12_page = page;
7647 * Load VMCS12 from guest memory since it is not already
7650 memcpy(vmx->nested.cached_vmcs12,
7651 vmx->nested.current_vmcs12, VMCS12_SIZE);
7652 set_current_vmptr(vmx, vmptr);
7655 nested_vmx_succeed(vcpu);
7656 return kvm_skip_emulated_instruction(vcpu);
7659 /* Emulate the VMPTRST instruction */
7660 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7662 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7663 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7665 struct x86_exception e;
7667 if (!nested_vmx_check_permission(vcpu))
7670 if (get_vmx_mem_address(vcpu, exit_qualification,
7671 vmx_instruction_info, true, &vmcs_gva))
7673 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
7674 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7675 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7677 kvm_inject_page_fault(vcpu, &e);
7680 nested_vmx_succeed(vcpu);
7681 return kvm_skip_emulated_instruction(vcpu);
7684 /* Emulate the INVEPT instruction */
7685 static int handle_invept(struct kvm_vcpu *vcpu)
7687 struct vcpu_vmx *vmx = to_vmx(vcpu);
7688 u32 vmx_instruction_info, types;
7691 struct x86_exception e;
7696 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7697 SECONDARY_EXEC_ENABLE_EPT) ||
7698 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7699 kvm_queue_exception(vcpu, UD_VECTOR);
7703 if (!nested_vmx_check_permission(vcpu))
7706 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
7707 kvm_queue_exception(vcpu, UD_VECTOR);
7711 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7712 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7714 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7716 if (type >= 32 || !(types & (1 << type))) {
7717 nested_vmx_failValid(vcpu,
7718 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7719 return kvm_skip_emulated_instruction(vcpu);
7722 /* According to the Intel VMX instruction reference, the memory
7723 * operand is read even if it isn't needed (e.g., for type==global)
7725 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7726 vmx_instruction_info, false, &gva))
7728 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7729 sizeof(operand), &e)) {
7730 kvm_inject_page_fault(vcpu, &e);
7735 case VMX_EPT_EXTENT_GLOBAL:
7737 * TODO: track mappings and invalidate
7738 * single context requests appropriately
7740 case VMX_EPT_EXTENT_CONTEXT:
7741 kvm_mmu_sync_roots(vcpu);
7742 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7743 nested_vmx_succeed(vcpu);
7750 return kvm_skip_emulated_instruction(vcpu);
7753 static int handle_invvpid(struct kvm_vcpu *vcpu)
7755 struct vcpu_vmx *vmx = to_vmx(vcpu);
7756 u32 vmx_instruction_info;
7757 unsigned long type, types;
7759 struct x86_exception e;
7762 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7763 SECONDARY_EXEC_ENABLE_VPID) ||
7764 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7765 kvm_queue_exception(vcpu, UD_VECTOR);
7769 if (!nested_vmx_check_permission(vcpu))
7772 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7773 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7775 types = (vmx->nested.nested_vmx_vpid_caps &
7776 VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
7778 if (type >= 32 || !(types & (1 << type))) {
7779 nested_vmx_failValid(vcpu,
7780 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7781 return kvm_skip_emulated_instruction(vcpu);
7784 /* according to the intel vmx instruction reference, the memory
7785 * operand is read even if it isn't needed (e.g., for type==global)
7787 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7788 vmx_instruction_info, false, &gva))
7790 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vpid,
7792 kvm_inject_page_fault(vcpu, &e);
7797 case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
7798 case VMX_VPID_EXTENT_SINGLE_CONTEXT:
7799 case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
7801 nested_vmx_failValid(vcpu,
7802 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7803 return kvm_skip_emulated_instruction(vcpu);
7806 case VMX_VPID_EXTENT_ALL_CONTEXT:
7810 return kvm_skip_emulated_instruction(vcpu);
7813 __vmx_flush_tlb(vcpu, vmx->nested.vpid02);
7814 nested_vmx_succeed(vcpu);
7816 return kvm_skip_emulated_instruction(vcpu);
7819 static int handle_pml_full(struct kvm_vcpu *vcpu)
7821 unsigned long exit_qualification;
7823 trace_kvm_pml_full(vcpu->vcpu_id);
7825 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7828 * PML buffer FULL happened while executing iret from NMI,
7829 * "blocked by NMI" bit has to be set before next VM entry.
7831 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7832 cpu_has_virtual_nmis() &&
7833 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7834 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7835 GUEST_INTR_STATE_NMI);
7838 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7839 * here.., and there's no userspace involvement needed for PML.
7844 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
7846 kvm_lapic_expired_hv_timer(vcpu);
7851 * The exit handlers return 1 if the exit was handled fully and guest execution
7852 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
7853 * to be done to userspace and return 0.
7855 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7856 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
7857 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
7858 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
7859 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
7860 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
7861 [EXIT_REASON_CR_ACCESS] = handle_cr,
7862 [EXIT_REASON_DR_ACCESS] = handle_dr,
7863 [EXIT_REASON_CPUID] = handle_cpuid,
7864 [EXIT_REASON_MSR_READ] = handle_rdmsr,
7865 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
7866 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
7867 [EXIT_REASON_HLT] = handle_halt,
7868 [EXIT_REASON_INVD] = handle_invd,
7869 [EXIT_REASON_INVLPG] = handle_invlpg,
7870 [EXIT_REASON_RDPMC] = handle_rdpmc,
7871 [EXIT_REASON_VMCALL] = handle_vmcall,
7872 [EXIT_REASON_VMCLEAR] = handle_vmclear,
7873 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
7874 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
7875 [EXIT_REASON_VMPTRST] = handle_vmptrst,
7876 [EXIT_REASON_VMREAD] = handle_vmread,
7877 [EXIT_REASON_VMRESUME] = handle_vmresume,
7878 [EXIT_REASON_VMWRITE] = handle_vmwrite,
7879 [EXIT_REASON_VMOFF] = handle_vmoff,
7880 [EXIT_REASON_VMON] = handle_vmon,
7881 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
7882 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
7883 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
7884 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
7885 [EXIT_REASON_WBINVD] = handle_wbinvd,
7886 [EXIT_REASON_XSETBV] = handle_xsetbv,
7887 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
7888 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
7889 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
7890 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
7891 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
7892 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
7893 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
7894 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
7895 [EXIT_REASON_INVEPT] = handle_invept,
7896 [EXIT_REASON_INVVPID] = handle_invvpid,
7897 [EXIT_REASON_XSAVES] = handle_xsaves,
7898 [EXIT_REASON_XRSTORS] = handle_xrstors,
7899 [EXIT_REASON_PML_FULL] = handle_pml_full,
7900 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
7903 static const int kvm_vmx_max_exit_handlers =
7904 ARRAY_SIZE(kvm_vmx_exit_handlers);
7906 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7907 struct vmcs12 *vmcs12)
7909 unsigned long exit_qualification;
7910 gpa_t bitmap, last_bitmap;
7915 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7916 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7918 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7920 port = exit_qualification >> 16;
7921 size = (exit_qualification & 7) + 1;
7923 last_bitmap = (gpa_t)-1;
7928 bitmap = vmcs12->io_bitmap_a;
7929 else if (port < 0x10000)
7930 bitmap = vmcs12->io_bitmap_b;
7933 bitmap += (port & 0x7fff) / 8;
7935 if (last_bitmap != bitmap)
7936 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
7938 if (b & (1 << (port & 7)))
7943 last_bitmap = bitmap;
7950 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7951 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7952 * disinterest in the current event (read or write a specific MSR) by using an
7953 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7955 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7956 struct vmcs12 *vmcs12, u32 exit_reason)
7958 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7961 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7965 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7966 * for the four combinations of read/write and low/high MSR numbers.
7967 * First we need to figure out which of the four to use:
7969 bitmap = vmcs12->msr_bitmap;
7970 if (exit_reason == EXIT_REASON_MSR_WRITE)
7972 if (msr_index >= 0xc0000000) {
7973 msr_index -= 0xc0000000;
7977 /* Then read the msr_index'th bit from this bitmap: */
7978 if (msr_index < 1024*8) {
7980 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
7982 return 1 & (b >> (msr_index & 7));
7984 return true; /* let L1 handle the wrong parameter */
7988 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7989 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7990 * intercept (via guest_host_mask etc.) the current event.
7992 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7993 struct vmcs12 *vmcs12)
7995 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7996 int cr = exit_qualification & 15;
7997 int reg = (exit_qualification >> 8) & 15;
7998 unsigned long val = kvm_register_readl(vcpu, reg);
8000 switch ((exit_qualification >> 4) & 3) {
8001 case 0: /* mov to cr */
8004 if (vmcs12->cr0_guest_host_mask &
8005 (val ^ vmcs12->cr0_read_shadow))
8009 if ((vmcs12->cr3_target_count >= 1 &&
8010 vmcs12->cr3_target_value0 == val) ||
8011 (vmcs12->cr3_target_count >= 2 &&
8012 vmcs12->cr3_target_value1 == val) ||
8013 (vmcs12->cr3_target_count >= 3 &&
8014 vmcs12->cr3_target_value2 == val) ||
8015 (vmcs12->cr3_target_count >= 4 &&
8016 vmcs12->cr3_target_value3 == val))
8018 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
8022 if (vmcs12->cr4_guest_host_mask &
8023 (vmcs12->cr4_read_shadow ^ val))
8027 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
8033 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
8034 (vmcs12->cr0_read_shadow & X86_CR0_TS))
8037 case 1: /* mov from cr */
8040 if (vmcs12->cpu_based_vm_exec_control &
8041 CPU_BASED_CR3_STORE_EXITING)
8045 if (vmcs12->cpu_based_vm_exec_control &
8046 CPU_BASED_CR8_STORE_EXITING)
8053 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
8054 * cr0. Other attempted changes are ignored, with no exit.
8056 if (vmcs12->cr0_guest_host_mask & 0xe &
8057 (val ^ vmcs12->cr0_read_shadow))
8059 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
8060 !(vmcs12->cr0_read_shadow & 0x1) &&
8069 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
8070 * should handle it ourselves in L0 (and then continue L2). Only call this
8071 * when in is_guest_mode (L2).
8073 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
8075 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8076 struct vcpu_vmx *vmx = to_vmx(vcpu);
8077 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8078 u32 exit_reason = vmx->exit_reason;
8080 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
8081 vmcs_readl(EXIT_QUALIFICATION),
8082 vmx->idt_vectoring_info,
8084 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8087 if (vmx->nested.nested_run_pending)
8090 if (unlikely(vmx->fail)) {
8091 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
8092 vmcs_read32(VM_INSTRUCTION_ERROR));
8096 switch (exit_reason) {
8097 case EXIT_REASON_EXCEPTION_NMI:
8098 if (is_nmi(intr_info))
8100 else if (is_page_fault(intr_info))
8102 else if (is_no_device(intr_info) &&
8103 !(vmcs12->guest_cr0 & X86_CR0_TS))
8105 else if (is_debug(intr_info) &&
8107 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
8109 else if (is_breakpoint(intr_info) &&
8110 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
8112 return vmcs12->exception_bitmap &
8113 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
8114 case EXIT_REASON_EXTERNAL_INTERRUPT:
8116 case EXIT_REASON_TRIPLE_FAULT:
8118 case EXIT_REASON_PENDING_INTERRUPT:
8119 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
8120 case EXIT_REASON_NMI_WINDOW:
8121 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
8122 case EXIT_REASON_TASK_SWITCH:
8124 case EXIT_REASON_CPUID:
8126 case EXIT_REASON_HLT:
8127 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
8128 case EXIT_REASON_INVD:
8130 case EXIT_REASON_INVLPG:
8131 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8132 case EXIT_REASON_RDPMC:
8133 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
8134 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
8135 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
8136 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
8137 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
8138 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
8139 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
8140 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
8141 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
8143 * VMX instructions trap unconditionally. This allows L1 to
8144 * emulate them for its L2 guest, i.e., allows 3-level nesting!
8147 case EXIT_REASON_CR_ACCESS:
8148 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
8149 case EXIT_REASON_DR_ACCESS:
8150 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
8151 case EXIT_REASON_IO_INSTRUCTION:
8152 return nested_vmx_exit_handled_io(vcpu, vmcs12);
8153 case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
8154 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
8155 case EXIT_REASON_MSR_READ:
8156 case EXIT_REASON_MSR_WRITE:
8157 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
8158 case EXIT_REASON_INVALID_STATE:
8160 case EXIT_REASON_MWAIT_INSTRUCTION:
8161 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
8162 case EXIT_REASON_MONITOR_TRAP_FLAG:
8163 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
8164 case EXIT_REASON_MONITOR_INSTRUCTION:
8165 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
8166 case EXIT_REASON_PAUSE_INSTRUCTION:
8167 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
8168 nested_cpu_has2(vmcs12,
8169 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
8170 case EXIT_REASON_MCE_DURING_VMENTRY:
8172 case EXIT_REASON_TPR_BELOW_THRESHOLD:
8173 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
8174 case EXIT_REASON_APIC_ACCESS:
8175 return nested_cpu_has2(vmcs12,
8176 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
8177 case EXIT_REASON_APIC_WRITE:
8178 case EXIT_REASON_EOI_INDUCED:
8179 /* apic_write and eoi_induced should exit unconditionally. */
8181 case EXIT_REASON_EPT_VIOLATION:
8183 * L0 always deals with the EPT violation. If nested EPT is
8184 * used, and the nested mmu code discovers that the address is
8185 * missing in the guest EPT table (EPT12), the EPT violation
8186 * will be injected with nested_ept_inject_page_fault()
8189 case EXIT_REASON_EPT_MISCONFIG:
8191 * L2 never uses directly L1's EPT, but rather L0's own EPT
8192 * table (shadow on EPT) or a merged EPT table that L0 built
8193 * (EPT on EPT). So any problems with the structure of the
8194 * table is L0's fault.
8197 case EXIT_REASON_WBINVD:
8198 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
8199 case EXIT_REASON_XSETBV:
8201 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
8203 * This should never happen, since it is not possible to
8204 * set XSS to a non-zero value---neither in L1 nor in L2.
8205 * If if it were, XSS would have to be checked against
8206 * the XSS exit bitmap in vmcs12.
8208 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
8209 case EXIT_REASON_PREEMPTION_TIMER:
8216 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
8218 *info1 = vmcs_readl(EXIT_QUALIFICATION);
8219 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
8222 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
8225 __free_page(vmx->pml_pg);
8230 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
8232 struct vcpu_vmx *vmx = to_vmx(vcpu);
8236 pml_idx = vmcs_read16(GUEST_PML_INDEX);
8238 /* Do nothing if PML buffer is empty */
8239 if (pml_idx == (PML_ENTITY_NUM - 1))
8242 /* PML index always points to next available PML buffer entity */
8243 if (pml_idx >= PML_ENTITY_NUM)
8248 pml_buf = page_address(vmx->pml_pg);
8249 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
8252 gpa = pml_buf[pml_idx];
8253 WARN_ON(gpa & (PAGE_SIZE - 1));
8254 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
8257 /* reset PML index */
8258 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8262 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8263 * Called before reporting dirty_bitmap to userspace.
8265 static void kvm_flush_pml_buffers(struct kvm *kvm)
8268 struct kvm_vcpu *vcpu;
8270 * We only need to kick vcpu out of guest mode here, as PML buffer
8271 * is flushed at beginning of all VMEXITs, and it's obvious that only
8272 * vcpus running in guest are possible to have unflushed GPAs in PML
8275 kvm_for_each_vcpu(i, vcpu, kvm)
8276 kvm_vcpu_kick(vcpu);
8279 static void vmx_dump_sel(char *name, uint32_t sel)
8281 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8282 name, vmcs_read32(sel),
8283 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
8284 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
8285 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
8288 static void vmx_dump_dtsel(char *name, uint32_t limit)
8290 pr_err("%s limit=0x%08x, base=0x%016lx\n",
8291 name, vmcs_read32(limit),
8292 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
8295 static void dump_vmcs(void)
8297 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
8298 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
8299 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
8300 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
8301 u32 secondary_exec_control = 0;
8302 unsigned long cr4 = vmcs_readl(GUEST_CR4);
8303 u64 efer = vmcs_read64(GUEST_IA32_EFER);
8306 if (cpu_has_secondary_exec_ctrls())
8307 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8309 pr_err("*** Guest State ***\n");
8310 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8311 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
8312 vmcs_readl(CR0_GUEST_HOST_MASK));
8313 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8314 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
8315 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
8316 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
8317 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
8319 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
8320 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
8321 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
8322 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
8324 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
8325 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
8326 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
8327 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
8328 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8329 vmcs_readl(GUEST_SYSENTER_ESP),
8330 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8331 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
8332 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
8333 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
8334 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
8335 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
8336 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
8337 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8338 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8339 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8340 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
8341 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8342 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8343 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8344 efer, vmcs_read64(GUEST_IA32_PAT));
8345 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
8346 vmcs_read64(GUEST_IA32_DEBUGCTL),
8347 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8348 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8349 pr_err("PerfGlobCtl = 0x%016llx\n",
8350 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8351 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8352 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8353 pr_err("Interruptibility = %08x ActivityState = %08x\n",
8354 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8355 vmcs_read32(GUEST_ACTIVITY_STATE));
8356 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8357 pr_err("InterruptStatus = %04x\n",
8358 vmcs_read16(GUEST_INTR_STATUS));
8360 pr_err("*** Host State ***\n");
8361 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
8362 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8363 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8364 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8365 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8366 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8367 vmcs_read16(HOST_TR_SELECTOR));
8368 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8369 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8370 vmcs_readl(HOST_TR_BASE));
8371 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8372 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8373 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8374 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8375 vmcs_readl(HOST_CR4));
8376 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8377 vmcs_readl(HOST_IA32_SYSENTER_ESP),
8378 vmcs_read32(HOST_IA32_SYSENTER_CS),
8379 vmcs_readl(HOST_IA32_SYSENTER_EIP));
8380 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8381 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8382 vmcs_read64(HOST_IA32_EFER),
8383 vmcs_read64(HOST_IA32_PAT));
8384 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8385 pr_err("PerfGlobCtl = 0x%016llx\n",
8386 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8388 pr_err("*** Control State ***\n");
8389 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8390 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8391 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8392 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8393 vmcs_read32(EXCEPTION_BITMAP),
8394 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8395 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8396 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8397 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8398 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8399 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8400 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8401 vmcs_read32(VM_EXIT_INTR_INFO),
8402 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8403 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8404 pr_err(" reason=%08x qualification=%016lx\n",
8405 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8406 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8407 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8408 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8409 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8410 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8411 pr_err("TSC Multiplier = 0x%016llx\n",
8412 vmcs_read64(TSC_MULTIPLIER));
8413 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8414 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8415 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8416 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8417 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8418 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8419 n = vmcs_read32(CR3_TARGET_COUNT);
8420 for (i = 0; i + 1 < n; i += 4)
8421 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8422 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8423 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8425 pr_err("CR3 target%u=%016lx\n",
8426 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8427 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8428 pr_err("PLE Gap=%08x Window=%08x\n",
8429 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8430 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8431 pr_err("Virtual processor ID = 0x%04x\n",
8432 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8436 * The guest has exited. See if we can fix it or if we need userspace
8439 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8441 struct vcpu_vmx *vmx = to_vmx(vcpu);
8442 u32 exit_reason = vmx->exit_reason;
8443 u32 vectoring_info = vmx->idt_vectoring_info;
8445 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8446 vcpu->arch.gpa_available = false;
8449 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8450 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8451 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8452 * mode as if vcpus is in root mode, the PML buffer must has been
8456 vmx_flush_pml_buffer(vcpu);
8458 /* If guest state is invalid, start emulating */
8459 if (vmx->emulation_required)
8460 return handle_invalid_guest_state(vcpu);
8462 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
8463 nested_vmx_vmexit(vcpu, exit_reason,
8464 vmcs_read32(VM_EXIT_INTR_INFO),
8465 vmcs_readl(EXIT_QUALIFICATION));
8469 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8471 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8472 vcpu->run->fail_entry.hardware_entry_failure_reason
8477 if (unlikely(vmx->fail)) {
8478 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8479 vcpu->run->fail_entry.hardware_entry_failure_reason
8480 = vmcs_read32(VM_INSTRUCTION_ERROR);
8486 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8487 * delivery event since it indicates guest is accessing MMIO.
8488 * The vm-exit can be triggered again after return to guest that
8489 * will cause infinite loop.
8491 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8492 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8493 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8494 exit_reason != EXIT_REASON_PML_FULL &&
8495 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8496 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8497 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8498 vcpu->run->internal.ndata = 2;
8499 vcpu->run->internal.data[0] = vectoring_info;
8500 vcpu->run->internal.data[1] = exit_reason;
8504 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
8505 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
8506 get_vmcs12(vcpu))))) {
8507 if (vmx_interrupt_allowed(vcpu)) {
8508 vmx->soft_vnmi_blocked = 0;
8509 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
8510 vcpu->arch.nmi_pending) {
8512 * This CPU don't support us in finding the end of an
8513 * NMI-blocked window if the guest runs with IRQs
8514 * disabled. So we pull the trigger after 1 s of
8515 * futile waiting, but inform the user about this.
8517 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
8518 "state on VCPU %d after 1 s timeout\n",
8519 __func__, vcpu->vcpu_id);
8520 vmx->soft_vnmi_blocked = 0;
8524 if (exit_reason < kvm_vmx_max_exit_handlers
8525 && kvm_vmx_exit_handlers[exit_reason])
8526 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8528 WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_reason);
8529 kvm_queue_exception(vcpu, UD_VECTOR);
8534 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8536 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8538 if (is_guest_mode(vcpu) &&
8539 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8542 if (irr == -1 || tpr < irr) {
8543 vmcs_write32(TPR_THRESHOLD, 0);
8547 vmcs_write32(TPR_THRESHOLD, irr);
8550 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8552 u32 sec_exec_control;
8554 /* Postpone execution until vmcs01 is the current VMCS. */
8555 if (is_guest_mode(vcpu)) {
8556 to_vmx(vcpu)->nested.change_vmcs01_virtual_x2apic_mode = true;
8560 if (!cpu_has_vmx_virtualize_x2apic_mode())
8563 if (!cpu_need_tpr_shadow(vcpu))
8566 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8569 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8570 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8572 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8573 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8575 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8577 vmx_set_msr_bitmap(vcpu);
8580 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8582 struct vcpu_vmx *vmx = to_vmx(vcpu);
8585 * Currently we do not handle the nested case where L2 has an
8586 * APIC access page of its own; that page is still pinned.
8587 * Hence, we skip the case where the VCPU is in guest mode _and_
8588 * L1 prepared an APIC access page for L2.
8590 * For the case where L1 and L2 share the same APIC access page
8591 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8592 * in the vmcs12), this function will only update either the vmcs01
8593 * or the vmcs02. If the former, the vmcs02 will be updated by
8594 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8595 * the next L2->L1 exit.
8597 if (!is_guest_mode(vcpu) ||
8598 !nested_cpu_has2(get_vmcs12(&vmx->vcpu),
8599 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
8600 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8603 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
8611 status = vmcs_read16(GUEST_INTR_STATUS);
8613 if (max_isr != old) {
8615 status |= max_isr << 8;
8616 vmcs_write16(GUEST_INTR_STATUS, status);
8620 static void vmx_set_rvi(int vector)
8628 status = vmcs_read16(GUEST_INTR_STATUS);
8629 old = (u8)status & 0xff;
8630 if ((u8)vector != old) {
8632 status |= (u8)vector;
8633 vmcs_write16(GUEST_INTR_STATUS, status);
8637 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8639 if (!is_guest_mode(vcpu)) {
8640 vmx_set_rvi(max_irr);
8648 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8651 if (nested_exit_on_intr(vcpu))
8655 * Else, fall back to pre-APICv interrupt injection since L2
8656 * is run without virtual interrupt delivery.
8658 if (!kvm_event_needs_reinjection(vcpu) &&
8659 vmx_interrupt_allowed(vcpu)) {
8660 kvm_queue_interrupt(vcpu, max_irr, false);
8661 vmx_inject_irq(vcpu);
8665 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
8667 struct vcpu_vmx *vmx = to_vmx(vcpu);
8670 WARN_ON(!vcpu->arch.apicv_active);
8671 if (pi_test_on(&vmx->pi_desc)) {
8672 pi_clear_on(&vmx->pi_desc);
8674 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
8675 * But on x86 this is just a compiler barrier anyway.
8677 smp_mb__after_atomic();
8678 max_irr = kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
8680 max_irr = kvm_lapic_find_highest_irr(vcpu);
8682 vmx_hwapic_irr_update(vcpu, max_irr);
8686 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
8688 if (!kvm_vcpu_apicv_active(vcpu))
8691 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8692 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8693 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8694 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8697 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
8699 struct vcpu_vmx *vmx = to_vmx(vcpu);
8701 pi_clear_on(&vmx->pi_desc);
8702 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
8705 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8709 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8710 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
8713 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8714 exit_intr_info = vmx->exit_intr_info;
8716 /* Handle machine checks before interrupts are enabled */
8717 if (is_machine_check(exit_intr_info))
8718 kvm_machine_check();
8720 /* We need to handle NMIs before interrupts are enabled */
8721 if (is_nmi(exit_intr_info)) {
8722 kvm_before_handle_nmi(&vmx->vcpu);
8724 kvm_after_handle_nmi(&vmx->vcpu);
8728 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8730 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8731 register void *__sp asm(_ASM_SP);
8733 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8734 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8735 unsigned int vector;
8736 unsigned long entry;
8738 struct vcpu_vmx *vmx = to_vmx(vcpu);
8739 #ifdef CONFIG_X86_64
8743 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8744 desc = (gate_desc *)vmx->host_idt_base + vector;
8745 entry = gate_offset(*desc);
8747 #ifdef CONFIG_X86_64
8748 "mov %%" _ASM_SP ", %[sp]\n\t"
8749 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8754 __ASM_SIZE(push) " $%c[cs]\n\t"
8755 "call *%[entry]\n\t"
8757 #ifdef CONFIG_X86_64
8763 [ss]"i"(__KERNEL_DS),
8764 [cs]"i"(__KERNEL_CS)
8769 static bool vmx_has_high_real_mode_segbase(void)
8771 return enable_unrestricted_guest || emulate_invalid_guest_state;
8774 static bool vmx_mpx_supported(void)
8776 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8777 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8780 static bool vmx_xsaves_supported(void)
8782 return vmcs_config.cpu_based_2nd_exec_ctrl &
8783 SECONDARY_EXEC_XSAVES;
8786 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8791 bool idtv_info_valid;
8793 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8795 if (cpu_has_virtual_nmis()) {
8796 if (vmx->nmi_known_unmasked)
8799 * Can't use vmx->exit_intr_info since we're not sure what
8800 * the exit reason is.
8802 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8803 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8804 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8806 * SDM 3: 27.7.1.2 (September 2008)
8807 * Re-set bit "block by NMI" before VM entry if vmexit caused by
8808 * a guest IRET fault.
8809 * SDM 3: 23.2.2 (September 2008)
8810 * Bit 12 is undefined in any of the following cases:
8811 * If the VM exit sets the valid bit in the IDT-vectoring
8812 * information field.
8813 * If the VM exit is due to a double fault.
8815 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8816 vector != DF_VECTOR && !idtv_info_valid)
8817 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8818 GUEST_INTR_STATE_NMI);
8820 vmx->nmi_known_unmasked =
8821 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8822 & GUEST_INTR_STATE_NMI);
8823 } else if (unlikely(vmx->soft_vnmi_blocked))
8824 vmx->vnmi_blocked_time +=
8825 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
8828 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8829 u32 idt_vectoring_info,
8830 int instr_len_field,
8831 int error_code_field)
8835 bool idtv_info_valid;
8837 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8839 vcpu->arch.nmi_injected = false;
8840 kvm_clear_exception_queue(vcpu);
8841 kvm_clear_interrupt_queue(vcpu);
8843 if (!idtv_info_valid)
8846 kvm_make_request(KVM_REQ_EVENT, vcpu);
8848 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8849 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8852 case INTR_TYPE_NMI_INTR:
8853 vcpu->arch.nmi_injected = true;
8855 * SDM 3: 27.7.1.2 (September 2008)
8856 * Clear bit "block by NMI" before VM entry if a NMI
8859 vmx_set_nmi_mask(vcpu, false);
8861 case INTR_TYPE_SOFT_EXCEPTION:
8862 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8864 case INTR_TYPE_HARD_EXCEPTION:
8865 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8866 u32 err = vmcs_read32(error_code_field);
8867 kvm_requeue_exception_e(vcpu, vector, err);
8869 kvm_requeue_exception(vcpu, vector);
8871 case INTR_TYPE_SOFT_INTR:
8872 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8874 case INTR_TYPE_EXT_INTR:
8875 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8882 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8884 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8885 VM_EXIT_INSTRUCTION_LEN,
8886 IDT_VECTORING_ERROR_CODE);
8889 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8891 __vmx_complete_interrupts(vcpu,
8892 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8893 VM_ENTRY_INSTRUCTION_LEN,
8894 VM_ENTRY_EXCEPTION_ERROR_CODE);
8896 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8899 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8902 struct perf_guest_switch_msr *msrs;
8904 msrs = perf_guest_get_msrs(&nr_msrs);
8909 for (i = 0; i < nr_msrs; i++)
8910 if (msrs[i].host == msrs[i].guest)
8911 clear_atomic_switch_msr(vmx, msrs[i].msr);
8913 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8917 static void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
8919 struct vcpu_vmx *vmx = to_vmx(vcpu);
8923 if (vmx->hv_deadline_tsc == -1)
8927 if (vmx->hv_deadline_tsc > tscl)
8928 /* sure to be 32 bit only because checked on set_hv_timer */
8929 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
8930 cpu_preemption_timer_multi);
8934 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
8937 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8939 struct vcpu_vmx *vmx = to_vmx(vcpu);
8940 unsigned long debugctlmsr, cr4;
8942 /* Record the guest's net vcpu time for enforced NMI injections. */
8943 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
8944 vmx->entry_time = ktime_get();
8946 /* Don't enter VMX if guest state is invalid, let the exit handler
8947 start emulation until we arrive back to a valid state */
8948 if (vmx->emulation_required)
8951 if (vmx->ple_window_dirty) {
8952 vmx->ple_window_dirty = false;
8953 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8956 if (vmx->nested.sync_shadow_vmcs) {
8957 copy_vmcs12_to_shadow(vmx);
8958 vmx->nested.sync_shadow_vmcs = false;
8961 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8962 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
8963 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
8964 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
8966 cr4 = cr4_read_shadow();
8967 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
8968 vmcs_writel(HOST_CR4, cr4);
8969 vmx->host_state.vmcs_host_cr4 = cr4;
8972 /* When single-stepping over STI and MOV SS, we must clear the
8973 * corresponding interruptibility bits in the guest state. Otherwise
8974 * vmentry fails as it then expects bit 14 (BS) in pending debug
8975 * exceptions being set, but that's not correct for the guest debugging
8977 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8978 vmx_set_interrupt_shadow(vcpu, 0);
8980 if (vmx->guest_pkru_valid)
8981 __write_pkru(vmx->guest_pkru);
8983 atomic_switch_perf_msrs(vmx);
8984 debugctlmsr = get_debugctlmsr();
8986 vmx_arm_hv_timer(vcpu);
8988 vmx->__launched = vmx->loaded_vmcs->launched;
8990 /* Store host registers */
8991 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
8992 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
8993 "push %%" _ASM_CX " \n\t"
8994 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8996 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8997 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
8999 /* Reload cr2 if changed */
9000 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
9001 "mov %%cr2, %%" _ASM_DX " \n\t"
9002 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
9004 "mov %%" _ASM_AX", %%cr2 \n\t"
9006 /* Check if vmlaunch of vmresume is needed */
9007 "cmpl $0, %c[launched](%0) \n\t"
9008 /* Load guest registers. Don't clobber flags. */
9009 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
9010 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
9011 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
9012 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
9013 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
9014 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
9015 #ifdef CONFIG_X86_64
9016 "mov %c[r8](%0), %%r8 \n\t"
9017 "mov %c[r9](%0), %%r9 \n\t"
9018 "mov %c[r10](%0), %%r10 \n\t"
9019 "mov %c[r11](%0), %%r11 \n\t"
9020 "mov %c[r12](%0), %%r12 \n\t"
9021 "mov %c[r13](%0), %%r13 \n\t"
9022 "mov %c[r14](%0), %%r14 \n\t"
9023 "mov %c[r15](%0), %%r15 \n\t"
9025 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
9027 /* Enter guest mode */
9029 __ex(ASM_VMX_VMLAUNCH) "\n\t"
9031 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
9033 /* Save guest registers, load host registers, keep flags */
9034 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
9036 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
9037 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
9038 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
9039 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
9040 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
9041 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
9042 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
9043 #ifdef CONFIG_X86_64
9044 "mov %%r8, %c[r8](%0) \n\t"
9045 "mov %%r9, %c[r9](%0) \n\t"
9046 "mov %%r10, %c[r10](%0) \n\t"
9047 "mov %%r11, %c[r11](%0) \n\t"
9048 "mov %%r12, %c[r12](%0) \n\t"
9049 "mov %%r13, %c[r13](%0) \n\t"
9050 "mov %%r14, %c[r14](%0) \n\t"
9051 "mov %%r15, %c[r15](%0) \n\t"
9053 "mov %%cr2, %%" _ASM_AX " \n\t"
9054 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
9056 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
9057 "setbe %c[fail](%0) \n\t"
9058 ".pushsection .rodata \n\t"
9059 ".global vmx_return \n\t"
9060 "vmx_return: " _ASM_PTR " 2b \n\t"
9062 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
9063 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
9064 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
9065 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
9066 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
9067 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
9068 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
9069 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
9070 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
9071 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
9072 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
9073 #ifdef CONFIG_X86_64
9074 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
9075 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
9076 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
9077 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
9078 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
9079 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
9080 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
9081 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
9083 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
9084 [wordsize]"i"(sizeof(ulong))
9086 #ifdef CONFIG_X86_64
9087 , "rax", "rbx", "rdi", "rsi"
9088 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
9090 , "eax", "ebx", "edi", "esi"
9094 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
9096 update_debugctlmsr(debugctlmsr);
9098 #ifndef CONFIG_X86_64
9100 * The sysexit path does not restore ds/es, so we must set them to
9101 * a reasonable value ourselves.
9103 * We can't defer this to vmx_load_host_state() since that function
9104 * may be executed in interrupt context, which saves and restore segments
9105 * around it, nullifying its effect.
9107 loadsegment(ds, __USER_DS);
9108 loadsegment(es, __USER_DS);
9111 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
9112 | (1 << VCPU_EXREG_RFLAGS)
9113 | (1 << VCPU_EXREG_PDPTR)
9114 | (1 << VCPU_EXREG_SEGMENTS)
9115 | (1 << VCPU_EXREG_CR3));
9116 vcpu->arch.regs_dirty = 0;
9118 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9120 vmx->loaded_vmcs->launched = 1;
9122 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
9125 * eager fpu is enabled if PKEY is supported and CR4 is switched
9126 * back on host, so it is safe to read guest PKRU from current
9129 if (boot_cpu_has(X86_FEATURE_OSPKE)) {
9130 vmx->guest_pkru = __read_pkru();
9131 if (vmx->guest_pkru != vmx->host_pkru) {
9132 vmx->guest_pkru_valid = true;
9133 __write_pkru(vmx->host_pkru);
9135 vmx->guest_pkru_valid = false;
9139 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
9140 * we did not inject a still-pending event to L1 now because of
9141 * nested_run_pending, we need to re-enable this bit.
9143 if (vmx->nested.nested_run_pending)
9144 kvm_make_request(KVM_REQ_EVENT, vcpu);
9146 vmx->nested.nested_run_pending = 0;
9148 vmx_complete_atomic_exit(vmx);
9149 vmx_recover_nmi_blocking(vmx);
9150 vmx_complete_interrupts(vmx);
9153 static void vmx_load_vmcs01(struct kvm_vcpu *vcpu)
9155 struct vcpu_vmx *vmx = to_vmx(vcpu);
9158 if (vmx->loaded_vmcs == &vmx->vmcs01)
9162 vmx->loaded_vmcs = &vmx->vmcs01;
9164 vmx_vcpu_load(vcpu, cpu);
9170 * Ensure that the current vmcs of the logical processor is the
9171 * vmcs01 of the vcpu before calling free_nested().
9173 static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
9175 struct vcpu_vmx *vmx = to_vmx(vcpu);
9178 r = vcpu_load(vcpu);
9180 vmx_load_vmcs01(vcpu);
9185 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
9187 struct vcpu_vmx *vmx = to_vmx(vcpu);
9190 vmx_destroy_pml_buffer(vmx);
9191 free_vpid(vmx->vpid);
9192 leave_guest_mode(vcpu);
9193 vmx_free_vcpu_nested(vcpu);
9194 free_loaded_vmcs(vmx->loaded_vmcs);
9195 kfree(vmx->guest_msrs);
9196 kvm_vcpu_uninit(vcpu);
9197 kmem_cache_free(kvm_vcpu_cache, vmx);
9200 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
9203 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
9207 return ERR_PTR(-ENOMEM);
9209 vmx->vpid = allocate_vpid();
9211 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
9218 * If PML is turned on, failure on enabling PML just results in failure
9219 * of creating the vcpu, therefore we can simplify PML logic (by
9220 * avoiding dealing with cases, such as enabling PML partially on vcpus
9221 * for the guest, etc.
9224 vmx->pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
9229 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
9230 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
9233 if (!vmx->guest_msrs)
9236 vmx->loaded_vmcs = &vmx->vmcs01;
9237 vmx->loaded_vmcs->vmcs = alloc_vmcs();
9238 vmx->loaded_vmcs->shadow_vmcs = NULL;
9239 if (!vmx->loaded_vmcs->vmcs)
9242 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
9243 loaded_vmcs_init(vmx->loaded_vmcs);
9248 vmx_vcpu_load(&vmx->vcpu, cpu);
9249 vmx->vcpu.cpu = cpu;
9250 err = vmx_vcpu_setup(vmx);
9251 vmx_vcpu_put(&vmx->vcpu);
9255 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9256 err = alloc_apic_access_page(kvm);
9262 if (!kvm->arch.ept_identity_map_addr)
9263 kvm->arch.ept_identity_map_addr =
9264 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
9265 err = init_rmode_identity_map(kvm);
9271 nested_vmx_setup_ctls_msrs(vmx);
9272 vmx->nested.vpid02 = allocate_vpid();
9275 vmx->nested.posted_intr_nv = -1;
9276 vmx->nested.current_vmptr = -1ull;
9277 vmx->nested.current_vmcs12 = NULL;
9279 vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
9284 free_vpid(vmx->nested.vpid02);
9285 free_loaded_vmcs(vmx->loaded_vmcs);
9287 kfree(vmx->guest_msrs);
9289 vmx_destroy_pml_buffer(vmx);
9291 kvm_vcpu_uninit(&vmx->vcpu);
9293 free_vpid(vmx->vpid);
9294 kmem_cache_free(kvm_vcpu_cache, vmx);
9295 return ERR_PTR(err);
9298 static void __init vmx_check_processor_compat(void *rtn)
9300 struct vmcs_config vmcs_conf;
9303 if (setup_vmcs_config(&vmcs_conf) < 0)
9305 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
9306 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
9307 smp_processor_id());
9312 static int get_ept_level(void)
9314 return VMX_EPT_DEFAULT_GAW + 1;
9317 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
9322 /* For VT-d and EPT combination
9323 * 1. MMIO: always map as UC
9325 * a. VT-d without snooping control feature: can't guarantee the
9326 * result, try to trust guest.
9327 * b. VT-d with snooping control feature: snooping control feature of
9328 * VT-d engine can guarantee the cache correctness. Just set it
9329 * to WB to keep consistent with host. So the same as item 3.
9330 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9331 * consistent with host MTRR
9334 cache = MTRR_TYPE_UNCACHABLE;
9338 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
9339 ipat = VMX_EPT_IPAT_BIT;
9340 cache = MTRR_TYPE_WRBACK;
9344 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
9345 ipat = VMX_EPT_IPAT_BIT;
9346 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
9347 cache = MTRR_TYPE_WRBACK;
9349 cache = MTRR_TYPE_UNCACHABLE;
9353 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
9356 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
9359 static int vmx_get_lpage_level(void)
9361 if (enable_ept && !cpu_has_vmx_ept_1g_page())
9362 return PT_DIRECTORY_LEVEL;
9364 /* For shadow and EPT supported 1GB page */
9365 return PT_PDPE_LEVEL;
9368 static void vmcs_set_secondary_exec_control(u32 new_ctl)
9371 * These bits in the secondary execution controls field
9372 * are dynamic, the others are mostly based on the hypervisor
9373 * architecture and the guest's CPUID. Do not touch the
9377 SECONDARY_EXEC_SHADOW_VMCS |
9378 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
9379 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9381 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9383 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
9384 (new_ctl & ~mask) | (cur_ctl & mask));
9388 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
9389 * (indicating "allowed-1") if they are supported in the guest's CPUID.
9391 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
9393 struct vcpu_vmx *vmx = to_vmx(vcpu);
9394 struct kvm_cpuid_entry2 *entry;
9396 vmx->nested.nested_vmx_cr0_fixed1 = 0xffffffff;
9397 vmx->nested.nested_vmx_cr4_fixed1 = X86_CR4_PCE;
9399 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
9400 if (entry && (entry->_reg & (_cpuid_mask))) \
9401 vmx->nested.nested_vmx_cr4_fixed1 |= (_cr4_mask); \
9404 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
9405 cr4_fixed1_update(X86_CR4_VME, edx, bit(X86_FEATURE_VME));
9406 cr4_fixed1_update(X86_CR4_PVI, edx, bit(X86_FEATURE_VME));
9407 cr4_fixed1_update(X86_CR4_TSD, edx, bit(X86_FEATURE_TSC));
9408 cr4_fixed1_update(X86_CR4_DE, edx, bit(X86_FEATURE_DE));
9409 cr4_fixed1_update(X86_CR4_PSE, edx, bit(X86_FEATURE_PSE));
9410 cr4_fixed1_update(X86_CR4_PAE, edx, bit(X86_FEATURE_PAE));
9411 cr4_fixed1_update(X86_CR4_MCE, edx, bit(X86_FEATURE_MCE));
9412 cr4_fixed1_update(X86_CR4_PGE, edx, bit(X86_FEATURE_PGE));
9413 cr4_fixed1_update(X86_CR4_OSFXSR, edx, bit(X86_FEATURE_FXSR));
9414 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
9415 cr4_fixed1_update(X86_CR4_VMXE, ecx, bit(X86_FEATURE_VMX));
9416 cr4_fixed1_update(X86_CR4_SMXE, ecx, bit(X86_FEATURE_SMX));
9417 cr4_fixed1_update(X86_CR4_PCIDE, ecx, bit(X86_FEATURE_PCID));
9418 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, bit(X86_FEATURE_XSAVE));
9420 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9421 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, bit(X86_FEATURE_FSGSBASE));
9422 cr4_fixed1_update(X86_CR4_SMEP, ebx, bit(X86_FEATURE_SMEP));
9423 cr4_fixed1_update(X86_CR4_SMAP, ebx, bit(X86_FEATURE_SMAP));
9424 cr4_fixed1_update(X86_CR4_PKE, ecx, bit(X86_FEATURE_PKU));
9425 /* TODO: Use X86_CR4_UMIP and X86_FEATURE_UMIP macros */
9426 cr4_fixed1_update(bit(11), ecx, bit(2));
9428 #undef cr4_fixed1_update
9431 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
9433 struct kvm_cpuid_entry2 *best;
9434 struct vcpu_vmx *vmx = to_vmx(vcpu);
9435 u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx);
9437 if (vmx_rdtscp_supported()) {
9438 bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu);
9439 if (!rdtscp_enabled)
9440 secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP;
9444 vmx->nested.nested_vmx_secondary_ctls_high |=
9445 SECONDARY_EXEC_RDTSCP;
9447 vmx->nested.nested_vmx_secondary_ctls_high &=
9448 ~SECONDARY_EXEC_RDTSCP;
9452 /* Exposing INVPCID only when PCID is exposed */
9453 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9454 if (vmx_invpcid_supported() &&
9455 (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) ||
9456 !guest_cpuid_has_pcid(vcpu))) {
9457 secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID;
9460 best->ebx &= ~bit(X86_FEATURE_INVPCID);
9463 if (cpu_has_secondary_exec_ctrls())
9464 vmcs_set_secondary_exec_control(secondary_exec_ctl);
9466 if (nested_vmx_allowed(vcpu))
9467 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
9468 FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9470 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
9471 ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9473 if (nested_vmx_allowed(vcpu))
9474 nested_vmx_cr_fixed1_bits_update(vcpu);
9477 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9479 if (func == 1 && nested)
9480 entry->ecx |= bit(X86_FEATURE_VMX);
9483 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9484 struct x86_exception *fault)
9486 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9489 if (fault->error_code & PFERR_RSVD_MASK)
9490 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9492 exit_reason = EXIT_REASON_EPT_VIOLATION;
9493 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
9494 vmcs12->guest_physical_address = fault->address;
9497 /* Callbacks for nested_ept_init_mmu_context: */
9499 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9501 /* return the page table to be shadowed - in our case, EPT12 */
9502 return get_vmcs12(vcpu)->ept_pointer;
9505 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9507 WARN_ON(mmu_is_nested(vcpu));
9508 kvm_init_shadow_ept_mmu(vcpu,
9509 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
9510 VMX_EPT_EXECUTE_ONLY_BIT);
9511 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
9512 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
9513 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
9515 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
9518 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
9520 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
9523 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9526 bool inequality, bit;
9528 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9530 (error_code & vmcs12->page_fault_error_code_mask) !=
9531 vmcs12->page_fault_error_code_match;
9532 return inequality ^ bit;
9535 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9536 struct x86_exception *fault)
9538 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9540 WARN_ON(!is_guest_mode(vcpu));
9542 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code))
9543 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
9544 vmcs_read32(VM_EXIT_INTR_INFO),
9545 vmcs_readl(EXIT_QUALIFICATION));
9547 kvm_inject_page_fault(vcpu, fault);
9550 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9551 struct vmcs12 *vmcs12);
9553 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9554 struct vmcs12 *vmcs12)
9556 struct vcpu_vmx *vmx = to_vmx(vcpu);
9559 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9561 * Translate L1 physical address to host physical
9562 * address for vmcs02. Keep the page pinned, so this
9563 * physical address remains valid. We keep a reference
9564 * to it so we can release it later.
9566 if (vmx->nested.apic_access_page) /* shouldn't happen */
9567 nested_release_page(vmx->nested.apic_access_page);
9568 vmx->nested.apic_access_page =
9569 nested_get_page(vcpu, vmcs12->apic_access_addr);
9571 * If translation failed, no matter: This feature asks
9572 * to exit when accessing the given address, and if it
9573 * can never be accessed, this feature won't do
9576 if (vmx->nested.apic_access_page) {
9577 hpa = page_to_phys(vmx->nested.apic_access_page);
9578 vmcs_write64(APIC_ACCESS_ADDR, hpa);
9580 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
9581 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9583 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9584 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9585 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
9586 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9587 kvm_vcpu_reload_apic_access_page(vcpu);
9590 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9591 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
9592 nested_release_page(vmx->nested.virtual_apic_page);
9593 vmx->nested.virtual_apic_page =
9594 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
9597 * If translation failed, VM entry will fail because
9598 * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
9599 * Failing the vm entry is _not_ what the processor
9600 * does but it's basically the only possibility we
9601 * have. We could still enter the guest if CR8 load
9602 * exits are enabled, CR8 store exits are enabled, and
9603 * virtualize APIC access is disabled; in this case
9604 * the processor would never use the TPR shadow and we
9605 * could simply clear the bit from the execution
9606 * control. But such a configuration is useless, so
9607 * let's keep the code simple.
9609 if (vmx->nested.virtual_apic_page) {
9610 hpa = page_to_phys(vmx->nested.virtual_apic_page);
9611 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
9615 if (nested_cpu_has_posted_intr(vmcs12)) {
9616 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9617 kunmap(vmx->nested.pi_desc_page);
9618 nested_release_page(vmx->nested.pi_desc_page);
9620 vmx->nested.pi_desc_page =
9621 nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
9622 vmx->nested.pi_desc =
9623 (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
9624 if (!vmx->nested.pi_desc) {
9625 nested_release_page_clean(vmx->nested.pi_desc_page);
9628 vmx->nested.pi_desc =
9629 (struct pi_desc *)((void *)vmx->nested.pi_desc +
9630 (unsigned long)(vmcs12->posted_intr_desc_addr &
9632 vmcs_write64(POSTED_INTR_DESC_ADDR,
9633 page_to_phys(vmx->nested.pi_desc_page) +
9634 (unsigned long)(vmcs12->posted_intr_desc_addr &
9637 if (cpu_has_vmx_msr_bitmap() &&
9638 nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS) &&
9639 nested_vmx_merge_msr_bitmap(vcpu, vmcs12))
9642 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
9643 CPU_BASED_USE_MSR_BITMAPS);
9646 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9648 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9649 struct vcpu_vmx *vmx = to_vmx(vcpu);
9651 if (vcpu->arch.virtual_tsc_khz == 0)
9654 /* Make sure short timeouts reliably trigger an immediate vmexit.
9655 * hrtimer_start does not guarantee this. */
9656 if (preemption_timeout <= 1) {
9657 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9661 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9662 preemption_timeout *= 1000000;
9663 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9664 hrtimer_start(&vmx->nested.preemption_timer,
9665 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9668 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9669 struct vmcs12 *vmcs12)
9674 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9677 if (vmcs12_read_any(vcpu, MSR_BITMAP, &addr)) {
9681 maxphyaddr = cpuid_maxphyaddr(vcpu);
9683 if (!PAGE_ALIGNED(vmcs12->msr_bitmap) ||
9684 ((addr + PAGE_SIZE) >> maxphyaddr))
9691 * Merge L0's and L1's MSR bitmap, return false to indicate that
9692 * we do not use the hardware.
9694 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9695 struct vmcs12 *vmcs12)
9699 unsigned long *msr_bitmap_l1;
9700 unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.msr_bitmap;
9702 /* This shortcut is ok because we support only x2APIC MSRs so far. */
9703 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
9706 page = nested_get_page(vcpu, vmcs12->msr_bitmap);
9711 msr_bitmap_l1 = (unsigned long *)kmap(page);
9713 memset(msr_bitmap_l0, 0xff, PAGE_SIZE);
9715 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
9716 if (nested_cpu_has_apic_reg_virt(vmcs12))
9717 for (msr = 0x800; msr <= 0x8ff; msr++)
9718 nested_vmx_disable_intercept_for_msr(
9719 msr_bitmap_l1, msr_bitmap_l0,
9722 nested_vmx_disable_intercept_for_msr(
9723 msr_bitmap_l1, msr_bitmap_l0,
9724 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9725 MSR_TYPE_R | MSR_TYPE_W);
9727 if (nested_cpu_has_vid(vmcs12)) {
9728 nested_vmx_disable_intercept_for_msr(
9729 msr_bitmap_l1, msr_bitmap_l0,
9730 APIC_BASE_MSR + (APIC_EOI >> 4),
9732 nested_vmx_disable_intercept_for_msr(
9733 msr_bitmap_l1, msr_bitmap_l0,
9734 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9739 nested_release_page_clean(page);
9744 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
9745 struct vmcs12 *vmcs12)
9747 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9748 !nested_cpu_has_apic_reg_virt(vmcs12) &&
9749 !nested_cpu_has_vid(vmcs12) &&
9750 !nested_cpu_has_posted_intr(vmcs12))
9754 * If virtualize x2apic mode is enabled,
9755 * virtualize apic access must be disabled.
9757 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9758 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9762 * If virtual interrupt delivery is enabled,
9763 * we must exit on external interrupts.
9765 if (nested_cpu_has_vid(vmcs12) &&
9766 !nested_exit_on_intr(vcpu))
9770 * bits 15:8 should be zero in posted_intr_nv,
9771 * the descriptor address has been already checked
9772 * in nested_get_vmcs12_pages.
9774 if (nested_cpu_has_posted_intr(vmcs12) &&
9775 (!nested_cpu_has_vid(vmcs12) ||
9776 !nested_exit_intr_ack_set(vcpu) ||
9777 vmcs12->posted_intr_nv & 0xff00))
9780 /* tpr shadow is needed by all apicv features. */
9781 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9787 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9788 unsigned long count_field,
9789 unsigned long addr_field)
9794 if (vmcs12_read_any(vcpu, count_field, &count) ||
9795 vmcs12_read_any(vcpu, addr_field, &addr)) {
9801 maxphyaddr = cpuid_maxphyaddr(vcpu);
9802 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9803 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9804 pr_debug_ratelimited(
9805 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9806 addr_field, maxphyaddr, count, addr);
9812 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9813 struct vmcs12 *vmcs12)
9815 if (vmcs12->vm_exit_msr_load_count == 0 &&
9816 vmcs12->vm_exit_msr_store_count == 0 &&
9817 vmcs12->vm_entry_msr_load_count == 0)
9818 return 0; /* Fast path */
9819 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9820 VM_EXIT_MSR_LOAD_ADDR) ||
9821 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9822 VM_EXIT_MSR_STORE_ADDR) ||
9823 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9824 VM_ENTRY_MSR_LOAD_ADDR))
9829 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9830 struct vmx_msr_entry *e)
9832 /* x2APIC MSR accesses are not allowed */
9833 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9835 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9836 e->index == MSR_IA32_UCODE_REV)
9838 if (e->reserved != 0)
9843 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9844 struct vmx_msr_entry *e)
9846 if (e->index == MSR_FS_BASE ||
9847 e->index == MSR_GS_BASE ||
9848 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9849 nested_vmx_msr_check_common(vcpu, e))
9854 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9855 struct vmx_msr_entry *e)
9857 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9858 nested_vmx_msr_check_common(vcpu, e))
9864 * Load guest's/host's msr at nested entry/exit.
9865 * return 0 for success, entry index for failure.
9867 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9870 struct vmx_msr_entry e;
9871 struct msr_data msr;
9873 msr.host_initiated = false;
9874 for (i = 0; i < count; i++) {
9875 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
9877 pr_debug_ratelimited(
9878 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9879 __func__, i, gpa + i * sizeof(e));
9882 if (nested_vmx_load_msr_check(vcpu, &e)) {
9883 pr_debug_ratelimited(
9884 "%s check failed (%u, 0x%x, 0x%x)\n",
9885 __func__, i, e.index, e.reserved);
9888 msr.index = e.index;
9890 if (kvm_set_msr(vcpu, &msr)) {
9891 pr_debug_ratelimited(
9892 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9893 __func__, i, e.index, e.value);
9902 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9905 struct vmx_msr_entry e;
9907 for (i = 0; i < count; i++) {
9908 struct msr_data msr_info;
9909 if (kvm_vcpu_read_guest(vcpu,
9910 gpa + i * sizeof(e),
9911 &e, 2 * sizeof(u32))) {
9912 pr_debug_ratelimited(
9913 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9914 __func__, i, gpa + i * sizeof(e));
9917 if (nested_vmx_store_msr_check(vcpu, &e)) {
9918 pr_debug_ratelimited(
9919 "%s check failed (%u, 0x%x, 0x%x)\n",
9920 __func__, i, e.index, e.reserved);
9923 msr_info.host_initiated = false;
9924 msr_info.index = e.index;
9925 if (kvm_get_msr(vcpu, &msr_info)) {
9926 pr_debug_ratelimited(
9927 "%s cannot read MSR (%u, 0x%x)\n",
9928 __func__, i, e.index);
9931 if (kvm_vcpu_write_guest(vcpu,
9932 gpa + i * sizeof(e) +
9933 offsetof(struct vmx_msr_entry, value),
9934 &msr_info.data, sizeof(msr_info.data))) {
9935 pr_debug_ratelimited(
9936 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9937 __func__, i, e.index, msr_info.data);
9944 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
9946 unsigned long invalid_mask;
9948 invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
9949 return (val & invalid_mask) == 0;
9953 * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
9954 * emulating VM entry into a guest with EPT enabled.
9955 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
9956 * is assigned to entry_failure_code on failure.
9958 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
9959 u32 *entry_failure_code)
9961 if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
9962 if (!nested_cr3_valid(vcpu, cr3)) {
9963 *entry_failure_code = ENTRY_FAIL_DEFAULT;
9968 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
9969 * must not be dereferenced.
9971 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
9973 if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
9974 *entry_failure_code = ENTRY_FAIL_PDPTE;
9979 vcpu->arch.cr3 = cr3;
9980 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
9983 kvm_mmu_reset_context(vcpu);
9988 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
9989 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
9990 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
9991 * guest in a way that will both be appropriate to L1's requests, and our
9992 * needs. In addition to modifying the active vmcs (which is vmcs02), this
9993 * function also has additional necessary side-effects, like setting various
9994 * vcpu->arch fields.
9995 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
9996 * is assigned to entry_failure_code on failure.
9998 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
9999 bool from_vmentry, u32 *entry_failure_code)
10001 struct vcpu_vmx *vmx = to_vmx(vcpu);
10003 bool nested_ept_enabled = false;
10005 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
10006 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
10007 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
10008 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
10009 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
10010 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
10011 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
10012 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
10013 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
10014 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
10015 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
10016 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
10017 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
10018 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
10019 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
10020 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
10021 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
10022 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
10023 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
10024 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
10025 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
10026 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
10027 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
10028 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
10029 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
10030 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
10031 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
10032 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
10033 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
10034 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
10035 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
10036 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
10037 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
10038 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
10039 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
10040 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
10042 if (from_vmentry &&
10043 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
10044 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
10045 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
10047 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
10048 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
10050 if (from_vmentry) {
10051 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
10052 vmcs12->vm_entry_intr_info_field);
10053 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
10054 vmcs12->vm_entry_exception_error_code);
10055 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
10056 vmcs12->vm_entry_instruction_len);
10057 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
10058 vmcs12->guest_interruptibility_info);
10060 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
10062 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
10063 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
10064 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
10065 vmcs12->guest_pending_dbg_exceptions);
10066 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
10067 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
10069 if (nested_cpu_has_xsaves(vmcs12))
10070 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
10071 vmcs_write64(VMCS_LINK_POINTER, -1ull);
10073 exec_control = vmcs12->pin_based_vm_exec_control;
10075 /* Preemption timer setting is only taken from vmcs01. */
10076 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10077 exec_control |= vmcs_config.pin_based_exec_ctrl;
10078 if (vmx->hv_deadline_tsc == -1)
10079 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10081 /* Posted interrupts setting is only taken from vmcs12. */
10082 if (nested_cpu_has_posted_intr(vmcs12)) {
10084 * Note that we use L0's vector here and in
10085 * vmx_deliver_nested_posted_interrupt.
10087 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
10088 vmx->nested.pi_pending = false;
10089 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
10091 exec_control &= ~PIN_BASED_POSTED_INTR;
10094 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
10096 vmx->nested.preemption_timer_expired = false;
10097 if (nested_cpu_has_preemption_timer(vmcs12))
10098 vmx_start_preemption_timer(vcpu);
10101 * Whether page-faults are trapped is determined by a combination of
10102 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
10103 * If enable_ept, L0 doesn't care about page faults and we should
10104 * set all of these to L1's desires. However, if !enable_ept, L0 does
10105 * care about (at least some) page faults, and because it is not easy
10106 * (if at all possible?) to merge L0 and L1's desires, we simply ask
10107 * to exit on each and every L2 page fault. This is done by setting
10108 * MASK=MATCH=0 and (see below) EB.PF=1.
10109 * Note that below we don't need special code to set EB.PF beyond the
10110 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
10111 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
10112 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
10114 * A problem with this approach (when !enable_ept) is that L1 may be
10115 * injected with more page faults than it asked for. This could have
10116 * caused problems, but in practice existing hypervisors don't care.
10117 * To fix this, we will need to emulate the PFEC checking (on the L1
10118 * page tables), using walk_addr(), when injecting PFs to L1.
10120 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
10121 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
10122 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
10123 enable_ept ? vmcs12->page_fault_error_code_match : 0);
10125 if (cpu_has_secondary_exec_ctrls()) {
10126 exec_control = vmx_secondary_exec_control(vmx);
10128 /* Take the following fields only from vmcs12 */
10129 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
10130 SECONDARY_EXEC_RDTSCP |
10131 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
10132 SECONDARY_EXEC_APIC_REGISTER_VIRT);
10133 if (nested_cpu_has(vmcs12,
10134 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
10135 exec_control |= vmcs12->secondary_vm_exec_control;
10137 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
10138 vmcs_write64(EOI_EXIT_BITMAP0,
10139 vmcs12->eoi_exit_bitmap0);
10140 vmcs_write64(EOI_EXIT_BITMAP1,
10141 vmcs12->eoi_exit_bitmap1);
10142 vmcs_write64(EOI_EXIT_BITMAP2,
10143 vmcs12->eoi_exit_bitmap2);
10144 vmcs_write64(EOI_EXIT_BITMAP3,
10145 vmcs12->eoi_exit_bitmap3);
10146 vmcs_write16(GUEST_INTR_STATUS,
10147 vmcs12->guest_intr_status);
10150 nested_ept_enabled = (exec_control & SECONDARY_EXEC_ENABLE_EPT) != 0;
10153 * Write an illegal value to APIC_ACCESS_ADDR. Later,
10154 * nested_get_vmcs12_pages will either fix it up or
10155 * remove the VM execution control.
10157 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
10158 vmcs_write64(APIC_ACCESS_ADDR, -1ull);
10160 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
10165 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
10166 * Some constant fields are set here by vmx_set_constant_host_state().
10167 * Other fields are different per CPU, and will be set later when
10168 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
10170 vmx_set_constant_host_state(vmx);
10173 * Set the MSR load/store lists to match L0's settings.
10175 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
10176 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10177 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
10178 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10179 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
10182 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
10183 * entry, but only if the current (host) sp changed from the value
10184 * we wrote last (vmx->host_rsp). This cache is no longer relevant
10185 * if we switch vmcs, and rather than hold a separate cache per vmcs,
10186 * here we just force the write to happen on entry.
10190 exec_control = vmx_exec_control(vmx); /* L0's desires */
10191 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
10192 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
10193 exec_control &= ~CPU_BASED_TPR_SHADOW;
10194 exec_control |= vmcs12->cpu_based_vm_exec_control;
10197 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
10198 * nested_get_vmcs12_pages can't fix it up, the illegal value
10199 * will result in a VM entry failure.
10201 if (exec_control & CPU_BASED_TPR_SHADOW) {
10202 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
10203 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
10207 * Merging of IO bitmap not currently supported.
10208 * Rather, exit every time.
10210 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
10211 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
10213 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
10215 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
10216 * bitwise-or of what L1 wants to trap for L2, and what we want to
10217 * trap. Note that CR0.TS also needs updating - we do this later.
10219 update_exception_bitmap(vcpu);
10220 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
10221 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10223 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
10224 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
10225 * bits are further modified by vmx_set_efer() below.
10227 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
10229 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
10230 * emulated by vmx_set_efer(), below.
10232 vm_entry_controls_init(vmx,
10233 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
10234 ~VM_ENTRY_IA32E_MODE) |
10235 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
10237 if (from_vmentry &&
10238 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
10239 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
10240 vcpu->arch.pat = vmcs12->guest_ia32_pat;
10241 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
10242 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
10245 set_cr4_guest_host_mask(vmx);
10247 if (from_vmentry &&
10248 vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
10249 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
10251 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
10252 vmcs_write64(TSC_OFFSET,
10253 vcpu->arch.tsc_offset + vmcs12->tsc_offset);
10255 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
10256 if (kvm_has_tsc_control)
10257 decache_tsc_multiplier(vmx);
10261 * There is no direct mapping between vpid02 and vpid12, the
10262 * vpid02 is per-vCPU for L0 and reused while the value of
10263 * vpid12 is changed w/ one invvpid during nested vmentry.
10264 * The vpid12 is allocated by L1 for L2, so it will not
10265 * influence global bitmap(for vpid01 and vpid02 allocation)
10266 * even if spawn a lot of nested vCPUs.
10268 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
10269 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
10270 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
10271 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
10272 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
10275 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
10276 vmx_flush_tlb(vcpu);
10281 if (nested_cpu_has_ept(vmcs12)) {
10282 kvm_mmu_unload(vcpu);
10283 nested_ept_init_mmu_context(vcpu);
10287 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
10288 * bits which we consider mandatory enabled.
10289 * The CR0_READ_SHADOW is what L2 should have expected to read given
10290 * the specifications by L1; It's not enough to take
10291 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10292 * have more bits than L1 expected.
10294 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
10295 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
10297 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
10298 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
10300 if (from_vmentry &&
10301 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
10302 vcpu->arch.efer = vmcs12->guest_ia32_efer;
10303 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
10304 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10306 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10307 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10308 vmx_set_efer(vcpu, vcpu->arch.efer);
10310 /* Shadow page tables on either EPT or shadow page tables. */
10311 if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_ept_enabled,
10312 entry_failure_code))
10315 kvm_mmu_reset_context(vcpu);
10318 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
10321 * L1 may access the L2's PDPTR, so save them to construct vmcs12
10324 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
10325 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
10326 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
10327 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
10330 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
10331 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
10335 static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10337 struct vcpu_vmx *vmx = to_vmx(vcpu);
10339 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
10340 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
10341 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10343 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12))
10344 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10346 if (nested_vmx_check_apicv_controls(vcpu, vmcs12))
10347 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10349 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12))
10350 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10352 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
10353 vmx->nested.nested_vmx_procbased_ctls_low,
10354 vmx->nested.nested_vmx_procbased_ctls_high) ||
10355 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
10356 vmx->nested.nested_vmx_secondary_ctls_low,
10357 vmx->nested.nested_vmx_secondary_ctls_high) ||
10358 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
10359 vmx->nested.nested_vmx_pinbased_ctls_low,
10360 vmx->nested.nested_vmx_pinbased_ctls_high) ||
10361 !vmx_control_verify(vmcs12->vm_exit_controls,
10362 vmx->nested.nested_vmx_exit_ctls_low,
10363 vmx->nested.nested_vmx_exit_ctls_high) ||
10364 !vmx_control_verify(vmcs12->vm_entry_controls,
10365 vmx->nested.nested_vmx_entry_ctls_low,
10366 vmx->nested.nested_vmx_entry_ctls_high))
10367 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10369 if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
10370 !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
10371 !nested_cr3_valid(vcpu, vmcs12->host_cr3))
10372 return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
10377 static int check_vmentry_postreqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10382 *exit_qual = ENTRY_FAIL_DEFAULT;
10384 if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
10385 !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
10388 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_SHADOW_VMCS) &&
10389 vmcs12->vmcs_link_pointer != -1ull) {
10390 *exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
10395 * If the load IA32_EFER VM-entry control is 1, the following checks
10396 * are performed on the field for the IA32_EFER MSR:
10397 * - Bits reserved in the IA32_EFER MSR must be 0.
10398 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
10399 * the IA-32e mode guest VM-exit control. It must also be identical
10400 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
10403 if (to_vmx(vcpu)->nested.nested_run_pending &&
10404 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
10405 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
10406 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
10407 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
10408 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
10409 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
10414 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
10415 * IA32_EFER MSR must be 0 in the field for that register. In addition,
10416 * the values of the LMA and LME bits in the field must each be that of
10417 * the host address-space size VM-exit control.
10419 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
10420 ia32e = (vmcs12->vm_exit_controls &
10421 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
10422 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
10423 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
10424 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
10431 static int enter_vmx_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
10433 struct vcpu_vmx *vmx = to_vmx(vcpu);
10434 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10435 struct loaded_vmcs *vmcs02;
10440 vmcs02 = nested_get_current_vmcs02(vmx);
10444 enter_guest_mode(vcpu);
10446 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
10447 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10450 vmx->loaded_vmcs = vmcs02;
10451 vmx_vcpu_put(vcpu);
10452 vmx_vcpu_load(vcpu, cpu);
10456 vmx_segment_cache_clear(vmx);
10458 if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &exit_qual)) {
10459 leave_guest_mode(vcpu);
10460 vmx_load_vmcs01(vcpu);
10461 nested_vmx_entry_failure(vcpu, vmcs12,
10462 EXIT_REASON_INVALID_STATE, exit_qual);
10466 nested_get_vmcs12_pages(vcpu, vmcs12);
10468 msr_entry_idx = nested_vmx_load_msr(vcpu,
10469 vmcs12->vm_entry_msr_load_addr,
10470 vmcs12->vm_entry_msr_load_count);
10471 if (msr_entry_idx) {
10472 leave_guest_mode(vcpu);
10473 vmx_load_vmcs01(vcpu);
10474 nested_vmx_entry_failure(vcpu, vmcs12,
10475 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
10479 vmcs12->launch_state = 1;
10482 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10483 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10484 * returned as far as L1 is concerned. It will only return (and set
10485 * the success flag) when L2 exits (see nested_vmx_vmexit()).
10491 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
10492 * for running an L2 nested guest.
10494 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
10496 struct vmcs12 *vmcs12;
10497 struct vcpu_vmx *vmx = to_vmx(vcpu);
10501 if (!nested_vmx_check_permission(vcpu))
10504 if (!nested_vmx_check_vmcs12(vcpu))
10507 vmcs12 = get_vmcs12(vcpu);
10509 if (enable_shadow_vmcs)
10510 copy_shadow_to_vmcs12(vmx);
10513 * The nested entry process starts with enforcing various prerequisites
10514 * on vmcs12 as required by the Intel SDM, and act appropriately when
10515 * they fail: As the SDM explains, some conditions should cause the
10516 * instruction to fail, while others will cause the instruction to seem
10517 * to succeed, but return an EXIT_REASON_INVALID_STATE.
10518 * To speed up the normal (success) code path, we should avoid checking
10519 * for misconfigurations which will anyway be caught by the processor
10520 * when using the merged vmcs02.
10522 if (vmcs12->launch_state == launch) {
10523 nested_vmx_failValid(vcpu,
10524 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
10525 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
10529 ret = check_vmentry_prereqs(vcpu, vmcs12);
10531 nested_vmx_failValid(vcpu, ret);
10536 * After this point, the trap flag no longer triggers a singlestep trap
10537 * on the vm entry instructions; don't call kvm_skip_emulated_instruction.
10538 * This is not 100% correct; for performance reasons, we delegate most
10539 * of the checks on host state to the processor. If those fail,
10540 * the singlestep trap is missed.
10542 skip_emulated_instruction(vcpu);
10544 ret = check_vmentry_postreqs(vcpu, vmcs12, &exit_qual);
10546 nested_vmx_entry_failure(vcpu, vmcs12,
10547 EXIT_REASON_INVALID_STATE, exit_qual);
10552 * We're finally done with prerequisite checking, and can start with
10553 * the nested entry.
10556 ret = enter_vmx_non_root_mode(vcpu, true);
10560 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
10561 return kvm_vcpu_halt(vcpu);
10563 vmx->nested.nested_run_pending = 1;
10568 return kvm_skip_emulated_instruction(vcpu);
10572 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
10573 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
10574 * This function returns the new value we should put in vmcs12.guest_cr0.
10575 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
10576 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
10577 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
10578 * didn't trap the bit, because if L1 did, so would L0).
10579 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
10580 * been modified by L2, and L1 knows it. So just leave the old value of
10581 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
10582 * isn't relevant, because if L0 traps this bit it can set it to anything.
10583 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
10584 * changed these bits, and therefore they need to be updated, but L0
10585 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
10586 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
10588 static inline unsigned long
10589 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10592 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
10593 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
10594 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
10595 vcpu->arch.cr0_guest_owned_bits));
10598 static inline unsigned long
10599 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10602 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
10603 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
10604 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
10605 vcpu->arch.cr4_guest_owned_bits));
10608 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
10609 struct vmcs12 *vmcs12)
10614 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
10615 nr = vcpu->arch.exception.nr;
10616 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10618 if (kvm_exception_is_soft(nr)) {
10619 vmcs12->vm_exit_instruction_len =
10620 vcpu->arch.event_exit_inst_len;
10621 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
10623 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
10625 if (vcpu->arch.exception.has_error_code) {
10626 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
10627 vmcs12->idt_vectoring_error_code =
10628 vcpu->arch.exception.error_code;
10631 vmcs12->idt_vectoring_info_field = idt_vectoring;
10632 } else if (vcpu->arch.nmi_injected) {
10633 vmcs12->idt_vectoring_info_field =
10634 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
10635 } else if (vcpu->arch.interrupt.pending) {
10636 nr = vcpu->arch.interrupt.nr;
10637 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10639 if (vcpu->arch.interrupt.soft) {
10640 idt_vectoring |= INTR_TYPE_SOFT_INTR;
10641 vmcs12->vm_entry_instruction_len =
10642 vcpu->arch.event_exit_inst_len;
10644 idt_vectoring |= INTR_TYPE_EXT_INTR;
10646 vmcs12->idt_vectoring_info_field = idt_vectoring;
10650 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
10652 struct vcpu_vmx *vmx = to_vmx(vcpu);
10654 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
10655 vmx->nested.preemption_timer_expired) {
10656 if (vmx->nested.nested_run_pending)
10658 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
10662 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
10663 if (vmx->nested.nested_run_pending ||
10664 vcpu->arch.interrupt.pending)
10666 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10667 NMI_VECTOR | INTR_TYPE_NMI_INTR |
10668 INTR_INFO_VALID_MASK, 0);
10670 * The NMI-triggered VM exit counts as injection:
10671 * clear this one and block further NMIs.
10673 vcpu->arch.nmi_pending = 0;
10674 vmx_set_nmi_mask(vcpu, true);
10678 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
10679 nested_exit_on_intr(vcpu)) {
10680 if (vmx->nested.nested_run_pending)
10682 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
10686 vmx_complete_nested_posted_interrupt(vcpu);
10690 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
10692 ktime_t remaining =
10693 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
10696 if (ktime_to_ns(remaining) <= 0)
10699 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
10700 do_div(value, 1000000);
10701 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10705 * Update the guest state fields of vmcs12 to reflect changes that
10706 * occurred while L2 was running. (The "IA-32e mode guest" bit of the
10707 * VM-entry controls is also updated, since this is really a guest
10710 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10712 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
10713 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
10715 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
10716 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
10717 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
10719 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
10720 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
10721 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
10722 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
10723 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
10724 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
10725 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
10726 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
10727 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
10728 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
10729 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
10730 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
10731 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
10732 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
10733 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
10734 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
10735 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
10736 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
10737 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
10738 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
10739 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
10740 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
10741 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
10742 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
10743 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
10744 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
10745 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
10746 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
10747 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
10748 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
10749 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
10750 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
10751 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
10752 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
10753 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
10754 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
10756 vmcs12->guest_interruptibility_info =
10757 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
10758 vmcs12->guest_pending_dbg_exceptions =
10759 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
10760 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10761 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
10763 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
10765 if (nested_cpu_has_preemption_timer(vmcs12)) {
10766 if (vmcs12->vm_exit_controls &
10767 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
10768 vmcs12->vmx_preemption_timer_value =
10769 vmx_get_preemption_timer_value(vcpu);
10770 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
10774 * In some cases (usually, nested EPT), L2 is allowed to change its
10775 * own CR3 without exiting. If it has changed it, we must keep it.
10776 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10777 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10779 * Additionally, restore L2's PDPTR to vmcs12.
10782 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
10783 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
10784 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
10785 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
10786 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
10789 if (nested_cpu_has_ept(vmcs12))
10790 vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
10792 if (nested_cpu_has_vid(vmcs12))
10793 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
10795 vmcs12->vm_entry_controls =
10796 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
10797 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
10799 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
10800 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
10801 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10804 /* TODO: These cannot have changed unless we have MSR bitmaps and
10805 * the relevant bit asks not to trap the change */
10806 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
10807 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
10808 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
10809 vmcs12->guest_ia32_efer = vcpu->arch.efer;
10810 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
10811 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
10812 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
10813 if (kvm_mpx_supported())
10814 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
10815 if (nested_cpu_has_xsaves(vmcs12))
10816 vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP);
10820 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10821 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10822 * and this function updates it to reflect the changes to the guest state while
10823 * L2 was running (and perhaps made some exits which were handled directly by L0
10824 * without going back to L1), and to reflect the exit reason.
10825 * Note that we do not have to copy here all VMCS fields, just those that
10826 * could have changed by the L2 guest or the exit - i.e., the guest-state and
10827 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10828 * which already writes to vmcs12 directly.
10830 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10831 u32 exit_reason, u32 exit_intr_info,
10832 unsigned long exit_qualification)
10834 /* update guest state fields: */
10835 sync_vmcs12(vcpu, vmcs12);
10837 /* update exit information fields: */
10839 vmcs12->vm_exit_reason = exit_reason;
10840 vmcs12->exit_qualification = exit_qualification;
10842 vmcs12->vm_exit_intr_info = exit_intr_info;
10843 if ((vmcs12->vm_exit_intr_info &
10844 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
10845 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
10846 vmcs12->vm_exit_intr_error_code =
10847 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
10848 vmcs12->idt_vectoring_info_field = 0;
10849 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
10850 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
10852 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
10853 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10854 * instead of reading the real value. */
10855 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
10858 * Transfer the event that L0 or L1 may wanted to inject into
10859 * L2 to IDT_VECTORING_INFO_FIELD.
10861 vmcs12_save_pending_event(vcpu, vmcs12);
10865 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10866 * preserved above and would only end up incorrectly in L1.
10868 vcpu->arch.nmi_injected = false;
10869 kvm_clear_exception_queue(vcpu);
10870 kvm_clear_interrupt_queue(vcpu);
10874 * A part of what we need to when the nested L2 guest exits and we want to
10875 * run its L1 parent, is to reset L1's guest state to the host state specified
10877 * This function is to be called not only on normal nested exit, but also on
10878 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10879 * Failures During or After Loading Guest State").
10880 * This function should be called when the active VMCS is L1's (vmcs01).
10882 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10883 struct vmcs12 *vmcs12)
10885 struct kvm_segment seg;
10886 u32 entry_failure_code;
10888 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10889 vcpu->arch.efer = vmcs12->host_ia32_efer;
10890 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10891 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10893 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10894 vmx_set_efer(vcpu, vcpu->arch.efer);
10896 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10897 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10898 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10900 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10901 * actually changed, because vmx_set_cr0 refers to efer set above.
10903 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
10904 * (KVM doesn't change it);
10906 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
10907 vmx_set_cr0(vcpu, vmcs12->host_cr0);
10909 /* Same as above - no reason to call set_cr4_guest_host_mask(). */
10910 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
10911 kvm_set_cr4(vcpu, vmcs12->host_cr4);
10913 nested_ept_uninit_mmu_context(vcpu);
10916 * Only PDPTE load can fail as the value of cr3 was checked on entry and
10917 * couldn't have changed.
10919 if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
10920 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
10923 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
10927 * Trivially support vpid by letting L2s share their parent
10928 * L1's vpid. TODO: move to a more elaborate solution, giving
10929 * each L2 its own vpid and exposing the vpid feature to L1.
10931 vmx_flush_tlb(vcpu);
10935 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
10936 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
10937 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
10938 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
10939 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
10941 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
10942 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
10943 vmcs_write64(GUEST_BNDCFGS, 0);
10945 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
10946 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
10947 vcpu->arch.pat = vmcs12->host_ia32_pat;
10949 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
10950 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
10951 vmcs12->host_ia32_perf_global_ctrl);
10953 /* Set L1 segment info according to Intel SDM
10954 27.5.2 Loading Host Segment and Descriptor-Table Registers */
10955 seg = (struct kvm_segment) {
10957 .limit = 0xFFFFFFFF,
10958 .selector = vmcs12->host_cs_selector,
10964 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10968 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
10969 seg = (struct kvm_segment) {
10971 .limit = 0xFFFFFFFF,
10978 seg.selector = vmcs12->host_ds_selector;
10979 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
10980 seg.selector = vmcs12->host_es_selector;
10981 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
10982 seg.selector = vmcs12->host_ss_selector;
10983 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
10984 seg.selector = vmcs12->host_fs_selector;
10985 seg.base = vmcs12->host_fs_base;
10986 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
10987 seg.selector = vmcs12->host_gs_selector;
10988 seg.base = vmcs12->host_gs_base;
10989 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
10990 seg = (struct kvm_segment) {
10991 .base = vmcs12->host_tr_base,
10993 .selector = vmcs12->host_tr_selector,
10997 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
10999 kvm_set_dr(vcpu, 7, 0x400);
11000 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
11002 if (cpu_has_vmx_msr_bitmap())
11003 vmx_set_msr_bitmap(vcpu);
11005 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
11006 vmcs12->vm_exit_msr_load_count))
11007 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
11011 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
11012 * and modify vmcs12 to make it see what it would expect to see there if
11013 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
11015 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
11016 u32 exit_intr_info,
11017 unsigned long exit_qualification)
11019 struct vcpu_vmx *vmx = to_vmx(vcpu);
11020 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
11021 u32 vm_inst_error = 0;
11023 /* trying to cancel vmlaunch/vmresume is a bug */
11024 WARN_ON_ONCE(vmx->nested.nested_run_pending);
11026 leave_guest_mode(vcpu);
11027 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
11028 exit_qualification);
11030 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
11031 vmcs12->vm_exit_msr_store_count))
11032 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
11034 if (unlikely(vmx->fail))
11035 vm_inst_error = vmcs_read32(VM_INSTRUCTION_ERROR);
11037 vmx_load_vmcs01(vcpu);
11039 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
11040 && nested_exit_intr_ack_set(vcpu)) {
11041 int irq = kvm_cpu_get_interrupt(vcpu);
11043 vmcs12->vm_exit_intr_info = irq |
11044 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
11047 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
11048 vmcs12->exit_qualification,
11049 vmcs12->idt_vectoring_info_field,
11050 vmcs12->vm_exit_intr_info,
11051 vmcs12->vm_exit_intr_error_code,
11054 vm_entry_controls_reset_shadow(vmx);
11055 vm_exit_controls_reset_shadow(vmx);
11056 vmx_segment_cache_clear(vmx);
11058 /* if no vmcs02 cache requested, remove the one we used */
11059 if (VMCS02_POOL_SIZE == 0)
11060 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
11062 load_vmcs12_host_state(vcpu, vmcs12);
11064 /* Update any VMCS fields that might have changed while L2 ran */
11065 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11066 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11067 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
11068 if (vmx->hv_deadline_tsc == -1)
11069 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11070 PIN_BASED_VMX_PREEMPTION_TIMER);
11072 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11073 PIN_BASED_VMX_PREEMPTION_TIMER);
11074 if (kvm_has_tsc_control)
11075 decache_tsc_multiplier(vmx);
11077 if (vmx->nested.change_vmcs01_virtual_x2apic_mode) {
11078 vmx->nested.change_vmcs01_virtual_x2apic_mode = false;
11079 vmx_set_virtual_x2apic_mode(vcpu,
11080 vcpu->arch.apic_base & X2APIC_ENABLE);
11083 /* This is needed for same reason as it was needed in prepare_vmcs02 */
11086 /* Unpin physical memory we referred to in vmcs02 */
11087 if (vmx->nested.apic_access_page) {
11088 nested_release_page(vmx->nested.apic_access_page);
11089 vmx->nested.apic_access_page = NULL;
11091 if (vmx->nested.virtual_apic_page) {
11092 nested_release_page(vmx->nested.virtual_apic_page);
11093 vmx->nested.virtual_apic_page = NULL;
11095 if (vmx->nested.pi_desc_page) {
11096 kunmap(vmx->nested.pi_desc_page);
11097 nested_release_page(vmx->nested.pi_desc_page);
11098 vmx->nested.pi_desc_page = NULL;
11099 vmx->nested.pi_desc = NULL;
11103 * We are now running in L2, mmu_notifier will force to reload the
11104 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
11106 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
11109 * Exiting from L2 to L1, we're now back to L1 which thinks it just
11110 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
11111 * success or failure flag accordingly.
11113 if (unlikely(vmx->fail)) {
11115 nested_vmx_failValid(vcpu, vm_inst_error);
11117 nested_vmx_succeed(vcpu);
11118 if (enable_shadow_vmcs)
11119 vmx->nested.sync_shadow_vmcs = true;
11121 /* in case we halted in L2 */
11122 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11126 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
11128 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
11130 if (is_guest_mode(vcpu))
11131 nested_vmx_vmexit(vcpu, -1, 0, 0);
11132 free_nested(to_vmx(vcpu));
11136 * L1's failure to enter L2 is a subset of a normal exit, as explained in
11137 * 23.7 "VM-entry failures during or after loading guest state" (this also
11138 * lists the acceptable exit-reason and exit-qualification parameters).
11139 * It should only be called before L2 actually succeeded to run, and when
11140 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
11142 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
11143 struct vmcs12 *vmcs12,
11144 u32 reason, unsigned long qualification)
11146 load_vmcs12_host_state(vcpu, vmcs12);
11147 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
11148 vmcs12->exit_qualification = qualification;
11149 nested_vmx_succeed(vcpu);
11150 if (enable_shadow_vmcs)
11151 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
11154 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
11155 struct x86_instruction_info *info,
11156 enum x86_intercept_stage stage)
11158 return X86EMUL_CONTINUE;
11161 #ifdef CONFIG_X86_64
11162 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
11163 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
11164 u64 divisor, u64 *result)
11166 u64 low = a << shift, high = a >> (64 - shift);
11168 /* To avoid the overflow on divq */
11169 if (high >= divisor)
11172 /* Low hold the result, high hold rem which is discarded */
11173 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
11174 "rm" (divisor), "0" (low), "1" (high));
11180 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
11182 struct vcpu_vmx *vmx = to_vmx(vcpu);
11183 u64 tscl = rdtsc();
11184 u64 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
11185 u64 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
11187 /* Convert to host delta tsc if tsc scaling is enabled */
11188 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
11189 u64_shl_div_u64(delta_tsc,
11190 kvm_tsc_scaling_ratio_frac_bits,
11191 vcpu->arch.tsc_scaling_ratio,
11196 * If the delta tsc can't fit in the 32 bit after the multi shift,
11197 * we can't use the preemption timer.
11198 * It's possible that it fits on later vmentries, but checking
11199 * on every vmentry is costly so we just use an hrtimer.
11201 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
11204 vmx->hv_deadline_tsc = tscl + delta_tsc;
11205 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11206 PIN_BASED_VMX_PREEMPTION_TIMER);
11210 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
11212 struct vcpu_vmx *vmx = to_vmx(vcpu);
11213 vmx->hv_deadline_tsc = -1;
11214 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11215 PIN_BASED_VMX_PREEMPTION_TIMER);
11219 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
11222 shrink_ple_window(vcpu);
11225 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
11226 struct kvm_memory_slot *slot)
11228 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
11229 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
11232 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
11233 struct kvm_memory_slot *slot)
11235 kvm_mmu_slot_set_dirty(kvm, slot);
11238 static void vmx_flush_log_dirty(struct kvm *kvm)
11240 kvm_flush_pml_buffers(kvm);
11243 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
11244 struct kvm_memory_slot *memslot,
11245 gfn_t offset, unsigned long mask)
11247 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
11251 * This routine does the following things for vCPU which is going
11252 * to be blocked if VT-d PI is enabled.
11253 * - Store the vCPU to the wakeup list, so when interrupts happen
11254 * we can find the right vCPU to wake up.
11255 * - Change the Posted-interrupt descriptor as below:
11256 * 'NDST' <-- vcpu->pre_pcpu
11257 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
11258 * - If 'ON' is set during this process, which means at least one
11259 * interrupt is posted for this vCPU, we cannot block it, in
11260 * this case, return 1, otherwise, return 0.
11263 static int pi_pre_block(struct kvm_vcpu *vcpu)
11265 unsigned long flags;
11267 struct pi_desc old, new;
11268 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11270 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11271 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11272 !kvm_vcpu_apicv_active(vcpu))
11275 vcpu->pre_pcpu = vcpu->cpu;
11276 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
11277 vcpu->pre_pcpu), flags);
11278 list_add_tail(&vcpu->blocked_vcpu_list,
11279 &per_cpu(blocked_vcpu_on_cpu,
11281 spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
11282 vcpu->pre_pcpu), flags);
11285 old.control = new.control = pi_desc->control;
11288 * We should not block the vCPU if
11289 * an interrupt is posted for it.
11291 if (pi_test_on(pi_desc) == 1) {
11292 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
11293 vcpu->pre_pcpu), flags);
11294 list_del(&vcpu->blocked_vcpu_list);
11295 spin_unlock_irqrestore(
11296 &per_cpu(blocked_vcpu_on_cpu_lock,
11297 vcpu->pre_pcpu), flags);
11298 vcpu->pre_pcpu = -1;
11303 WARN((pi_desc->sn == 1),
11304 "Warning: SN field of posted-interrupts "
11305 "is set before blocking\n");
11308 * Since vCPU can be preempted during this process,
11309 * vcpu->cpu could be different with pre_pcpu, we
11310 * need to set pre_pcpu as the destination of wakeup
11311 * notification event, then we can find the right vCPU
11312 * to wakeup in wakeup handler if interrupts happen
11313 * when the vCPU is in blocked state.
11315 dest = cpu_physical_id(vcpu->pre_pcpu);
11317 if (x2apic_enabled())
11320 new.ndst = (dest << 8) & 0xFF00;
11322 /* set 'NV' to 'wakeup vector' */
11323 new.nv = POSTED_INTR_WAKEUP_VECTOR;
11324 } while (cmpxchg(&pi_desc->control, old.control,
11325 new.control) != old.control);
11330 static int vmx_pre_block(struct kvm_vcpu *vcpu)
11332 if (pi_pre_block(vcpu))
11335 if (kvm_lapic_hv_timer_in_use(vcpu))
11336 kvm_lapic_switch_to_sw_timer(vcpu);
11341 static void pi_post_block(struct kvm_vcpu *vcpu)
11343 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11344 struct pi_desc old, new;
11346 unsigned long flags;
11348 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11349 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11350 !kvm_vcpu_apicv_active(vcpu))
11354 old.control = new.control = pi_desc->control;
11356 dest = cpu_physical_id(vcpu->cpu);
11358 if (x2apic_enabled())
11361 new.ndst = (dest << 8) & 0xFF00;
11363 /* Allow posting non-urgent interrupts */
11366 /* set 'NV' to 'notification vector' */
11367 new.nv = POSTED_INTR_VECTOR;
11368 } while (cmpxchg(&pi_desc->control, old.control,
11369 new.control) != old.control);
11371 if(vcpu->pre_pcpu != -1) {
11373 &per_cpu(blocked_vcpu_on_cpu_lock,
11374 vcpu->pre_pcpu), flags);
11375 list_del(&vcpu->blocked_vcpu_list);
11376 spin_unlock_irqrestore(
11377 &per_cpu(blocked_vcpu_on_cpu_lock,
11378 vcpu->pre_pcpu), flags);
11379 vcpu->pre_pcpu = -1;
11383 static void vmx_post_block(struct kvm_vcpu *vcpu)
11385 if (kvm_x86_ops->set_hv_timer)
11386 kvm_lapic_switch_to_hv_timer(vcpu);
11388 pi_post_block(vcpu);
11392 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
11395 * @host_irq: host irq of the interrupt
11396 * @guest_irq: gsi of the interrupt
11397 * @set: set or unset PI
11398 * returns 0 on success, < 0 on failure
11400 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
11401 uint32_t guest_irq, bool set)
11403 struct kvm_kernel_irq_routing_entry *e;
11404 struct kvm_irq_routing_table *irq_rt;
11405 struct kvm_lapic_irq irq;
11406 struct kvm_vcpu *vcpu;
11407 struct vcpu_data vcpu_info;
11408 int idx, ret = -EINVAL;
11410 if (!kvm_arch_has_assigned_device(kvm) ||
11411 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11412 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
11415 idx = srcu_read_lock(&kvm->irq_srcu);
11416 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
11417 BUG_ON(guest_irq >= irq_rt->nr_rt_entries);
11419 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
11420 if (e->type != KVM_IRQ_ROUTING_MSI)
11423 * VT-d PI cannot support posting multicast/broadcast
11424 * interrupts to a vCPU, we still use interrupt remapping
11425 * for these kind of interrupts.
11427 * For lowest-priority interrupts, we only support
11428 * those with single CPU as the destination, e.g. user
11429 * configures the interrupts via /proc/irq or uses
11430 * irqbalance to make the interrupts single-CPU.
11432 * We will support full lowest-priority interrupt later.
11435 kvm_set_msi_irq(kvm, e, &irq);
11436 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
11438 * Make sure the IRTE is in remapped mode if
11439 * we don't handle it in posted mode.
11441 ret = irq_set_vcpu_affinity(host_irq, NULL);
11444 "failed to back to remapped mode, irq: %u\n",
11452 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
11453 vcpu_info.vector = irq.vector;
11455 trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
11456 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
11459 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
11461 /* suppress notification event before unposting */
11462 pi_set_sn(vcpu_to_pi_desc(vcpu));
11463 ret = irq_set_vcpu_affinity(host_irq, NULL);
11464 pi_clear_sn(vcpu_to_pi_desc(vcpu));
11468 printk(KERN_INFO "%s: failed to update PI IRTE\n",
11476 srcu_read_unlock(&kvm->irq_srcu, idx);
11480 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
11482 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
11483 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
11484 FEATURE_CONTROL_LMCE;
11486 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
11487 ~FEATURE_CONTROL_LMCE;
11490 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
11491 .cpu_has_kvm_support = cpu_has_kvm_support,
11492 .disabled_by_bios = vmx_disabled_by_bios,
11493 .hardware_setup = hardware_setup,
11494 .hardware_unsetup = hardware_unsetup,
11495 .check_processor_compatibility = vmx_check_processor_compat,
11496 .hardware_enable = hardware_enable,
11497 .hardware_disable = hardware_disable,
11498 .cpu_has_accelerated_tpr = report_flexpriority,
11499 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
11501 .vcpu_create = vmx_create_vcpu,
11502 .vcpu_free = vmx_free_vcpu,
11503 .vcpu_reset = vmx_vcpu_reset,
11505 .prepare_guest_switch = vmx_save_host_state,
11506 .vcpu_load = vmx_vcpu_load,
11507 .vcpu_put = vmx_vcpu_put,
11509 .update_bp_intercept = update_exception_bitmap,
11510 .get_msr = vmx_get_msr,
11511 .set_msr = vmx_set_msr,
11512 .get_segment_base = vmx_get_segment_base,
11513 .get_segment = vmx_get_segment,
11514 .set_segment = vmx_set_segment,
11515 .get_cpl = vmx_get_cpl,
11516 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
11517 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
11518 .decache_cr3 = vmx_decache_cr3,
11519 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
11520 .set_cr0 = vmx_set_cr0,
11521 .set_cr3 = vmx_set_cr3,
11522 .set_cr4 = vmx_set_cr4,
11523 .set_efer = vmx_set_efer,
11524 .get_idt = vmx_get_idt,
11525 .set_idt = vmx_set_idt,
11526 .get_gdt = vmx_get_gdt,
11527 .set_gdt = vmx_set_gdt,
11528 .get_dr6 = vmx_get_dr6,
11529 .set_dr6 = vmx_set_dr6,
11530 .set_dr7 = vmx_set_dr7,
11531 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
11532 .cache_reg = vmx_cache_reg,
11533 .get_rflags = vmx_get_rflags,
11534 .set_rflags = vmx_set_rflags,
11536 .get_pkru = vmx_get_pkru,
11538 .tlb_flush = vmx_flush_tlb,
11540 .run = vmx_vcpu_run,
11541 .handle_exit = vmx_handle_exit,
11542 .skip_emulated_instruction = skip_emulated_instruction,
11543 .set_interrupt_shadow = vmx_set_interrupt_shadow,
11544 .get_interrupt_shadow = vmx_get_interrupt_shadow,
11545 .patch_hypercall = vmx_patch_hypercall,
11546 .set_irq = vmx_inject_irq,
11547 .set_nmi = vmx_inject_nmi,
11548 .queue_exception = vmx_queue_exception,
11549 .cancel_injection = vmx_cancel_injection,
11550 .interrupt_allowed = vmx_interrupt_allowed,
11551 .nmi_allowed = vmx_nmi_allowed,
11552 .get_nmi_mask = vmx_get_nmi_mask,
11553 .set_nmi_mask = vmx_set_nmi_mask,
11554 .enable_nmi_window = enable_nmi_window,
11555 .enable_irq_window = enable_irq_window,
11556 .update_cr8_intercept = update_cr8_intercept,
11557 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
11558 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
11559 .get_enable_apicv = vmx_get_enable_apicv,
11560 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
11561 .load_eoi_exitmap = vmx_load_eoi_exitmap,
11562 .apicv_post_state_restore = vmx_apicv_post_state_restore,
11563 .hwapic_irr_update = vmx_hwapic_irr_update,
11564 .hwapic_isr_update = vmx_hwapic_isr_update,
11565 .sync_pir_to_irr = vmx_sync_pir_to_irr,
11566 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
11568 .set_tss_addr = vmx_set_tss_addr,
11569 .get_tdp_level = get_ept_level,
11570 .get_mt_mask = vmx_get_mt_mask,
11572 .get_exit_info = vmx_get_exit_info,
11574 .get_lpage_level = vmx_get_lpage_level,
11576 .cpuid_update = vmx_cpuid_update,
11578 .rdtscp_supported = vmx_rdtscp_supported,
11579 .invpcid_supported = vmx_invpcid_supported,
11581 .set_supported_cpuid = vmx_set_supported_cpuid,
11583 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
11585 .write_tsc_offset = vmx_write_tsc_offset,
11587 .set_tdp_cr3 = vmx_set_cr3,
11589 .check_intercept = vmx_check_intercept,
11590 .handle_external_intr = vmx_handle_external_intr,
11591 .mpx_supported = vmx_mpx_supported,
11592 .xsaves_supported = vmx_xsaves_supported,
11594 .check_nested_events = vmx_check_nested_events,
11596 .sched_in = vmx_sched_in,
11598 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
11599 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
11600 .flush_log_dirty = vmx_flush_log_dirty,
11601 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
11603 .pre_block = vmx_pre_block,
11604 .post_block = vmx_post_block,
11606 .pmu_ops = &intel_pmu_ops,
11608 .update_pi_irte = vmx_update_pi_irte,
11610 #ifdef CONFIG_X86_64
11611 .set_hv_timer = vmx_set_hv_timer,
11612 .cancel_hv_timer = vmx_cancel_hv_timer,
11615 .setup_mce = vmx_setup_mce,
11618 static int __init vmx_init(void)
11620 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
11621 __alignof__(struct vcpu_vmx), THIS_MODULE);
11625 #ifdef CONFIG_KEXEC_CORE
11626 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
11627 crash_vmclear_local_loaded_vmcss);
11633 static void __exit vmx_exit(void)
11635 #ifdef CONFIG_KEXEC_CORE
11636 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
11643 module_init(vmx_init)
11644 module_exit(vmx_exit)