2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
24 #include <linux/kvm_host.h>
25 #include <linux/module.h>
26 #include <linux/kernel.h>
28 #include <linux/highmem.h>
29 #include <linux/sched.h>
30 #include <linux/moduleparam.h>
31 #include <linux/mod_devicetable.h>
32 #include <linux/trace_events.h>
33 #include <linux/slab.h>
34 #include <linux/tboot.h>
35 #include <linux/hrtimer.h>
36 #include <linux/frame.h>
37 #include "kvm_cache_regs.h"
44 #include <asm/virtext.h>
46 #include <asm/fpu/internal.h>
47 #include <asm/perf_event.h>
48 #include <asm/debugreg.h>
49 #include <asm/kexec.h>
51 #include <asm/irq_remapping.h>
52 #include <asm/mmu_context.h>
53 #include <asm/nospec-branch.h>
58 #define __ex(x) __kvm_handle_fault_on_reboot(x)
59 #define __ex_clear(x, reg) \
60 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
62 MODULE_AUTHOR("Qumranet");
63 MODULE_LICENSE("GPL");
65 static const struct x86_cpu_id vmx_cpu_id[] = {
66 X86_FEATURE_MATCH(X86_FEATURE_VMX),
69 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
71 static bool __read_mostly enable_vpid = 1;
72 module_param_named(vpid, enable_vpid, bool, 0444);
74 static bool __read_mostly flexpriority_enabled = 1;
75 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
77 static bool __read_mostly enable_ept = 1;
78 module_param_named(ept, enable_ept, bool, S_IRUGO);
80 static bool __read_mostly enable_unrestricted_guest = 1;
81 module_param_named(unrestricted_guest,
82 enable_unrestricted_guest, bool, S_IRUGO);
84 static bool __read_mostly enable_ept_ad_bits = 1;
85 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
87 static bool __read_mostly emulate_invalid_guest_state = true;
88 module_param(emulate_invalid_guest_state, 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_LA57 | 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 bool nmi_known_unmasked;
204 unsigned long vmcs_host_cr3; /* May not match real cr3 */
205 unsigned long vmcs_host_cr4; /* May not match real cr4 */
206 struct list_head loaded_vmcss_on_cpu_link;
209 struct shared_msr_entry {
216 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
217 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
218 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
219 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
220 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
221 * More than one of these structures may exist, if L1 runs multiple L2 guests.
222 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
223 * underlying hardware which will be used to run L2.
224 * This structure is packed to ensure that its layout is identical across
225 * machines (necessary for live migration).
226 * If there are changes in this struct, VMCS12_REVISION must be changed.
228 typedef u64 natural_width;
229 struct __packed vmcs12 {
230 /* According to the Intel spec, a VMCS region must start with the
231 * following two fields. Then follow implementation-specific data.
236 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
237 u32 padding[7]; /* room for future expansion */
242 u64 vm_exit_msr_store_addr;
243 u64 vm_exit_msr_load_addr;
244 u64 vm_entry_msr_load_addr;
246 u64 virtual_apic_page_addr;
247 u64 apic_access_addr;
248 u64 posted_intr_desc_addr;
249 u64 vm_function_control;
251 u64 eoi_exit_bitmap0;
252 u64 eoi_exit_bitmap1;
253 u64 eoi_exit_bitmap2;
254 u64 eoi_exit_bitmap3;
255 u64 eptp_list_address;
257 u64 guest_physical_address;
258 u64 vmcs_link_pointer;
260 u64 guest_ia32_debugctl;
263 u64 guest_ia32_perf_global_ctrl;
271 u64 host_ia32_perf_global_ctrl;
272 u64 padding64[8]; /* room for future expansion */
274 * To allow migration of L1 (complete with its L2 guests) between
275 * machines of different natural widths (32 or 64 bit), we cannot have
276 * unsigned long fields with no explict size. We use u64 (aliased
277 * natural_width) instead. Luckily, x86 is little-endian.
279 natural_width cr0_guest_host_mask;
280 natural_width cr4_guest_host_mask;
281 natural_width cr0_read_shadow;
282 natural_width cr4_read_shadow;
283 natural_width cr3_target_value0;
284 natural_width cr3_target_value1;
285 natural_width cr3_target_value2;
286 natural_width cr3_target_value3;
287 natural_width exit_qualification;
288 natural_width guest_linear_address;
289 natural_width guest_cr0;
290 natural_width guest_cr3;
291 natural_width guest_cr4;
292 natural_width guest_es_base;
293 natural_width guest_cs_base;
294 natural_width guest_ss_base;
295 natural_width guest_ds_base;
296 natural_width guest_fs_base;
297 natural_width guest_gs_base;
298 natural_width guest_ldtr_base;
299 natural_width guest_tr_base;
300 natural_width guest_gdtr_base;
301 natural_width guest_idtr_base;
302 natural_width guest_dr7;
303 natural_width guest_rsp;
304 natural_width guest_rip;
305 natural_width guest_rflags;
306 natural_width guest_pending_dbg_exceptions;
307 natural_width guest_sysenter_esp;
308 natural_width guest_sysenter_eip;
309 natural_width host_cr0;
310 natural_width host_cr3;
311 natural_width host_cr4;
312 natural_width host_fs_base;
313 natural_width host_gs_base;
314 natural_width host_tr_base;
315 natural_width host_gdtr_base;
316 natural_width host_idtr_base;
317 natural_width host_ia32_sysenter_esp;
318 natural_width host_ia32_sysenter_eip;
319 natural_width host_rsp;
320 natural_width host_rip;
321 natural_width paddingl[8]; /* room for future expansion */
322 u32 pin_based_vm_exec_control;
323 u32 cpu_based_vm_exec_control;
324 u32 exception_bitmap;
325 u32 page_fault_error_code_mask;
326 u32 page_fault_error_code_match;
327 u32 cr3_target_count;
328 u32 vm_exit_controls;
329 u32 vm_exit_msr_store_count;
330 u32 vm_exit_msr_load_count;
331 u32 vm_entry_controls;
332 u32 vm_entry_msr_load_count;
333 u32 vm_entry_intr_info_field;
334 u32 vm_entry_exception_error_code;
335 u32 vm_entry_instruction_len;
337 u32 secondary_vm_exec_control;
338 u32 vm_instruction_error;
340 u32 vm_exit_intr_info;
341 u32 vm_exit_intr_error_code;
342 u32 idt_vectoring_info_field;
343 u32 idt_vectoring_error_code;
344 u32 vm_exit_instruction_len;
345 u32 vmx_instruction_info;
352 u32 guest_ldtr_limit;
354 u32 guest_gdtr_limit;
355 u32 guest_idtr_limit;
356 u32 guest_es_ar_bytes;
357 u32 guest_cs_ar_bytes;
358 u32 guest_ss_ar_bytes;
359 u32 guest_ds_ar_bytes;
360 u32 guest_fs_ar_bytes;
361 u32 guest_gs_ar_bytes;
362 u32 guest_ldtr_ar_bytes;
363 u32 guest_tr_ar_bytes;
364 u32 guest_interruptibility_info;
365 u32 guest_activity_state;
366 u32 guest_sysenter_cs;
367 u32 host_ia32_sysenter_cs;
368 u32 vmx_preemption_timer_value;
369 u32 padding32[7]; /* room for future expansion */
370 u16 virtual_processor_id;
372 u16 guest_es_selector;
373 u16 guest_cs_selector;
374 u16 guest_ss_selector;
375 u16 guest_ds_selector;
376 u16 guest_fs_selector;
377 u16 guest_gs_selector;
378 u16 guest_ldtr_selector;
379 u16 guest_tr_selector;
380 u16 guest_intr_status;
382 u16 host_es_selector;
383 u16 host_cs_selector;
384 u16 host_ss_selector;
385 u16 host_ds_selector;
386 u16 host_fs_selector;
387 u16 host_gs_selector;
388 u16 host_tr_selector;
392 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
393 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
394 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
396 #define VMCS12_REVISION 0x11e57ed0
399 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
400 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
401 * current implementation, 4K are reserved to avoid future complications.
403 #define VMCS12_SIZE 0x1000
405 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
407 struct list_head list;
409 struct loaded_vmcs vmcs02;
413 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
414 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
417 /* Has the level1 guest done vmxon? */
422 /* The guest-physical address of the current VMCS L1 keeps for L2 */
425 * Cache of the guest's VMCS, existing outside of guest memory.
426 * Loaded from guest memory during VMPTRLD. Flushed to guest
427 * memory during VMCLEAR and VMPTRLD.
429 struct vmcs12 *cached_vmcs12;
431 * Indicates if the shadow vmcs must be updated with the
432 * data hold by vmcs12
434 bool sync_shadow_vmcs;
436 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
437 struct list_head vmcs02_pool;
439 bool change_vmcs01_virtual_x2apic_mode;
440 /* L2 must run next, and mustn't decide to exit to L1. */
441 bool nested_run_pending;
443 * Guest pages referred to in vmcs02 with host-physical pointers, so
444 * we must keep them pinned while L2 runs.
446 struct page *apic_access_page;
447 struct page *virtual_apic_page;
448 struct page *pi_desc_page;
449 struct pi_desc *pi_desc;
453 unsigned long *msr_bitmap;
455 struct hrtimer preemption_timer;
456 bool preemption_timer_expired;
458 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
465 * We only store the "true" versions of the VMX capability MSRs. We
466 * generate the "non-true" versions by setting the must-be-1 bits
467 * according to the SDM.
469 u32 nested_vmx_procbased_ctls_low;
470 u32 nested_vmx_procbased_ctls_high;
471 u32 nested_vmx_secondary_ctls_low;
472 u32 nested_vmx_secondary_ctls_high;
473 u32 nested_vmx_pinbased_ctls_low;
474 u32 nested_vmx_pinbased_ctls_high;
475 u32 nested_vmx_exit_ctls_low;
476 u32 nested_vmx_exit_ctls_high;
477 u32 nested_vmx_entry_ctls_low;
478 u32 nested_vmx_entry_ctls_high;
479 u32 nested_vmx_misc_low;
480 u32 nested_vmx_misc_high;
481 u32 nested_vmx_ept_caps;
482 u32 nested_vmx_vpid_caps;
483 u64 nested_vmx_basic;
484 u64 nested_vmx_cr0_fixed0;
485 u64 nested_vmx_cr0_fixed1;
486 u64 nested_vmx_cr4_fixed0;
487 u64 nested_vmx_cr4_fixed1;
488 u64 nested_vmx_vmcs_enum;
489 u64 nested_vmx_vmfunc_controls;
492 #define POSTED_INTR_ON 0
493 #define POSTED_INTR_SN 1
495 /* Posted-Interrupt Descriptor */
497 u32 pir[8]; /* Posted interrupt requested */
500 /* bit 256 - Outstanding Notification */
502 /* bit 257 - Suppress Notification */
504 /* bit 271:258 - Reserved */
506 /* bit 279:272 - Notification Vector */
508 /* bit 287:280 - Reserved */
510 /* bit 319:288 - Notification Destination */
518 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
520 return test_and_set_bit(POSTED_INTR_ON,
521 (unsigned long *)&pi_desc->control);
524 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
526 return test_and_clear_bit(POSTED_INTR_ON,
527 (unsigned long *)&pi_desc->control);
530 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
532 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
535 static inline void pi_clear_sn(struct pi_desc *pi_desc)
537 return clear_bit(POSTED_INTR_SN,
538 (unsigned long *)&pi_desc->control);
541 static inline void pi_set_sn(struct pi_desc *pi_desc)
543 return set_bit(POSTED_INTR_SN,
544 (unsigned long *)&pi_desc->control);
547 static inline void pi_clear_on(struct pi_desc *pi_desc)
549 clear_bit(POSTED_INTR_ON,
550 (unsigned long *)&pi_desc->control);
553 static inline int pi_test_on(struct pi_desc *pi_desc)
555 return test_bit(POSTED_INTR_ON,
556 (unsigned long *)&pi_desc->control);
559 static inline int pi_test_sn(struct pi_desc *pi_desc)
561 return test_bit(POSTED_INTR_SN,
562 (unsigned long *)&pi_desc->control);
566 struct kvm_vcpu vcpu;
567 unsigned long host_rsp;
570 u32 idt_vectoring_info;
572 struct shared_msr_entry *guest_msrs;
575 unsigned long host_idt_base;
577 u64 msr_host_kernel_gs_base;
578 u64 msr_guest_kernel_gs_base;
580 u32 vm_entry_controls_shadow;
581 u32 vm_exit_controls_shadow;
582 u32 secondary_exec_control;
585 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
586 * non-nested (L1) guest, it always points to vmcs01. For a nested
587 * guest (L2), it points to a different VMCS.
589 struct loaded_vmcs vmcs01;
590 struct loaded_vmcs *loaded_vmcs;
591 bool __launched; /* temporary, used in vmx_vcpu_run */
592 struct msr_autoload {
594 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
595 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
599 u16 fs_sel, gs_sel, ldt_sel;
603 int gs_ldt_reload_needed;
604 int fs_reload_needed;
605 u64 msr_host_bndcfgs;
610 struct kvm_segment segs[8];
613 u32 bitmask; /* 4 bits per segment (1 bit per field) */
614 struct kvm_save_segment {
622 bool emulation_required;
626 /* Posted interrupt descriptor */
627 struct pi_desc pi_desc;
629 /* Support for a guest hypervisor (nested VMX) */
630 struct nested_vmx nested;
632 /* Dynamic PLE window. */
634 bool ple_window_dirty;
636 /* Support for PML */
637 #define PML_ENTITY_NUM 512
640 /* apic deadline value in host tsc */
643 u64 current_tsc_ratio;
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(GUEST_PML_INDEX, guest_pml_index),
753 FIELD(HOST_ES_SELECTOR, host_es_selector),
754 FIELD(HOST_CS_SELECTOR, host_cs_selector),
755 FIELD(HOST_SS_SELECTOR, host_ss_selector),
756 FIELD(HOST_DS_SELECTOR, host_ds_selector),
757 FIELD(HOST_FS_SELECTOR, host_fs_selector),
758 FIELD(HOST_GS_SELECTOR, host_gs_selector),
759 FIELD(HOST_TR_SELECTOR, host_tr_selector),
760 FIELD64(IO_BITMAP_A, io_bitmap_a),
761 FIELD64(IO_BITMAP_B, io_bitmap_b),
762 FIELD64(MSR_BITMAP, msr_bitmap),
763 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
764 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
765 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
766 FIELD64(TSC_OFFSET, tsc_offset),
767 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
768 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
769 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
770 FIELD64(VM_FUNCTION_CONTROL, vm_function_control),
771 FIELD64(EPT_POINTER, ept_pointer),
772 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
773 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
774 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
775 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
776 FIELD64(EPTP_LIST_ADDRESS, eptp_list_address),
777 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
778 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
779 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
780 FIELD64(PML_ADDRESS, pml_address),
781 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
782 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
783 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
784 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
785 FIELD64(GUEST_PDPTR0, guest_pdptr0),
786 FIELD64(GUEST_PDPTR1, guest_pdptr1),
787 FIELD64(GUEST_PDPTR2, guest_pdptr2),
788 FIELD64(GUEST_PDPTR3, guest_pdptr3),
789 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
790 FIELD64(HOST_IA32_PAT, host_ia32_pat),
791 FIELD64(HOST_IA32_EFER, host_ia32_efer),
792 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
793 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
794 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
795 FIELD(EXCEPTION_BITMAP, exception_bitmap),
796 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
797 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
798 FIELD(CR3_TARGET_COUNT, cr3_target_count),
799 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
800 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
801 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
802 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
803 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
804 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
805 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
806 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
807 FIELD(TPR_THRESHOLD, tpr_threshold),
808 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
809 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
810 FIELD(VM_EXIT_REASON, vm_exit_reason),
811 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
812 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
813 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
814 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
815 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
816 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
817 FIELD(GUEST_ES_LIMIT, guest_es_limit),
818 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
819 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
820 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
821 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
822 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
823 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
824 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
825 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
826 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
827 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
828 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
829 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
830 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
831 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
832 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
833 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
834 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
835 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
836 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
837 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
838 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
839 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
840 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
841 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
842 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
843 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
844 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
845 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
846 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
847 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
848 FIELD(EXIT_QUALIFICATION, exit_qualification),
849 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
850 FIELD(GUEST_CR0, guest_cr0),
851 FIELD(GUEST_CR3, guest_cr3),
852 FIELD(GUEST_CR4, guest_cr4),
853 FIELD(GUEST_ES_BASE, guest_es_base),
854 FIELD(GUEST_CS_BASE, guest_cs_base),
855 FIELD(GUEST_SS_BASE, guest_ss_base),
856 FIELD(GUEST_DS_BASE, guest_ds_base),
857 FIELD(GUEST_FS_BASE, guest_fs_base),
858 FIELD(GUEST_GS_BASE, guest_gs_base),
859 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
860 FIELD(GUEST_TR_BASE, guest_tr_base),
861 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
862 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
863 FIELD(GUEST_DR7, guest_dr7),
864 FIELD(GUEST_RSP, guest_rsp),
865 FIELD(GUEST_RIP, guest_rip),
866 FIELD(GUEST_RFLAGS, guest_rflags),
867 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
868 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
869 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
870 FIELD(HOST_CR0, host_cr0),
871 FIELD(HOST_CR3, host_cr3),
872 FIELD(HOST_CR4, host_cr4),
873 FIELD(HOST_FS_BASE, host_fs_base),
874 FIELD(HOST_GS_BASE, host_gs_base),
875 FIELD(HOST_TR_BASE, host_tr_base),
876 FIELD(HOST_GDTR_BASE, host_gdtr_base),
877 FIELD(HOST_IDTR_BASE, host_idtr_base),
878 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
879 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
880 FIELD(HOST_RSP, host_rsp),
881 FIELD(HOST_RIP, host_rip),
884 static inline short vmcs_field_to_offset(unsigned long field)
886 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
888 if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
889 vmcs_field_to_offset_table[field] == 0)
892 return vmcs_field_to_offset_table[field];
895 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
897 return to_vmx(vcpu)->nested.cached_vmcs12;
900 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu);
901 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
902 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
903 static bool vmx_xsaves_supported(void);
904 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
905 static void vmx_set_segment(struct kvm_vcpu *vcpu,
906 struct kvm_segment *var, int seg);
907 static void vmx_get_segment(struct kvm_vcpu *vcpu,
908 struct kvm_segment *var, int seg);
909 static bool guest_state_valid(struct kvm_vcpu *vcpu);
910 static u32 vmx_segment_access_rights(struct kvm_segment *var);
911 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
912 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
913 static int alloc_identity_pagetable(struct kvm *kvm);
914 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
915 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
916 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
919 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
920 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
922 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
923 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
925 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
928 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
929 * can find which vCPU should be waken up.
931 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
932 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
937 VMX_MSR_BITMAP_LEGACY,
938 VMX_MSR_BITMAP_LONGMODE,
939 VMX_MSR_BITMAP_LEGACY_X2APIC_APICV,
940 VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV,
941 VMX_MSR_BITMAP_LEGACY_X2APIC,
942 VMX_MSR_BITMAP_LONGMODE_X2APIC,
948 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
950 #define vmx_io_bitmap_a (vmx_bitmap[VMX_IO_BITMAP_A])
951 #define vmx_io_bitmap_b (vmx_bitmap[VMX_IO_BITMAP_B])
952 #define vmx_msr_bitmap_legacy (vmx_bitmap[VMX_MSR_BITMAP_LEGACY])
953 #define vmx_msr_bitmap_longmode (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE])
954 #define vmx_msr_bitmap_legacy_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC_APICV])
955 #define vmx_msr_bitmap_longmode_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV])
956 #define vmx_msr_bitmap_legacy_x2apic (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC])
957 #define vmx_msr_bitmap_longmode_x2apic (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC])
958 #define vmx_vmread_bitmap (vmx_bitmap[VMX_VMREAD_BITMAP])
959 #define vmx_vmwrite_bitmap (vmx_bitmap[VMX_VMWRITE_BITMAP])
961 static bool cpu_has_load_ia32_efer;
962 static bool cpu_has_load_perf_global_ctrl;
964 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
965 static DEFINE_SPINLOCK(vmx_vpid_lock);
967 static struct vmcs_config {
972 u32 pin_based_exec_ctrl;
973 u32 cpu_based_exec_ctrl;
974 u32 cpu_based_2nd_exec_ctrl;
979 static struct vmx_capability {
984 #define VMX_SEGMENT_FIELD(seg) \
985 [VCPU_SREG_##seg] = { \
986 .selector = GUEST_##seg##_SELECTOR, \
987 .base = GUEST_##seg##_BASE, \
988 .limit = GUEST_##seg##_LIMIT, \
989 .ar_bytes = GUEST_##seg##_AR_BYTES, \
992 static const struct kvm_vmx_segment_field {
997 } kvm_vmx_segment_fields[] = {
998 VMX_SEGMENT_FIELD(CS),
999 VMX_SEGMENT_FIELD(DS),
1000 VMX_SEGMENT_FIELD(ES),
1001 VMX_SEGMENT_FIELD(FS),
1002 VMX_SEGMENT_FIELD(GS),
1003 VMX_SEGMENT_FIELD(SS),
1004 VMX_SEGMENT_FIELD(TR),
1005 VMX_SEGMENT_FIELD(LDTR),
1008 static u64 host_efer;
1010 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
1013 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
1014 * away by decrementing the array size.
1016 static const u32 vmx_msr_index[] = {
1017 #ifdef CONFIG_X86_64
1018 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
1020 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
1023 static inline bool is_exception_n(u32 intr_info, u8 vector)
1025 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1026 INTR_INFO_VALID_MASK)) ==
1027 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
1030 static inline bool is_debug(u32 intr_info)
1032 return is_exception_n(intr_info, DB_VECTOR);
1035 static inline bool is_breakpoint(u32 intr_info)
1037 return is_exception_n(intr_info, BP_VECTOR);
1040 static inline bool is_page_fault(u32 intr_info)
1042 return is_exception_n(intr_info, PF_VECTOR);
1045 static inline bool is_no_device(u32 intr_info)
1047 return is_exception_n(intr_info, NM_VECTOR);
1050 static inline bool is_invalid_opcode(u32 intr_info)
1052 return is_exception_n(intr_info, UD_VECTOR);
1055 static inline bool is_external_interrupt(u32 intr_info)
1057 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1058 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1061 static inline bool is_machine_check(u32 intr_info)
1063 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1064 INTR_INFO_VALID_MASK)) ==
1065 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1068 static inline bool cpu_has_vmx_msr_bitmap(void)
1070 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1073 static inline bool cpu_has_vmx_tpr_shadow(void)
1075 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1078 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1080 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1083 static inline bool cpu_has_secondary_exec_ctrls(void)
1085 return vmcs_config.cpu_based_exec_ctrl &
1086 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1089 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1091 return vmcs_config.cpu_based_2nd_exec_ctrl &
1092 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1095 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1097 return vmcs_config.cpu_based_2nd_exec_ctrl &
1098 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1101 static inline bool cpu_has_vmx_apic_register_virt(void)
1103 return vmcs_config.cpu_based_2nd_exec_ctrl &
1104 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1107 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1109 return vmcs_config.cpu_based_2nd_exec_ctrl &
1110 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1114 * Comment's format: document - errata name - stepping - processor name.
1116 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1118 static u32 vmx_preemption_cpu_tfms[] = {
1119 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
1121 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
1122 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1123 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1125 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1127 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
1128 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
1130 * 320767.pdf - AAP86 - B1 -
1131 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1134 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1136 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1138 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1140 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1141 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1142 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1146 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1148 u32 eax = cpuid_eax(0x00000001), i;
1150 /* Clear the reserved bits */
1151 eax &= ~(0x3U << 14 | 0xfU << 28);
1152 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
1153 if (eax == vmx_preemption_cpu_tfms[i])
1159 static inline bool cpu_has_vmx_preemption_timer(void)
1161 return vmcs_config.pin_based_exec_ctrl &
1162 PIN_BASED_VMX_PREEMPTION_TIMER;
1165 static inline bool cpu_has_vmx_posted_intr(void)
1167 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1168 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1171 static inline bool cpu_has_vmx_apicv(void)
1173 return cpu_has_vmx_apic_register_virt() &&
1174 cpu_has_vmx_virtual_intr_delivery() &&
1175 cpu_has_vmx_posted_intr();
1178 static inline bool cpu_has_vmx_flexpriority(void)
1180 return cpu_has_vmx_tpr_shadow() &&
1181 cpu_has_vmx_virtualize_apic_accesses();
1184 static inline bool cpu_has_vmx_ept_execute_only(void)
1186 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1189 static inline bool cpu_has_vmx_ept_2m_page(void)
1191 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1194 static inline bool cpu_has_vmx_ept_1g_page(void)
1196 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1199 static inline bool cpu_has_vmx_ept_4levels(void)
1201 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1204 static inline bool cpu_has_vmx_ept_mt_wb(void)
1206 return vmx_capability.ept & VMX_EPTP_WB_BIT;
1209 static inline bool cpu_has_vmx_ept_5levels(void)
1211 return vmx_capability.ept & VMX_EPT_PAGE_WALK_5_BIT;
1214 static inline bool cpu_has_vmx_ept_ad_bits(void)
1216 return vmx_capability.ept & VMX_EPT_AD_BIT;
1219 static inline bool cpu_has_vmx_invept_context(void)
1221 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1224 static inline bool cpu_has_vmx_invept_global(void)
1226 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1229 static inline bool cpu_has_vmx_invvpid_single(void)
1231 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1234 static inline bool cpu_has_vmx_invvpid_global(void)
1236 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1239 static inline bool cpu_has_vmx_invvpid(void)
1241 return vmx_capability.vpid & VMX_VPID_INVVPID_BIT;
1244 static inline bool cpu_has_vmx_ept(void)
1246 return vmcs_config.cpu_based_2nd_exec_ctrl &
1247 SECONDARY_EXEC_ENABLE_EPT;
1250 static inline bool cpu_has_vmx_unrestricted_guest(void)
1252 return vmcs_config.cpu_based_2nd_exec_ctrl &
1253 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1256 static inline bool cpu_has_vmx_ple(void)
1258 return vmcs_config.cpu_based_2nd_exec_ctrl &
1259 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1262 static inline bool cpu_has_vmx_basic_inout(void)
1264 return (((u64)vmcs_config.basic_cap << 32) & VMX_BASIC_INOUT);
1267 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1269 return flexpriority_enabled && lapic_in_kernel(vcpu);
1272 static inline bool cpu_has_vmx_vpid(void)
1274 return vmcs_config.cpu_based_2nd_exec_ctrl &
1275 SECONDARY_EXEC_ENABLE_VPID;
1278 static inline bool cpu_has_vmx_rdtscp(void)
1280 return vmcs_config.cpu_based_2nd_exec_ctrl &
1281 SECONDARY_EXEC_RDTSCP;
1284 static inline bool cpu_has_vmx_invpcid(void)
1286 return vmcs_config.cpu_based_2nd_exec_ctrl &
1287 SECONDARY_EXEC_ENABLE_INVPCID;
1290 static inline bool cpu_has_vmx_wbinvd_exit(void)
1292 return vmcs_config.cpu_based_2nd_exec_ctrl &
1293 SECONDARY_EXEC_WBINVD_EXITING;
1296 static inline bool cpu_has_vmx_shadow_vmcs(void)
1299 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1300 /* check if the cpu supports writing r/o exit information fields */
1301 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1304 return vmcs_config.cpu_based_2nd_exec_ctrl &
1305 SECONDARY_EXEC_SHADOW_VMCS;
1308 static inline bool cpu_has_vmx_pml(void)
1310 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1313 static inline bool cpu_has_vmx_tsc_scaling(void)
1315 return vmcs_config.cpu_based_2nd_exec_ctrl &
1316 SECONDARY_EXEC_TSC_SCALING;
1319 static inline bool cpu_has_vmx_vmfunc(void)
1321 return vmcs_config.cpu_based_2nd_exec_ctrl &
1322 SECONDARY_EXEC_ENABLE_VMFUNC;
1325 static inline bool report_flexpriority(void)
1327 return flexpriority_enabled;
1330 static inline unsigned nested_cpu_vmx_misc_cr3_count(struct kvm_vcpu *vcpu)
1332 return vmx_misc_cr3_count(to_vmx(vcpu)->nested.nested_vmx_misc_low);
1335 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1337 return vmcs12->cpu_based_vm_exec_control & bit;
1340 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1342 return (vmcs12->cpu_based_vm_exec_control &
1343 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1344 (vmcs12->secondary_vm_exec_control & bit);
1347 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1349 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1352 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1354 return vmcs12->pin_based_vm_exec_control &
1355 PIN_BASED_VMX_PREEMPTION_TIMER;
1358 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1360 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1363 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1365 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
1368 static inline bool nested_cpu_has_pml(struct vmcs12 *vmcs12)
1370 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML);
1373 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1375 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1378 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1380 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1383 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1385 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1388 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1390 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1393 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1395 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1398 static inline bool nested_cpu_has_vmfunc(struct vmcs12 *vmcs12)
1400 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VMFUNC);
1403 static inline bool nested_cpu_has_eptp_switching(struct vmcs12 *vmcs12)
1405 return nested_cpu_has_vmfunc(vmcs12) &&
1406 (vmcs12->vm_function_control &
1407 VMX_VMFUNC_EPTP_SWITCHING);
1410 static inline bool is_nmi(u32 intr_info)
1412 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1413 == (INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK);
1416 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1418 unsigned long exit_qualification);
1419 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1420 struct vmcs12 *vmcs12,
1421 u32 reason, unsigned long qualification);
1423 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1427 for (i = 0; i < vmx->nmsrs; ++i)
1428 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1433 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1439 } operand = { vpid, 0, gva };
1441 asm volatile (__ex(ASM_VMX_INVVPID)
1442 /* CF==1 or ZF==1 --> rc = -1 */
1443 "; ja 1f ; ud2 ; 1:"
1444 : : "a"(&operand), "c"(ext) : "cc", "memory");
1447 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1451 } operand = {eptp, gpa};
1453 asm volatile (__ex(ASM_VMX_INVEPT)
1454 /* CF==1 or ZF==1 --> rc = -1 */
1455 "; ja 1f ; ud2 ; 1:\n"
1456 : : "a" (&operand), "c" (ext) : "cc", "memory");
1459 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1463 i = __find_msr_index(vmx, msr);
1465 return &vmx->guest_msrs[i];
1469 static void vmcs_clear(struct vmcs *vmcs)
1471 u64 phys_addr = __pa(vmcs);
1474 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1475 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1478 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1482 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1484 vmcs_clear(loaded_vmcs->vmcs);
1485 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
1486 vmcs_clear(loaded_vmcs->shadow_vmcs);
1487 loaded_vmcs->cpu = -1;
1488 loaded_vmcs->launched = 0;
1491 static void vmcs_load(struct vmcs *vmcs)
1493 u64 phys_addr = __pa(vmcs);
1496 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1497 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1500 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1504 #ifdef CONFIG_KEXEC_CORE
1506 * This bitmap is used to indicate whether the vmclear
1507 * operation is enabled on all cpus. All disabled by
1510 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1512 static inline void crash_enable_local_vmclear(int cpu)
1514 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1517 static inline void crash_disable_local_vmclear(int cpu)
1519 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1522 static inline int crash_local_vmclear_enabled(int cpu)
1524 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1527 static void crash_vmclear_local_loaded_vmcss(void)
1529 int cpu = raw_smp_processor_id();
1530 struct loaded_vmcs *v;
1532 if (!crash_local_vmclear_enabled(cpu))
1535 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1536 loaded_vmcss_on_cpu_link)
1537 vmcs_clear(v->vmcs);
1540 static inline void crash_enable_local_vmclear(int cpu) { }
1541 static inline void crash_disable_local_vmclear(int cpu) { }
1542 #endif /* CONFIG_KEXEC_CORE */
1544 static void __loaded_vmcs_clear(void *arg)
1546 struct loaded_vmcs *loaded_vmcs = arg;
1547 int cpu = raw_smp_processor_id();
1549 if (loaded_vmcs->cpu != cpu)
1550 return; /* vcpu migration can race with cpu offline */
1551 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1552 per_cpu(current_vmcs, cpu) = NULL;
1553 crash_disable_local_vmclear(cpu);
1554 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1557 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1558 * is before setting loaded_vmcs->vcpu to -1 which is done in
1559 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1560 * then adds the vmcs into percpu list before it is deleted.
1564 loaded_vmcs_init(loaded_vmcs);
1565 crash_enable_local_vmclear(cpu);
1568 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1570 int cpu = loaded_vmcs->cpu;
1573 smp_call_function_single(cpu,
1574 __loaded_vmcs_clear, loaded_vmcs, 1);
1577 static inline void vpid_sync_vcpu_single(int vpid)
1582 if (cpu_has_vmx_invvpid_single())
1583 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1586 static inline void vpid_sync_vcpu_global(void)
1588 if (cpu_has_vmx_invvpid_global())
1589 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1592 static inline void vpid_sync_context(int vpid)
1594 if (cpu_has_vmx_invvpid_single())
1595 vpid_sync_vcpu_single(vpid);
1597 vpid_sync_vcpu_global();
1600 static inline void ept_sync_global(void)
1602 if (cpu_has_vmx_invept_global())
1603 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1606 static inline void ept_sync_context(u64 eptp)
1609 if (cpu_has_vmx_invept_context())
1610 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1616 static __always_inline void vmcs_check16(unsigned long field)
1618 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1619 "16-bit accessor invalid for 64-bit field");
1620 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1621 "16-bit accessor invalid for 64-bit high field");
1622 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1623 "16-bit accessor invalid for 32-bit high field");
1624 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1625 "16-bit accessor invalid for natural width field");
1628 static __always_inline void vmcs_check32(unsigned long field)
1630 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1631 "32-bit accessor invalid for 16-bit field");
1632 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1633 "32-bit accessor invalid for natural width field");
1636 static __always_inline void vmcs_check64(unsigned long field)
1638 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1639 "64-bit accessor invalid for 16-bit field");
1640 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1641 "64-bit accessor invalid for 64-bit high field");
1642 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1643 "64-bit accessor invalid for 32-bit field");
1644 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1645 "64-bit accessor invalid for natural width field");
1648 static __always_inline void vmcs_checkl(unsigned long field)
1650 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1651 "Natural width accessor invalid for 16-bit field");
1652 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1653 "Natural width accessor invalid for 64-bit field");
1654 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1655 "Natural width accessor invalid for 64-bit high field");
1656 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1657 "Natural width accessor invalid for 32-bit field");
1660 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1662 unsigned long value;
1664 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1665 : "=a"(value) : "d"(field) : "cc");
1669 static __always_inline u16 vmcs_read16(unsigned long field)
1671 vmcs_check16(field);
1672 return __vmcs_readl(field);
1675 static __always_inline u32 vmcs_read32(unsigned long field)
1677 vmcs_check32(field);
1678 return __vmcs_readl(field);
1681 static __always_inline u64 vmcs_read64(unsigned long field)
1683 vmcs_check64(field);
1684 #ifdef CONFIG_X86_64
1685 return __vmcs_readl(field);
1687 return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1691 static __always_inline unsigned long vmcs_readl(unsigned long field)
1694 return __vmcs_readl(field);
1697 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1699 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1700 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1704 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1708 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1709 : "=q"(error) : "a"(value), "d"(field) : "cc");
1710 if (unlikely(error))
1711 vmwrite_error(field, value);
1714 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1716 vmcs_check16(field);
1717 __vmcs_writel(field, value);
1720 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1722 vmcs_check32(field);
1723 __vmcs_writel(field, value);
1726 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1728 vmcs_check64(field);
1729 __vmcs_writel(field, value);
1730 #ifndef CONFIG_X86_64
1732 __vmcs_writel(field+1, value >> 32);
1736 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1739 __vmcs_writel(field, value);
1742 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1744 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1745 "vmcs_clear_bits does not support 64-bit fields");
1746 __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1749 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1751 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1752 "vmcs_set_bits does not support 64-bit fields");
1753 __vmcs_writel(field, __vmcs_readl(field) | mask);
1756 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
1758 vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
1761 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1763 vmcs_write32(VM_ENTRY_CONTROLS, val);
1764 vmx->vm_entry_controls_shadow = val;
1767 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1769 if (vmx->vm_entry_controls_shadow != val)
1770 vm_entry_controls_init(vmx, val);
1773 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1775 return vmx->vm_entry_controls_shadow;
1779 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1781 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1784 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1786 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1789 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
1791 vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
1794 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1796 vmcs_write32(VM_EXIT_CONTROLS, val);
1797 vmx->vm_exit_controls_shadow = val;
1800 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1802 if (vmx->vm_exit_controls_shadow != val)
1803 vm_exit_controls_init(vmx, val);
1806 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1808 return vmx->vm_exit_controls_shadow;
1812 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1814 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1817 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1819 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1822 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1824 vmx->segment_cache.bitmask = 0;
1827 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1831 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1833 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1834 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1835 vmx->segment_cache.bitmask = 0;
1837 ret = vmx->segment_cache.bitmask & mask;
1838 vmx->segment_cache.bitmask |= mask;
1842 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1844 u16 *p = &vmx->segment_cache.seg[seg].selector;
1846 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1847 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1851 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1853 ulong *p = &vmx->segment_cache.seg[seg].base;
1855 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1856 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1860 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1862 u32 *p = &vmx->segment_cache.seg[seg].limit;
1864 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1865 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1869 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1871 u32 *p = &vmx->segment_cache.seg[seg].ar;
1873 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1874 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1878 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1882 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1883 (1u << DB_VECTOR) | (1u << AC_VECTOR);
1884 if ((vcpu->guest_debug &
1885 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1886 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1887 eb |= 1u << BP_VECTOR;
1888 if (to_vmx(vcpu)->rmode.vm86_active)
1891 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1893 /* When we are running a nested L2 guest and L1 specified for it a
1894 * certain exception bitmap, we must trap the same exceptions and pass
1895 * them to L1. When running L2, we will only handle the exceptions
1896 * specified above if L1 did not want them.
1898 if (is_guest_mode(vcpu))
1899 eb |= get_vmcs12(vcpu)->exception_bitmap;
1901 vmcs_write32(EXCEPTION_BITMAP, eb);
1904 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1905 unsigned long entry, unsigned long exit)
1907 vm_entry_controls_clearbit(vmx, entry);
1908 vm_exit_controls_clearbit(vmx, exit);
1911 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1914 struct msr_autoload *m = &vmx->msr_autoload;
1918 if (cpu_has_load_ia32_efer) {
1919 clear_atomic_switch_msr_special(vmx,
1920 VM_ENTRY_LOAD_IA32_EFER,
1921 VM_EXIT_LOAD_IA32_EFER);
1925 case MSR_CORE_PERF_GLOBAL_CTRL:
1926 if (cpu_has_load_perf_global_ctrl) {
1927 clear_atomic_switch_msr_special(vmx,
1928 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1929 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1935 for (i = 0; i < m->nr; ++i)
1936 if (m->guest[i].index == msr)
1942 m->guest[i] = m->guest[m->nr];
1943 m->host[i] = m->host[m->nr];
1944 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1945 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1948 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1949 unsigned long entry, unsigned long exit,
1950 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1951 u64 guest_val, u64 host_val)
1953 vmcs_write64(guest_val_vmcs, guest_val);
1954 vmcs_write64(host_val_vmcs, host_val);
1955 vm_entry_controls_setbit(vmx, entry);
1956 vm_exit_controls_setbit(vmx, exit);
1959 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1960 u64 guest_val, u64 host_val)
1963 struct msr_autoload *m = &vmx->msr_autoload;
1967 if (cpu_has_load_ia32_efer) {
1968 add_atomic_switch_msr_special(vmx,
1969 VM_ENTRY_LOAD_IA32_EFER,
1970 VM_EXIT_LOAD_IA32_EFER,
1973 guest_val, host_val);
1977 case MSR_CORE_PERF_GLOBAL_CTRL:
1978 if (cpu_has_load_perf_global_ctrl) {
1979 add_atomic_switch_msr_special(vmx,
1980 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1981 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1982 GUEST_IA32_PERF_GLOBAL_CTRL,
1983 HOST_IA32_PERF_GLOBAL_CTRL,
1984 guest_val, host_val);
1988 case MSR_IA32_PEBS_ENABLE:
1989 /* PEBS needs a quiescent period after being disabled (to write
1990 * a record). Disabling PEBS through VMX MSR swapping doesn't
1991 * provide that period, so a CPU could write host's record into
1994 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1997 for (i = 0; i < m->nr; ++i)
1998 if (m->guest[i].index == msr)
2001 if (i == NR_AUTOLOAD_MSRS) {
2002 printk_once(KERN_WARNING "Not enough msr switch entries. "
2003 "Can't add msr %x\n", msr);
2005 } else if (i == m->nr) {
2007 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
2008 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
2011 m->guest[i].index = msr;
2012 m->guest[i].value = guest_val;
2013 m->host[i].index = msr;
2014 m->host[i].value = host_val;
2017 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
2019 u64 guest_efer = vmx->vcpu.arch.efer;
2020 u64 ignore_bits = 0;
2024 * NX is needed to handle CR0.WP=1, CR4.SMEP=1. Testing
2025 * host CPUID is more efficient than testing guest CPUID
2026 * or CR4. Host SMEP is anyway a requirement for guest SMEP.
2028 if (boot_cpu_has(X86_FEATURE_SMEP))
2029 guest_efer |= EFER_NX;
2030 else if (!(guest_efer & EFER_NX))
2031 ignore_bits |= EFER_NX;
2035 * LMA and LME handled by hardware; SCE meaningless outside long mode.
2037 ignore_bits |= EFER_SCE;
2038 #ifdef CONFIG_X86_64
2039 ignore_bits |= EFER_LMA | EFER_LME;
2040 /* SCE is meaningful only in long mode on Intel */
2041 if (guest_efer & EFER_LMA)
2042 ignore_bits &= ~(u64)EFER_SCE;
2045 clear_atomic_switch_msr(vmx, MSR_EFER);
2048 * On EPT, we can't emulate NX, so we must switch EFER atomically.
2049 * On CPUs that support "load IA32_EFER", always switch EFER
2050 * atomically, since it's faster than switching it manually.
2052 if (cpu_has_load_ia32_efer ||
2053 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
2054 if (!(guest_efer & EFER_LMA))
2055 guest_efer &= ~EFER_LME;
2056 if (guest_efer != host_efer)
2057 add_atomic_switch_msr(vmx, MSR_EFER,
2058 guest_efer, host_efer);
2061 guest_efer &= ~ignore_bits;
2062 guest_efer |= host_efer & ignore_bits;
2064 vmx->guest_msrs[efer_offset].data = guest_efer;
2065 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
2071 #ifdef CONFIG_X86_32
2073 * On 32-bit kernels, VM exits still load the FS and GS bases from the
2074 * VMCS rather than the segment table. KVM uses this helper to figure
2075 * out the current bases to poke them into the VMCS before entry.
2077 static unsigned long segment_base(u16 selector)
2079 struct desc_struct *table;
2082 if (!(selector & ~SEGMENT_RPL_MASK))
2085 table = get_current_gdt_ro();
2087 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
2088 u16 ldt_selector = kvm_read_ldt();
2090 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
2093 table = (struct desc_struct *)segment_base(ldt_selector);
2095 v = get_desc_base(&table[selector >> 3]);
2100 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
2102 struct vcpu_vmx *vmx = to_vmx(vcpu);
2105 if (vmx->host_state.loaded)
2108 vmx->host_state.loaded = 1;
2110 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
2111 * allow segment selectors with cpl > 0 or ti == 1.
2113 vmx->host_state.ldt_sel = kvm_read_ldt();
2114 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
2115 savesegment(fs, vmx->host_state.fs_sel);
2116 if (!(vmx->host_state.fs_sel & 7)) {
2117 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
2118 vmx->host_state.fs_reload_needed = 0;
2120 vmcs_write16(HOST_FS_SELECTOR, 0);
2121 vmx->host_state.fs_reload_needed = 1;
2123 savesegment(gs, vmx->host_state.gs_sel);
2124 if (!(vmx->host_state.gs_sel & 7))
2125 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
2127 vmcs_write16(HOST_GS_SELECTOR, 0);
2128 vmx->host_state.gs_ldt_reload_needed = 1;
2131 #ifdef CONFIG_X86_64
2132 savesegment(ds, vmx->host_state.ds_sel);
2133 savesegment(es, vmx->host_state.es_sel);
2136 #ifdef CONFIG_X86_64
2137 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
2138 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
2140 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
2141 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
2144 #ifdef CONFIG_X86_64
2145 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2146 if (is_long_mode(&vmx->vcpu))
2147 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2149 if (boot_cpu_has(X86_FEATURE_MPX))
2150 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2151 for (i = 0; i < vmx->save_nmsrs; ++i)
2152 kvm_set_shared_msr(vmx->guest_msrs[i].index,
2153 vmx->guest_msrs[i].data,
2154 vmx->guest_msrs[i].mask);
2157 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
2159 if (!vmx->host_state.loaded)
2162 ++vmx->vcpu.stat.host_state_reload;
2163 vmx->host_state.loaded = 0;
2164 #ifdef CONFIG_X86_64
2165 if (is_long_mode(&vmx->vcpu))
2166 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2168 if (vmx->host_state.gs_ldt_reload_needed) {
2169 kvm_load_ldt(vmx->host_state.ldt_sel);
2170 #ifdef CONFIG_X86_64
2171 load_gs_index(vmx->host_state.gs_sel);
2173 loadsegment(gs, vmx->host_state.gs_sel);
2176 if (vmx->host_state.fs_reload_needed)
2177 loadsegment(fs, vmx->host_state.fs_sel);
2178 #ifdef CONFIG_X86_64
2179 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2180 loadsegment(ds, vmx->host_state.ds_sel);
2181 loadsegment(es, vmx->host_state.es_sel);
2184 invalidate_tss_limit();
2185 #ifdef CONFIG_X86_64
2186 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2188 if (vmx->host_state.msr_host_bndcfgs)
2189 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2190 load_fixmap_gdt(raw_smp_processor_id());
2193 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2196 __vmx_load_host_state(vmx);
2200 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2202 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2203 struct pi_desc old, new;
2207 * In case of hot-plug or hot-unplug, we may have to undo
2208 * vmx_vcpu_pi_put even if there is no assigned device. And we
2209 * always keep PI.NDST up to date for simplicity: it makes the
2210 * code easier, and CPU migration is not a fast path.
2212 if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
2216 * First handle the simple case where no cmpxchg is necessary; just
2217 * allow posting non-urgent interrupts.
2219 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
2220 * PI.NDST: pi_post_block will do it for us and the wakeup_handler
2221 * expects the VCPU to be on the blocked_vcpu_list that matches
2224 if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR ||
2226 pi_clear_sn(pi_desc);
2230 /* The full case. */
2232 old.control = new.control = pi_desc->control;
2234 dest = cpu_physical_id(cpu);
2236 if (x2apic_enabled())
2239 new.ndst = (dest << 8) & 0xFF00;
2242 } while (cmpxchg64(&pi_desc->control, old.control,
2243 new.control) != old.control);
2246 static void decache_tsc_multiplier(struct vcpu_vmx *vmx)
2248 vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
2249 vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2253 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2254 * vcpu mutex is already taken.
2256 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2258 struct vcpu_vmx *vmx = to_vmx(vcpu);
2259 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
2261 if (!already_loaded) {
2262 loaded_vmcs_clear(vmx->loaded_vmcs);
2263 local_irq_disable();
2264 crash_disable_local_vmclear(cpu);
2267 * Read loaded_vmcs->cpu should be before fetching
2268 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2269 * See the comments in __loaded_vmcs_clear().
2273 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2274 &per_cpu(loaded_vmcss_on_cpu, cpu));
2275 crash_enable_local_vmclear(cpu);
2279 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2280 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2281 vmcs_load(vmx->loaded_vmcs->vmcs);
2284 if (!already_loaded) {
2285 void *gdt = get_current_gdt_ro();
2286 unsigned long sysenter_esp;
2288 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2291 * Linux uses per-cpu TSS and GDT, so set these when switching
2292 * processors. See 22.2.4.
2294 vmcs_writel(HOST_TR_BASE,
2295 (unsigned long)&get_cpu_entry_area(cpu)->tss.x86_tss);
2296 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
2299 * VM exits change the host TR limit to 0x67 after a VM
2300 * exit. This is okay, since 0x67 covers everything except
2301 * the IO bitmap and have have code to handle the IO bitmap
2302 * being lost after a VM exit.
2304 BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
2306 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2307 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2309 vmx->loaded_vmcs->cpu = cpu;
2312 /* Setup TSC multiplier */
2313 if (kvm_has_tsc_control &&
2314 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
2315 decache_tsc_multiplier(vmx);
2317 vmx_vcpu_pi_load(vcpu, cpu);
2318 vmx->host_pkru = read_pkru();
2321 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2323 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2325 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2326 !irq_remapping_cap(IRQ_POSTING_CAP) ||
2327 !kvm_vcpu_apicv_active(vcpu))
2330 /* Set SN when the vCPU is preempted */
2331 if (vcpu->preempted)
2335 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2337 vmx_vcpu_pi_put(vcpu);
2339 __vmx_load_host_state(to_vmx(vcpu));
2342 static bool emulation_required(struct kvm_vcpu *vcpu)
2344 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
2347 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2350 * Return the cr0 value that a nested guest would read. This is a combination
2351 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2352 * its hypervisor (cr0_read_shadow).
2354 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2356 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2357 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2359 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2361 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2362 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2365 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2367 unsigned long rflags, save_rflags;
2369 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2370 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2371 rflags = vmcs_readl(GUEST_RFLAGS);
2372 if (to_vmx(vcpu)->rmode.vm86_active) {
2373 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2374 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2375 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2377 to_vmx(vcpu)->rflags = rflags;
2379 return to_vmx(vcpu)->rflags;
2382 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2384 unsigned long old_rflags = vmx_get_rflags(vcpu);
2386 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2387 to_vmx(vcpu)->rflags = rflags;
2388 if (to_vmx(vcpu)->rmode.vm86_active) {
2389 to_vmx(vcpu)->rmode.save_rflags = rflags;
2390 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2392 vmcs_writel(GUEST_RFLAGS, rflags);
2394 if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
2395 to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
2398 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2400 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2403 if (interruptibility & GUEST_INTR_STATE_STI)
2404 ret |= KVM_X86_SHADOW_INT_STI;
2405 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2406 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2411 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2413 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2414 u32 interruptibility = interruptibility_old;
2416 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2418 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2419 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2420 else if (mask & KVM_X86_SHADOW_INT_STI)
2421 interruptibility |= GUEST_INTR_STATE_STI;
2423 if ((interruptibility != interruptibility_old))
2424 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2427 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2431 rip = kvm_rip_read(vcpu);
2432 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2433 kvm_rip_write(vcpu, rip);
2435 /* skipping an emulated instruction also counts */
2436 vmx_set_interrupt_shadow(vcpu, 0);
2439 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
2440 unsigned long exit_qual)
2442 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2443 unsigned int nr = vcpu->arch.exception.nr;
2444 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2446 if (vcpu->arch.exception.has_error_code) {
2447 vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
2448 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2451 if (kvm_exception_is_soft(nr))
2452 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2454 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2456 if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
2457 vmx_get_nmi_mask(vcpu))
2458 intr_info |= INTR_INFO_UNBLOCK_NMI;
2460 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
2464 * KVM wants to inject page-faults which it got to the guest. This function
2465 * checks whether in a nested guest, we need to inject them to L1 or L2.
2467 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
2469 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2470 unsigned int nr = vcpu->arch.exception.nr;
2472 if (nr == PF_VECTOR) {
2473 if (vcpu->arch.exception.nested_apf) {
2474 *exit_qual = vcpu->arch.apf.nested_apf_token;
2478 * FIXME: we must not write CR2 when L1 intercepts an L2 #PF exception.
2479 * The fix is to add the ancillary datum (CR2 or DR6) to structs
2480 * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6
2481 * can be written only when inject_pending_event runs. This should be
2482 * conditional on a new capability---if the capability is disabled,
2483 * kvm_multiple_exception would write the ancillary information to
2484 * CR2 or DR6, for backwards ABI-compatibility.
2486 if (nested_vmx_is_page_fault_vmexit(vmcs12,
2487 vcpu->arch.exception.error_code)) {
2488 *exit_qual = vcpu->arch.cr2;
2492 if (vmcs12->exception_bitmap & (1u << nr)) {
2493 if (nr == DB_VECTOR)
2494 *exit_qual = vcpu->arch.dr6;
2504 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
2506 struct vcpu_vmx *vmx = to_vmx(vcpu);
2507 unsigned nr = vcpu->arch.exception.nr;
2508 bool has_error_code = vcpu->arch.exception.has_error_code;
2509 u32 error_code = vcpu->arch.exception.error_code;
2510 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2512 if (has_error_code) {
2513 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2514 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2517 if (vmx->rmode.vm86_active) {
2519 if (kvm_exception_is_soft(nr))
2520 inc_eip = vcpu->arch.event_exit_inst_len;
2521 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2522 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2526 if (kvm_exception_is_soft(nr)) {
2527 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2528 vmx->vcpu.arch.event_exit_inst_len);
2529 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2531 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2533 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2536 static bool vmx_rdtscp_supported(void)
2538 return cpu_has_vmx_rdtscp();
2541 static bool vmx_invpcid_supported(void)
2543 return cpu_has_vmx_invpcid() && enable_ept;
2547 * Swap MSR entry in host/guest MSR entry array.
2549 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2551 struct shared_msr_entry tmp;
2553 tmp = vmx->guest_msrs[to];
2554 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2555 vmx->guest_msrs[from] = tmp;
2558 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2560 unsigned long *msr_bitmap;
2562 if (is_guest_mode(vcpu))
2563 msr_bitmap = to_vmx(vcpu)->nested.msr_bitmap;
2564 else if (cpu_has_secondary_exec_ctrls() &&
2565 (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
2566 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
2567 if (enable_apicv && kvm_vcpu_apicv_active(vcpu)) {
2568 if (is_long_mode(vcpu))
2569 msr_bitmap = vmx_msr_bitmap_longmode_x2apic_apicv;
2571 msr_bitmap = vmx_msr_bitmap_legacy_x2apic_apicv;
2573 if (is_long_mode(vcpu))
2574 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2576 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2579 if (is_long_mode(vcpu))
2580 msr_bitmap = vmx_msr_bitmap_longmode;
2582 msr_bitmap = vmx_msr_bitmap_legacy;
2585 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2589 * Set up the vmcs to automatically save and restore system
2590 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2591 * mode, as fiddling with msrs is very expensive.
2593 static void setup_msrs(struct vcpu_vmx *vmx)
2595 int save_nmsrs, index;
2598 #ifdef CONFIG_X86_64
2599 if (is_long_mode(&vmx->vcpu)) {
2600 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2602 move_msr_up(vmx, index, save_nmsrs++);
2603 index = __find_msr_index(vmx, MSR_LSTAR);
2605 move_msr_up(vmx, index, save_nmsrs++);
2606 index = __find_msr_index(vmx, MSR_CSTAR);
2608 move_msr_up(vmx, index, save_nmsrs++);
2609 index = __find_msr_index(vmx, MSR_TSC_AUX);
2610 if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
2611 move_msr_up(vmx, index, save_nmsrs++);
2613 * MSR_STAR is only needed on long mode guests, and only
2614 * if efer.sce is enabled.
2616 index = __find_msr_index(vmx, MSR_STAR);
2617 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2618 move_msr_up(vmx, index, save_nmsrs++);
2621 index = __find_msr_index(vmx, MSR_EFER);
2622 if (index >= 0 && update_transition_efer(vmx, index))
2623 move_msr_up(vmx, index, save_nmsrs++);
2625 vmx->save_nmsrs = save_nmsrs;
2627 if (cpu_has_vmx_msr_bitmap())
2628 vmx_set_msr_bitmap(&vmx->vcpu);
2632 * reads and returns guest's timestamp counter "register"
2633 * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2634 * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2636 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2638 u64 host_tsc, tsc_offset;
2641 tsc_offset = vmcs_read64(TSC_OFFSET);
2642 return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2646 * writes 'offset' into guest's timestamp counter offset register
2648 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2650 if (is_guest_mode(vcpu)) {
2652 * We're here if L1 chose not to trap WRMSR to TSC. According
2653 * to the spec, this should set L1's TSC; The offset that L1
2654 * set for L2 remains unchanged, and still needs to be added
2655 * to the newly set TSC to get L2's TSC.
2657 struct vmcs12 *vmcs12;
2658 /* recalculate vmcs02.TSC_OFFSET: */
2659 vmcs12 = get_vmcs12(vcpu);
2660 vmcs_write64(TSC_OFFSET, offset +
2661 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2662 vmcs12->tsc_offset : 0));
2664 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2665 vmcs_read64(TSC_OFFSET), offset);
2666 vmcs_write64(TSC_OFFSET, offset);
2671 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2672 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2673 * all guests if the "nested" module option is off, and can also be disabled
2674 * for a single guest by disabling its VMX cpuid bit.
2676 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2678 return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
2682 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2683 * returned for the various VMX controls MSRs when nested VMX is enabled.
2684 * The same values should also be used to verify that vmcs12 control fields are
2685 * valid during nested entry from L1 to L2.
2686 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2687 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2688 * bit in the high half is on if the corresponding bit in the control field
2689 * may be on. See also vmx_control_verify().
2691 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2694 * Note that as a general rule, the high half of the MSRs (bits in
2695 * the control fields which may be 1) should be initialized by the
2696 * intersection of the underlying hardware's MSR (i.e., features which
2697 * can be supported) and the list of features we want to expose -
2698 * because they are known to be properly supported in our code.
2699 * Also, usually, the low half of the MSRs (bits which must be 1) can
2700 * be set to 0, meaning that L1 may turn off any of these bits. The
2701 * reason is that if one of these bits is necessary, it will appear
2702 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2703 * fields of vmcs01 and vmcs02, will turn these bits off - and
2704 * nested_vmx_exit_reflected() will not pass related exits to L1.
2705 * These rules have exceptions below.
2708 /* pin-based controls */
2709 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2710 vmx->nested.nested_vmx_pinbased_ctls_low,
2711 vmx->nested.nested_vmx_pinbased_ctls_high);
2712 vmx->nested.nested_vmx_pinbased_ctls_low |=
2713 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2714 vmx->nested.nested_vmx_pinbased_ctls_high &=
2715 PIN_BASED_EXT_INTR_MASK |
2716 PIN_BASED_NMI_EXITING |
2717 PIN_BASED_VIRTUAL_NMIS;
2718 vmx->nested.nested_vmx_pinbased_ctls_high |=
2719 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2720 PIN_BASED_VMX_PREEMPTION_TIMER;
2721 if (kvm_vcpu_apicv_active(&vmx->vcpu))
2722 vmx->nested.nested_vmx_pinbased_ctls_high |=
2723 PIN_BASED_POSTED_INTR;
2726 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2727 vmx->nested.nested_vmx_exit_ctls_low,
2728 vmx->nested.nested_vmx_exit_ctls_high);
2729 vmx->nested.nested_vmx_exit_ctls_low =
2730 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2732 vmx->nested.nested_vmx_exit_ctls_high &=
2733 #ifdef CONFIG_X86_64
2734 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2736 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2737 vmx->nested.nested_vmx_exit_ctls_high |=
2738 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2739 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2740 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2742 if (kvm_mpx_supported())
2743 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2745 /* We support free control of debug control saving. */
2746 vmx->nested.nested_vmx_exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2748 /* entry controls */
2749 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2750 vmx->nested.nested_vmx_entry_ctls_low,
2751 vmx->nested.nested_vmx_entry_ctls_high);
2752 vmx->nested.nested_vmx_entry_ctls_low =
2753 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2754 vmx->nested.nested_vmx_entry_ctls_high &=
2755 #ifdef CONFIG_X86_64
2756 VM_ENTRY_IA32E_MODE |
2758 VM_ENTRY_LOAD_IA32_PAT;
2759 vmx->nested.nested_vmx_entry_ctls_high |=
2760 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2761 if (kvm_mpx_supported())
2762 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2764 /* We support free control of debug control loading. */
2765 vmx->nested.nested_vmx_entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2767 /* cpu-based controls */
2768 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2769 vmx->nested.nested_vmx_procbased_ctls_low,
2770 vmx->nested.nested_vmx_procbased_ctls_high);
2771 vmx->nested.nested_vmx_procbased_ctls_low =
2772 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2773 vmx->nested.nested_vmx_procbased_ctls_high &=
2774 CPU_BASED_VIRTUAL_INTR_PENDING |
2775 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2776 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2777 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2778 CPU_BASED_CR3_STORE_EXITING |
2779 #ifdef CONFIG_X86_64
2780 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2782 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2783 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2784 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2785 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2786 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2788 * We can allow some features even when not supported by the
2789 * hardware. For example, L1 can specify an MSR bitmap - and we
2790 * can use it to avoid exits to L1 - even when L0 runs L2
2791 * without MSR bitmaps.
2793 vmx->nested.nested_vmx_procbased_ctls_high |=
2794 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2795 CPU_BASED_USE_MSR_BITMAPS;
2797 /* We support free control of CR3 access interception. */
2798 vmx->nested.nested_vmx_procbased_ctls_low &=
2799 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2802 * secondary cpu-based controls. Do not include those that
2803 * depend on CPUID bits, they are added later by vmx_cpuid_update.
2805 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2806 vmx->nested.nested_vmx_secondary_ctls_low,
2807 vmx->nested.nested_vmx_secondary_ctls_high);
2808 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2809 vmx->nested.nested_vmx_secondary_ctls_high &=
2810 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2811 SECONDARY_EXEC_DESC |
2812 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2813 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2814 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2815 SECONDARY_EXEC_WBINVD_EXITING;
2818 /* nested EPT: emulate EPT also to L1 */
2819 vmx->nested.nested_vmx_secondary_ctls_high |=
2820 SECONDARY_EXEC_ENABLE_EPT;
2821 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2822 VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
2823 if (cpu_has_vmx_ept_execute_only())
2824 vmx->nested.nested_vmx_ept_caps |=
2825 VMX_EPT_EXECUTE_ONLY_BIT;
2826 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2827 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
2828 VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
2829 VMX_EPT_1GB_PAGE_BIT;
2830 if (enable_ept_ad_bits) {
2831 vmx->nested.nested_vmx_secondary_ctls_high |=
2832 SECONDARY_EXEC_ENABLE_PML;
2833 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_AD_BIT;
2836 vmx->nested.nested_vmx_ept_caps = 0;
2838 if (cpu_has_vmx_vmfunc()) {
2839 vmx->nested.nested_vmx_secondary_ctls_high |=
2840 SECONDARY_EXEC_ENABLE_VMFUNC;
2842 * Advertise EPTP switching unconditionally
2843 * since we emulate it
2845 vmx->nested.nested_vmx_vmfunc_controls =
2846 VMX_VMFUNC_EPTP_SWITCHING;
2850 * Old versions of KVM use the single-context version without
2851 * checking for support, so declare that it is supported even
2852 * though it is treated as global context. The alternative is
2853 * not failing the single-context invvpid, and it is worse.
2856 vmx->nested.nested_vmx_secondary_ctls_high |=
2857 SECONDARY_EXEC_ENABLE_VPID;
2858 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2859 VMX_VPID_EXTENT_SUPPORTED_MASK;
2861 vmx->nested.nested_vmx_vpid_caps = 0;
2863 if (enable_unrestricted_guest)
2864 vmx->nested.nested_vmx_secondary_ctls_high |=
2865 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2867 /* miscellaneous data */
2868 rdmsr(MSR_IA32_VMX_MISC,
2869 vmx->nested.nested_vmx_misc_low,
2870 vmx->nested.nested_vmx_misc_high);
2871 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2872 vmx->nested.nested_vmx_misc_low |=
2873 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2874 VMX_MISC_ACTIVITY_HLT;
2875 vmx->nested.nested_vmx_misc_high = 0;
2878 * This MSR reports some information about VMX support. We
2879 * should return information about the VMX we emulate for the
2880 * guest, and the VMCS structure we give it - not about the
2881 * VMX support of the underlying hardware.
2883 vmx->nested.nested_vmx_basic =
2885 VMX_BASIC_TRUE_CTLS |
2886 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2887 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2889 if (cpu_has_vmx_basic_inout())
2890 vmx->nested.nested_vmx_basic |= VMX_BASIC_INOUT;
2893 * These MSRs specify bits which the guest must keep fixed on
2894 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2895 * We picked the standard core2 setting.
2897 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2898 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2899 vmx->nested.nested_vmx_cr0_fixed0 = VMXON_CR0_ALWAYSON;
2900 vmx->nested.nested_vmx_cr4_fixed0 = VMXON_CR4_ALWAYSON;
2902 /* These MSRs specify bits which the guest must keep fixed off. */
2903 rdmsrl(MSR_IA32_VMX_CR0_FIXED1, vmx->nested.nested_vmx_cr0_fixed1);
2904 rdmsrl(MSR_IA32_VMX_CR4_FIXED1, vmx->nested.nested_vmx_cr4_fixed1);
2906 /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2907 vmx->nested.nested_vmx_vmcs_enum = 0x2e;
2911 * if fixed0[i] == 1: val[i] must be 1
2912 * if fixed1[i] == 0: val[i] must be 0
2914 static inline bool fixed_bits_valid(u64 val, u64 fixed0, u64 fixed1)
2916 return ((val & fixed1) | fixed0) == val;
2919 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2921 return fixed_bits_valid(control, low, high);
2924 static inline u64 vmx_control_msr(u32 low, u32 high)
2926 return low | ((u64)high << 32);
2929 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
2934 return (superset | subset) == superset;
2937 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
2939 const u64 feature_and_reserved =
2940 /* feature (except bit 48; see below) */
2941 BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
2943 BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
2944 u64 vmx_basic = vmx->nested.nested_vmx_basic;
2946 if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
2950 * KVM does not emulate a version of VMX that constrains physical
2951 * addresses of VMX structures (e.g. VMCS) to 32-bits.
2953 if (data & BIT_ULL(48))
2956 if (vmx_basic_vmcs_revision_id(vmx_basic) !=
2957 vmx_basic_vmcs_revision_id(data))
2960 if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
2963 vmx->nested.nested_vmx_basic = data;
2968 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
2973 switch (msr_index) {
2974 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2975 lowp = &vmx->nested.nested_vmx_pinbased_ctls_low;
2976 highp = &vmx->nested.nested_vmx_pinbased_ctls_high;
2978 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2979 lowp = &vmx->nested.nested_vmx_procbased_ctls_low;
2980 highp = &vmx->nested.nested_vmx_procbased_ctls_high;
2982 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2983 lowp = &vmx->nested.nested_vmx_exit_ctls_low;
2984 highp = &vmx->nested.nested_vmx_exit_ctls_high;
2986 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2987 lowp = &vmx->nested.nested_vmx_entry_ctls_low;
2988 highp = &vmx->nested.nested_vmx_entry_ctls_high;
2990 case MSR_IA32_VMX_PROCBASED_CTLS2:
2991 lowp = &vmx->nested.nested_vmx_secondary_ctls_low;
2992 highp = &vmx->nested.nested_vmx_secondary_ctls_high;
2998 supported = vmx_control_msr(*lowp, *highp);
3000 /* Check must-be-1 bits are still 1. */
3001 if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
3004 /* Check must-be-0 bits are still 0. */
3005 if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
3009 *highp = data >> 32;
3013 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
3015 const u64 feature_and_reserved_bits =
3017 BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
3018 BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
3020 GENMASK_ULL(13, 9) | BIT_ULL(31);
3023 vmx_misc = vmx_control_msr(vmx->nested.nested_vmx_misc_low,
3024 vmx->nested.nested_vmx_misc_high);
3026 if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
3029 if ((vmx->nested.nested_vmx_pinbased_ctls_high &
3030 PIN_BASED_VMX_PREEMPTION_TIMER) &&
3031 vmx_misc_preemption_timer_rate(data) !=
3032 vmx_misc_preemption_timer_rate(vmx_misc))
3035 if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
3038 if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
3041 if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
3044 vmx->nested.nested_vmx_misc_low = data;
3045 vmx->nested.nested_vmx_misc_high = data >> 32;
3049 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
3051 u64 vmx_ept_vpid_cap;
3053 vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.nested_vmx_ept_caps,
3054 vmx->nested.nested_vmx_vpid_caps);
3056 /* Every bit is either reserved or a feature bit. */
3057 if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
3060 vmx->nested.nested_vmx_ept_caps = data;
3061 vmx->nested.nested_vmx_vpid_caps = data >> 32;
3065 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
3069 switch (msr_index) {
3070 case MSR_IA32_VMX_CR0_FIXED0:
3071 msr = &vmx->nested.nested_vmx_cr0_fixed0;
3073 case MSR_IA32_VMX_CR4_FIXED0:
3074 msr = &vmx->nested.nested_vmx_cr4_fixed0;
3081 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
3082 * must be 1 in the restored value.
3084 if (!is_bitwise_subset(data, *msr, -1ULL))
3092 * Called when userspace is restoring VMX MSRs.
3094 * Returns 0 on success, non-0 otherwise.
3096 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
3098 struct vcpu_vmx *vmx = to_vmx(vcpu);
3100 switch (msr_index) {
3101 case MSR_IA32_VMX_BASIC:
3102 return vmx_restore_vmx_basic(vmx, data);
3103 case MSR_IA32_VMX_PINBASED_CTLS:
3104 case MSR_IA32_VMX_PROCBASED_CTLS:
3105 case MSR_IA32_VMX_EXIT_CTLS:
3106 case MSR_IA32_VMX_ENTRY_CTLS:
3108 * The "non-true" VMX capability MSRs are generated from the
3109 * "true" MSRs, so we do not support restoring them directly.
3111 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
3112 * should restore the "true" MSRs with the must-be-1 bits
3113 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
3114 * DEFAULT SETTINGS".
3117 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3118 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3119 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3120 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3121 case MSR_IA32_VMX_PROCBASED_CTLS2:
3122 return vmx_restore_control_msr(vmx, msr_index, data);
3123 case MSR_IA32_VMX_MISC:
3124 return vmx_restore_vmx_misc(vmx, data);
3125 case MSR_IA32_VMX_CR0_FIXED0:
3126 case MSR_IA32_VMX_CR4_FIXED0:
3127 return vmx_restore_fixed0_msr(vmx, msr_index, data);
3128 case MSR_IA32_VMX_CR0_FIXED1:
3129 case MSR_IA32_VMX_CR4_FIXED1:
3131 * These MSRs are generated based on the vCPU's CPUID, so we
3132 * do not support restoring them directly.
3135 case MSR_IA32_VMX_EPT_VPID_CAP:
3136 return vmx_restore_vmx_ept_vpid_cap(vmx, data);
3137 case MSR_IA32_VMX_VMCS_ENUM:
3138 vmx->nested.nested_vmx_vmcs_enum = data;
3142 * The rest of the VMX capability MSRs do not support restore.
3148 /* Returns 0 on success, non-0 otherwise. */
3149 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
3151 struct vcpu_vmx *vmx = to_vmx(vcpu);
3153 switch (msr_index) {
3154 case MSR_IA32_VMX_BASIC:
3155 *pdata = vmx->nested.nested_vmx_basic;
3157 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3158 case MSR_IA32_VMX_PINBASED_CTLS:
3159 *pdata = vmx_control_msr(
3160 vmx->nested.nested_vmx_pinbased_ctls_low,
3161 vmx->nested.nested_vmx_pinbased_ctls_high);
3162 if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
3163 *pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3165 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3166 case MSR_IA32_VMX_PROCBASED_CTLS:
3167 *pdata = vmx_control_msr(
3168 vmx->nested.nested_vmx_procbased_ctls_low,
3169 vmx->nested.nested_vmx_procbased_ctls_high);
3170 if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
3171 *pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3173 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3174 case MSR_IA32_VMX_EXIT_CTLS:
3175 *pdata = vmx_control_msr(
3176 vmx->nested.nested_vmx_exit_ctls_low,
3177 vmx->nested.nested_vmx_exit_ctls_high);
3178 if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
3179 *pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
3181 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3182 case MSR_IA32_VMX_ENTRY_CTLS:
3183 *pdata = vmx_control_msr(
3184 vmx->nested.nested_vmx_entry_ctls_low,
3185 vmx->nested.nested_vmx_entry_ctls_high);
3186 if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
3187 *pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
3189 case MSR_IA32_VMX_MISC:
3190 *pdata = vmx_control_msr(
3191 vmx->nested.nested_vmx_misc_low,
3192 vmx->nested.nested_vmx_misc_high);
3194 case MSR_IA32_VMX_CR0_FIXED0:
3195 *pdata = vmx->nested.nested_vmx_cr0_fixed0;
3197 case MSR_IA32_VMX_CR0_FIXED1:
3198 *pdata = vmx->nested.nested_vmx_cr0_fixed1;
3200 case MSR_IA32_VMX_CR4_FIXED0:
3201 *pdata = vmx->nested.nested_vmx_cr4_fixed0;
3203 case MSR_IA32_VMX_CR4_FIXED1:
3204 *pdata = vmx->nested.nested_vmx_cr4_fixed1;
3206 case MSR_IA32_VMX_VMCS_ENUM:
3207 *pdata = vmx->nested.nested_vmx_vmcs_enum;
3209 case MSR_IA32_VMX_PROCBASED_CTLS2:
3210 *pdata = vmx_control_msr(
3211 vmx->nested.nested_vmx_secondary_ctls_low,
3212 vmx->nested.nested_vmx_secondary_ctls_high);
3214 case MSR_IA32_VMX_EPT_VPID_CAP:
3215 *pdata = vmx->nested.nested_vmx_ept_caps |
3216 ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
3218 case MSR_IA32_VMX_VMFUNC:
3219 *pdata = vmx->nested.nested_vmx_vmfunc_controls;
3228 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
3231 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
3233 return !(val & ~valid_bits);
3237 * Reads an msr value (of 'msr_index') into 'pdata'.
3238 * Returns 0 on success, non-0 otherwise.
3239 * Assumes vcpu_load() was already called.
3241 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3243 struct shared_msr_entry *msr;
3245 switch (msr_info->index) {
3246 #ifdef CONFIG_X86_64
3248 msr_info->data = vmcs_readl(GUEST_FS_BASE);
3251 msr_info->data = vmcs_readl(GUEST_GS_BASE);
3253 case MSR_KERNEL_GS_BASE:
3254 vmx_load_host_state(to_vmx(vcpu));
3255 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
3259 return kvm_get_msr_common(vcpu, msr_info);
3261 msr_info->data = guest_read_tsc(vcpu);
3263 case MSR_IA32_SYSENTER_CS:
3264 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
3266 case MSR_IA32_SYSENTER_EIP:
3267 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
3269 case MSR_IA32_SYSENTER_ESP:
3270 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
3272 case MSR_IA32_BNDCFGS:
3273 if (!kvm_mpx_supported() ||
3274 (!msr_info->host_initiated &&
3275 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
3277 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
3279 case MSR_IA32_MCG_EXT_CTL:
3280 if (!msr_info->host_initiated &&
3281 !(to_vmx(vcpu)->msr_ia32_feature_control &
3282 FEATURE_CONTROL_LMCE))
3284 msr_info->data = vcpu->arch.mcg_ext_ctl;
3286 case MSR_IA32_FEATURE_CONTROL:
3287 msr_info->data = to_vmx(vcpu)->msr_ia32_feature_control;
3289 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3290 if (!nested_vmx_allowed(vcpu))
3292 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
3294 if (!vmx_xsaves_supported())
3296 msr_info->data = vcpu->arch.ia32_xss;
3299 if (!msr_info->host_initiated &&
3300 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
3302 /* Otherwise falls through */
3304 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
3306 msr_info->data = msr->data;
3309 return kvm_get_msr_common(vcpu, msr_info);
3315 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3318 * Writes msr value into into the appropriate "register".
3319 * Returns 0 on success, non-0 otherwise.
3320 * Assumes vcpu_load() was already called.
3322 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3324 struct vcpu_vmx *vmx = to_vmx(vcpu);
3325 struct shared_msr_entry *msr;
3327 u32 msr_index = msr_info->index;
3328 u64 data = msr_info->data;
3330 switch (msr_index) {
3332 ret = kvm_set_msr_common(vcpu, msr_info);
3334 #ifdef CONFIG_X86_64
3336 vmx_segment_cache_clear(vmx);
3337 vmcs_writel(GUEST_FS_BASE, data);
3340 vmx_segment_cache_clear(vmx);
3341 vmcs_writel(GUEST_GS_BASE, data);
3343 case MSR_KERNEL_GS_BASE:
3344 vmx_load_host_state(vmx);
3345 vmx->msr_guest_kernel_gs_base = data;
3348 case MSR_IA32_SYSENTER_CS:
3349 vmcs_write32(GUEST_SYSENTER_CS, data);
3351 case MSR_IA32_SYSENTER_EIP:
3352 vmcs_writel(GUEST_SYSENTER_EIP, data);
3354 case MSR_IA32_SYSENTER_ESP:
3355 vmcs_writel(GUEST_SYSENTER_ESP, data);
3357 case MSR_IA32_BNDCFGS:
3358 if (!kvm_mpx_supported() ||
3359 (!msr_info->host_initiated &&
3360 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
3362 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
3363 (data & MSR_IA32_BNDCFGS_RSVD))
3365 vmcs_write64(GUEST_BNDCFGS, data);
3368 kvm_write_tsc(vcpu, msr_info);
3370 case MSR_IA32_CR_PAT:
3371 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3372 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3374 vmcs_write64(GUEST_IA32_PAT, data);
3375 vcpu->arch.pat = data;
3378 ret = kvm_set_msr_common(vcpu, msr_info);
3380 case MSR_IA32_TSC_ADJUST:
3381 ret = kvm_set_msr_common(vcpu, msr_info);
3383 case MSR_IA32_MCG_EXT_CTL:
3384 if ((!msr_info->host_initiated &&
3385 !(to_vmx(vcpu)->msr_ia32_feature_control &
3386 FEATURE_CONTROL_LMCE)) ||
3387 (data & ~MCG_EXT_CTL_LMCE_EN))
3389 vcpu->arch.mcg_ext_ctl = data;
3391 case MSR_IA32_FEATURE_CONTROL:
3392 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3393 (to_vmx(vcpu)->msr_ia32_feature_control &
3394 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3396 vmx->msr_ia32_feature_control = data;
3397 if (msr_info->host_initiated && data == 0)
3398 vmx_leave_nested(vcpu);
3400 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3401 if (!msr_info->host_initiated)
3402 return 1; /* they are read-only */
3403 if (!nested_vmx_allowed(vcpu))
3405 return vmx_set_vmx_msr(vcpu, msr_index, data);
3407 if (!vmx_xsaves_supported())
3410 * The only supported bit as of Skylake is bit 8, but
3411 * it is not supported on KVM.
3415 vcpu->arch.ia32_xss = data;
3416 if (vcpu->arch.ia32_xss != host_xss)
3417 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3418 vcpu->arch.ia32_xss, host_xss);
3420 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3423 if (!msr_info->host_initiated &&
3424 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
3426 /* Check reserved bit, higher 32 bits should be zero */
3427 if ((data >> 32) != 0)
3429 /* Otherwise falls through */
3431 msr = find_msr_entry(vmx, msr_index);
3433 u64 old_msr_data = msr->data;
3435 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3437 ret = kvm_set_shared_msr(msr->index, msr->data,
3441 msr->data = old_msr_data;
3445 ret = kvm_set_msr_common(vcpu, msr_info);
3451 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3453 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3456 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3459 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3461 case VCPU_EXREG_PDPTR:
3463 ept_save_pdptrs(vcpu);
3470 static __init int cpu_has_kvm_support(void)
3472 return cpu_has_vmx();
3475 static __init int vmx_disabled_by_bios(void)
3479 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3480 if (msr & FEATURE_CONTROL_LOCKED) {
3481 /* launched w/ TXT and VMX disabled */
3482 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3485 /* launched w/o TXT and VMX only enabled w/ TXT */
3486 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3487 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3488 && !tboot_enabled()) {
3489 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3490 "activate TXT before enabling KVM\n");
3493 /* launched w/o TXT and VMX disabled */
3494 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3495 && !tboot_enabled())
3502 static void kvm_cpu_vmxon(u64 addr)
3504 cr4_set_bits(X86_CR4_VMXE);
3505 intel_pt_handle_vmx(1);
3507 asm volatile (ASM_VMX_VMXON_RAX
3508 : : "a"(&addr), "m"(addr)
3512 static int hardware_enable(void)
3514 int cpu = raw_smp_processor_id();
3515 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3518 if (cr4_read_shadow() & X86_CR4_VMXE)
3521 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3522 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3523 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3526 * Now we can enable the vmclear operation in kdump
3527 * since the loaded_vmcss_on_cpu list on this cpu
3528 * has been initialized.
3530 * Though the cpu is not in VMX operation now, there
3531 * is no problem to enable the vmclear operation
3532 * for the loaded_vmcss_on_cpu list is empty!
3534 crash_enable_local_vmclear(cpu);
3536 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3538 test_bits = FEATURE_CONTROL_LOCKED;
3539 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3540 if (tboot_enabled())
3541 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3543 if ((old & test_bits) != test_bits) {
3544 /* enable and lock */
3545 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3547 kvm_cpu_vmxon(phys_addr);
3553 static void vmclear_local_loaded_vmcss(void)
3555 int cpu = raw_smp_processor_id();
3556 struct loaded_vmcs *v, *n;
3558 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3559 loaded_vmcss_on_cpu_link)
3560 __loaded_vmcs_clear(v);
3564 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3567 static void kvm_cpu_vmxoff(void)
3569 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3571 intel_pt_handle_vmx(0);
3572 cr4_clear_bits(X86_CR4_VMXE);
3575 static void hardware_disable(void)
3577 vmclear_local_loaded_vmcss();
3581 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3582 u32 msr, u32 *result)
3584 u32 vmx_msr_low, vmx_msr_high;
3585 u32 ctl = ctl_min | ctl_opt;
3587 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3589 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3590 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3592 /* Ensure minimum (required) set of control bits are supported. */
3600 static __init bool allow_1_setting(u32 msr, u32 ctl)
3602 u32 vmx_msr_low, vmx_msr_high;
3604 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3605 return vmx_msr_high & ctl;
3608 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3610 u32 vmx_msr_low, vmx_msr_high;
3611 u32 min, opt, min2, opt2;
3612 u32 _pin_based_exec_control = 0;
3613 u32 _cpu_based_exec_control = 0;
3614 u32 _cpu_based_2nd_exec_control = 0;
3615 u32 _vmexit_control = 0;
3616 u32 _vmentry_control = 0;
3618 min = CPU_BASED_HLT_EXITING |
3619 #ifdef CONFIG_X86_64
3620 CPU_BASED_CR8_LOAD_EXITING |
3621 CPU_BASED_CR8_STORE_EXITING |
3623 CPU_BASED_CR3_LOAD_EXITING |
3624 CPU_BASED_CR3_STORE_EXITING |
3625 CPU_BASED_USE_IO_BITMAPS |
3626 CPU_BASED_MOV_DR_EXITING |
3627 CPU_BASED_USE_TSC_OFFSETING |
3628 CPU_BASED_INVLPG_EXITING |
3629 CPU_BASED_RDPMC_EXITING;
3631 if (!kvm_mwait_in_guest())
3632 min |= CPU_BASED_MWAIT_EXITING |
3633 CPU_BASED_MONITOR_EXITING;
3635 opt = CPU_BASED_TPR_SHADOW |
3636 CPU_BASED_USE_MSR_BITMAPS |
3637 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3638 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3639 &_cpu_based_exec_control) < 0)
3641 #ifdef CONFIG_X86_64
3642 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3643 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3644 ~CPU_BASED_CR8_STORE_EXITING;
3646 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3648 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3649 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3650 SECONDARY_EXEC_WBINVD_EXITING |
3651 SECONDARY_EXEC_ENABLE_VPID |
3652 SECONDARY_EXEC_ENABLE_EPT |
3653 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3654 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3655 SECONDARY_EXEC_RDTSCP |
3656 SECONDARY_EXEC_ENABLE_INVPCID |
3657 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3658 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3659 SECONDARY_EXEC_SHADOW_VMCS |
3660 SECONDARY_EXEC_XSAVES |
3661 SECONDARY_EXEC_RDSEED |
3662 SECONDARY_EXEC_RDRAND |
3663 SECONDARY_EXEC_ENABLE_PML |
3664 SECONDARY_EXEC_TSC_SCALING |
3665 SECONDARY_EXEC_ENABLE_VMFUNC;
3666 if (adjust_vmx_controls(min2, opt2,
3667 MSR_IA32_VMX_PROCBASED_CTLS2,
3668 &_cpu_based_2nd_exec_control) < 0)
3671 #ifndef CONFIG_X86_64
3672 if (!(_cpu_based_2nd_exec_control &
3673 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3674 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3677 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3678 _cpu_based_2nd_exec_control &= ~(
3679 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3680 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3681 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3683 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3684 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3686 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3687 CPU_BASED_CR3_STORE_EXITING |
3688 CPU_BASED_INVLPG_EXITING);
3689 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3690 vmx_capability.ept, vmx_capability.vpid);
3693 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
3694 #ifdef CONFIG_X86_64
3695 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3697 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3698 VM_EXIT_CLEAR_BNDCFGS;
3699 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3700 &_vmexit_control) < 0)
3703 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
3704 PIN_BASED_VIRTUAL_NMIS;
3705 opt = PIN_BASED_POSTED_INTR | PIN_BASED_VMX_PREEMPTION_TIMER;
3706 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3707 &_pin_based_exec_control) < 0)
3710 if (cpu_has_broken_vmx_preemption_timer())
3711 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3712 if (!(_cpu_based_2nd_exec_control &
3713 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
3714 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3716 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3717 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3718 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3719 &_vmentry_control) < 0)
3722 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3724 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3725 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3728 #ifdef CONFIG_X86_64
3729 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3730 if (vmx_msr_high & (1u<<16))
3734 /* Require Write-Back (WB) memory type for VMCS accesses. */
3735 if (((vmx_msr_high >> 18) & 15) != 6)
3738 vmcs_conf->size = vmx_msr_high & 0x1fff;
3739 vmcs_conf->order = get_order(vmcs_conf->size);
3740 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
3741 vmcs_conf->revision_id = vmx_msr_low;
3743 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3744 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3745 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3746 vmcs_conf->vmexit_ctrl = _vmexit_control;
3747 vmcs_conf->vmentry_ctrl = _vmentry_control;
3749 cpu_has_load_ia32_efer =
3750 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3751 VM_ENTRY_LOAD_IA32_EFER)
3752 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3753 VM_EXIT_LOAD_IA32_EFER);
3755 cpu_has_load_perf_global_ctrl =
3756 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3757 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3758 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3759 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3762 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3763 * but due to errata below it can't be used. Workaround is to use
3764 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3766 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3771 * BC86,AAY89,BD102 (model 44)
3775 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3776 switch (boot_cpu_data.x86_model) {
3782 cpu_has_load_perf_global_ctrl = false;
3783 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3784 "does not work properly. Using workaround\n");
3791 if (boot_cpu_has(X86_FEATURE_XSAVES))
3792 rdmsrl(MSR_IA32_XSS, host_xss);
3797 static struct vmcs *alloc_vmcs_cpu(int cpu)
3799 int node = cpu_to_node(cpu);
3803 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3806 vmcs = page_address(pages);
3807 memset(vmcs, 0, vmcs_config.size);
3808 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3812 static struct vmcs *alloc_vmcs(void)
3814 return alloc_vmcs_cpu(raw_smp_processor_id());
3817 static void free_vmcs(struct vmcs *vmcs)
3819 free_pages((unsigned long)vmcs, vmcs_config.order);
3823 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3825 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3827 if (!loaded_vmcs->vmcs)
3829 loaded_vmcs_clear(loaded_vmcs);
3830 free_vmcs(loaded_vmcs->vmcs);
3831 loaded_vmcs->vmcs = NULL;
3832 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
3835 static void free_kvm_area(void)
3839 for_each_possible_cpu(cpu) {
3840 free_vmcs(per_cpu(vmxarea, cpu));
3841 per_cpu(vmxarea, cpu) = NULL;
3845 enum vmcs_field_type {
3846 VMCS_FIELD_TYPE_U16 = 0,
3847 VMCS_FIELD_TYPE_U64 = 1,
3848 VMCS_FIELD_TYPE_U32 = 2,
3849 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
3852 static inline int vmcs_field_type(unsigned long field)
3854 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
3855 return VMCS_FIELD_TYPE_U32;
3856 return (field >> 13) & 0x3 ;
3859 static inline int vmcs_field_readonly(unsigned long field)
3861 return (((field >> 10) & 0x3) == 1);
3864 static void init_vmcs_shadow_fields(void)
3868 /* No checks for read only fields yet */
3870 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3871 switch (shadow_read_write_fields[i]) {
3873 if (!kvm_mpx_supported())
3881 shadow_read_write_fields[j] =
3882 shadow_read_write_fields[i];
3885 max_shadow_read_write_fields = j;
3887 /* shadowed fields guest access without vmexit */
3888 for (i = 0; i < max_shadow_read_write_fields; i++) {
3889 unsigned long field = shadow_read_write_fields[i];
3891 clear_bit(field, vmx_vmwrite_bitmap);
3892 clear_bit(field, vmx_vmread_bitmap);
3893 if (vmcs_field_type(field) == VMCS_FIELD_TYPE_U64) {
3894 clear_bit(field + 1, vmx_vmwrite_bitmap);
3895 clear_bit(field + 1, vmx_vmread_bitmap);
3898 for (i = 0; i < max_shadow_read_only_fields; i++) {
3899 unsigned long field = shadow_read_only_fields[i];
3901 clear_bit(field, vmx_vmread_bitmap);
3902 if (vmcs_field_type(field) == VMCS_FIELD_TYPE_U64)
3903 clear_bit(field + 1, vmx_vmread_bitmap);
3907 static __init int alloc_kvm_area(void)
3911 for_each_possible_cpu(cpu) {
3914 vmcs = alloc_vmcs_cpu(cpu);
3920 per_cpu(vmxarea, cpu) = vmcs;
3925 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3926 struct kvm_segment *save)
3928 if (!emulate_invalid_guest_state) {
3930 * CS and SS RPL should be equal during guest entry according
3931 * to VMX spec, but in reality it is not always so. Since vcpu
3932 * is in the middle of the transition from real mode to
3933 * protected mode it is safe to assume that RPL 0 is a good
3936 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3937 save->selector &= ~SEGMENT_RPL_MASK;
3938 save->dpl = save->selector & SEGMENT_RPL_MASK;
3941 vmx_set_segment(vcpu, save, seg);
3944 static void enter_pmode(struct kvm_vcpu *vcpu)
3946 unsigned long flags;
3947 struct vcpu_vmx *vmx = to_vmx(vcpu);
3950 * Update real mode segment cache. It may be not up-to-date if sement
3951 * register was written while vcpu was in a guest mode.
3953 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3954 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3955 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3956 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3957 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3958 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3960 vmx->rmode.vm86_active = 0;
3962 vmx_segment_cache_clear(vmx);
3964 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3966 flags = vmcs_readl(GUEST_RFLAGS);
3967 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3968 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3969 vmcs_writel(GUEST_RFLAGS, flags);
3971 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3972 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3974 update_exception_bitmap(vcpu);
3976 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3977 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3978 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3979 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3980 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3981 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3984 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3986 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3987 struct kvm_segment var = *save;
3990 if (seg == VCPU_SREG_CS)
3993 if (!emulate_invalid_guest_state) {
3994 var.selector = var.base >> 4;
3995 var.base = var.base & 0xffff0;
4005 if (save->base & 0xf)
4006 printk_once(KERN_WARNING "kvm: segment base is not "
4007 "paragraph aligned when entering "
4008 "protected mode (seg=%d)", seg);
4011 vmcs_write16(sf->selector, var.selector);
4012 vmcs_writel(sf->base, var.base);
4013 vmcs_write32(sf->limit, var.limit);
4014 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
4017 static void enter_rmode(struct kvm_vcpu *vcpu)
4019 unsigned long flags;
4020 struct vcpu_vmx *vmx = to_vmx(vcpu);
4022 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
4023 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
4024 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
4025 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
4026 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
4027 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
4028 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
4030 vmx->rmode.vm86_active = 1;
4033 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
4034 * vcpu. Warn the user that an update is overdue.
4036 if (!vcpu->kvm->arch.tss_addr)
4037 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
4038 "called before entering vcpu\n");
4040 vmx_segment_cache_clear(vmx);
4042 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
4043 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
4044 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4046 flags = vmcs_readl(GUEST_RFLAGS);
4047 vmx->rmode.save_rflags = flags;
4049 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
4051 vmcs_writel(GUEST_RFLAGS, flags);
4052 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
4053 update_exception_bitmap(vcpu);
4055 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
4056 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
4057 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
4058 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
4059 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
4060 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
4062 kvm_mmu_reset_context(vcpu);
4065 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
4067 struct vcpu_vmx *vmx = to_vmx(vcpu);
4068 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
4074 * Force kernel_gs_base reloading before EFER changes, as control
4075 * of this msr depends on is_long_mode().
4077 vmx_load_host_state(to_vmx(vcpu));
4078 vcpu->arch.efer = efer;
4079 if (efer & EFER_LMA) {
4080 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4083 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4085 msr->data = efer & ~EFER_LME;
4090 #ifdef CONFIG_X86_64
4092 static void enter_lmode(struct kvm_vcpu *vcpu)
4096 vmx_segment_cache_clear(to_vmx(vcpu));
4098 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
4099 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
4100 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
4102 vmcs_write32(GUEST_TR_AR_BYTES,
4103 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
4104 | VMX_AR_TYPE_BUSY_64_TSS);
4106 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
4109 static void exit_lmode(struct kvm_vcpu *vcpu)
4111 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4112 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
4117 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
4120 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
4122 ept_sync_context(construct_eptp(vcpu, vcpu->arch.mmu.root_hpa));
4124 vpid_sync_context(vpid);
4128 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
4130 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
4133 static void vmx_flush_tlb_ept_only(struct kvm_vcpu *vcpu)
4136 vmx_flush_tlb(vcpu);
4139 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
4141 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
4143 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
4144 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
4147 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
4149 if (enable_ept && is_paging(vcpu))
4150 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4151 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
4154 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
4156 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
4158 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
4159 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
4162 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
4164 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4166 if (!test_bit(VCPU_EXREG_PDPTR,
4167 (unsigned long *)&vcpu->arch.regs_dirty))
4170 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4171 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
4172 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
4173 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
4174 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
4178 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
4180 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4182 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4183 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
4184 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
4185 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
4186 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
4189 __set_bit(VCPU_EXREG_PDPTR,
4190 (unsigned long *)&vcpu->arch.regs_avail);
4191 __set_bit(VCPU_EXREG_PDPTR,
4192 (unsigned long *)&vcpu->arch.regs_dirty);
4195 static bool nested_guest_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4197 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4198 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4199 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4201 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
4202 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4203 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4204 fixed0 &= ~(X86_CR0_PE | X86_CR0_PG);
4206 return fixed_bits_valid(val, fixed0, fixed1);
4209 static bool nested_host_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4211 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4212 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4214 return fixed_bits_valid(val, fixed0, fixed1);
4217 static bool nested_cr4_valid(struct kvm_vcpu *vcpu, unsigned long val)
4219 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed0;
4220 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed1;
4222 return fixed_bits_valid(val, fixed0, fixed1);
4225 /* No difference in the restrictions on guest and host CR4 in VMX operation. */
4226 #define nested_guest_cr4_valid nested_cr4_valid
4227 #define nested_host_cr4_valid nested_cr4_valid
4229 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
4231 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
4233 struct kvm_vcpu *vcpu)
4235 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
4236 vmx_decache_cr3(vcpu);
4237 if (!(cr0 & X86_CR0_PG)) {
4238 /* From paging/starting to nonpaging */
4239 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4240 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
4241 (CPU_BASED_CR3_LOAD_EXITING |
4242 CPU_BASED_CR3_STORE_EXITING));
4243 vcpu->arch.cr0 = cr0;
4244 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4245 } else if (!is_paging(vcpu)) {
4246 /* From nonpaging to paging */
4247 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4248 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
4249 ~(CPU_BASED_CR3_LOAD_EXITING |
4250 CPU_BASED_CR3_STORE_EXITING));
4251 vcpu->arch.cr0 = cr0;
4252 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4255 if (!(cr0 & X86_CR0_WP))
4256 *hw_cr0 &= ~X86_CR0_WP;
4259 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
4261 struct vcpu_vmx *vmx = to_vmx(vcpu);
4262 unsigned long hw_cr0;
4264 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
4265 if (enable_unrestricted_guest)
4266 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
4268 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
4270 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
4273 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
4277 #ifdef CONFIG_X86_64
4278 if (vcpu->arch.efer & EFER_LME) {
4279 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
4281 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
4287 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
4289 vmcs_writel(CR0_READ_SHADOW, cr0);
4290 vmcs_writel(GUEST_CR0, hw_cr0);
4291 vcpu->arch.cr0 = cr0;
4293 /* depends on vcpu->arch.cr0 to be set to a new value */
4294 vmx->emulation_required = emulation_required(vcpu);
4297 static int get_ept_level(struct kvm_vcpu *vcpu)
4299 if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
4304 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
4306 u64 eptp = VMX_EPTP_MT_WB;
4308 eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
4310 if (enable_ept_ad_bits &&
4311 (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
4312 eptp |= VMX_EPTP_AD_ENABLE_BIT;
4313 eptp |= (root_hpa & PAGE_MASK);
4318 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
4320 unsigned long guest_cr3;
4325 eptp = construct_eptp(vcpu, cr3);
4326 vmcs_write64(EPT_POINTER, eptp);
4327 if (is_paging(vcpu) || is_guest_mode(vcpu))
4328 guest_cr3 = kvm_read_cr3(vcpu);
4330 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
4331 ept_load_pdptrs(vcpu);
4334 vmx_flush_tlb(vcpu);
4335 vmcs_writel(GUEST_CR3, guest_cr3);
4338 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
4341 * Pass through host's Machine Check Enable value to hw_cr4, which
4342 * is in force while we are in guest mode. Do not let guests control
4343 * this bit, even if host CR4.MCE == 0.
4345 unsigned long hw_cr4 =
4346 (cr4_read_shadow() & X86_CR4_MCE) |
4347 (cr4 & ~X86_CR4_MCE) |
4348 (to_vmx(vcpu)->rmode.vm86_active ?
4349 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
4351 if (cr4 & X86_CR4_VMXE) {
4353 * To use VMXON (and later other VMX instructions), a guest
4354 * must first be able to turn on cr4.VMXE (see handle_vmon()).
4355 * So basically the check on whether to allow nested VMX
4358 if (!nested_vmx_allowed(vcpu))
4362 if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
4365 vcpu->arch.cr4 = cr4;
4367 if (!is_paging(vcpu)) {
4368 hw_cr4 &= ~X86_CR4_PAE;
4369 hw_cr4 |= X86_CR4_PSE;
4370 } else if (!(cr4 & X86_CR4_PAE)) {
4371 hw_cr4 &= ~X86_CR4_PAE;
4375 if (!enable_unrestricted_guest && !is_paging(vcpu))
4377 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4378 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
4379 * to be manually disabled when guest switches to non-paging
4382 * If !enable_unrestricted_guest, the CPU is always running
4383 * with CR0.PG=1 and CR4 needs to be modified.
4384 * If enable_unrestricted_guest, the CPU automatically
4385 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4387 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
4389 vmcs_writel(CR4_READ_SHADOW, cr4);
4390 vmcs_writel(GUEST_CR4, hw_cr4);
4394 static void vmx_get_segment(struct kvm_vcpu *vcpu,
4395 struct kvm_segment *var, int seg)
4397 struct vcpu_vmx *vmx = to_vmx(vcpu);
4400 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4401 *var = vmx->rmode.segs[seg];
4402 if (seg == VCPU_SREG_TR
4403 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
4405 var->base = vmx_read_guest_seg_base(vmx, seg);
4406 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4409 var->base = vmx_read_guest_seg_base(vmx, seg);
4410 var->limit = vmx_read_guest_seg_limit(vmx, seg);
4411 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4412 ar = vmx_read_guest_seg_ar(vmx, seg);
4413 var->unusable = (ar >> 16) & 1;
4414 var->type = ar & 15;
4415 var->s = (ar >> 4) & 1;
4416 var->dpl = (ar >> 5) & 3;
4418 * Some userspaces do not preserve unusable property. Since usable
4419 * segment has to be present according to VMX spec we can use present
4420 * property to amend userspace bug by making unusable segment always
4421 * nonpresent. vmx_segment_access_rights() already marks nonpresent
4422 * segment as unusable.
4424 var->present = !var->unusable;
4425 var->avl = (ar >> 12) & 1;
4426 var->l = (ar >> 13) & 1;
4427 var->db = (ar >> 14) & 1;
4428 var->g = (ar >> 15) & 1;
4431 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
4433 struct kvm_segment s;
4435 if (to_vmx(vcpu)->rmode.vm86_active) {
4436 vmx_get_segment(vcpu, &s, seg);
4439 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
4442 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
4444 struct vcpu_vmx *vmx = to_vmx(vcpu);
4446 if (unlikely(vmx->rmode.vm86_active))
4449 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
4450 return VMX_AR_DPL(ar);
4454 static u32 vmx_segment_access_rights(struct kvm_segment *var)
4458 if (var->unusable || !var->present)
4461 ar = var->type & 15;
4462 ar |= (var->s & 1) << 4;
4463 ar |= (var->dpl & 3) << 5;
4464 ar |= (var->present & 1) << 7;
4465 ar |= (var->avl & 1) << 12;
4466 ar |= (var->l & 1) << 13;
4467 ar |= (var->db & 1) << 14;
4468 ar |= (var->g & 1) << 15;
4474 static void vmx_set_segment(struct kvm_vcpu *vcpu,
4475 struct kvm_segment *var, int seg)
4477 struct vcpu_vmx *vmx = to_vmx(vcpu);
4478 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4480 vmx_segment_cache_clear(vmx);
4482 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4483 vmx->rmode.segs[seg] = *var;
4484 if (seg == VCPU_SREG_TR)
4485 vmcs_write16(sf->selector, var->selector);
4487 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
4491 vmcs_writel(sf->base, var->base);
4492 vmcs_write32(sf->limit, var->limit);
4493 vmcs_write16(sf->selector, var->selector);
4496 * Fix the "Accessed" bit in AR field of segment registers for older
4498 * IA32 arch specifies that at the time of processor reset the
4499 * "Accessed" bit in the AR field of segment registers is 1. And qemu
4500 * is setting it to 0 in the userland code. This causes invalid guest
4501 * state vmexit when "unrestricted guest" mode is turned on.
4502 * Fix for this setup issue in cpu_reset is being pushed in the qemu
4503 * tree. Newer qemu binaries with that qemu fix would not need this
4506 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
4507 var->type |= 0x1; /* Accessed */
4509 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4512 vmx->emulation_required = emulation_required(vcpu);
4515 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4517 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4519 *db = (ar >> 14) & 1;
4520 *l = (ar >> 13) & 1;
4523 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4525 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4526 dt->address = vmcs_readl(GUEST_IDTR_BASE);
4529 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4531 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4532 vmcs_writel(GUEST_IDTR_BASE, dt->address);
4535 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4537 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4538 dt->address = vmcs_readl(GUEST_GDTR_BASE);
4541 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4543 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4544 vmcs_writel(GUEST_GDTR_BASE, dt->address);
4547 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4549 struct kvm_segment var;
4552 vmx_get_segment(vcpu, &var, seg);
4554 if (seg == VCPU_SREG_CS)
4556 ar = vmx_segment_access_rights(&var);
4558 if (var.base != (var.selector << 4))
4560 if (var.limit != 0xffff)
4568 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4570 struct kvm_segment cs;
4571 unsigned int cs_rpl;
4573 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4574 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4578 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4582 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4583 if (cs.dpl > cs_rpl)
4586 if (cs.dpl != cs_rpl)
4592 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4596 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4598 struct kvm_segment ss;
4599 unsigned int ss_rpl;
4601 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4602 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4606 if (ss.type != 3 && ss.type != 7)
4610 if (ss.dpl != ss_rpl) /* DPL != RPL */
4618 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4620 struct kvm_segment var;
4623 vmx_get_segment(vcpu, &var, seg);
4624 rpl = var.selector & SEGMENT_RPL_MASK;
4632 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4633 if (var.dpl < rpl) /* DPL < RPL */
4637 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4643 static bool tr_valid(struct kvm_vcpu *vcpu)
4645 struct kvm_segment tr;
4647 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4651 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4653 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4661 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4663 struct kvm_segment ldtr;
4665 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4669 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4679 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4681 struct kvm_segment cs, ss;
4683 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4684 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4686 return ((cs.selector & SEGMENT_RPL_MASK) ==
4687 (ss.selector & SEGMENT_RPL_MASK));
4691 * Check if guest state is valid. Returns true if valid, false if
4693 * We assume that registers are always usable
4695 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4697 if (enable_unrestricted_guest)
4700 /* real mode guest state checks */
4701 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4702 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4704 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4706 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4708 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4710 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4712 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4715 /* protected mode guest state checks */
4716 if (!cs_ss_rpl_check(vcpu))
4718 if (!code_segment_valid(vcpu))
4720 if (!stack_segment_valid(vcpu))
4722 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4724 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4726 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4728 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4730 if (!tr_valid(vcpu))
4732 if (!ldtr_valid(vcpu))
4736 * - Add checks on RIP
4737 * - Add checks on RFLAGS
4743 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
4745 return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
4748 static int init_rmode_tss(struct kvm *kvm)
4754 idx = srcu_read_lock(&kvm->srcu);
4755 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4756 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4759 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4760 r = kvm_write_guest_page(kvm, fn++, &data,
4761 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4764 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4767 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4771 r = kvm_write_guest_page(kvm, fn, &data,
4772 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4775 srcu_read_unlock(&kvm->srcu, idx);
4779 static int init_rmode_identity_map(struct kvm *kvm)
4782 kvm_pfn_t identity_map_pfn;
4788 /* Protect kvm->arch.ept_identity_pagetable_done. */
4789 mutex_lock(&kvm->slots_lock);
4791 if (likely(kvm->arch.ept_identity_pagetable_done))
4794 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4796 r = alloc_identity_pagetable(kvm);
4800 idx = srcu_read_lock(&kvm->srcu);
4801 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4804 /* Set up identity-mapping pagetable for EPT in real mode */
4805 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4806 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4807 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4808 r = kvm_write_guest_page(kvm, identity_map_pfn,
4809 &tmp, i * sizeof(tmp), sizeof(tmp));
4813 kvm->arch.ept_identity_pagetable_done = true;
4816 srcu_read_unlock(&kvm->srcu, idx);
4819 mutex_unlock(&kvm->slots_lock);
4823 static void seg_setup(int seg)
4825 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4828 vmcs_write16(sf->selector, 0);
4829 vmcs_writel(sf->base, 0);
4830 vmcs_write32(sf->limit, 0xffff);
4832 if (seg == VCPU_SREG_CS)
4833 ar |= 0x08; /* code segment */
4835 vmcs_write32(sf->ar_bytes, ar);
4838 static int alloc_apic_access_page(struct kvm *kvm)
4843 mutex_lock(&kvm->slots_lock);
4844 if (kvm->arch.apic_access_page_done)
4846 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4847 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4851 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4852 if (is_error_page(page)) {
4858 * Do not pin the page in memory, so that memory hot-unplug
4859 * is able to migrate it.
4862 kvm->arch.apic_access_page_done = true;
4864 mutex_unlock(&kvm->slots_lock);
4868 static int alloc_identity_pagetable(struct kvm *kvm)
4870 /* Called with kvm->slots_lock held. */
4874 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4876 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4877 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4882 static int allocate_vpid(void)
4888 spin_lock(&vmx_vpid_lock);
4889 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4890 if (vpid < VMX_NR_VPIDS)
4891 __set_bit(vpid, vmx_vpid_bitmap);
4894 spin_unlock(&vmx_vpid_lock);
4898 static void free_vpid(int vpid)
4900 if (!enable_vpid || vpid == 0)
4902 spin_lock(&vmx_vpid_lock);
4903 __clear_bit(vpid, vmx_vpid_bitmap);
4904 spin_unlock(&vmx_vpid_lock);
4907 #define MSR_TYPE_R 1
4908 #define MSR_TYPE_W 2
4909 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4912 int f = sizeof(unsigned long);
4914 if (!cpu_has_vmx_msr_bitmap())
4918 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4919 * have the write-low and read-high bitmap offsets the wrong way round.
4920 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4922 if (msr <= 0x1fff) {
4923 if (type & MSR_TYPE_R)
4925 __clear_bit(msr, msr_bitmap + 0x000 / f);
4927 if (type & MSR_TYPE_W)
4929 __clear_bit(msr, msr_bitmap + 0x800 / f);
4931 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4933 if (type & MSR_TYPE_R)
4935 __clear_bit(msr, msr_bitmap + 0x400 / f);
4937 if (type & MSR_TYPE_W)
4939 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4945 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4946 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4948 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4949 unsigned long *msr_bitmap_nested,
4952 int f = sizeof(unsigned long);
4954 if (!cpu_has_vmx_msr_bitmap()) {
4960 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4961 * have the write-low and read-high bitmap offsets the wrong way round.
4962 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4964 if (msr <= 0x1fff) {
4965 if (type & MSR_TYPE_R &&
4966 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4968 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4970 if (type & MSR_TYPE_W &&
4971 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4973 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4975 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4977 if (type & MSR_TYPE_R &&
4978 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4980 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4982 if (type & MSR_TYPE_W &&
4983 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4985 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4990 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4993 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4994 msr, MSR_TYPE_R | MSR_TYPE_W);
4995 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4996 msr, MSR_TYPE_R | MSR_TYPE_W);
4999 static void vmx_disable_intercept_msr_x2apic(u32 msr, int type, bool apicv_active)
5002 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic_apicv,
5004 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic_apicv,
5007 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
5009 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
5014 static bool vmx_get_enable_apicv(struct kvm_vcpu *vcpu)
5016 return enable_apicv;
5019 static void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
5021 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5025 * Don't need to mark the APIC access page dirty; it is never
5026 * written to by the CPU during APIC virtualization.
5029 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
5030 gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
5031 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5034 if (nested_cpu_has_posted_intr(vmcs12)) {
5035 gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
5036 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5041 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
5043 struct vcpu_vmx *vmx = to_vmx(vcpu);
5048 if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
5051 vmx->nested.pi_pending = false;
5052 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
5055 max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
5056 if (max_irr != 256) {
5057 vapic_page = kmap(vmx->nested.virtual_apic_page);
5058 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
5059 kunmap(vmx->nested.virtual_apic_page);
5061 status = vmcs_read16(GUEST_INTR_STATUS);
5062 if ((u8)max_irr > ((u8)status & 0xff)) {
5064 status |= (u8)max_irr;
5065 vmcs_write16(GUEST_INTR_STATUS, status);
5069 nested_mark_vmcs12_pages_dirty(vcpu);
5072 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
5076 int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
5078 if (vcpu->mode == IN_GUEST_MODE) {
5080 * The vector of interrupt to be delivered to vcpu had
5081 * been set in PIR before this function.
5083 * Following cases will be reached in this block, and
5084 * we always send a notification event in all cases as
5087 * Case 1: vcpu keeps in non-root mode. Sending a
5088 * notification event posts the interrupt to vcpu.
5090 * Case 2: vcpu exits to root mode and is still
5091 * runnable. PIR will be synced to vIRR before the
5092 * next vcpu entry. Sending a notification event in
5093 * this case has no effect, as vcpu is not in root
5096 * Case 3: vcpu exits to root mode and is blocked.
5097 * vcpu_block() has already synced PIR to vIRR and
5098 * never blocks vcpu if vIRR is not cleared. Therefore,
5099 * a blocked vcpu here does not wait for any requested
5100 * interrupts in PIR, and sending a notification event
5101 * which has no effect is safe here.
5104 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
5111 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
5114 struct vcpu_vmx *vmx = to_vmx(vcpu);
5116 if (is_guest_mode(vcpu) &&
5117 vector == vmx->nested.posted_intr_nv) {
5118 /* the PIR and ON have been set by L1. */
5119 kvm_vcpu_trigger_posted_interrupt(vcpu, true);
5121 * If a posted intr is not recognized by hardware,
5122 * we will accomplish it in the next vmentry.
5124 vmx->nested.pi_pending = true;
5125 kvm_make_request(KVM_REQ_EVENT, vcpu);
5131 * Send interrupt to vcpu via posted interrupt way.
5132 * 1. If target vcpu is running(non-root mode), send posted interrupt
5133 * notification to vcpu and hardware will sync PIR to vIRR atomically.
5134 * 2. If target vcpu isn't running(root mode), kick it to pick up the
5135 * interrupt from PIR in next vmentry.
5137 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
5139 struct vcpu_vmx *vmx = to_vmx(vcpu);
5142 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
5146 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
5149 /* If a previous notification has sent the IPI, nothing to do. */
5150 if (pi_test_and_set_on(&vmx->pi_desc))
5153 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
5154 kvm_vcpu_kick(vcpu);
5158 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
5159 * will not change in the lifetime of the guest.
5160 * Note that host-state that does change is set elsewhere. E.g., host-state
5161 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
5163 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
5168 unsigned long cr0, cr3, cr4;
5171 WARN_ON(cr0 & X86_CR0_TS);
5172 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
5175 * Save the most likely value for this task's CR3 in the VMCS.
5176 * We can't use __get_current_cr3_fast() because we're not atomic.
5179 vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */
5180 vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
5182 /* Save the most likely value for this task's CR4 in the VMCS. */
5183 cr4 = cr4_read_shadow();
5184 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
5185 vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
5187 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
5188 #ifdef CONFIG_X86_64
5190 * Load null selectors, so we can avoid reloading them in
5191 * __vmx_load_host_state(), in case userspace uses the null selectors
5192 * too (the expected case).
5194 vmcs_write16(HOST_DS_SELECTOR, 0);
5195 vmcs_write16(HOST_ES_SELECTOR, 0);
5197 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5198 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5200 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5201 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
5204 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
5205 vmx->host_idt_base = dt.address;
5207 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
5209 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
5210 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
5211 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
5212 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
5214 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
5215 rdmsr(MSR_IA32_CR_PAT, low32, high32);
5216 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
5220 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
5222 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
5224 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
5225 if (is_guest_mode(&vmx->vcpu))
5226 vmx->vcpu.arch.cr4_guest_owned_bits &=
5227 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
5228 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
5231 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
5233 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
5235 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5236 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
5237 /* Enable the preemption timer dynamically */
5238 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
5239 return pin_based_exec_ctrl;
5242 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5244 struct vcpu_vmx *vmx = to_vmx(vcpu);
5246 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5247 if (cpu_has_secondary_exec_ctrls()) {
5248 if (kvm_vcpu_apicv_active(vcpu))
5249 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
5250 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5251 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5253 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
5254 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5255 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5258 if (cpu_has_vmx_msr_bitmap())
5259 vmx_set_msr_bitmap(vcpu);
5262 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
5264 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
5266 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
5267 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5269 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
5270 exec_control &= ~CPU_BASED_TPR_SHADOW;
5271 #ifdef CONFIG_X86_64
5272 exec_control |= CPU_BASED_CR8_STORE_EXITING |
5273 CPU_BASED_CR8_LOAD_EXITING;
5277 exec_control |= CPU_BASED_CR3_STORE_EXITING |
5278 CPU_BASED_CR3_LOAD_EXITING |
5279 CPU_BASED_INVLPG_EXITING;
5280 return exec_control;
5283 static bool vmx_rdrand_supported(void)
5285 return vmcs_config.cpu_based_2nd_exec_ctrl &
5286 SECONDARY_EXEC_RDRAND;
5289 static bool vmx_rdseed_supported(void)
5291 return vmcs_config.cpu_based_2nd_exec_ctrl &
5292 SECONDARY_EXEC_RDSEED;
5295 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
5297 struct kvm_vcpu *vcpu = &vmx->vcpu;
5299 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
5300 if (!cpu_need_virtualize_apic_accesses(vcpu))
5301 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5303 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
5305 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
5306 enable_unrestricted_guest = 0;
5307 /* Enable INVPCID for non-ept guests may cause performance regression. */
5308 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5310 if (!enable_unrestricted_guest)
5311 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
5313 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
5314 if (!kvm_vcpu_apicv_active(vcpu))
5315 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
5316 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5317 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
5318 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
5320 We can NOT enable shadow_vmcs here because we don't have yet
5323 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
5326 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
5328 if (vmx_xsaves_supported()) {
5329 /* Exposing XSAVES only when XSAVE is exposed */
5330 bool xsaves_enabled =
5331 guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
5332 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
5334 if (!xsaves_enabled)
5335 exec_control &= ~SECONDARY_EXEC_XSAVES;
5339 vmx->nested.nested_vmx_secondary_ctls_high |=
5340 SECONDARY_EXEC_XSAVES;
5342 vmx->nested.nested_vmx_secondary_ctls_high &=
5343 ~SECONDARY_EXEC_XSAVES;
5347 if (vmx_rdtscp_supported()) {
5348 bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
5349 if (!rdtscp_enabled)
5350 exec_control &= ~SECONDARY_EXEC_RDTSCP;
5354 vmx->nested.nested_vmx_secondary_ctls_high |=
5355 SECONDARY_EXEC_RDTSCP;
5357 vmx->nested.nested_vmx_secondary_ctls_high &=
5358 ~SECONDARY_EXEC_RDTSCP;
5362 if (vmx_invpcid_supported()) {
5363 /* Exposing INVPCID only when PCID is exposed */
5364 bool invpcid_enabled =
5365 guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
5366 guest_cpuid_has(vcpu, X86_FEATURE_PCID);
5368 if (!invpcid_enabled) {
5369 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5370 guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
5374 if (invpcid_enabled)
5375 vmx->nested.nested_vmx_secondary_ctls_high |=
5376 SECONDARY_EXEC_ENABLE_INVPCID;
5378 vmx->nested.nested_vmx_secondary_ctls_high &=
5379 ~SECONDARY_EXEC_ENABLE_INVPCID;
5383 if (vmx_rdrand_supported()) {
5384 bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
5386 exec_control &= ~SECONDARY_EXEC_RDRAND;
5390 vmx->nested.nested_vmx_secondary_ctls_high |=
5391 SECONDARY_EXEC_RDRAND;
5393 vmx->nested.nested_vmx_secondary_ctls_high &=
5394 ~SECONDARY_EXEC_RDRAND;
5398 if (vmx_rdseed_supported()) {
5399 bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
5401 exec_control &= ~SECONDARY_EXEC_RDSEED;
5405 vmx->nested.nested_vmx_secondary_ctls_high |=
5406 SECONDARY_EXEC_RDSEED;
5408 vmx->nested.nested_vmx_secondary_ctls_high &=
5409 ~SECONDARY_EXEC_RDSEED;
5413 vmx->secondary_exec_control = exec_control;
5416 static void ept_set_mmio_spte_mask(void)
5419 * EPT Misconfigurations can be generated if the value of bits 2:0
5420 * of an EPT paging-structure entry is 110b (write/execute).
5422 kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
5423 VMX_EPT_MISCONFIG_WX_VALUE);
5426 #define VMX_XSS_EXIT_BITMAP 0
5428 * Sets up the vmcs for emulated real mode.
5430 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
5432 #ifdef CONFIG_X86_64
5438 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
5439 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
5441 if (enable_shadow_vmcs) {
5442 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5443 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5445 if (cpu_has_vmx_msr_bitmap())
5446 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
5448 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
5451 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5452 vmx->hv_deadline_tsc = -1;
5454 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
5456 if (cpu_has_secondary_exec_ctrls()) {
5457 vmx_compute_secondary_exec_control(vmx);
5458 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5459 vmx->secondary_exec_control);
5462 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
5463 vmcs_write64(EOI_EXIT_BITMAP0, 0);
5464 vmcs_write64(EOI_EXIT_BITMAP1, 0);
5465 vmcs_write64(EOI_EXIT_BITMAP2, 0);
5466 vmcs_write64(EOI_EXIT_BITMAP3, 0);
5468 vmcs_write16(GUEST_INTR_STATUS, 0);
5470 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
5471 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
5475 vmcs_write32(PLE_GAP, ple_gap);
5476 vmx->ple_window = ple_window;
5477 vmx->ple_window_dirty = true;
5480 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
5481 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
5482 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
5484 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
5485 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
5486 vmx_set_constant_host_state(vmx);
5487 #ifdef CONFIG_X86_64
5488 rdmsrl(MSR_FS_BASE, a);
5489 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
5490 rdmsrl(MSR_GS_BASE, a);
5491 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
5493 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
5494 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
5497 if (cpu_has_vmx_vmfunc())
5498 vmcs_write64(VM_FUNCTION_CONTROL, 0);
5500 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
5501 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
5502 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
5503 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
5504 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
5506 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
5507 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
5509 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
5510 u32 index = vmx_msr_index[i];
5511 u32 data_low, data_high;
5514 if (rdmsr_safe(index, &data_low, &data_high) < 0)
5516 if (wrmsr_safe(index, data_low, data_high) < 0)
5518 vmx->guest_msrs[j].index = i;
5519 vmx->guest_msrs[j].data = 0;
5520 vmx->guest_msrs[j].mask = -1ull;
5525 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
5527 /* 22.2.1, 20.8.1 */
5528 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
5530 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
5531 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
5533 set_cr4_guest_host_mask(vmx);
5535 if (vmx_xsaves_supported())
5536 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
5539 ASSERT(vmx->pml_pg);
5540 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
5541 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5547 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
5549 struct vcpu_vmx *vmx = to_vmx(vcpu);
5550 struct msr_data apic_base_msr;
5553 vmx->rmode.vm86_active = 0;
5555 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
5556 kvm_set_cr8(vcpu, 0);
5559 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
5560 MSR_IA32_APICBASE_ENABLE;
5561 if (kvm_vcpu_is_reset_bsp(vcpu))
5562 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
5563 apic_base_msr.host_initiated = true;
5564 kvm_set_apic_base(vcpu, &apic_base_msr);
5567 vmx_segment_cache_clear(vmx);
5569 seg_setup(VCPU_SREG_CS);
5570 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
5571 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
5573 seg_setup(VCPU_SREG_DS);
5574 seg_setup(VCPU_SREG_ES);
5575 seg_setup(VCPU_SREG_FS);
5576 seg_setup(VCPU_SREG_GS);
5577 seg_setup(VCPU_SREG_SS);
5579 vmcs_write16(GUEST_TR_SELECTOR, 0);
5580 vmcs_writel(GUEST_TR_BASE, 0);
5581 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
5582 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
5584 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
5585 vmcs_writel(GUEST_LDTR_BASE, 0);
5586 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
5587 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
5590 vmcs_write32(GUEST_SYSENTER_CS, 0);
5591 vmcs_writel(GUEST_SYSENTER_ESP, 0);
5592 vmcs_writel(GUEST_SYSENTER_EIP, 0);
5593 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
5596 vmcs_writel(GUEST_RFLAGS, 0x02);
5597 kvm_rip_write(vcpu, 0xfff0);
5599 vmcs_writel(GUEST_GDTR_BASE, 0);
5600 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
5602 vmcs_writel(GUEST_IDTR_BASE, 0);
5603 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
5605 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
5606 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
5607 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
5611 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
5613 if (cpu_has_vmx_tpr_shadow() && !init_event) {
5614 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
5615 if (cpu_need_tpr_shadow(vcpu))
5616 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
5617 __pa(vcpu->arch.apic->regs));
5618 vmcs_write32(TPR_THRESHOLD, 0);
5621 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5624 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5626 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5627 vmx->vcpu.arch.cr0 = cr0;
5628 vmx_set_cr0(vcpu, cr0); /* enter rmode */
5629 vmx_set_cr4(vcpu, 0);
5630 vmx_set_efer(vcpu, 0);
5632 update_exception_bitmap(vcpu);
5634 vpid_sync_context(vmx->vpid);
5638 * In nested virtualization, check if L1 asked to exit on external interrupts.
5639 * For most existing hypervisors, this will always return true.
5641 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5643 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5644 PIN_BASED_EXT_INTR_MASK;
5648 * In nested virtualization, check if L1 has set
5649 * VM_EXIT_ACK_INTR_ON_EXIT
5651 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5653 return get_vmcs12(vcpu)->vm_exit_controls &
5654 VM_EXIT_ACK_INTR_ON_EXIT;
5657 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5659 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5660 PIN_BASED_NMI_EXITING;
5663 static void enable_irq_window(struct kvm_vcpu *vcpu)
5665 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5666 CPU_BASED_VIRTUAL_INTR_PENDING);
5669 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5671 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5672 enable_irq_window(vcpu);
5676 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5677 CPU_BASED_VIRTUAL_NMI_PENDING);
5680 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5682 struct vcpu_vmx *vmx = to_vmx(vcpu);
5684 int irq = vcpu->arch.interrupt.nr;
5686 trace_kvm_inj_virq(irq);
5688 ++vcpu->stat.irq_injections;
5689 if (vmx->rmode.vm86_active) {
5691 if (vcpu->arch.interrupt.soft)
5692 inc_eip = vcpu->arch.event_exit_inst_len;
5693 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5694 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5697 intr = irq | INTR_INFO_VALID_MASK;
5698 if (vcpu->arch.interrupt.soft) {
5699 intr |= INTR_TYPE_SOFT_INTR;
5700 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5701 vmx->vcpu.arch.event_exit_inst_len);
5703 intr |= INTR_TYPE_EXT_INTR;
5704 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5707 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5709 struct vcpu_vmx *vmx = to_vmx(vcpu);
5711 ++vcpu->stat.nmi_injections;
5712 vmx->loaded_vmcs->nmi_known_unmasked = false;
5714 if (vmx->rmode.vm86_active) {
5715 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5716 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5720 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5721 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5724 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5726 struct vcpu_vmx *vmx = to_vmx(vcpu);
5729 if (vmx->loaded_vmcs->nmi_known_unmasked)
5731 masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5732 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5736 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5738 struct vcpu_vmx *vmx = to_vmx(vcpu);
5740 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5742 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5743 GUEST_INTR_STATE_NMI);
5745 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5746 GUEST_INTR_STATE_NMI);
5749 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5751 if (to_vmx(vcpu)->nested.nested_run_pending)
5754 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5755 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5756 | GUEST_INTR_STATE_NMI));
5759 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5761 return (!to_vmx(vcpu)->nested.nested_run_pending &&
5762 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5763 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5764 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5767 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5771 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5775 kvm->arch.tss_addr = addr;
5776 return init_rmode_tss(kvm);
5779 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5784 * Update instruction length as we may reinject the exception
5785 * from user space while in guest debugging mode.
5787 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5788 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5789 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5793 if (vcpu->guest_debug &
5794 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5811 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5812 int vec, u32 err_code)
5815 * Instruction with address size override prefix opcode 0x67
5816 * Cause the #SS fault with 0 error code in VM86 mode.
5818 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5819 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5820 if (vcpu->arch.halt_request) {
5821 vcpu->arch.halt_request = 0;
5822 return kvm_vcpu_halt(vcpu);
5830 * Forward all other exceptions that are valid in real mode.
5831 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5832 * the required debugging infrastructure rework.
5834 kvm_queue_exception(vcpu, vec);
5839 * Trigger machine check on the host. We assume all the MSRs are already set up
5840 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5841 * We pass a fake environment to the machine check handler because we want
5842 * the guest to be always treated like user space, no matter what context
5843 * it used internally.
5845 static void kvm_machine_check(void)
5847 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5848 struct pt_regs regs = {
5849 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5850 .flags = X86_EFLAGS_IF,
5853 do_machine_check(®s, 0);
5857 static int handle_machine_check(struct kvm_vcpu *vcpu)
5859 /* already handled by vcpu_run */
5863 static int handle_exception(struct kvm_vcpu *vcpu)
5865 struct vcpu_vmx *vmx = to_vmx(vcpu);
5866 struct kvm_run *kvm_run = vcpu->run;
5867 u32 intr_info, ex_no, error_code;
5868 unsigned long cr2, rip, dr6;
5870 enum emulation_result er;
5872 vect_info = vmx->idt_vectoring_info;
5873 intr_info = vmx->exit_intr_info;
5875 if (is_machine_check(intr_info))
5876 return handle_machine_check(vcpu);
5878 if (is_nmi(intr_info))
5879 return 1; /* already handled by vmx_vcpu_run() */
5881 if (is_invalid_opcode(intr_info)) {
5882 if (is_guest_mode(vcpu)) {
5883 kvm_queue_exception(vcpu, UD_VECTOR);
5886 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5887 if (er != EMULATE_DONE)
5888 kvm_queue_exception(vcpu, UD_VECTOR);
5893 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5894 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5897 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5898 * MMIO, it is better to report an internal error.
5899 * See the comments in vmx_handle_exit.
5901 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5902 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5903 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5904 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5905 vcpu->run->internal.ndata = 3;
5906 vcpu->run->internal.data[0] = vect_info;
5907 vcpu->run->internal.data[1] = intr_info;
5908 vcpu->run->internal.data[2] = error_code;
5912 if (is_page_fault(intr_info)) {
5913 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5914 /* EPT won't cause page fault directly */
5915 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
5916 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0,
5920 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5922 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5923 return handle_rmode_exception(vcpu, ex_no, error_code);
5927 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5930 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5931 if (!(vcpu->guest_debug &
5932 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5933 vcpu->arch.dr6 &= ~15;
5934 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5935 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5936 skip_emulated_instruction(vcpu);
5938 kvm_queue_exception(vcpu, DB_VECTOR);
5941 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5942 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5946 * Update instruction length as we may reinject #BP from
5947 * user space while in guest debugging mode. Reading it for
5948 * #DB as well causes no harm, it is not used in that case.
5950 vmx->vcpu.arch.event_exit_inst_len =
5951 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5952 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5953 rip = kvm_rip_read(vcpu);
5954 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5955 kvm_run->debug.arch.exception = ex_no;
5958 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5959 kvm_run->ex.exception = ex_no;
5960 kvm_run->ex.error_code = error_code;
5966 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5968 ++vcpu->stat.irq_exits;
5972 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5974 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5975 vcpu->mmio_needed = 0;
5979 static int handle_io(struct kvm_vcpu *vcpu)
5981 unsigned long exit_qualification;
5982 int size, in, string, ret;
5985 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5986 string = (exit_qualification & 16) != 0;
5987 in = (exit_qualification & 8) != 0;
5989 ++vcpu->stat.io_exits;
5992 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5994 port = exit_qualification >> 16;
5995 size = (exit_qualification & 7) + 1;
5997 ret = kvm_skip_emulated_instruction(vcpu);
6000 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6001 * KVM_EXIT_DEBUG here.
6003 return kvm_fast_pio_out(vcpu, size, port) && ret;
6007 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
6010 * Patch in the VMCALL instruction:
6012 hypercall[0] = 0x0f;
6013 hypercall[1] = 0x01;
6014 hypercall[2] = 0xc1;
6017 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
6018 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
6020 if (is_guest_mode(vcpu)) {
6021 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6022 unsigned long orig_val = val;
6025 * We get here when L2 changed cr0 in a way that did not change
6026 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
6027 * but did change L0 shadowed bits. So we first calculate the
6028 * effective cr0 value that L1 would like to write into the
6029 * hardware. It consists of the L2-owned bits from the new
6030 * value combined with the L1-owned bits from L1's guest_cr0.
6032 val = (val & ~vmcs12->cr0_guest_host_mask) |
6033 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
6035 if (!nested_guest_cr0_valid(vcpu, val))
6038 if (kvm_set_cr0(vcpu, val))
6040 vmcs_writel(CR0_READ_SHADOW, orig_val);
6043 if (to_vmx(vcpu)->nested.vmxon &&
6044 !nested_host_cr0_valid(vcpu, val))
6047 return kvm_set_cr0(vcpu, val);
6051 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
6053 if (is_guest_mode(vcpu)) {
6054 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6055 unsigned long orig_val = val;
6057 /* analogously to handle_set_cr0 */
6058 val = (val & ~vmcs12->cr4_guest_host_mask) |
6059 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
6060 if (kvm_set_cr4(vcpu, val))
6062 vmcs_writel(CR4_READ_SHADOW, orig_val);
6065 return kvm_set_cr4(vcpu, val);
6068 static int handle_cr(struct kvm_vcpu *vcpu)
6070 unsigned long exit_qualification, val;
6076 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6077 cr = exit_qualification & 15;
6078 reg = (exit_qualification >> 8) & 15;
6079 switch ((exit_qualification >> 4) & 3) {
6080 case 0: /* mov to cr */
6081 val = kvm_register_readl(vcpu, reg);
6082 trace_kvm_cr_write(cr, val);
6085 err = handle_set_cr0(vcpu, val);
6086 return kvm_complete_insn_gp(vcpu, err);
6088 err = kvm_set_cr3(vcpu, val);
6089 return kvm_complete_insn_gp(vcpu, err);
6091 err = handle_set_cr4(vcpu, val);
6092 return kvm_complete_insn_gp(vcpu, err);
6094 u8 cr8_prev = kvm_get_cr8(vcpu);
6096 err = kvm_set_cr8(vcpu, cr8);
6097 ret = kvm_complete_insn_gp(vcpu, err);
6098 if (lapic_in_kernel(vcpu))
6100 if (cr8_prev <= cr8)
6103 * TODO: we might be squashing a
6104 * KVM_GUESTDBG_SINGLESTEP-triggered
6105 * KVM_EXIT_DEBUG here.
6107 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
6113 WARN_ONCE(1, "Guest should always own CR0.TS");
6114 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
6115 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
6116 return kvm_skip_emulated_instruction(vcpu);
6117 case 1: /*mov from cr*/
6120 val = kvm_read_cr3(vcpu);
6121 kvm_register_write(vcpu, reg, val);
6122 trace_kvm_cr_read(cr, val);
6123 return kvm_skip_emulated_instruction(vcpu);
6125 val = kvm_get_cr8(vcpu);
6126 kvm_register_write(vcpu, reg, val);
6127 trace_kvm_cr_read(cr, val);
6128 return kvm_skip_emulated_instruction(vcpu);
6132 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
6133 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
6134 kvm_lmsw(vcpu, val);
6136 return kvm_skip_emulated_instruction(vcpu);
6140 vcpu->run->exit_reason = 0;
6141 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
6142 (int)(exit_qualification >> 4) & 3, cr);
6146 static int handle_dr(struct kvm_vcpu *vcpu)
6148 unsigned long exit_qualification;
6151 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6152 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
6154 /* First, if DR does not exist, trigger UD */
6155 if (!kvm_require_dr(vcpu, dr))
6158 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
6159 if (!kvm_require_cpl(vcpu, 0))
6161 dr7 = vmcs_readl(GUEST_DR7);
6164 * As the vm-exit takes precedence over the debug trap, we
6165 * need to emulate the latter, either for the host or the
6166 * guest debugging itself.
6168 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6169 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
6170 vcpu->run->debug.arch.dr7 = dr7;
6171 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
6172 vcpu->run->debug.arch.exception = DB_VECTOR;
6173 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
6176 vcpu->arch.dr6 &= ~15;
6177 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
6178 kvm_queue_exception(vcpu, DB_VECTOR);
6183 if (vcpu->guest_debug == 0) {
6184 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6185 CPU_BASED_MOV_DR_EXITING);
6188 * No more DR vmexits; force a reload of the debug registers
6189 * and reenter on this instruction. The next vmexit will
6190 * retrieve the full state of the debug registers.
6192 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
6196 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
6197 if (exit_qualification & TYPE_MOV_FROM_DR) {
6200 if (kvm_get_dr(vcpu, dr, &val))
6202 kvm_register_write(vcpu, reg, val);
6204 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
6207 return kvm_skip_emulated_instruction(vcpu);
6210 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
6212 return vcpu->arch.dr6;
6215 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
6219 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
6221 get_debugreg(vcpu->arch.db[0], 0);
6222 get_debugreg(vcpu->arch.db[1], 1);
6223 get_debugreg(vcpu->arch.db[2], 2);
6224 get_debugreg(vcpu->arch.db[3], 3);
6225 get_debugreg(vcpu->arch.dr6, 6);
6226 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
6228 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
6229 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
6232 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
6234 vmcs_writel(GUEST_DR7, val);
6237 static int handle_cpuid(struct kvm_vcpu *vcpu)
6239 return kvm_emulate_cpuid(vcpu);
6242 static int handle_rdmsr(struct kvm_vcpu *vcpu)
6244 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6245 struct msr_data msr_info;
6247 msr_info.index = ecx;
6248 msr_info.host_initiated = false;
6249 if (vmx_get_msr(vcpu, &msr_info)) {
6250 trace_kvm_msr_read_ex(ecx);
6251 kvm_inject_gp(vcpu, 0);
6255 trace_kvm_msr_read(ecx, msr_info.data);
6257 /* FIXME: handling of bits 32:63 of rax, rdx */
6258 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
6259 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
6260 return kvm_skip_emulated_instruction(vcpu);
6263 static int handle_wrmsr(struct kvm_vcpu *vcpu)
6265 struct msr_data msr;
6266 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6267 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
6268 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6272 msr.host_initiated = false;
6273 if (kvm_set_msr(vcpu, &msr) != 0) {
6274 trace_kvm_msr_write_ex(ecx, data);
6275 kvm_inject_gp(vcpu, 0);
6279 trace_kvm_msr_write(ecx, data);
6280 return kvm_skip_emulated_instruction(vcpu);
6283 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
6285 kvm_apic_update_ppr(vcpu);
6289 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
6291 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6292 CPU_BASED_VIRTUAL_INTR_PENDING);
6294 kvm_make_request(KVM_REQ_EVENT, vcpu);
6296 ++vcpu->stat.irq_window_exits;
6300 static int handle_halt(struct kvm_vcpu *vcpu)
6302 return kvm_emulate_halt(vcpu);
6305 static int handle_vmcall(struct kvm_vcpu *vcpu)
6307 return kvm_emulate_hypercall(vcpu);
6310 static int handle_invd(struct kvm_vcpu *vcpu)
6312 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6315 static int handle_invlpg(struct kvm_vcpu *vcpu)
6317 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6319 kvm_mmu_invlpg(vcpu, exit_qualification);
6320 return kvm_skip_emulated_instruction(vcpu);
6323 static int handle_rdpmc(struct kvm_vcpu *vcpu)
6327 err = kvm_rdpmc(vcpu);
6328 return kvm_complete_insn_gp(vcpu, err);
6331 static int handle_wbinvd(struct kvm_vcpu *vcpu)
6333 return kvm_emulate_wbinvd(vcpu);
6336 static int handle_xsetbv(struct kvm_vcpu *vcpu)
6338 u64 new_bv = kvm_read_edx_eax(vcpu);
6339 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
6341 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
6342 return kvm_skip_emulated_instruction(vcpu);
6346 static int handle_xsaves(struct kvm_vcpu *vcpu)
6348 kvm_skip_emulated_instruction(vcpu);
6349 WARN(1, "this should never happen\n");
6353 static int handle_xrstors(struct kvm_vcpu *vcpu)
6355 kvm_skip_emulated_instruction(vcpu);
6356 WARN(1, "this should never happen\n");
6360 static int handle_apic_access(struct kvm_vcpu *vcpu)
6362 if (likely(fasteoi)) {
6363 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6364 int access_type, offset;
6366 access_type = exit_qualification & APIC_ACCESS_TYPE;
6367 offset = exit_qualification & APIC_ACCESS_OFFSET;
6369 * Sane guest uses MOV to write EOI, with written value
6370 * not cared. So make a short-circuit here by avoiding
6371 * heavy instruction emulation.
6373 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
6374 (offset == APIC_EOI)) {
6375 kvm_lapic_set_eoi(vcpu);
6376 return kvm_skip_emulated_instruction(vcpu);
6379 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6382 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
6384 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6385 int vector = exit_qualification & 0xff;
6387 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
6388 kvm_apic_set_eoi_accelerated(vcpu, vector);
6392 static int handle_apic_write(struct kvm_vcpu *vcpu)
6394 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6395 u32 offset = exit_qualification & 0xfff;
6397 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6398 kvm_apic_write_nodecode(vcpu, offset);
6402 static int handle_task_switch(struct kvm_vcpu *vcpu)
6404 struct vcpu_vmx *vmx = to_vmx(vcpu);
6405 unsigned long exit_qualification;
6406 bool has_error_code = false;
6409 int reason, type, idt_v, idt_index;
6411 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
6412 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
6413 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
6415 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6417 reason = (u32)exit_qualification >> 30;
6418 if (reason == TASK_SWITCH_GATE && idt_v) {
6420 case INTR_TYPE_NMI_INTR:
6421 vcpu->arch.nmi_injected = false;
6422 vmx_set_nmi_mask(vcpu, true);
6424 case INTR_TYPE_EXT_INTR:
6425 case INTR_TYPE_SOFT_INTR:
6426 kvm_clear_interrupt_queue(vcpu);
6428 case INTR_TYPE_HARD_EXCEPTION:
6429 if (vmx->idt_vectoring_info &
6430 VECTORING_INFO_DELIVER_CODE_MASK) {
6431 has_error_code = true;
6433 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6436 case INTR_TYPE_SOFT_EXCEPTION:
6437 kvm_clear_exception_queue(vcpu);
6443 tss_selector = exit_qualification;
6445 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
6446 type != INTR_TYPE_EXT_INTR &&
6447 type != INTR_TYPE_NMI_INTR))
6448 skip_emulated_instruction(vcpu);
6450 if (kvm_task_switch(vcpu, tss_selector,
6451 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
6452 has_error_code, error_code) == EMULATE_FAIL) {
6453 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6454 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6455 vcpu->run->internal.ndata = 0;
6460 * TODO: What about debug traps on tss switch?
6461 * Are we supposed to inject them and update dr6?
6467 static int handle_ept_violation(struct kvm_vcpu *vcpu)
6469 unsigned long exit_qualification;
6473 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6476 * EPT violation happened while executing iret from NMI,
6477 * "blocked by NMI" bit has to be set before next VM entry.
6478 * There are errata that may cause this bit to not be set:
6481 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
6482 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
6483 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
6485 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6486 trace_kvm_page_fault(gpa, exit_qualification);
6488 /* Is it a read fault? */
6489 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
6490 ? PFERR_USER_MASK : 0;
6491 /* Is it a write fault? */
6492 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
6493 ? PFERR_WRITE_MASK : 0;
6494 /* Is it a fetch fault? */
6495 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
6496 ? PFERR_FETCH_MASK : 0;
6497 /* ept page table entry is present? */
6498 error_code |= (exit_qualification &
6499 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
6500 EPT_VIOLATION_EXECUTABLE))
6501 ? PFERR_PRESENT_MASK : 0;
6503 error_code |= (exit_qualification & 0x100) != 0 ?
6504 PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
6506 vcpu->arch.exit_qualification = exit_qualification;
6507 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
6510 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
6516 * A nested guest cannot optimize MMIO vmexits, because we have an
6517 * nGPA here instead of the required GPA.
6519 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6520 if (!is_guest_mode(vcpu) &&
6521 !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
6522 trace_kvm_fast_mmio(gpa);
6523 return kvm_skip_emulated_instruction(vcpu);
6526 ret = kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
6530 /* It is the real ept misconfig */
6533 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6534 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6539 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6541 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6542 CPU_BASED_VIRTUAL_NMI_PENDING);
6543 ++vcpu->stat.nmi_window_exits;
6544 kvm_make_request(KVM_REQ_EVENT, vcpu);
6549 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6551 struct vcpu_vmx *vmx = to_vmx(vcpu);
6552 enum emulation_result err = EMULATE_DONE;
6555 bool intr_window_requested;
6556 unsigned count = 130;
6558 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6559 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6561 while (vmx->emulation_required && count-- != 0) {
6562 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6563 return handle_interrupt_window(&vmx->vcpu);
6565 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
6568 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
6570 if (err == EMULATE_USER_EXIT) {
6571 ++vcpu->stat.mmio_exits;
6576 if (err != EMULATE_DONE) {
6577 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6578 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6579 vcpu->run->internal.ndata = 0;
6583 if (vcpu->arch.halt_request) {
6584 vcpu->arch.halt_request = 0;
6585 ret = kvm_vcpu_halt(vcpu);
6589 if (signal_pending(current))
6599 static int __grow_ple_window(int val)
6601 if (ple_window_grow < 1)
6604 val = min(val, ple_window_actual_max);
6606 if (ple_window_grow < ple_window)
6607 val *= ple_window_grow;
6609 val += ple_window_grow;
6614 static int __shrink_ple_window(int val, int modifier, int minimum)
6619 if (modifier < ple_window)
6624 return max(val, minimum);
6627 static void grow_ple_window(struct kvm_vcpu *vcpu)
6629 struct vcpu_vmx *vmx = to_vmx(vcpu);
6630 int old = vmx->ple_window;
6632 vmx->ple_window = __grow_ple_window(old);
6634 if (vmx->ple_window != old)
6635 vmx->ple_window_dirty = true;
6637 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6640 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6642 struct vcpu_vmx *vmx = to_vmx(vcpu);
6643 int old = vmx->ple_window;
6645 vmx->ple_window = __shrink_ple_window(old,
6646 ple_window_shrink, ple_window);
6648 if (vmx->ple_window != old)
6649 vmx->ple_window_dirty = true;
6651 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6655 * ple_window_actual_max is computed to be one grow_ple_window() below
6656 * ple_window_max. (See __grow_ple_window for the reason.)
6657 * This prevents overflows, because ple_window_max is int.
6658 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6660 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6662 static void update_ple_window_actual_max(void)
6664 ple_window_actual_max =
6665 __shrink_ple_window(max(ple_window_max, ple_window),
6666 ple_window_grow, INT_MIN);
6670 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6672 static void wakeup_handler(void)
6674 struct kvm_vcpu *vcpu;
6675 int cpu = smp_processor_id();
6677 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6678 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6679 blocked_vcpu_list) {
6680 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6682 if (pi_test_on(pi_desc) == 1)
6683 kvm_vcpu_kick(vcpu);
6685 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6688 void vmx_enable_tdp(void)
6690 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
6691 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
6692 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
6693 0ull, VMX_EPT_EXECUTABLE_MASK,
6694 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
6695 VMX_EPT_RWX_MASK, 0ull);
6697 ept_set_mmio_spte_mask();
6701 static __init int hardware_setup(void)
6703 int r = -ENOMEM, i, msr;
6705 rdmsrl_safe(MSR_EFER, &host_efer);
6707 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6708 kvm_define_shared_msr(i, vmx_msr_index[i]);
6710 for (i = 0; i < VMX_BITMAP_NR; i++) {
6711 vmx_bitmap[i] = (unsigned long *)__get_free_page(GFP_KERNEL);
6716 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6717 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6718 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6721 * Allow direct access to the PC debug port (it is often used for I/O
6722 * delays, but the vmexits simply slow things down).
6724 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6725 clear_bit(0x80, vmx_io_bitmap_a);
6727 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6729 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6730 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6732 if (setup_vmcs_config(&vmcs_config) < 0) {
6737 if (boot_cpu_has(X86_FEATURE_NX))
6738 kvm_enable_efer_bits(EFER_NX);
6740 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
6741 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
6744 if (!cpu_has_vmx_shadow_vmcs())
6745 enable_shadow_vmcs = 0;
6746 if (enable_shadow_vmcs)
6747 init_vmcs_shadow_fields();
6749 if (!cpu_has_vmx_ept() ||
6750 !cpu_has_vmx_ept_4levels() ||
6751 !cpu_has_vmx_ept_mt_wb()) {
6753 enable_unrestricted_guest = 0;
6754 enable_ept_ad_bits = 0;
6757 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
6758 enable_ept_ad_bits = 0;
6760 if (!cpu_has_vmx_unrestricted_guest())
6761 enable_unrestricted_guest = 0;
6763 if (!cpu_has_vmx_flexpriority())
6764 flexpriority_enabled = 0;
6767 * set_apic_access_page_addr() is used to reload apic access
6768 * page upon invalidation. No need to do anything if not
6769 * using the APIC_ACCESS_ADDR VMCS field.
6771 if (!flexpriority_enabled)
6772 kvm_x86_ops->set_apic_access_page_addr = NULL;
6774 if (!cpu_has_vmx_tpr_shadow())
6775 kvm_x86_ops->update_cr8_intercept = NULL;
6777 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6778 kvm_disable_largepages();
6780 if (!cpu_has_vmx_ple())
6783 if (!cpu_has_vmx_apicv()) {
6785 kvm_x86_ops->sync_pir_to_irr = NULL;
6788 if (cpu_has_vmx_tsc_scaling()) {
6789 kvm_has_tsc_control = true;
6790 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6791 kvm_tsc_scaling_ratio_frac_bits = 48;
6794 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6795 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6796 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6797 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6798 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6799 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6801 memcpy(vmx_msr_bitmap_legacy_x2apic_apicv,
6802 vmx_msr_bitmap_legacy, PAGE_SIZE);
6803 memcpy(vmx_msr_bitmap_longmode_x2apic_apicv,
6804 vmx_msr_bitmap_longmode, PAGE_SIZE);
6805 memcpy(vmx_msr_bitmap_legacy_x2apic,
6806 vmx_msr_bitmap_legacy, PAGE_SIZE);
6807 memcpy(vmx_msr_bitmap_longmode_x2apic,
6808 vmx_msr_bitmap_longmode, PAGE_SIZE);
6810 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6812 for (msr = 0x800; msr <= 0x8ff; msr++) {
6813 if (msr == 0x839 /* TMCCT */)
6815 vmx_disable_intercept_msr_x2apic(msr, MSR_TYPE_R, true);
6819 * TPR reads and writes can be virtualized even if virtual interrupt
6820 * delivery is not in use.
6822 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_W, true);
6823 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_R | MSR_TYPE_W, false);
6826 vmx_disable_intercept_msr_x2apic(0x80b, MSR_TYPE_W, true);
6828 vmx_disable_intercept_msr_x2apic(0x83f, MSR_TYPE_W, true);
6835 update_ple_window_actual_max();
6838 * Only enable PML when hardware supports PML feature, and both EPT
6839 * and EPT A/D bit features are enabled -- PML depends on them to work.
6841 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6845 kvm_x86_ops->slot_enable_log_dirty = NULL;
6846 kvm_x86_ops->slot_disable_log_dirty = NULL;
6847 kvm_x86_ops->flush_log_dirty = NULL;
6848 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6851 if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
6854 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
6855 cpu_preemption_timer_multi =
6856 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
6858 kvm_x86_ops->set_hv_timer = NULL;
6859 kvm_x86_ops->cancel_hv_timer = NULL;
6862 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6864 kvm_mce_cap_supported |= MCG_LMCE_P;
6866 return alloc_kvm_area();
6869 for (i = 0; i < VMX_BITMAP_NR; i++)
6870 free_page((unsigned long)vmx_bitmap[i]);
6875 static __exit void hardware_unsetup(void)
6879 for (i = 0; i < VMX_BITMAP_NR; i++)
6880 free_page((unsigned long)vmx_bitmap[i]);
6886 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6887 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6889 static int handle_pause(struct kvm_vcpu *vcpu)
6892 grow_ple_window(vcpu);
6895 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
6896 * VM-execution control is ignored if CPL > 0. OTOH, KVM
6897 * never set PAUSE_EXITING and just set PLE if supported,
6898 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
6900 kvm_vcpu_on_spin(vcpu, true);
6901 return kvm_skip_emulated_instruction(vcpu);
6904 static int handle_nop(struct kvm_vcpu *vcpu)
6906 return kvm_skip_emulated_instruction(vcpu);
6909 static int handle_mwait(struct kvm_vcpu *vcpu)
6911 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6912 return handle_nop(vcpu);
6915 static int handle_invalid_op(struct kvm_vcpu *vcpu)
6917 kvm_queue_exception(vcpu, UD_VECTOR);
6921 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6926 static int handle_monitor(struct kvm_vcpu *vcpu)
6928 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6929 return handle_nop(vcpu);
6933 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6934 * We could reuse a single VMCS for all the L2 guests, but we also want the
6935 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6936 * allows keeping them loaded on the processor, and in the future will allow
6937 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6938 * every entry if they never change.
6939 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6940 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6942 * The following functions allocate and free a vmcs02 in this pool.
6945 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6946 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6948 struct vmcs02_list *item;
6949 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6950 if (item->vmptr == vmx->nested.current_vmptr) {
6951 list_move(&item->list, &vmx->nested.vmcs02_pool);
6952 return &item->vmcs02;
6955 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6956 /* Recycle the least recently used VMCS. */
6957 item = list_last_entry(&vmx->nested.vmcs02_pool,
6958 struct vmcs02_list, list);
6959 item->vmptr = vmx->nested.current_vmptr;
6960 list_move(&item->list, &vmx->nested.vmcs02_pool);
6961 return &item->vmcs02;
6964 /* Create a new VMCS */
6965 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6968 item->vmcs02.vmcs = alloc_vmcs();
6969 item->vmcs02.shadow_vmcs = NULL;
6970 if (!item->vmcs02.vmcs) {
6974 loaded_vmcs_init(&item->vmcs02);
6975 item->vmptr = vmx->nested.current_vmptr;
6976 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6977 vmx->nested.vmcs02_num++;
6978 return &item->vmcs02;
6981 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6982 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6984 struct vmcs02_list *item;
6985 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6986 if (item->vmptr == vmptr) {
6987 free_loaded_vmcs(&item->vmcs02);
6988 list_del(&item->list);
6990 vmx->nested.vmcs02_num--;
6996 * Free all VMCSs saved for this vcpu, except the one pointed by
6997 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6998 * must be &vmx->vmcs01.
7000 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
7002 struct vmcs02_list *item, *n;
7004 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
7005 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
7007 * Something will leak if the above WARN triggers. Better than
7010 if (vmx->loaded_vmcs == &item->vmcs02)
7013 free_loaded_vmcs(&item->vmcs02);
7014 list_del(&item->list);
7016 vmx->nested.vmcs02_num--;
7021 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
7022 * set the success or error code of an emulated VMX instruction, as specified
7023 * by Vol 2B, VMX Instruction Reference, "Conventions".
7025 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
7027 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
7028 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
7029 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
7032 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
7034 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
7035 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
7036 X86_EFLAGS_SF | X86_EFLAGS_OF))
7040 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
7041 u32 vm_instruction_error)
7043 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
7045 * failValid writes the error number to the current VMCS, which
7046 * can't be done there isn't a current VMCS.
7048 nested_vmx_failInvalid(vcpu);
7051 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
7052 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
7053 X86_EFLAGS_SF | X86_EFLAGS_OF))
7055 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
7057 * We don't need to force a shadow sync because
7058 * VM_INSTRUCTION_ERROR is not shadowed
7062 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
7064 /* TODO: not to reset guest simply here. */
7065 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7066 pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
7069 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
7071 struct vcpu_vmx *vmx =
7072 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
7074 vmx->nested.preemption_timer_expired = true;
7075 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
7076 kvm_vcpu_kick(&vmx->vcpu);
7078 return HRTIMER_NORESTART;
7082 * Decode the memory-address operand of a vmx instruction, as recorded on an
7083 * exit caused by such an instruction (run by a guest hypervisor).
7084 * On success, returns 0. When the operand is invalid, returns 1 and throws
7087 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
7088 unsigned long exit_qualification,
7089 u32 vmx_instruction_info, bool wr, gva_t *ret)
7093 struct kvm_segment s;
7096 * According to Vol. 3B, "Information for VM Exits Due to Instruction
7097 * Execution", on an exit, vmx_instruction_info holds most of the
7098 * addressing components of the operand. Only the displacement part
7099 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
7100 * For how an actual address is calculated from all these components,
7101 * refer to Vol. 1, "Operand Addressing".
7103 int scaling = vmx_instruction_info & 3;
7104 int addr_size = (vmx_instruction_info >> 7) & 7;
7105 bool is_reg = vmx_instruction_info & (1u << 10);
7106 int seg_reg = (vmx_instruction_info >> 15) & 7;
7107 int index_reg = (vmx_instruction_info >> 18) & 0xf;
7108 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
7109 int base_reg = (vmx_instruction_info >> 23) & 0xf;
7110 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
7113 kvm_queue_exception(vcpu, UD_VECTOR);
7117 /* Addr = segment_base + offset */
7118 /* offset = base + [index * scale] + displacement */
7119 off = exit_qualification; /* holds the displacement */
7121 off += kvm_register_read(vcpu, base_reg);
7123 off += kvm_register_read(vcpu, index_reg)<<scaling;
7124 vmx_get_segment(vcpu, &s, seg_reg);
7125 *ret = s.base + off;
7127 if (addr_size == 1) /* 32 bit */
7130 /* Checks for #GP/#SS exceptions. */
7132 if (is_long_mode(vcpu)) {
7133 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
7134 * non-canonical form. This is the only check on the memory
7135 * destination for long mode!
7137 exn = is_noncanonical_address(*ret, vcpu);
7138 } else if (is_protmode(vcpu)) {
7139 /* Protected mode: apply checks for segment validity in the
7141 * - segment type check (#GP(0) may be thrown)
7142 * - usability check (#GP(0)/#SS(0))
7143 * - limit check (#GP(0)/#SS(0))
7146 /* #GP(0) if the destination operand is located in a
7147 * read-only data segment or any code segment.
7149 exn = ((s.type & 0xa) == 0 || (s.type & 8));
7151 /* #GP(0) if the source operand is located in an
7152 * execute-only code segment
7154 exn = ((s.type & 0xa) == 8);
7156 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7159 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
7161 exn = (s.unusable != 0);
7162 /* Protected mode: #GP(0)/#SS(0) if the memory
7163 * operand is outside the segment limit.
7165 exn = exn || (off + sizeof(u64) > s.limit);
7168 kvm_queue_exception_e(vcpu,
7169 seg_reg == VCPU_SREG_SS ?
7170 SS_VECTOR : GP_VECTOR,
7178 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
7181 struct x86_exception e;
7183 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7184 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
7187 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, vmpointer,
7188 sizeof(*vmpointer), &e)) {
7189 kvm_inject_page_fault(vcpu, &e);
7196 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
7198 struct vcpu_vmx *vmx = to_vmx(vcpu);
7199 struct vmcs *shadow_vmcs;
7201 if (cpu_has_vmx_msr_bitmap()) {
7202 vmx->nested.msr_bitmap =
7203 (unsigned long *)__get_free_page(GFP_KERNEL);
7204 if (!vmx->nested.msr_bitmap)
7205 goto out_msr_bitmap;
7208 vmx->nested.cached_vmcs12 = kmalloc(VMCS12_SIZE, GFP_KERNEL);
7209 if (!vmx->nested.cached_vmcs12)
7210 goto out_cached_vmcs12;
7212 if (enable_shadow_vmcs) {
7213 shadow_vmcs = alloc_vmcs();
7215 goto out_shadow_vmcs;
7216 /* mark vmcs as shadow */
7217 shadow_vmcs->revision_id |= (1u << 31);
7218 /* init shadow vmcs */
7219 vmcs_clear(shadow_vmcs);
7220 vmx->vmcs01.shadow_vmcs = shadow_vmcs;
7223 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
7224 vmx->nested.vmcs02_num = 0;
7226 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
7227 HRTIMER_MODE_REL_PINNED);
7228 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
7230 vmx->nested.vmxon = true;
7234 kfree(vmx->nested.cached_vmcs12);
7237 free_page((unsigned long)vmx->nested.msr_bitmap);
7244 * Emulate the VMXON instruction.
7245 * Currently, we just remember that VMX is active, and do not save or even
7246 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
7247 * do not currently need to store anything in that guest-allocated memory
7248 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
7249 * argument is different from the VMXON pointer (which the spec says they do).
7251 static int handle_vmon(struct kvm_vcpu *vcpu)
7256 struct vcpu_vmx *vmx = to_vmx(vcpu);
7257 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
7258 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7261 * The Intel VMX Instruction Reference lists a bunch of bits that are
7262 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
7263 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
7264 * Otherwise, we should fail with #UD. But most faulting conditions
7265 * have already been checked by hardware, prior to the VM-exit for
7266 * VMXON. We do test guest cr4.VMXE because processor CR4 always has
7267 * that bit set to 1 in non-root mode.
7269 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
7270 kvm_queue_exception(vcpu, UD_VECTOR);
7274 if (vmx->nested.vmxon) {
7275 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
7276 return kvm_skip_emulated_instruction(vcpu);
7279 if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
7280 != VMXON_NEEDED_FEATURES) {
7281 kvm_inject_gp(vcpu, 0);
7285 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7290 * The first 4 bytes of VMXON region contain the supported
7291 * VMCS revision identifier
7293 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
7294 * which replaces physical address width with 32
7296 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7297 nested_vmx_failInvalid(vcpu);
7298 return kvm_skip_emulated_instruction(vcpu);
7301 page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
7302 if (is_error_page(page)) {
7303 nested_vmx_failInvalid(vcpu);
7304 return kvm_skip_emulated_instruction(vcpu);
7306 if (*(u32 *)kmap(page) != VMCS12_REVISION) {
7308 kvm_release_page_clean(page);
7309 nested_vmx_failInvalid(vcpu);
7310 return kvm_skip_emulated_instruction(vcpu);
7313 kvm_release_page_clean(page);
7315 vmx->nested.vmxon_ptr = vmptr;
7316 ret = enter_vmx_operation(vcpu);
7320 nested_vmx_succeed(vcpu);
7321 return kvm_skip_emulated_instruction(vcpu);
7325 * Intel's VMX Instruction Reference specifies a common set of prerequisites
7326 * for running VMX instructions (except VMXON, whose prerequisites are
7327 * slightly different). It also specifies what exception to inject otherwise.
7328 * Note that many of these exceptions have priority over VM exits, so they
7329 * don't have to be checked again here.
7331 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
7333 if (!to_vmx(vcpu)->nested.vmxon) {
7334 kvm_queue_exception(vcpu, UD_VECTOR);
7340 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
7342 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL, SECONDARY_EXEC_SHADOW_VMCS);
7343 vmcs_write64(VMCS_LINK_POINTER, -1ull);
7346 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
7348 if (vmx->nested.current_vmptr == -1ull)
7351 if (enable_shadow_vmcs) {
7352 /* copy to memory all shadowed fields in case
7353 they were modified */
7354 copy_shadow_to_vmcs12(vmx);
7355 vmx->nested.sync_shadow_vmcs = false;
7356 vmx_disable_shadow_vmcs(vmx);
7358 vmx->nested.posted_intr_nv = -1;
7360 /* Flush VMCS12 to guest memory */
7361 kvm_vcpu_write_guest_page(&vmx->vcpu,
7362 vmx->nested.current_vmptr >> PAGE_SHIFT,
7363 vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
7365 vmx->nested.current_vmptr = -1ull;
7369 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7370 * just stops using VMX.
7372 static void free_nested(struct vcpu_vmx *vmx)
7374 if (!vmx->nested.vmxon)
7377 vmx->nested.vmxon = false;
7378 free_vpid(vmx->nested.vpid02);
7379 vmx->nested.posted_intr_nv = -1;
7380 vmx->nested.current_vmptr = -1ull;
7381 if (vmx->nested.msr_bitmap) {
7382 free_page((unsigned long)vmx->nested.msr_bitmap);
7383 vmx->nested.msr_bitmap = NULL;
7385 if (enable_shadow_vmcs) {
7386 vmx_disable_shadow_vmcs(vmx);
7387 vmcs_clear(vmx->vmcs01.shadow_vmcs);
7388 free_vmcs(vmx->vmcs01.shadow_vmcs);
7389 vmx->vmcs01.shadow_vmcs = NULL;
7391 kfree(vmx->nested.cached_vmcs12);
7392 /* Unpin physical memory we referred to in current vmcs02 */
7393 if (vmx->nested.apic_access_page) {
7394 kvm_release_page_dirty(vmx->nested.apic_access_page);
7395 vmx->nested.apic_access_page = NULL;
7397 if (vmx->nested.virtual_apic_page) {
7398 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
7399 vmx->nested.virtual_apic_page = NULL;
7401 if (vmx->nested.pi_desc_page) {
7402 kunmap(vmx->nested.pi_desc_page);
7403 kvm_release_page_dirty(vmx->nested.pi_desc_page);
7404 vmx->nested.pi_desc_page = NULL;
7405 vmx->nested.pi_desc = NULL;
7408 nested_free_all_saved_vmcss(vmx);
7411 /* Emulate the VMXOFF instruction */
7412 static int handle_vmoff(struct kvm_vcpu *vcpu)
7414 if (!nested_vmx_check_permission(vcpu))
7416 free_nested(to_vmx(vcpu));
7417 nested_vmx_succeed(vcpu);
7418 return kvm_skip_emulated_instruction(vcpu);
7421 /* Emulate the VMCLEAR instruction */
7422 static int handle_vmclear(struct kvm_vcpu *vcpu)
7424 struct vcpu_vmx *vmx = to_vmx(vcpu);
7428 if (!nested_vmx_check_permission(vcpu))
7431 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7434 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7435 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
7436 return kvm_skip_emulated_instruction(vcpu);
7439 if (vmptr == vmx->nested.vmxon_ptr) {
7440 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
7441 return kvm_skip_emulated_instruction(vcpu);
7444 if (vmptr == vmx->nested.current_vmptr)
7445 nested_release_vmcs12(vmx);
7447 kvm_vcpu_write_guest(vcpu,
7448 vmptr + offsetof(struct vmcs12, launch_state),
7449 &zero, sizeof(zero));
7451 nested_free_vmcs02(vmx, vmptr);
7453 nested_vmx_succeed(vcpu);
7454 return kvm_skip_emulated_instruction(vcpu);
7457 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
7459 /* Emulate the VMLAUNCH instruction */
7460 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
7462 return nested_vmx_run(vcpu, true);
7465 /* Emulate the VMRESUME instruction */
7466 static int handle_vmresume(struct kvm_vcpu *vcpu)
7469 return nested_vmx_run(vcpu, false);
7473 * Read a vmcs12 field. Since these can have varying lengths and we return
7474 * one type, we chose the biggest type (u64) and zero-extend the return value
7475 * to that size. Note that the caller, handle_vmread, might need to use only
7476 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7477 * 64-bit fields are to be returned).
7479 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7480 unsigned long field, u64 *ret)
7482 short offset = vmcs_field_to_offset(field);
7488 p = ((char *)(get_vmcs12(vcpu))) + offset;
7490 switch (vmcs_field_type(field)) {
7491 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7492 *ret = *((natural_width *)p);
7494 case VMCS_FIELD_TYPE_U16:
7497 case VMCS_FIELD_TYPE_U32:
7500 case VMCS_FIELD_TYPE_U64:
7510 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7511 unsigned long field, u64 field_value){
7512 short offset = vmcs_field_to_offset(field);
7513 char *p = ((char *) get_vmcs12(vcpu)) + offset;
7517 switch (vmcs_field_type(field)) {
7518 case VMCS_FIELD_TYPE_U16:
7519 *(u16 *)p = field_value;
7521 case VMCS_FIELD_TYPE_U32:
7522 *(u32 *)p = field_value;
7524 case VMCS_FIELD_TYPE_U64:
7525 *(u64 *)p = field_value;
7527 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7528 *(natural_width *)p = field_value;
7537 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7540 unsigned long field;
7542 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7543 const unsigned long *fields = shadow_read_write_fields;
7544 const int num_fields = max_shadow_read_write_fields;
7548 vmcs_load(shadow_vmcs);
7550 for (i = 0; i < num_fields; i++) {
7552 switch (vmcs_field_type(field)) {
7553 case VMCS_FIELD_TYPE_U16:
7554 field_value = vmcs_read16(field);
7556 case VMCS_FIELD_TYPE_U32:
7557 field_value = vmcs_read32(field);
7559 case VMCS_FIELD_TYPE_U64:
7560 field_value = vmcs_read64(field);
7562 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7563 field_value = vmcs_readl(field);
7569 vmcs12_write_any(&vmx->vcpu, field, field_value);
7572 vmcs_clear(shadow_vmcs);
7573 vmcs_load(vmx->loaded_vmcs->vmcs);
7578 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7580 const unsigned long *fields[] = {
7581 shadow_read_write_fields,
7582 shadow_read_only_fields
7584 const int max_fields[] = {
7585 max_shadow_read_write_fields,
7586 max_shadow_read_only_fields
7589 unsigned long field;
7590 u64 field_value = 0;
7591 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7593 vmcs_load(shadow_vmcs);
7595 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7596 for (i = 0; i < max_fields[q]; i++) {
7597 field = fields[q][i];
7598 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7600 switch (vmcs_field_type(field)) {
7601 case VMCS_FIELD_TYPE_U16:
7602 vmcs_write16(field, (u16)field_value);
7604 case VMCS_FIELD_TYPE_U32:
7605 vmcs_write32(field, (u32)field_value);
7607 case VMCS_FIELD_TYPE_U64:
7608 vmcs_write64(field, (u64)field_value);
7610 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7611 vmcs_writel(field, (long)field_value);
7620 vmcs_clear(shadow_vmcs);
7621 vmcs_load(vmx->loaded_vmcs->vmcs);
7625 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7626 * used before) all generate the same failure when it is missing.
7628 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7630 struct vcpu_vmx *vmx = to_vmx(vcpu);
7631 if (vmx->nested.current_vmptr == -1ull) {
7632 nested_vmx_failInvalid(vcpu);
7638 static int handle_vmread(struct kvm_vcpu *vcpu)
7640 unsigned long field;
7642 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7643 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7646 if (!nested_vmx_check_permission(vcpu))
7649 if (!nested_vmx_check_vmcs12(vcpu))
7650 return kvm_skip_emulated_instruction(vcpu);
7652 /* Decode instruction info and find the field to read */
7653 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7654 /* Read the field, zero-extended to a u64 field_value */
7655 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7656 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7657 return kvm_skip_emulated_instruction(vcpu);
7660 * Now copy part of this value to register or memory, as requested.
7661 * Note that the number of bits actually copied is 32 or 64 depending
7662 * on the guest's mode (32 or 64 bit), not on the given field's length.
7664 if (vmx_instruction_info & (1u << 10)) {
7665 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7668 if (get_vmx_mem_address(vcpu, exit_qualification,
7669 vmx_instruction_info, true, &gva))
7671 /* _system ok, as hardware has verified cpl=0 */
7672 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7673 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7676 nested_vmx_succeed(vcpu);
7677 return kvm_skip_emulated_instruction(vcpu);
7681 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7683 unsigned long field;
7685 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7686 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7687 /* The value to write might be 32 or 64 bits, depending on L1's long
7688 * mode, and eventually we need to write that into a field of several
7689 * possible lengths. The code below first zero-extends the value to 64
7690 * bit (field_value), and then copies only the appropriate number of
7691 * bits into the vmcs12 field.
7693 u64 field_value = 0;
7694 struct x86_exception e;
7696 if (!nested_vmx_check_permission(vcpu))
7699 if (!nested_vmx_check_vmcs12(vcpu))
7700 return kvm_skip_emulated_instruction(vcpu);
7702 if (vmx_instruction_info & (1u << 10))
7703 field_value = kvm_register_readl(vcpu,
7704 (((vmx_instruction_info) >> 3) & 0xf));
7706 if (get_vmx_mem_address(vcpu, exit_qualification,
7707 vmx_instruction_info, false, &gva))
7709 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7710 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7711 kvm_inject_page_fault(vcpu, &e);
7717 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7718 if (vmcs_field_readonly(field)) {
7719 nested_vmx_failValid(vcpu,
7720 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7721 return kvm_skip_emulated_instruction(vcpu);
7724 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7725 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7726 return kvm_skip_emulated_instruction(vcpu);
7729 nested_vmx_succeed(vcpu);
7730 return kvm_skip_emulated_instruction(vcpu);
7733 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
7735 vmx->nested.current_vmptr = vmptr;
7736 if (enable_shadow_vmcs) {
7737 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7738 SECONDARY_EXEC_SHADOW_VMCS);
7739 vmcs_write64(VMCS_LINK_POINTER,
7740 __pa(vmx->vmcs01.shadow_vmcs));
7741 vmx->nested.sync_shadow_vmcs = true;
7745 /* Emulate the VMPTRLD instruction */
7746 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7748 struct vcpu_vmx *vmx = to_vmx(vcpu);
7751 if (!nested_vmx_check_permission(vcpu))
7754 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7757 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7758 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
7759 return kvm_skip_emulated_instruction(vcpu);
7762 if (vmptr == vmx->nested.vmxon_ptr) {
7763 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
7764 return kvm_skip_emulated_instruction(vcpu);
7767 if (vmx->nested.current_vmptr != vmptr) {
7768 struct vmcs12 *new_vmcs12;
7770 page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
7771 if (is_error_page(page)) {
7772 nested_vmx_failInvalid(vcpu);
7773 return kvm_skip_emulated_instruction(vcpu);
7775 new_vmcs12 = kmap(page);
7776 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7778 kvm_release_page_clean(page);
7779 nested_vmx_failValid(vcpu,
7780 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7781 return kvm_skip_emulated_instruction(vcpu);
7784 nested_release_vmcs12(vmx);
7786 * Load VMCS12 from guest memory since it is not already
7789 memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
7791 kvm_release_page_clean(page);
7793 set_current_vmptr(vmx, vmptr);
7796 nested_vmx_succeed(vcpu);
7797 return kvm_skip_emulated_instruction(vcpu);
7800 /* Emulate the VMPTRST instruction */
7801 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7803 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7804 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7806 struct x86_exception e;
7808 if (!nested_vmx_check_permission(vcpu))
7811 if (get_vmx_mem_address(vcpu, exit_qualification,
7812 vmx_instruction_info, true, &vmcs_gva))
7814 /* ok to use *_system, as hardware has verified cpl=0 */
7815 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7816 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7818 kvm_inject_page_fault(vcpu, &e);
7821 nested_vmx_succeed(vcpu);
7822 return kvm_skip_emulated_instruction(vcpu);
7825 /* Emulate the INVEPT instruction */
7826 static int handle_invept(struct kvm_vcpu *vcpu)
7828 struct vcpu_vmx *vmx = to_vmx(vcpu);
7829 u32 vmx_instruction_info, types;
7832 struct x86_exception e;
7837 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7838 SECONDARY_EXEC_ENABLE_EPT) ||
7839 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7840 kvm_queue_exception(vcpu, UD_VECTOR);
7844 if (!nested_vmx_check_permission(vcpu))
7847 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7848 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7850 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7852 if (type >= 32 || !(types & (1 << type))) {
7853 nested_vmx_failValid(vcpu,
7854 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7855 return kvm_skip_emulated_instruction(vcpu);
7858 /* According to the Intel VMX instruction reference, the memory
7859 * operand is read even if it isn't needed (e.g., for type==global)
7861 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7862 vmx_instruction_info, false, &gva))
7864 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7865 sizeof(operand), &e)) {
7866 kvm_inject_page_fault(vcpu, &e);
7871 case VMX_EPT_EXTENT_GLOBAL:
7873 * TODO: track mappings and invalidate
7874 * single context requests appropriately
7876 case VMX_EPT_EXTENT_CONTEXT:
7877 kvm_mmu_sync_roots(vcpu);
7878 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7879 nested_vmx_succeed(vcpu);
7886 return kvm_skip_emulated_instruction(vcpu);
7889 static int handle_invvpid(struct kvm_vcpu *vcpu)
7891 struct vcpu_vmx *vmx = to_vmx(vcpu);
7892 u32 vmx_instruction_info;
7893 unsigned long type, types;
7895 struct x86_exception e;
7901 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7902 SECONDARY_EXEC_ENABLE_VPID) ||
7903 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7904 kvm_queue_exception(vcpu, UD_VECTOR);
7908 if (!nested_vmx_check_permission(vcpu))
7911 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7912 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7914 types = (vmx->nested.nested_vmx_vpid_caps &
7915 VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
7917 if (type >= 32 || !(types & (1 << type))) {
7918 nested_vmx_failValid(vcpu,
7919 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7920 return kvm_skip_emulated_instruction(vcpu);
7923 /* according to the intel vmx instruction reference, the memory
7924 * operand is read even if it isn't needed (e.g., for type==global)
7926 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7927 vmx_instruction_info, false, &gva))
7929 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7930 sizeof(operand), &e)) {
7931 kvm_inject_page_fault(vcpu, &e);
7934 if (operand.vpid >> 16) {
7935 nested_vmx_failValid(vcpu,
7936 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7937 return kvm_skip_emulated_instruction(vcpu);
7941 case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
7942 if (is_noncanonical_address(operand.gla, vcpu)) {
7943 nested_vmx_failValid(vcpu,
7944 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7945 return kvm_skip_emulated_instruction(vcpu);
7948 case VMX_VPID_EXTENT_SINGLE_CONTEXT:
7949 case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
7950 if (!operand.vpid) {
7951 nested_vmx_failValid(vcpu,
7952 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7953 return kvm_skip_emulated_instruction(vcpu);
7956 case VMX_VPID_EXTENT_ALL_CONTEXT:
7960 return kvm_skip_emulated_instruction(vcpu);
7963 __vmx_flush_tlb(vcpu, vmx->nested.vpid02);
7964 nested_vmx_succeed(vcpu);
7966 return kvm_skip_emulated_instruction(vcpu);
7969 static int handle_pml_full(struct kvm_vcpu *vcpu)
7971 unsigned long exit_qualification;
7973 trace_kvm_pml_full(vcpu->vcpu_id);
7975 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7978 * PML buffer FULL happened while executing iret from NMI,
7979 * "blocked by NMI" bit has to be set before next VM entry.
7981 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7982 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7983 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7984 GUEST_INTR_STATE_NMI);
7987 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7988 * here.., and there's no userspace involvement needed for PML.
7993 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
7995 kvm_lapic_expired_hv_timer(vcpu);
7999 static bool valid_ept_address(struct kvm_vcpu *vcpu, u64 address)
8001 struct vcpu_vmx *vmx = to_vmx(vcpu);
8002 int maxphyaddr = cpuid_maxphyaddr(vcpu);
8004 /* Check for memory type validity */
8005 switch (address & VMX_EPTP_MT_MASK) {
8006 case VMX_EPTP_MT_UC:
8007 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPTP_UC_BIT))
8010 case VMX_EPTP_MT_WB:
8011 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPTP_WB_BIT))
8018 /* only 4 levels page-walk length are valid */
8019 if ((address & VMX_EPTP_PWL_MASK) != VMX_EPTP_PWL_4)
8022 /* Reserved bits should not be set */
8023 if (address >> maxphyaddr || ((address >> 7) & 0x1f))
8026 /* AD, if set, should be supported */
8027 if (address & VMX_EPTP_AD_ENABLE_BIT) {
8028 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPT_AD_BIT))
8035 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
8036 struct vmcs12 *vmcs12)
8038 u32 index = vcpu->arch.regs[VCPU_REGS_RCX];
8040 bool accessed_dirty;
8041 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
8043 if (!nested_cpu_has_eptp_switching(vmcs12) ||
8044 !nested_cpu_has_ept(vmcs12))
8047 if (index >= VMFUNC_EPTP_ENTRIES)
8051 if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
8052 &address, index * 8, 8))
8055 accessed_dirty = !!(address & VMX_EPTP_AD_ENABLE_BIT);
8058 * If the (L2) guest does a vmfunc to the currently
8059 * active ept pointer, we don't have to do anything else
8061 if (vmcs12->ept_pointer != address) {
8062 if (!valid_ept_address(vcpu, address))
8065 kvm_mmu_unload(vcpu);
8066 mmu->ept_ad = accessed_dirty;
8067 mmu->base_role.ad_disabled = !accessed_dirty;
8068 vmcs12->ept_pointer = address;
8070 * TODO: Check what's the correct approach in case
8071 * mmu reload fails. Currently, we just let the next
8072 * reload potentially fail
8074 kvm_mmu_reload(vcpu);
8080 static int handle_vmfunc(struct kvm_vcpu *vcpu)
8082 struct vcpu_vmx *vmx = to_vmx(vcpu);
8083 struct vmcs12 *vmcs12;
8084 u32 function = vcpu->arch.regs[VCPU_REGS_RAX];
8087 * VMFUNC is only supported for nested guests, but we always enable the
8088 * secondary control for simplicity; for non-nested mode, fake that we
8089 * didn't by injecting #UD.
8091 if (!is_guest_mode(vcpu)) {
8092 kvm_queue_exception(vcpu, UD_VECTOR);
8096 vmcs12 = get_vmcs12(vcpu);
8097 if ((vmcs12->vm_function_control & (1 << function)) == 0)
8102 if (nested_vmx_eptp_switching(vcpu, vmcs12))
8108 return kvm_skip_emulated_instruction(vcpu);
8111 nested_vmx_vmexit(vcpu, vmx->exit_reason,
8112 vmcs_read32(VM_EXIT_INTR_INFO),
8113 vmcs_readl(EXIT_QUALIFICATION));
8118 * The exit handlers return 1 if the exit was handled fully and guest execution
8119 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
8120 * to be done to userspace and return 0.
8122 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
8123 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
8124 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
8125 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
8126 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
8127 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
8128 [EXIT_REASON_CR_ACCESS] = handle_cr,
8129 [EXIT_REASON_DR_ACCESS] = handle_dr,
8130 [EXIT_REASON_CPUID] = handle_cpuid,
8131 [EXIT_REASON_MSR_READ] = handle_rdmsr,
8132 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
8133 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
8134 [EXIT_REASON_HLT] = handle_halt,
8135 [EXIT_REASON_INVD] = handle_invd,
8136 [EXIT_REASON_INVLPG] = handle_invlpg,
8137 [EXIT_REASON_RDPMC] = handle_rdpmc,
8138 [EXIT_REASON_VMCALL] = handle_vmcall,
8139 [EXIT_REASON_VMCLEAR] = handle_vmclear,
8140 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
8141 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
8142 [EXIT_REASON_VMPTRST] = handle_vmptrst,
8143 [EXIT_REASON_VMREAD] = handle_vmread,
8144 [EXIT_REASON_VMRESUME] = handle_vmresume,
8145 [EXIT_REASON_VMWRITE] = handle_vmwrite,
8146 [EXIT_REASON_VMOFF] = handle_vmoff,
8147 [EXIT_REASON_VMON] = handle_vmon,
8148 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
8149 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
8150 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
8151 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
8152 [EXIT_REASON_WBINVD] = handle_wbinvd,
8153 [EXIT_REASON_XSETBV] = handle_xsetbv,
8154 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
8155 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
8156 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
8157 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
8158 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
8159 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
8160 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
8161 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
8162 [EXIT_REASON_INVEPT] = handle_invept,
8163 [EXIT_REASON_INVVPID] = handle_invvpid,
8164 [EXIT_REASON_RDRAND] = handle_invalid_op,
8165 [EXIT_REASON_RDSEED] = handle_invalid_op,
8166 [EXIT_REASON_XSAVES] = handle_xsaves,
8167 [EXIT_REASON_XRSTORS] = handle_xrstors,
8168 [EXIT_REASON_PML_FULL] = handle_pml_full,
8169 [EXIT_REASON_VMFUNC] = handle_vmfunc,
8170 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
8173 static const int kvm_vmx_max_exit_handlers =
8174 ARRAY_SIZE(kvm_vmx_exit_handlers);
8176 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
8177 struct vmcs12 *vmcs12)
8179 unsigned long exit_qualification;
8180 gpa_t bitmap, last_bitmap;
8185 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
8186 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
8188 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8190 port = exit_qualification >> 16;
8191 size = (exit_qualification & 7) + 1;
8193 last_bitmap = (gpa_t)-1;
8198 bitmap = vmcs12->io_bitmap_a;
8199 else if (port < 0x10000)
8200 bitmap = vmcs12->io_bitmap_b;
8203 bitmap += (port & 0x7fff) / 8;
8205 if (last_bitmap != bitmap)
8206 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
8208 if (b & (1 << (port & 7)))
8213 last_bitmap = bitmap;
8220 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
8221 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
8222 * disinterest in the current event (read or write a specific MSR) by using an
8223 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
8225 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
8226 struct vmcs12 *vmcs12, u32 exit_reason)
8228 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
8231 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
8235 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
8236 * for the four combinations of read/write and low/high MSR numbers.
8237 * First we need to figure out which of the four to use:
8239 bitmap = vmcs12->msr_bitmap;
8240 if (exit_reason == EXIT_REASON_MSR_WRITE)
8242 if (msr_index >= 0xc0000000) {
8243 msr_index -= 0xc0000000;
8247 /* Then read the msr_index'th bit from this bitmap: */
8248 if (msr_index < 1024*8) {
8250 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
8252 return 1 & (b >> (msr_index & 7));
8254 return true; /* let L1 handle the wrong parameter */
8258 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
8259 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
8260 * intercept (via guest_host_mask etc.) the current event.
8262 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
8263 struct vmcs12 *vmcs12)
8265 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8266 int cr = exit_qualification & 15;
8270 switch ((exit_qualification >> 4) & 3) {
8271 case 0: /* mov to cr */
8272 reg = (exit_qualification >> 8) & 15;
8273 val = kvm_register_readl(vcpu, reg);
8276 if (vmcs12->cr0_guest_host_mask &
8277 (val ^ vmcs12->cr0_read_shadow))
8281 if ((vmcs12->cr3_target_count >= 1 &&
8282 vmcs12->cr3_target_value0 == val) ||
8283 (vmcs12->cr3_target_count >= 2 &&
8284 vmcs12->cr3_target_value1 == val) ||
8285 (vmcs12->cr3_target_count >= 3 &&
8286 vmcs12->cr3_target_value2 == val) ||
8287 (vmcs12->cr3_target_count >= 4 &&
8288 vmcs12->cr3_target_value3 == val))
8290 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
8294 if (vmcs12->cr4_guest_host_mask &
8295 (vmcs12->cr4_read_shadow ^ val))
8299 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
8305 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
8306 (vmcs12->cr0_read_shadow & X86_CR0_TS))
8309 case 1: /* mov from cr */
8312 if (vmcs12->cpu_based_vm_exec_control &
8313 CPU_BASED_CR3_STORE_EXITING)
8317 if (vmcs12->cpu_based_vm_exec_control &
8318 CPU_BASED_CR8_STORE_EXITING)
8325 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
8326 * cr0. Other attempted changes are ignored, with no exit.
8328 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
8329 if (vmcs12->cr0_guest_host_mask & 0xe &
8330 (val ^ vmcs12->cr0_read_shadow))
8332 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
8333 !(vmcs12->cr0_read_shadow & 0x1) &&
8342 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
8343 * should handle it ourselves in L0 (and then continue L2). Only call this
8344 * when in is_guest_mode (L2).
8346 static bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
8348 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8349 struct vcpu_vmx *vmx = to_vmx(vcpu);
8350 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8352 if (vmx->nested.nested_run_pending)
8355 if (unlikely(vmx->fail)) {
8356 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
8357 vmcs_read32(VM_INSTRUCTION_ERROR));
8362 * The host physical addresses of some pages of guest memory
8363 * are loaded into VMCS02 (e.g. L1's Virtual APIC Page). The CPU
8364 * may write to these pages via their host physical address while
8365 * L2 is running, bypassing any address-translation-based dirty
8366 * tracking (e.g. EPT write protection).
8368 * Mark them dirty on every exit from L2 to prevent them from
8369 * getting out of sync with dirty tracking.
8371 nested_mark_vmcs12_pages_dirty(vcpu);
8373 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
8374 vmcs_readl(EXIT_QUALIFICATION),
8375 vmx->idt_vectoring_info,
8377 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8380 switch (exit_reason) {
8381 case EXIT_REASON_EXCEPTION_NMI:
8382 if (is_nmi(intr_info))
8384 else if (is_page_fault(intr_info))
8385 return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
8386 else if (is_no_device(intr_info) &&
8387 !(vmcs12->guest_cr0 & X86_CR0_TS))
8389 else if (is_debug(intr_info) &&
8391 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
8393 else if (is_breakpoint(intr_info) &&
8394 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
8396 return vmcs12->exception_bitmap &
8397 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
8398 case EXIT_REASON_EXTERNAL_INTERRUPT:
8400 case EXIT_REASON_TRIPLE_FAULT:
8402 case EXIT_REASON_PENDING_INTERRUPT:
8403 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
8404 case EXIT_REASON_NMI_WINDOW:
8405 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
8406 case EXIT_REASON_TASK_SWITCH:
8408 case EXIT_REASON_CPUID:
8410 case EXIT_REASON_HLT:
8411 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
8412 case EXIT_REASON_INVD:
8414 case EXIT_REASON_INVLPG:
8415 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8416 case EXIT_REASON_RDPMC:
8417 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
8418 case EXIT_REASON_RDRAND:
8419 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND);
8420 case EXIT_REASON_RDSEED:
8421 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED);
8422 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
8423 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
8424 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
8425 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
8426 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
8427 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
8428 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
8429 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
8431 * VMX instructions trap unconditionally. This allows L1 to
8432 * emulate them for its L2 guest, i.e., allows 3-level nesting!
8435 case EXIT_REASON_CR_ACCESS:
8436 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
8437 case EXIT_REASON_DR_ACCESS:
8438 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
8439 case EXIT_REASON_IO_INSTRUCTION:
8440 return nested_vmx_exit_handled_io(vcpu, vmcs12);
8441 case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
8442 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
8443 case EXIT_REASON_MSR_READ:
8444 case EXIT_REASON_MSR_WRITE:
8445 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
8446 case EXIT_REASON_INVALID_STATE:
8448 case EXIT_REASON_MWAIT_INSTRUCTION:
8449 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
8450 case EXIT_REASON_MONITOR_TRAP_FLAG:
8451 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
8452 case EXIT_REASON_MONITOR_INSTRUCTION:
8453 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
8454 case EXIT_REASON_PAUSE_INSTRUCTION:
8455 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
8456 nested_cpu_has2(vmcs12,
8457 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
8458 case EXIT_REASON_MCE_DURING_VMENTRY:
8460 case EXIT_REASON_TPR_BELOW_THRESHOLD:
8461 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
8462 case EXIT_REASON_APIC_ACCESS:
8463 return nested_cpu_has2(vmcs12,
8464 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
8465 case EXIT_REASON_APIC_WRITE:
8466 case EXIT_REASON_EOI_INDUCED:
8467 /* apic_write and eoi_induced should exit unconditionally. */
8469 case EXIT_REASON_EPT_VIOLATION:
8471 * L0 always deals with the EPT violation. If nested EPT is
8472 * used, and the nested mmu code discovers that the address is
8473 * missing in the guest EPT table (EPT12), the EPT violation
8474 * will be injected with nested_ept_inject_page_fault()
8477 case EXIT_REASON_EPT_MISCONFIG:
8479 * L2 never uses directly L1's EPT, but rather L0's own EPT
8480 * table (shadow on EPT) or a merged EPT table that L0 built
8481 * (EPT on EPT). So any problems with the structure of the
8482 * table is L0's fault.
8485 case EXIT_REASON_INVPCID:
8487 nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
8488 nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8489 case EXIT_REASON_WBINVD:
8490 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
8491 case EXIT_REASON_XSETBV:
8493 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
8495 * This should never happen, since it is not possible to
8496 * set XSS to a non-zero value---neither in L1 nor in L2.
8497 * If if it were, XSS would have to be checked against
8498 * the XSS exit bitmap in vmcs12.
8500 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
8501 case EXIT_REASON_PREEMPTION_TIMER:
8503 case EXIT_REASON_PML_FULL:
8504 /* We emulate PML support to L1. */
8506 case EXIT_REASON_VMFUNC:
8507 /* VM functions are emulated through L2->L0 vmexits. */
8514 static int nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason)
8516 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8519 * At this point, the exit interruption info in exit_intr_info
8520 * is only valid for EXCEPTION_NMI exits. For EXTERNAL_INTERRUPT
8521 * we need to query the in-kernel LAPIC.
8523 WARN_ON(exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT);
8524 if ((exit_intr_info &
8525 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
8526 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) {
8527 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8528 vmcs12->vm_exit_intr_error_code =
8529 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
8532 nested_vmx_vmexit(vcpu, exit_reason, exit_intr_info,
8533 vmcs_readl(EXIT_QUALIFICATION));
8537 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
8539 *info1 = vmcs_readl(EXIT_QUALIFICATION);
8540 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
8543 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
8546 __free_page(vmx->pml_pg);
8551 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
8553 struct vcpu_vmx *vmx = to_vmx(vcpu);
8557 pml_idx = vmcs_read16(GUEST_PML_INDEX);
8559 /* Do nothing if PML buffer is empty */
8560 if (pml_idx == (PML_ENTITY_NUM - 1))
8563 /* PML index always points to next available PML buffer entity */
8564 if (pml_idx >= PML_ENTITY_NUM)
8569 pml_buf = page_address(vmx->pml_pg);
8570 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
8573 gpa = pml_buf[pml_idx];
8574 WARN_ON(gpa & (PAGE_SIZE - 1));
8575 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
8578 /* reset PML index */
8579 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8583 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8584 * Called before reporting dirty_bitmap to userspace.
8586 static void kvm_flush_pml_buffers(struct kvm *kvm)
8589 struct kvm_vcpu *vcpu;
8591 * We only need to kick vcpu out of guest mode here, as PML buffer
8592 * is flushed at beginning of all VMEXITs, and it's obvious that only
8593 * vcpus running in guest are possible to have unflushed GPAs in PML
8596 kvm_for_each_vcpu(i, vcpu, kvm)
8597 kvm_vcpu_kick(vcpu);
8600 static void vmx_dump_sel(char *name, uint32_t sel)
8602 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8603 name, vmcs_read16(sel),
8604 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
8605 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
8606 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
8609 static void vmx_dump_dtsel(char *name, uint32_t limit)
8611 pr_err("%s limit=0x%08x, base=0x%016lx\n",
8612 name, vmcs_read32(limit),
8613 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
8616 static void dump_vmcs(void)
8618 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
8619 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
8620 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
8621 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
8622 u32 secondary_exec_control = 0;
8623 unsigned long cr4 = vmcs_readl(GUEST_CR4);
8624 u64 efer = vmcs_read64(GUEST_IA32_EFER);
8627 if (cpu_has_secondary_exec_ctrls())
8628 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8630 pr_err("*** Guest State ***\n");
8631 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8632 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
8633 vmcs_readl(CR0_GUEST_HOST_MASK));
8634 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8635 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
8636 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
8637 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
8638 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
8640 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
8641 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
8642 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
8643 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
8645 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
8646 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
8647 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
8648 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
8649 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8650 vmcs_readl(GUEST_SYSENTER_ESP),
8651 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8652 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
8653 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
8654 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
8655 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
8656 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
8657 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
8658 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8659 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8660 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8661 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
8662 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8663 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8664 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8665 efer, vmcs_read64(GUEST_IA32_PAT));
8666 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
8667 vmcs_read64(GUEST_IA32_DEBUGCTL),
8668 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8669 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8670 pr_err("PerfGlobCtl = 0x%016llx\n",
8671 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8672 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8673 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8674 pr_err("Interruptibility = %08x ActivityState = %08x\n",
8675 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8676 vmcs_read32(GUEST_ACTIVITY_STATE));
8677 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8678 pr_err("InterruptStatus = %04x\n",
8679 vmcs_read16(GUEST_INTR_STATUS));
8681 pr_err("*** Host State ***\n");
8682 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
8683 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8684 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8685 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8686 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8687 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8688 vmcs_read16(HOST_TR_SELECTOR));
8689 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8690 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8691 vmcs_readl(HOST_TR_BASE));
8692 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8693 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8694 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8695 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8696 vmcs_readl(HOST_CR4));
8697 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8698 vmcs_readl(HOST_IA32_SYSENTER_ESP),
8699 vmcs_read32(HOST_IA32_SYSENTER_CS),
8700 vmcs_readl(HOST_IA32_SYSENTER_EIP));
8701 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8702 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8703 vmcs_read64(HOST_IA32_EFER),
8704 vmcs_read64(HOST_IA32_PAT));
8705 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8706 pr_err("PerfGlobCtl = 0x%016llx\n",
8707 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8709 pr_err("*** Control State ***\n");
8710 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8711 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8712 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8713 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8714 vmcs_read32(EXCEPTION_BITMAP),
8715 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8716 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8717 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8718 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8719 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8720 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8721 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8722 vmcs_read32(VM_EXIT_INTR_INFO),
8723 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8724 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8725 pr_err(" reason=%08x qualification=%016lx\n",
8726 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8727 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8728 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8729 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8730 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8731 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8732 pr_err("TSC Multiplier = 0x%016llx\n",
8733 vmcs_read64(TSC_MULTIPLIER));
8734 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8735 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8736 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8737 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8738 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8739 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8740 n = vmcs_read32(CR3_TARGET_COUNT);
8741 for (i = 0; i + 1 < n; i += 4)
8742 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8743 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8744 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8746 pr_err("CR3 target%u=%016lx\n",
8747 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8748 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8749 pr_err("PLE Gap=%08x Window=%08x\n",
8750 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8751 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8752 pr_err("Virtual processor ID = 0x%04x\n",
8753 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8757 * The guest has exited. See if we can fix it or if we need userspace
8760 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8762 struct vcpu_vmx *vmx = to_vmx(vcpu);
8763 u32 exit_reason = vmx->exit_reason;
8764 u32 vectoring_info = vmx->idt_vectoring_info;
8766 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8769 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8770 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8771 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8772 * mode as if vcpus is in root mode, the PML buffer must has been
8776 vmx_flush_pml_buffer(vcpu);
8778 /* If guest state is invalid, start emulating */
8779 if (vmx->emulation_required)
8780 return handle_invalid_guest_state(vcpu);
8782 if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
8783 return nested_vmx_reflect_vmexit(vcpu, exit_reason);
8785 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8787 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8788 vcpu->run->fail_entry.hardware_entry_failure_reason
8793 if (unlikely(vmx->fail)) {
8794 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8795 vcpu->run->fail_entry.hardware_entry_failure_reason
8796 = vmcs_read32(VM_INSTRUCTION_ERROR);
8802 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8803 * delivery event since it indicates guest is accessing MMIO.
8804 * The vm-exit can be triggered again after return to guest that
8805 * will cause infinite loop.
8807 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8808 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8809 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8810 exit_reason != EXIT_REASON_PML_FULL &&
8811 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8812 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8813 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8814 vcpu->run->internal.ndata = 3;
8815 vcpu->run->internal.data[0] = vectoring_info;
8816 vcpu->run->internal.data[1] = exit_reason;
8817 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
8818 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
8819 vcpu->run->internal.ndata++;
8820 vcpu->run->internal.data[3] =
8821 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
8826 if (exit_reason < kvm_vmx_max_exit_handlers
8827 && kvm_vmx_exit_handlers[exit_reason])
8828 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8830 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
8832 kvm_queue_exception(vcpu, UD_VECTOR);
8837 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8839 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8841 if (is_guest_mode(vcpu) &&
8842 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8845 if (irr == -1 || tpr < irr) {
8846 vmcs_write32(TPR_THRESHOLD, 0);
8850 vmcs_write32(TPR_THRESHOLD, irr);
8853 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8855 u32 sec_exec_control;
8857 /* Postpone execution until vmcs01 is the current VMCS. */
8858 if (is_guest_mode(vcpu)) {
8859 to_vmx(vcpu)->nested.change_vmcs01_virtual_x2apic_mode = true;
8863 if (!cpu_has_vmx_virtualize_x2apic_mode())
8866 if (!cpu_need_tpr_shadow(vcpu))
8869 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8872 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8873 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8875 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8876 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8877 vmx_flush_tlb_ept_only(vcpu);
8879 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8881 vmx_set_msr_bitmap(vcpu);
8884 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8886 struct vcpu_vmx *vmx = to_vmx(vcpu);
8889 * Currently we do not handle the nested case where L2 has an
8890 * APIC access page of its own; that page is still pinned.
8891 * Hence, we skip the case where the VCPU is in guest mode _and_
8892 * L1 prepared an APIC access page for L2.
8894 * For the case where L1 and L2 share the same APIC access page
8895 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8896 * in the vmcs12), this function will only update either the vmcs01
8897 * or the vmcs02. If the former, the vmcs02 will be updated by
8898 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8899 * the next L2->L1 exit.
8901 if (!is_guest_mode(vcpu) ||
8902 !nested_cpu_has2(get_vmcs12(&vmx->vcpu),
8903 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
8904 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8905 vmx_flush_tlb_ept_only(vcpu);
8909 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
8917 status = vmcs_read16(GUEST_INTR_STATUS);
8919 if (max_isr != old) {
8921 status |= max_isr << 8;
8922 vmcs_write16(GUEST_INTR_STATUS, status);
8926 static void vmx_set_rvi(int vector)
8934 status = vmcs_read16(GUEST_INTR_STATUS);
8935 old = (u8)status & 0xff;
8936 if ((u8)vector != old) {
8938 status |= (u8)vector;
8939 vmcs_write16(GUEST_INTR_STATUS, status);
8943 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8945 if (!is_guest_mode(vcpu)) {
8946 vmx_set_rvi(max_irr);
8954 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8957 if (nested_exit_on_intr(vcpu))
8961 * Else, fall back to pre-APICv interrupt injection since L2
8962 * is run without virtual interrupt delivery.
8964 if (!kvm_event_needs_reinjection(vcpu) &&
8965 vmx_interrupt_allowed(vcpu)) {
8966 kvm_queue_interrupt(vcpu, max_irr, false);
8967 vmx_inject_irq(vcpu);
8971 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
8973 struct vcpu_vmx *vmx = to_vmx(vcpu);
8976 WARN_ON(!vcpu->arch.apicv_active);
8977 if (pi_test_on(&vmx->pi_desc)) {
8978 pi_clear_on(&vmx->pi_desc);
8980 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
8981 * But on x86 this is just a compiler barrier anyway.
8983 smp_mb__after_atomic();
8984 max_irr = kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
8986 max_irr = kvm_lapic_find_highest_irr(vcpu);
8988 vmx_hwapic_irr_update(vcpu, max_irr);
8992 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
8994 if (!kvm_vcpu_apicv_active(vcpu))
8997 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8998 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8999 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
9000 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
9003 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
9005 struct vcpu_vmx *vmx = to_vmx(vcpu);
9007 pi_clear_on(&vmx->pi_desc);
9008 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
9011 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
9013 u32 exit_intr_info = 0;
9014 u16 basic_exit_reason = (u16)vmx->exit_reason;
9016 if (!(basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
9017 || basic_exit_reason == EXIT_REASON_EXCEPTION_NMI))
9020 if (!(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
9021 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9022 vmx->exit_intr_info = exit_intr_info;
9024 /* if exit due to PF check for async PF */
9025 if (is_page_fault(exit_intr_info))
9026 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
9028 /* Handle machine checks before interrupts are enabled */
9029 if (basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY ||
9030 is_machine_check(exit_intr_info))
9031 kvm_machine_check();
9033 /* We need to handle NMIs before interrupts are enabled */
9034 if (is_nmi(exit_intr_info)) {
9035 kvm_before_handle_nmi(&vmx->vcpu);
9037 kvm_after_handle_nmi(&vmx->vcpu);
9041 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
9043 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9045 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
9046 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
9047 unsigned int vector;
9048 unsigned long entry;
9050 struct vcpu_vmx *vmx = to_vmx(vcpu);
9051 #ifdef CONFIG_X86_64
9055 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
9056 desc = (gate_desc *)vmx->host_idt_base + vector;
9057 entry = gate_offset(desc);
9059 #ifdef CONFIG_X86_64
9060 "mov %%" _ASM_SP ", %[sp]\n\t"
9061 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
9066 __ASM_SIZE(push) " $%c[cs]\n\t"
9067 "call *%[entry]\n\t"
9069 #ifdef CONFIG_X86_64
9075 [ss]"i"(__KERNEL_DS),
9076 [cs]"i"(__KERNEL_CS)
9080 STACK_FRAME_NON_STANDARD(vmx_handle_external_intr);
9082 static bool vmx_has_high_real_mode_segbase(void)
9084 return enable_unrestricted_guest || emulate_invalid_guest_state;
9087 static bool vmx_mpx_supported(void)
9089 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
9090 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
9093 static bool vmx_xsaves_supported(void)
9095 return vmcs_config.cpu_based_2nd_exec_ctrl &
9096 SECONDARY_EXEC_XSAVES;
9099 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
9104 bool idtv_info_valid;
9106 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
9108 if (vmx->loaded_vmcs->nmi_known_unmasked)
9111 * Can't use vmx->exit_intr_info since we're not sure what
9112 * the exit reason is.
9114 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9115 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
9116 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
9118 * SDM 3: 27.7.1.2 (September 2008)
9119 * Re-set bit "block by NMI" before VM entry if vmexit caused by
9120 * a guest IRET fault.
9121 * SDM 3: 23.2.2 (September 2008)
9122 * Bit 12 is undefined in any of the following cases:
9123 * If the VM exit sets the valid bit in the IDT-vectoring
9124 * information field.
9125 * If the VM exit is due to a double fault.
9127 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
9128 vector != DF_VECTOR && !idtv_info_valid)
9129 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
9130 GUEST_INTR_STATE_NMI);
9132 vmx->loaded_vmcs->nmi_known_unmasked =
9133 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
9134 & GUEST_INTR_STATE_NMI);
9137 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
9138 u32 idt_vectoring_info,
9139 int instr_len_field,
9140 int error_code_field)
9144 bool idtv_info_valid;
9146 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
9148 vcpu->arch.nmi_injected = false;
9149 kvm_clear_exception_queue(vcpu);
9150 kvm_clear_interrupt_queue(vcpu);
9152 if (!idtv_info_valid)
9155 kvm_make_request(KVM_REQ_EVENT, vcpu);
9157 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
9158 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
9161 case INTR_TYPE_NMI_INTR:
9162 vcpu->arch.nmi_injected = true;
9164 * SDM 3: 27.7.1.2 (September 2008)
9165 * Clear bit "block by NMI" before VM entry if a NMI
9168 vmx_set_nmi_mask(vcpu, false);
9170 case INTR_TYPE_SOFT_EXCEPTION:
9171 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
9173 case INTR_TYPE_HARD_EXCEPTION:
9174 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
9175 u32 err = vmcs_read32(error_code_field);
9176 kvm_requeue_exception_e(vcpu, vector, err);
9178 kvm_requeue_exception(vcpu, vector);
9180 case INTR_TYPE_SOFT_INTR:
9181 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
9183 case INTR_TYPE_EXT_INTR:
9184 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
9191 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
9193 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
9194 VM_EXIT_INSTRUCTION_LEN,
9195 IDT_VECTORING_ERROR_CODE);
9198 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
9200 __vmx_complete_interrupts(vcpu,
9201 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
9202 VM_ENTRY_INSTRUCTION_LEN,
9203 VM_ENTRY_EXCEPTION_ERROR_CODE);
9205 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
9208 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
9211 struct perf_guest_switch_msr *msrs;
9213 msrs = perf_guest_get_msrs(&nr_msrs);
9218 for (i = 0; i < nr_msrs; i++)
9219 if (msrs[i].host == msrs[i].guest)
9220 clear_atomic_switch_msr(vmx, msrs[i].msr);
9222 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
9226 static void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
9228 struct vcpu_vmx *vmx = to_vmx(vcpu);
9232 if (vmx->hv_deadline_tsc == -1)
9236 if (vmx->hv_deadline_tsc > tscl)
9237 /* sure to be 32 bit only because checked on set_hv_timer */
9238 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
9239 cpu_preemption_timer_multi);
9243 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
9246 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
9248 struct vcpu_vmx *vmx = to_vmx(vcpu);
9249 unsigned long debugctlmsr, cr3, cr4;
9251 /* Don't enter VMX if guest state is invalid, let the exit handler
9252 start emulation until we arrive back to a valid state */
9253 if (vmx->emulation_required)
9256 if (vmx->ple_window_dirty) {
9257 vmx->ple_window_dirty = false;
9258 vmcs_write32(PLE_WINDOW, vmx->ple_window);
9261 if (vmx->nested.sync_shadow_vmcs) {
9262 copy_vmcs12_to_shadow(vmx);
9263 vmx->nested.sync_shadow_vmcs = false;
9266 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
9267 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
9268 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
9269 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
9271 cr3 = __get_current_cr3_fast();
9272 if (unlikely(cr3 != vmx->loaded_vmcs->vmcs_host_cr3)) {
9273 vmcs_writel(HOST_CR3, cr3);
9274 vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
9277 cr4 = cr4_read_shadow();
9278 if (unlikely(cr4 != vmx->loaded_vmcs->vmcs_host_cr4)) {
9279 vmcs_writel(HOST_CR4, cr4);
9280 vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
9283 /* When single-stepping over STI and MOV SS, we must clear the
9284 * corresponding interruptibility bits in the guest state. Otherwise
9285 * vmentry fails as it then expects bit 14 (BS) in pending debug
9286 * exceptions being set, but that's not correct for the guest debugging
9288 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9289 vmx_set_interrupt_shadow(vcpu, 0);
9291 if (static_cpu_has(X86_FEATURE_PKU) &&
9292 kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
9293 vcpu->arch.pkru != vmx->host_pkru)
9294 __write_pkru(vcpu->arch.pkru);
9296 atomic_switch_perf_msrs(vmx);
9297 debugctlmsr = get_debugctlmsr();
9299 vmx_arm_hv_timer(vcpu);
9301 vmx->__launched = vmx->loaded_vmcs->launched;
9303 /* Store host registers */
9304 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
9305 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
9306 "push %%" _ASM_CX " \n\t"
9307 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9309 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9310 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
9312 /* Reload cr2 if changed */
9313 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
9314 "mov %%cr2, %%" _ASM_DX " \n\t"
9315 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
9317 "mov %%" _ASM_AX", %%cr2 \n\t"
9319 /* Check if vmlaunch of vmresume is needed */
9320 "cmpl $0, %c[launched](%0) \n\t"
9321 /* Load guest registers. Don't clobber flags. */
9322 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
9323 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
9324 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
9325 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
9326 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
9327 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
9328 #ifdef CONFIG_X86_64
9329 "mov %c[r8](%0), %%r8 \n\t"
9330 "mov %c[r9](%0), %%r9 \n\t"
9331 "mov %c[r10](%0), %%r10 \n\t"
9332 "mov %c[r11](%0), %%r11 \n\t"
9333 "mov %c[r12](%0), %%r12 \n\t"
9334 "mov %c[r13](%0), %%r13 \n\t"
9335 "mov %c[r14](%0), %%r14 \n\t"
9336 "mov %c[r15](%0), %%r15 \n\t"
9338 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
9340 /* Enter guest mode */
9342 __ex(ASM_VMX_VMLAUNCH) "\n\t"
9344 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
9346 /* Save guest registers, load host registers, keep flags */
9347 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
9349 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
9350 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
9351 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
9352 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
9353 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
9354 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
9355 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
9356 #ifdef CONFIG_X86_64
9357 "mov %%r8, %c[r8](%0) \n\t"
9358 "mov %%r9, %c[r9](%0) \n\t"
9359 "mov %%r10, %c[r10](%0) \n\t"
9360 "mov %%r11, %c[r11](%0) \n\t"
9361 "mov %%r12, %c[r12](%0) \n\t"
9362 "mov %%r13, %c[r13](%0) \n\t"
9363 "mov %%r14, %c[r14](%0) \n\t"
9364 "mov %%r15, %c[r15](%0) \n\t"
9366 "mov %%cr2, %%" _ASM_AX " \n\t"
9367 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
9369 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
9370 "setbe %c[fail](%0) \n\t"
9371 ".pushsection .rodata \n\t"
9372 ".global vmx_return \n\t"
9373 "vmx_return: " _ASM_PTR " 2b \n\t"
9375 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
9376 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
9377 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
9378 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
9379 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
9380 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
9381 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
9382 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
9383 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
9384 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
9385 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
9386 #ifdef CONFIG_X86_64
9387 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
9388 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
9389 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
9390 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
9391 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
9392 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
9393 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
9394 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
9396 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
9397 [wordsize]"i"(sizeof(ulong))
9399 #ifdef CONFIG_X86_64
9400 , "rax", "rbx", "rdi", "rsi"
9401 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
9403 , "eax", "ebx", "edi", "esi"
9407 /* Eliminate branch target predictions from guest mode */
9410 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
9412 update_debugctlmsr(debugctlmsr);
9414 #ifndef CONFIG_X86_64
9416 * The sysexit path does not restore ds/es, so we must set them to
9417 * a reasonable value ourselves.
9419 * We can't defer this to vmx_load_host_state() since that function
9420 * may be executed in interrupt context, which saves and restore segments
9421 * around it, nullifying its effect.
9423 loadsegment(ds, __USER_DS);
9424 loadsegment(es, __USER_DS);
9427 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
9428 | (1 << VCPU_EXREG_RFLAGS)
9429 | (1 << VCPU_EXREG_PDPTR)
9430 | (1 << VCPU_EXREG_SEGMENTS)
9431 | (1 << VCPU_EXREG_CR3));
9432 vcpu->arch.regs_dirty = 0;
9435 * eager fpu is enabled if PKEY is supported and CR4 is switched
9436 * back on host, so it is safe to read guest PKRU from current
9439 if (static_cpu_has(X86_FEATURE_PKU) &&
9440 kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
9441 vcpu->arch.pkru = __read_pkru();
9442 if (vcpu->arch.pkru != vmx->host_pkru)
9443 __write_pkru(vmx->host_pkru);
9447 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
9448 * we did not inject a still-pending event to L1 now because of
9449 * nested_run_pending, we need to re-enable this bit.
9451 if (vmx->nested.nested_run_pending)
9452 kvm_make_request(KVM_REQ_EVENT, vcpu);
9454 vmx->nested.nested_run_pending = 0;
9455 vmx->idt_vectoring_info = 0;
9457 vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
9458 if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
9461 vmx->loaded_vmcs->launched = 1;
9462 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9464 vmx_complete_atomic_exit(vmx);
9465 vmx_recover_nmi_blocking(vmx);
9466 vmx_complete_interrupts(vmx);
9468 STACK_FRAME_NON_STANDARD(vmx_vcpu_run);
9470 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
9472 struct vcpu_vmx *vmx = to_vmx(vcpu);
9475 if (vmx->loaded_vmcs == vmcs)
9479 vmx->loaded_vmcs = vmcs;
9481 vmx_vcpu_load(vcpu, cpu);
9487 * Ensure that the current vmcs of the logical processor is the
9488 * vmcs01 of the vcpu before calling free_nested().
9490 static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
9492 struct vcpu_vmx *vmx = to_vmx(vcpu);
9495 r = vcpu_load(vcpu);
9497 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
9502 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
9504 struct vcpu_vmx *vmx = to_vmx(vcpu);
9507 vmx_destroy_pml_buffer(vmx);
9508 free_vpid(vmx->vpid);
9509 leave_guest_mode(vcpu);
9510 vmx_free_vcpu_nested(vcpu);
9511 free_loaded_vmcs(vmx->loaded_vmcs);
9512 kfree(vmx->guest_msrs);
9513 kvm_vcpu_uninit(vcpu);
9514 kmem_cache_free(kvm_vcpu_cache, vmx);
9517 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
9520 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
9524 return ERR_PTR(-ENOMEM);
9526 vmx->vpid = allocate_vpid();
9528 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
9535 * If PML is turned on, failure on enabling PML just results in failure
9536 * of creating the vcpu, therefore we can simplify PML logic (by
9537 * avoiding dealing with cases, such as enabling PML partially on vcpus
9538 * for the guest, etc.
9541 vmx->pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
9546 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
9547 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
9550 if (!vmx->guest_msrs)
9553 vmx->loaded_vmcs = &vmx->vmcs01;
9554 vmx->loaded_vmcs->vmcs = alloc_vmcs();
9555 vmx->loaded_vmcs->shadow_vmcs = NULL;
9556 if (!vmx->loaded_vmcs->vmcs)
9558 loaded_vmcs_init(vmx->loaded_vmcs);
9561 vmx_vcpu_load(&vmx->vcpu, cpu);
9562 vmx->vcpu.cpu = cpu;
9563 err = vmx_vcpu_setup(vmx);
9564 vmx_vcpu_put(&vmx->vcpu);
9568 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9569 err = alloc_apic_access_page(kvm);
9575 if (!kvm->arch.ept_identity_map_addr)
9576 kvm->arch.ept_identity_map_addr =
9577 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
9578 err = init_rmode_identity_map(kvm);
9584 nested_vmx_setup_ctls_msrs(vmx);
9585 vmx->nested.vpid02 = allocate_vpid();
9588 vmx->nested.posted_intr_nv = -1;
9589 vmx->nested.current_vmptr = -1ull;
9591 vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
9594 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
9595 * or POSTED_INTR_WAKEUP_VECTOR.
9597 vmx->pi_desc.nv = POSTED_INTR_VECTOR;
9598 vmx->pi_desc.sn = 1;
9603 free_vpid(vmx->nested.vpid02);
9604 free_loaded_vmcs(vmx->loaded_vmcs);
9606 kfree(vmx->guest_msrs);
9608 vmx_destroy_pml_buffer(vmx);
9610 kvm_vcpu_uninit(&vmx->vcpu);
9612 free_vpid(vmx->vpid);
9613 kmem_cache_free(kvm_vcpu_cache, vmx);
9614 return ERR_PTR(err);
9617 static void __init vmx_check_processor_compat(void *rtn)
9619 struct vmcs_config vmcs_conf;
9622 if (setup_vmcs_config(&vmcs_conf) < 0)
9624 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
9625 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
9626 smp_processor_id());
9631 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
9636 /* For VT-d and EPT combination
9637 * 1. MMIO: always map as UC
9639 * a. VT-d without snooping control feature: can't guarantee the
9640 * result, try to trust guest.
9641 * b. VT-d with snooping control feature: snooping control feature of
9642 * VT-d engine can guarantee the cache correctness. Just set it
9643 * to WB to keep consistent with host. So the same as item 3.
9644 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9645 * consistent with host MTRR
9648 cache = MTRR_TYPE_UNCACHABLE;
9652 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
9653 ipat = VMX_EPT_IPAT_BIT;
9654 cache = MTRR_TYPE_WRBACK;
9658 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
9659 ipat = VMX_EPT_IPAT_BIT;
9660 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
9661 cache = MTRR_TYPE_WRBACK;
9663 cache = MTRR_TYPE_UNCACHABLE;
9667 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
9670 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
9673 static int vmx_get_lpage_level(void)
9675 if (enable_ept && !cpu_has_vmx_ept_1g_page())
9676 return PT_DIRECTORY_LEVEL;
9678 /* For shadow and EPT supported 1GB page */
9679 return PT_PDPE_LEVEL;
9682 static void vmcs_set_secondary_exec_control(u32 new_ctl)
9685 * These bits in the secondary execution controls field
9686 * are dynamic, the others are mostly based on the hypervisor
9687 * architecture and the guest's CPUID. Do not touch the
9691 SECONDARY_EXEC_SHADOW_VMCS |
9692 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
9693 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9695 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9697 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
9698 (new_ctl & ~mask) | (cur_ctl & mask));
9702 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
9703 * (indicating "allowed-1") if they are supported in the guest's CPUID.
9705 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
9707 struct vcpu_vmx *vmx = to_vmx(vcpu);
9708 struct kvm_cpuid_entry2 *entry;
9710 vmx->nested.nested_vmx_cr0_fixed1 = 0xffffffff;
9711 vmx->nested.nested_vmx_cr4_fixed1 = X86_CR4_PCE;
9713 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
9714 if (entry && (entry->_reg & (_cpuid_mask))) \
9715 vmx->nested.nested_vmx_cr4_fixed1 |= (_cr4_mask); \
9718 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
9719 cr4_fixed1_update(X86_CR4_VME, edx, bit(X86_FEATURE_VME));
9720 cr4_fixed1_update(X86_CR4_PVI, edx, bit(X86_FEATURE_VME));
9721 cr4_fixed1_update(X86_CR4_TSD, edx, bit(X86_FEATURE_TSC));
9722 cr4_fixed1_update(X86_CR4_DE, edx, bit(X86_FEATURE_DE));
9723 cr4_fixed1_update(X86_CR4_PSE, edx, bit(X86_FEATURE_PSE));
9724 cr4_fixed1_update(X86_CR4_PAE, edx, bit(X86_FEATURE_PAE));
9725 cr4_fixed1_update(X86_CR4_MCE, edx, bit(X86_FEATURE_MCE));
9726 cr4_fixed1_update(X86_CR4_PGE, edx, bit(X86_FEATURE_PGE));
9727 cr4_fixed1_update(X86_CR4_OSFXSR, edx, bit(X86_FEATURE_FXSR));
9728 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
9729 cr4_fixed1_update(X86_CR4_VMXE, ecx, bit(X86_FEATURE_VMX));
9730 cr4_fixed1_update(X86_CR4_SMXE, ecx, bit(X86_FEATURE_SMX));
9731 cr4_fixed1_update(X86_CR4_PCIDE, ecx, bit(X86_FEATURE_PCID));
9732 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, bit(X86_FEATURE_XSAVE));
9734 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9735 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, bit(X86_FEATURE_FSGSBASE));
9736 cr4_fixed1_update(X86_CR4_SMEP, ebx, bit(X86_FEATURE_SMEP));
9737 cr4_fixed1_update(X86_CR4_SMAP, ebx, bit(X86_FEATURE_SMAP));
9738 cr4_fixed1_update(X86_CR4_PKE, ecx, bit(X86_FEATURE_PKU));
9739 /* TODO: Use X86_CR4_UMIP and X86_FEATURE_UMIP macros */
9740 cr4_fixed1_update(bit(11), ecx, bit(2));
9742 #undef cr4_fixed1_update
9745 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
9747 struct vcpu_vmx *vmx = to_vmx(vcpu);
9749 if (cpu_has_secondary_exec_ctrls()) {
9750 vmx_compute_secondary_exec_control(vmx);
9751 vmcs_set_secondary_exec_control(vmx->secondary_exec_control);
9754 if (nested_vmx_allowed(vcpu))
9755 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
9756 FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9758 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
9759 ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9761 if (nested_vmx_allowed(vcpu))
9762 nested_vmx_cr_fixed1_bits_update(vcpu);
9765 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9767 if (func == 1 && nested)
9768 entry->ecx |= bit(X86_FEATURE_VMX);
9771 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9772 struct x86_exception *fault)
9774 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9775 struct vcpu_vmx *vmx = to_vmx(vcpu);
9777 unsigned long exit_qualification = vcpu->arch.exit_qualification;
9779 if (vmx->nested.pml_full) {
9780 exit_reason = EXIT_REASON_PML_FULL;
9781 vmx->nested.pml_full = false;
9782 exit_qualification &= INTR_INFO_UNBLOCK_NMI;
9783 } else if (fault->error_code & PFERR_RSVD_MASK)
9784 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9786 exit_reason = EXIT_REASON_EPT_VIOLATION;
9788 nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
9789 vmcs12->guest_physical_address = fault->address;
9792 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu)
9794 return nested_ept_get_cr3(vcpu) & VMX_EPTP_AD_ENABLE_BIT;
9797 /* Callbacks for nested_ept_init_mmu_context: */
9799 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9801 /* return the page table to be shadowed - in our case, EPT12 */
9802 return get_vmcs12(vcpu)->ept_pointer;
9805 static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9807 WARN_ON(mmu_is_nested(vcpu));
9808 if (!valid_ept_address(vcpu, nested_ept_get_cr3(vcpu)))
9811 kvm_mmu_unload(vcpu);
9812 kvm_init_shadow_ept_mmu(vcpu,
9813 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
9814 VMX_EPT_EXECUTE_ONLY_BIT,
9815 nested_ept_ad_enabled(vcpu));
9816 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
9817 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
9818 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
9820 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
9824 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
9826 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
9829 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9832 bool inequality, bit;
9834 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9836 (error_code & vmcs12->page_fault_error_code_mask) !=
9837 vmcs12->page_fault_error_code_match;
9838 return inequality ^ bit;
9841 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9842 struct x86_exception *fault)
9844 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9846 WARN_ON(!is_guest_mode(vcpu));
9848 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
9849 !to_vmx(vcpu)->nested.nested_run_pending) {
9850 vmcs12->vm_exit_intr_error_code = fault->error_code;
9851 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
9852 PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
9853 INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
9856 kvm_inject_page_fault(vcpu, fault);
9860 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9861 struct vmcs12 *vmcs12);
9863 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9864 struct vmcs12 *vmcs12)
9866 struct vcpu_vmx *vmx = to_vmx(vcpu);
9870 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9872 * Translate L1 physical address to host physical
9873 * address for vmcs02. Keep the page pinned, so this
9874 * physical address remains valid. We keep a reference
9875 * to it so we can release it later.
9877 if (vmx->nested.apic_access_page) { /* shouldn't happen */
9878 kvm_release_page_dirty(vmx->nested.apic_access_page);
9879 vmx->nested.apic_access_page = NULL;
9881 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
9883 * If translation failed, no matter: This feature asks
9884 * to exit when accessing the given address, and if it
9885 * can never be accessed, this feature won't do
9888 if (!is_error_page(page)) {
9889 vmx->nested.apic_access_page = page;
9890 hpa = page_to_phys(vmx->nested.apic_access_page);
9891 vmcs_write64(APIC_ACCESS_ADDR, hpa);
9893 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
9894 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9896 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9897 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9898 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
9899 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9900 kvm_vcpu_reload_apic_access_page(vcpu);
9903 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9904 if (vmx->nested.virtual_apic_page) { /* shouldn't happen */
9905 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
9906 vmx->nested.virtual_apic_page = NULL;
9908 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->virtual_apic_page_addr);
9911 * If translation failed, VM entry will fail because
9912 * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
9913 * Failing the vm entry is _not_ what the processor
9914 * does but it's basically the only possibility we
9915 * have. We could still enter the guest if CR8 load
9916 * exits are enabled, CR8 store exits are enabled, and
9917 * virtualize APIC access is disabled; in this case
9918 * the processor would never use the TPR shadow and we
9919 * could simply clear the bit from the execution
9920 * control. But such a configuration is useless, so
9921 * let's keep the code simple.
9923 if (!is_error_page(page)) {
9924 vmx->nested.virtual_apic_page = page;
9925 hpa = page_to_phys(vmx->nested.virtual_apic_page);
9926 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
9930 if (nested_cpu_has_posted_intr(vmcs12)) {
9931 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9932 kunmap(vmx->nested.pi_desc_page);
9933 kvm_release_page_dirty(vmx->nested.pi_desc_page);
9934 vmx->nested.pi_desc_page = NULL;
9936 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->posted_intr_desc_addr);
9937 if (is_error_page(page))
9939 vmx->nested.pi_desc_page = page;
9940 vmx->nested.pi_desc = kmap(vmx->nested.pi_desc_page);
9941 vmx->nested.pi_desc =
9942 (struct pi_desc *)((void *)vmx->nested.pi_desc +
9943 (unsigned long)(vmcs12->posted_intr_desc_addr &
9945 vmcs_write64(POSTED_INTR_DESC_ADDR,
9946 page_to_phys(vmx->nested.pi_desc_page) +
9947 (unsigned long)(vmcs12->posted_intr_desc_addr &
9950 if (cpu_has_vmx_msr_bitmap() &&
9951 nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS) &&
9952 nested_vmx_merge_msr_bitmap(vcpu, vmcs12))
9955 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
9956 CPU_BASED_USE_MSR_BITMAPS);
9959 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9961 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9962 struct vcpu_vmx *vmx = to_vmx(vcpu);
9964 if (vcpu->arch.virtual_tsc_khz == 0)
9967 /* Make sure short timeouts reliably trigger an immediate vmexit.
9968 * hrtimer_start does not guarantee this. */
9969 if (preemption_timeout <= 1) {
9970 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9974 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9975 preemption_timeout *= 1000000;
9976 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9977 hrtimer_start(&vmx->nested.preemption_timer,
9978 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9981 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
9982 struct vmcs12 *vmcs12)
9984 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
9987 if (!page_address_valid(vcpu, vmcs12->io_bitmap_a) ||
9988 !page_address_valid(vcpu, vmcs12->io_bitmap_b))
9994 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9995 struct vmcs12 *vmcs12)
9997 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
10000 if (!page_address_valid(vcpu, vmcs12->msr_bitmap))
10006 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
10007 struct vmcs12 *vmcs12)
10009 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
10012 if (!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr))
10019 * Merge L0's and L1's MSR bitmap, return false to indicate that
10020 * we do not use the hardware.
10022 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
10023 struct vmcs12 *vmcs12)
10027 unsigned long *msr_bitmap_l1;
10028 unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.msr_bitmap;
10030 /* This shortcut is ok because we support only x2APIC MSRs so far. */
10031 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
10034 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->msr_bitmap);
10035 if (is_error_page(page))
10037 msr_bitmap_l1 = (unsigned long *)kmap(page);
10039 memset(msr_bitmap_l0, 0xff, PAGE_SIZE);
10041 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
10042 if (nested_cpu_has_apic_reg_virt(vmcs12))
10043 for (msr = 0x800; msr <= 0x8ff; msr++)
10044 nested_vmx_disable_intercept_for_msr(
10045 msr_bitmap_l1, msr_bitmap_l0,
10048 nested_vmx_disable_intercept_for_msr(
10049 msr_bitmap_l1, msr_bitmap_l0,
10050 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
10051 MSR_TYPE_R | MSR_TYPE_W);
10053 if (nested_cpu_has_vid(vmcs12)) {
10054 nested_vmx_disable_intercept_for_msr(
10055 msr_bitmap_l1, msr_bitmap_l0,
10056 APIC_BASE_MSR + (APIC_EOI >> 4),
10058 nested_vmx_disable_intercept_for_msr(
10059 msr_bitmap_l1, msr_bitmap_l0,
10060 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
10065 kvm_release_page_clean(page);
10070 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
10071 struct vmcs12 *vmcs12)
10073 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10074 !nested_cpu_has_apic_reg_virt(vmcs12) &&
10075 !nested_cpu_has_vid(vmcs12) &&
10076 !nested_cpu_has_posted_intr(vmcs12))
10080 * If virtualize x2apic mode is enabled,
10081 * virtualize apic access must be disabled.
10083 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10084 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
10088 * If virtual interrupt delivery is enabled,
10089 * we must exit on external interrupts.
10091 if (nested_cpu_has_vid(vmcs12) &&
10092 !nested_exit_on_intr(vcpu))
10096 * bits 15:8 should be zero in posted_intr_nv,
10097 * the descriptor address has been already checked
10098 * in nested_get_vmcs12_pages.
10100 if (nested_cpu_has_posted_intr(vmcs12) &&
10101 (!nested_cpu_has_vid(vmcs12) ||
10102 !nested_exit_intr_ack_set(vcpu) ||
10103 vmcs12->posted_intr_nv & 0xff00))
10106 /* tpr shadow is needed by all apicv features. */
10107 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
10113 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
10114 unsigned long count_field,
10115 unsigned long addr_field)
10120 if (vmcs12_read_any(vcpu, count_field, &count) ||
10121 vmcs12_read_any(vcpu, addr_field, &addr)) {
10127 maxphyaddr = cpuid_maxphyaddr(vcpu);
10128 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
10129 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
10130 pr_debug_ratelimited(
10131 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
10132 addr_field, maxphyaddr, count, addr);
10138 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
10139 struct vmcs12 *vmcs12)
10141 if (vmcs12->vm_exit_msr_load_count == 0 &&
10142 vmcs12->vm_exit_msr_store_count == 0 &&
10143 vmcs12->vm_entry_msr_load_count == 0)
10144 return 0; /* Fast path */
10145 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
10146 VM_EXIT_MSR_LOAD_ADDR) ||
10147 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
10148 VM_EXIT_MSR_STORE_ADDR) ||
10149 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
10150 VM_ENTRY_MSR_LOAD_ADDR))
10155 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
10156 struct vmcs12 *vmcs12)
10158 u64 address = vmcs12->pml_address;
10159 int maxphyaddr = cpuid_maxphyaddr(vcpu);
10161 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML)) {
10162 if (!nested_cpu_has_ept(vmcs12) ||
10163 !IS_ALIGNED(address, 4096) ||
10164 address >> maxphyaddr)
10171 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
10172 struct vmx_msr_entry *e)
10174 /* x2APIC MSR accesses are not allowed */
10175 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
10177 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
10178 e->index == MSR_IA32_UCODE_REV)
10180 if (e->reserved != 0)
10185 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
10186 struct vmx_msr_entry *e)
10188 if (e->index == MSR_FS_BASE ||
10189 e->index == MSR_GS_BASE ||
10190 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
10191 nested_vmx_msr_check_common(vcpu, e))
10196 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
10197 struct vmx_msr_entry *e)
10199 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
10200 nested_vmx_msr_check_common(vcpu, e))
10206 * Load guest's/host's msr at nested entry/exit.
10207 * return 0 for success, entry index for failure.
10209 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
10212 struct vmx_msr_entry e;
10213 struct msr_data msr;
10215 msr.host_initiated = false;
10216 for (i = 0; i < count; i++) {
10217 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
10219 pr_debug_ratelimited(
10220 "%s cannot read MSR entry (%u, 0x%08llx)\n",
10221 __func__, i, gpa + i * sizeof(e));
10224 if (nested_vmx_load_msr_check(vcpu, &e)) {
10225 pr_debug_ratelimited(
10226 "%s check failed (%u, 0x%x, 0x%x)\n",
10227 __func__, i, e.index, e.reserved);
10230 msr.index = e.index;
10231 msr.data = e.value;
10232 if (kvm_set_msr(vcpu, &msr)) {
10233 pr_debug_ratelimited(
10234 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
10235 __func__, i, e.index, e.value);
10244 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
10247 struct vmx_msr_entry e;
10249 for (i = 0; i < count; i++) {
10250 struct msr_data msr_info;
10251 if (kvm_vcpu_read_guest(vcpu,
10252 gpa + i * sizeof(e),
10253 &e, 2 * sizeof(u32))) {
10254 pr_debug_ratelimited(
10255 "%s cannot read MSR entry (%u, 0x%08llx)\n",
10256 __func__, i, gpa + i * sizeof(e));
10259 if (nested_vmx_store_msr_check(vcpu, &e)) {
10260 pr_debug_ratelimited(
10261 "%s check failed (%u, 0x%x, 0x%x)\n",
10262 __func__, i, e.index, e.reserved);
10265 msr_info.host_initiated = false;
10266 msr_info.index = e.index;
10267 if (kvm_get_msr(vcpu, &msr_info)) {
10268 pr_debug_ratelimited(
10269 "%s cannot read MSR (%u, 0x%x)\n",
10270 __func__, i, e.index);
10273 if (kvm_vcpu_write_guest(vcpu,
10274 gpa + i * sizeof(e) +
10275 offsetof(struct vmx_msr_entry, value),
10276 &msr_info.data, sizeof(msr_info.data))) {
10277 pr_debug_ratelimited(
10278 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
10279 __func__, i, e.index, msr_info.data);
10286 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
10288 unsigned long invalid_mask;
10290 invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
10291 return (val & invalid_mask) == 0;
10295 * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
10296 * emulating VM entry into a guest with EPT enabled.
10297 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10298 * is assigned to entry_failure_code on failure.
10300 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
10301 u32 *entry_failure_code)
10303 if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
10304 if (!nested_cr3_valid(vcpu, cr3)) {
10305 *entry_failure_code = ENTRY_FAIL_DEFAULT;
10310 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
10311 * must not be dereferenced.
10313 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
10315 if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
10316 *entry_failure_code = ENTRY_FAIL_PDPTE;
10321 vcpu->arch.cr3 = cr3;
10322 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
10325 kvm_mmu_reset_context(vcpu);
10330 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
10331 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
10332 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
10333 * guest in a way that will both be appropriate to L1's requests, and our
10334 * needs. In addition to modifying the active vmcs (which is vmcs02), this
10335 * function also has additional necessary side-effects, like setting various
10336 * vcpu->arch fields.
10337 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10338 * is assigned to entry_failure_code on failure.
10340 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10341 bool from_vmentry, u32 *entry_failure_code)
10343 struct vcpu_vmx *vmx = to_vmx(vcpu);
10344 u32 exec_control, vmcs12_exec_ctrl;
10346 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
10347 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
10348 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
10349 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
10350 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
10351 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
10352 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
10353 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
10354 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
10355 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
10356 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
10357 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
10358 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
10359 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
10360 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
10361 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
10362 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
10363 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
10364 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
10365 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
10366 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
10367 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
10368 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
10369 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
10370 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
10371 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
10372 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
10373 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
10374 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
10375 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
10376 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
10377 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
10378 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
10379 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
10380 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
10381 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
10383 if (from_vmentry &&
10384 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
10385 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
10386 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
10388 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
10389 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
10391 if (from_vmentry) {
10392 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
10393 vmcs12->vm_entry_intr_info_field);
10394 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
10395 vmcs12->vm_entry_exception_error_code);
10396 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
10397 vmcs12->vm_entry_instruction_len);
10398 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
10399 vmcs12->guest_interruptibility_info);
10400 vmx->loaded_vmcs->nmi_known_unmasked =
10401 !(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
10403 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
10405 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
10406 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
10407 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
10408 vmcs12->guest_pending_dbg_exceptions);
10409 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
10410 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
10412 if (nested_cpu_has_xsaves(vmcs12))
10413 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
10414 vmcs_write64(VMCS_LINK_POINTER, -1ull);
10416 exec_control = vmcs12->pin_based_vm_exec_control;
10418 /* Preemption timer setting is only taken from vmcs01. */
10419 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10420 exec_control |= vmcs_config.pin_based_exec_ctrl;
10421 if (vmx->hv_deadline_tsc == -1)
10422 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10424 /* Posted interrupts setting is only taken from vmcs12. */
10425 if (nested_cpu_has_posted_intr(vmcs12)) {
10426 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
10427 vmx->nested.pi_pending = false;
10428 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
10430 exec_control &= ~PIN_BASED_POSTED_INTR;
10433 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
10435 vmx->nested.preemption_timer_expired = false;
10436 if (nested_cpu_has_preemption_timer(vmcs12))
10437 vmx_start_preemption_timer(vcpu);
10440 * Whether page-faults are trapped is determined by a combination of
10441 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
10442 * If enable_ept, L0 doesn't care about page faults and we should
10443 * set all of these to L1's desires. However, if !enable_ept, L0 does
10444 * care about (at least some) page faults, and because it is not easy
10445 * (if at all possible?) to merge L0 and L1's desires, we simply ask
10446 * to exit on each and every L2 page fault. This is done by setting
10447 * MASK=MATCH=0 and (see below) EB.PF=1.
10448 * Note that below we don't need special code to set EB.PF beyond the
10449 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
10450 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
10451 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
10453 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
10454 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
10455 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
10456 enable_ept ? vmcs12->page_fault_error_code_match : 0);
10458 if (cpu_has_secondary_exec_ctrls()) {
10459 exec_control = vmx->secondary_exec_control;
10461 /* Take the following fields only from vmcs12 */
10462 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
10463 SECONDARY_EXEC_ENABLE_INVPCID |
10464 SECONDARY_EXEC_RDTSCP |
10465 SECONDARY_EXEC_XSAVES |
10466 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
10467 SECONDARY_EXEC_APIC_REGISTER_VIRT |
10468 SECONDARY_EXEC_ENABLE_VMFUNC);
10469 if (nested_cpu_has(vmcs12,
10470 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
10471 vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
10472 ~SECONDARY_EXEC_ENABLE_PML;
10473 exec_control |= vmcs12_exec_ctrl;
10476 /* All VMFUNCs are currently emulated through L0 vmexits. */
10477 if (exec_control & SECONDARY_EXEC_ENABLE_VMFUNC)
10478 vmcs_write64(VM_FUNCTION_CONTROL, 0);
10480 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
10481 vmcs_write64(EOI_EXIT_BITMAP0,
10482 vmcs12->eoi_exit_bitmap0);
10483 vmcs_write64(EOI_EXIT_BITMAP1,
10484 vmcs12->eoi_exit_bitmap1);
10485 vmcs_write64(EOI_EXIT_BITMAP2,
10486 vmcs12->eoi_exit_bitmap2);
10487 vmcs_write64(EOI_EXIT_BITMAP3,
10488 vmcs12->eoi_exit_bitmap3);
10489 vmcs_write16(GUEST_INTR_STATUS,
10490 vmcs12->guest_intr_status);
10494 * Write an illegal value to APIC_ACCESS_ADDR. Later,
10495 * nested_get_vmcs12_pages will either fix it up or
10496 * remove the VM execution control.
10498 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
10499 vmcs_write64(APIC_ACCESS_ADDR, -1ull);
10501 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
10506 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
10507 * Some constant fields are set here by vmx_set_constant_host_state().
10508 * Other fields are different per CPU, and will be set later when
10509 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
10511 vmx_set_constant_host_state(vmx);
10514 * Set the MSR load/store lists to match L0's settings.
10516 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
10517 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10518 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
10519 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10520 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
10523 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
10524 * entry, but only if the current (host) sp changed from the value
10525 * we wrote last (vmx->host_rsp). This cache is no longer relevant
10526 * if we switch vmcs, and rather than hold a separate cache per vmcs,
10527 * here we just force the write to happen on entry.
10531 exec_control = vmx_exec_control(vmx); /* L0's desires */
10532 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
10533 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
10534 exec_control &= ~CPU_BASED_TPR_SHADOW;
10535 exec_control |= vmcs12->cpu_based_vm_exec_control;
10538 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
10539 * nested_get_vmcs12_pages can't fix it up, the illegal value
10540 * will result in a VM entry failure.
10542 if (exec_control & CPU_BASED_TPR_SHADOW) {
10543 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
10544 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
10546 #ifdef CONFIG_X86_64
10547 exec_control |= CPU_BASED_CR8_LOAD_EXITING |
10548 CPU_BASED_CR8_STORE_EXITING;
10553 * Merging of IO bitmap not currently supported.
10554 * Rather, exit every time.
10556 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
10557 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
10559 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
10561 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
10562 * bitwise-or of what L1 wants to trap for L2, and what we want to
10563 * trap. Note that CR0.TS also needs updating - we do this later.
10565 update_exception_bitmap(vcpu);
10566 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
10567 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10569 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
10570 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
10571 * bits are further modified by vmx_set_efer() below.
10573 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
10575 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
10576 * emulated by vmx_set_efer(), below.
10578 vm_entry_controls_init(vmx,
10579 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
10580 ~VM_ENTRY_IA32E_MODE) |
10581 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
10583 if (from_vmentry &&
10584 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
10585 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
10586 vcpu->arch.pat = vmcs12->guest_ia32_pat;
10587 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
10588 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
10591 set_cr4_guest_host_mask(vmx);
10593 if (from_vmentry &&
10594 vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
10595 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
10597 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
10598 vmcs_write64(TSC_OFFSET,
10599 vcpu->arch.tsc_offset + vmcs12->tsc_offset);
10601 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
10602 if (kvm_has_tsc_control)
10603 decache_tsc_multiplier(vmx);
10607 * There is no direct mapping between vpid02 and vpid12, the
10608 * vpid02 is per-vCPU for L0 and reused while the value of
10609 * vpid12 is changed w/ one invvpid during nested vmentry.
10610 * The vpid12 is allocated by L1 for L2, so it will not
10611 * influence global bitmap(for vpid01 and vpid02 allocation)
10612 * even if spawn a lot of nested vCPUs.
10614 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
10615 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
10616 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
10617 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
10618 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
10621 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
10622 vmx_flush_tlb(vcpu);
10629 * Conceptually we want to copy the PML address and index from
10630 * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
10631 * since we always flush the log on each vmexit, this happens
10632 * to be equivalent to simply resetting the fields in vmcs02.
10634 ASSERT(vmx->pml_pg);
10635 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
10636 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
10639 if (nested_cpu_has_ept(vmcs12)) {
10640 if (nested_ept_init_mmu_context(vcpu)) {
10641 *entry_failure_code = ENTRY_FAIL_DEFAULT;
10644 } else if (nested_cpu_has2(vmcs12,
10645 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10646 vmx_flush_tlb_ept_only(vcpu);
10650 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
10651 * bits which we consider mandatory enabled.
10652 * The CR0_READ_SHADOW is what L2 should have expected to read given
10653 * the specifications by L1; It's not enough to take
10654 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10655 * have more bits than L1 expected.
10657 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
10658 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
10660 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
10661 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
10663 if (from_vmentry &&
10664 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
10665 vcpu->arch.efer = vmcs12->guest_ia32_efer;
10666 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
10667 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10669 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10670 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10671 vmx_set_efer(vcpu, vcpu->arch.efer);
10673 /* Shadow page tables on either EPT or shadow page tables. */
10674 if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
10675 entry_failure_code))
10679 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
10682 * L1 may access the L2's PDPTR, so save them to construct vmcs12
10685 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
10686 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
10687 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
10688 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
10691 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
10692 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
10696 static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10698 struct vcpu_vmx *vmx = to_vmx(vcpu);
10700 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
10701 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
10702 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10704 if (nested_vmx_check_io_bitmap_controls(vcpu, vmcs12))
10705 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10707 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12))
10708 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10710 if (nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12))
10711 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10713 if (nested_vmx_check_apicv_controls(vcpu, vmcs12))
10714 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10716 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12))
10717 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10719 if (nested_vmx_check_pml_controls(vcpu, vmcs12))
10720 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10722 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
10723 vmx->nested.nested_vmx_procbased_ctls_low,
10724 vmx->nested.nested_vmx_procbased_ctls_high) ||
10725 (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
10726 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
10727 vmx->nested.nested_vmx_secondary_ctls_low,
10728 vmx->nested.nested_vmx_secondary_ctls_high)) ||
10729 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
10730 vmx->nested.nested_vmx_pinbased_ctls_low,
10731 vmx->nested.nested_vmx_pinbased_ctls_high) ||
10732 !vmx_control_verify(vmcs12->vm_exit_controls,
10733 vmx->nested.nested_vmx_exit_ctls_low,
10734 vmx->nested.nested_vmx_exit_ctls_high) ||
10735 !vmx_control_verify(vmcs12->vm_entry_controls,
10736 vmx->nested.nested_vmx_entry_ctls_low,
10737 vmx->nested.nested_vmx_entry_ctls_high))
10738 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10740 if (nested_cpu_has_vmfunc(vmcs12)) {
10741 if (vmcs12->vm_function_control &
10742 ~vmx->nested.nested_vmx_vmfunc_controls)
10743 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10745 if (nested_cpu_has_eptp_switching(vmcs12)) {
10746 if (!nested_cpu_has_ept(vmcs12) ||
10747 !page_address_valid(vcpu, vmcs12->eptp_list_address))
10748 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10752 if (vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu))
10753 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10755 if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
10756 !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
10757 !nested_cr3_valid(vcpu, vmcs12->host_cr3))
10758 return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
10763 static int check_vmentry_postreqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10768 *exit_qual = ENTRY_FAIL_DEFAULT;
10770 if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
10771 !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
10774 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_SHADOW_VMCS) &&
10775 vmcs12->vmcs_link_pointer != -1ull) {
10776 *exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
10781 * If the load IA32_EFER VM-entry control is 1, the following checks
10782 * are performed on the field for the IA32_EFER MSR:
10783 * - Bits reserved in the IA32_EFER MSR must be 0.
10784 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
10785 * the IA-32e mode guest VM-exit control. It must also be identical
10786 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
10789 if (to_vmx(vcpu)->nested.nested_run_pending &&
10790 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
10791 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
10792 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
10793 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
10794 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
10795 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
10800 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
10801 * IA32_EFER MSR must be 0 in the field for that register. In addition,
10802 * the values of the LMA and LME bits in the field must each be that of
10803 * the host address-space size VM-exit control.
10805 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
10806 ia32e = (vmcs12->vm_exit_controls &
10807 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
10808 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
10809 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
10810 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
10817 static int enter_vmx_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
10819 struct vcpu_vmx *vmx = to_vmx(vcpu);
10820 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10821 struct loaded_vmcs *vmcs02;
10825 vmcs02 = nested_get_current_vmcs02(vmx);
10829 enter_guest_mode(vcpu);
10831 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
10832 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10834 vmx_switch_vmcs(vcpu, vmcs02);
10835 vmx_segment_cache_clear(vmx);
10837 if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &exit_qual)) {
10838 leave_guest_mode(vcpu);
10839 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10840 nested_vmx_entry_failure(vcpu, vmcs12,
10841 EXIT_REASON_INVALID_STATE, exit_qual);
10845 nested_get_vmcs12_pages(vcpu, vmcs12);
10847 msr_entry_idx = nested_vmx_load_msr(vcpu,
10848 vmcs12->vm_entry_msr_load_addr,
10849 vmcs12->vm_entry_msr_load_count);
10850 if (msr_entry_idx) {
10851 leave_guest_mode(vcpu);
10852 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10853 nested_vmx_entry_failure(vcpu, vmcs12,
10854 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
10859 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10860 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10861 * returned as far as L1 is concerned. It will only return (and set
10862 * the success flag) when L2 exits (see nested_vmx_vmexit()).
10868 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
10869 * for running an L2 nested guest.
10871 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
10873 struct vmcs12 *vmcs12;
10874 struct vcpu_vmx *vmx = to_vmx(vcpu);
10875 u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
10879 if (!nested_vmx_check_permission(vcpu))
10882 if (!nested_vmx_check_vmcs12(vcpu))
10885 vmcs12 = get_vmcs12(vcpu);
10887 if (enable_shadow_vmcs)
10888 copy_shadow_to_vmcs12(vmx);
10891 * The nested entry process starts with enforcing various prerequisites
10892 * on vmcs12 as required by the Intel SDM, and act appropriately when
10893 * they fail: As the SDM explains, some conditions should cause the
10894 * instruction to fail, while others will cause the instruction to seem
10895 * to succeed, but return an EXIT_REASON_INVALID_STATE.
10896 * To speed up the normal (success) code path, we should avoid checking
10897 * for misconfigurations which will anyway be caught by the processor
10898 * when using the merged vmcs02.
10900 if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS) {
10901 nested_vmx_failValid(vcpu,
10902 VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
10906 if (vmcs12->launch_state == launch) {
10907 nested_vmx_failValid(vcpu,
10908 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
10909 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
10913 ret = check_vmentry_prereqs(vcpu, vmcs12);
10915 nested_vmx_failValid(vcpu, ret);
10920 * After this point, the trap flag no longer triggers a singlestep trap
10921 * on the vm entry instructions; don't call kvm_skip_emulated_instruction.
10922 * This is not 100% correct; for performance reasons, we delegate most
10923 * of the checks on host state to the processor. If those fail,
10924 * the singlestep trap is missed.
10926 skip_emulated_instruction(vcpu);
10928 ret = check_vmentry_postreqs(vcpu, vmcs12, &exit_qual);
10930 nested_vmx_entry_failure(vcpu, vmcs12,
10931 EXIT_REASON_INVALID_STATE, exit_qual);
10936 * We're finally done with prerequisite checking, and can start with
10937 * the nested entry.
10940 ret = enter_vmx_non_root_mode(vcpu, true);
10944 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
10945 return kvm_vcpu_halt(vcpu);
10947 vmx->nested.nested_run_pending = 1;
10952 return kvm_skip_emulated_instruction(vcpu);
10956 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
10957 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
10958 * This function returns the new value we should put in vmcs12.guest_cr0.
10959 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
10960 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
10961 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
10962 * didn't trap the bit, because if L1 did, so would L0).
10963 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
10964 * been modified by L2, and L1 knows it. So just leave the old value of
10965 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
10966 * isn't relevant, because if L0 traps this bit it can set it to anything.
10967 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
10968 * changed these bits, and therefore they need to be updated, but L0
10969 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
10970 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
10972 static inline unsigned long
10973 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10976 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
10977 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
10978 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
10979 vcpu->arch.cr0_guest_owned_bits));
10982 static inline unsigned long
10983 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10986 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
10987 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
10988 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
10989 vcpu->arch.cr4_guest_owned_bits));
10992 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
10993 struct vmcs12 *vmcs12)
10998 if (vcpu->arch.exception.injected) {
10999 nr = vcpu->arch.exception.nr;
11000 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
11002 if (kvm_exception_is_soft(nr)) {
11003 vmcs12->vm_exit_instruction_len =
11004 vcpu->arch.event_exit_inst_len;
11005 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
11007 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
11009 if (vcpu->arch.exception.has_error_code) {
11010 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
11011 vmcs12->idt_vectoring_error_code =
11012 vcpu->arch.exception.error_code;
11015 vmcs12->idt_vectoring_info_field = idt_vectoring;
11016 } else if (vcpu->arch.nmi_injected) {
11017 vmcs12->idt_vectoring_info_field =
11018 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
11019 } else if (vcpu->arch.interrupt.pending) {
11020 nr = vcpu->arch.interrupt.nr;
11021 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
11023 if (vcpu->arch.interrupt.soft) {
11024 idt_vectoring |= INTR_TYPE_SOFT_INTR;
11025 vmcs12->vm_entry_instruction_len =
11026 vcpu->arch.event_exit_inst_len;
11028 idt_vectoring |= INTR_TYPE_EXT_INTR;
11030 vmcs12->idt_vectoring_info_field = idt_vectoring;
11034 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
11036 struct vcpu_vmx *vmx = to_vmx(vcpu);
11037 unsigned long exit_qual;
11039 if (kvm_event_needs_reinjection(vcpu))
11042 if (vcpu->arch.exception.pending &&
11043 nested_vmx_check_exception(vcpu, &exit_qual)) {
11044 if (vmx->nested.nested_run_pending)
11046 nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
11047 vcpu->arch.exception.pending = false;
11051 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
11052 vmx->nested.preemption_timer_expired) {
11053 if (vmx->nested.nested_run_pending)
11055 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
11059 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
11060 if (vmx->nested.nested_run_pending)
11062 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
11063 NMI_VECTOR | INTR_TYPE_NMI_INTR |
11064 INTR_INFO_VALID_MASK, 0);
11066 * The NMI-triggered VM exit counts as injection:
11067 * clear this one and block further NMIs.
11069 vcpu->arch.nmi_pending = 0;
11070 vmx_set_nmi_mask(vcpu, true);
11074 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
11075 nested_exit_on_intr(vcpu)) {
11076 if (vmx->nested.nested_run_pending)
11078 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
11082 vmx_complete_nested_posted_interrupt(vcpu);
11086 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
11088 ktime_t remaining =
11089 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
11092 if (ktime_to_ns(remaining) <= 0)
11095 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
11096 do_div(value, 1000000);
11097 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
11101 * Update the guest state fields of vmcs12 to reflect changes that
11102 * occurred while L2 was running. (The "IA-32e mode guest" bit of the
11103 * VM-entry controls is also updated, since this is really a guest
11106 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11108 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
11109 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
11111 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
11112 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
11113 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
11115 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
11116 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
11117 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
11118 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
11119 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
11120 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
11121 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
11122 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
11123 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
11124 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
11125 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
11126 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
11127 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
11128 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
11129 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
11130 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
11131 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
11132 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
11133 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
11134 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
11135 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
11136 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
11137 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
11138 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
11139 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
11140 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
11141 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
11142 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
11143 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
11144 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
11145 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
11146 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
11147 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
11148 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
11149 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
11150 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
11152 vmcs12->guest_interruptibility_info =
11153 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
11154 vmcs12->guest_pending_dbg_exceptions =
11155 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
11156 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
11157 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
11159 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
11161 if (nested_cpu_has_preemption_timer(vmcs12)) {
11162 if (vmcs12->vm_exit_controls &
11163 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
11164 vmcs12->vmx_preemption_timer_value =
11165 vmx_get_preemption_timer_value(vcpu);
11166 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
11170 * In some cases (usually, nested EPT), L2 is allowed to change its
11171 * own CR3 without exiting. If it has changed it, we must keep it.
11172 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
11173 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
11175 * Additionally, restore L2's PDPTR to vmcs12.
11178 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
11179 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
11180 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
11181 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
11182 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
11185 vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
11187 if (nested_cpu_has_vid(vmcs12))
11188 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
11190 vmcs12->vm_entry_controls =
11191 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
11192 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
11194 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
11195 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
11196 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
11199 /* TODO: These cannot have changed unless we have MSR bitmaps and
11200 * the relevant bit asks not to trap the change */
11201 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
11202 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
11203 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
11204 vmcs12->guest_ia32_efer = vcpu->arch.efer;
11205 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
11206 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
11207 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
11208 if (kvm_mpx_supported())
11209 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
11213 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
11214 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
11215 * and this function updates it to reflect the changes to the guest state while
11216 * L2 was running (and perhaps made some exits which were handled directly by L0
11217 * without going back to L1), and to reflect the exit reason.
11218 * Note that we do not have to copy here all VMCS fields, just those that
11219 * could have changed by the L2 guest or the exit - i.e., the guest-state and
11220 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
11221 * which already writes to vmcs12 directly.
11223 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
11224 u32 exit_reason, u32 exit_intr_info,
11225 unsigned long exit_qualification)
11227 /* update guest state fields: */
11228 sync_vmcs12(vcpu, vmcs12);
11230 /* update exit information fields: */
11232 vmcs12->vm_exit_reason = exit_reason;
11233 vmcs12->exit_qualification = exit_qualification;
11234 vmcs12->vm_exit_intr_info = exit_intr_info;
11236 vmcs12->idt_vectoring_info_field = 0;
11237 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
11238 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
11240 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
11241 vmcs12->launch_state = 1;
11243 /* vm_entry_intr_info_field is cleared on exit. Emulate this
11244 * instead of reading the real value. */
11245 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
11248 * Transfer the event that L0 or L1 may wanted to inject into
11249 * L2 to IDT_VECTORING_INFO_FIELD.
11251 vmcs12_save_pending_event(vcpu, vmcs12);
11255 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
11256 * preserved above and would only end up incorrectly in L1.
11258 vcpu->arch.nmi_injected = false;
11259 kvm_clear_exception_queue(vcpu);
11260 kvm_clear_interrupt_queue(vcpu);
11264 * A part of what we need to when the nested L2 guest exits and we want to
11265 * run its L1 parent, is to reset L1's guest state to the host state specified
11267 * This function is to be called not only on normal nested exit, but also on
11268 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
11269 * Failures During or After Loading Guest State").
11270 * This function should be called when the active VMCS is L1's (vmcs01).
11272 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
11273 struct vmcs12 *vmcs12)
11275 struct kvm_segment seg;
11276 u32 entry_failure_code;
11278 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
11279 vcpu->arch.efer = vmcs12->host_ia32_efer;
11280 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
11281 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
11283 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
11284 vmx_set_efer(vcpu, vcpu->arch.efer);
11286 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
11287 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
11288 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
11290 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
11291 * actually changed, because vmx_set_cr0 refers to efer set above.
11293 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
11294 * (KVM doesn't change it);
11296 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
11297 vmx_set_cr0(vcpu, vmcs12->host_cr0);
11299 /* Same as above - no reason to call set_cr4_guest_host_mask(). */
11300 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
11301 vmx_set_cr4(vcpu, vmcs12->host_cr4);
11303 nested_ept_uninit_mmu_context(vcpu);
11306 * Only PDPTE load can fail as the value of cr3 was checked on entry and
11307 * couldn't have changed.
11309 if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
11310 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
11313 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
11317 * Trivially support vpid by letting L2s share their parent
11318 * L1's vpid. TODO: move to a more elaborate solution, giving
11319 * each L2 its own vpid and exposing the vpid feature to L1.
11321 vmx_flush_tlb(vcpu);
11323 /* Restore posted intr vector. */
11324 if (nested_cpu_has_posted_intr(vmcs12))
11325 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
11327 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
11328 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
11329 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
11330 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
11331 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
11333 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
11334 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
11335 vmcs_write64(GUEST_BNDCFGS, 0);
11337 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
11338 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
11339 vcpu->arch.pat = vmcs12->host_ia32_pat;
11341 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
11342 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
11343 vmcs12->host_ia32_perf_global_ctrl);
11345 /* Set L1 segment info according to Intel SDM
11346 27.5.2 Loading Host Segment and Descriptor-Table Registers */
11347 seg = (struct kvm_segment) {
11349 .limit = 0xFFFFFFFF,
11350 .selector = vmcs12->host_cs_selector,
11356 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
11360 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
11361 seg = (struct kvm_segment) {
11363 .limit = 0xFFFFFFFF,
11370 seg.selector = vmcs12->host_ds_selector;
11371 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
11372 seg.selector = vmcs12->host_es_selector;
11373 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
11374 seg.selector = vmcs12->host_ss_selector;
11375 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
11376 seg.selector = vmcs12->host_fs_selector;
11377 seg.base = vmcs12->host_fs_base;
11378 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
11379 seg.selector = vmcs12->host_gs_selector;
11380 seg.base = vmcs12->host_gs_base;
11381 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
11382 seg = (struct kvm_segment) {
11383 .base = vmcs12->host_tr_base,
11385 .selector = vmcs12->host_tr_selector,
11389 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
11391 kvm_set_dr(vcpu, 7, 0x400);
11392 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
11394 if (cpu_has_vmx_msr_bitmap())
11395 vmx_set_msr_bitmap(vcpu);
11397 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
11398 vmcs12->vm_exit_msr_load_count))
11399 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
11403 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
11404 * and modify vmcs12 to make it see what it would expect to see there if
11405 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
11407 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
11408 u32 exit_intr_info,
11409 unsigned long exit_qualification)
11411 struct vcpu_vmx *vmx = to_vmx(vcpu);
11412 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
11414 /* trying to cancel vmlaunch/vmresume is a bug */
11415 WARN_ON_ONCE(vmx->nested.nested_run_pending);
11418 * The only expected VM-instruction error is "VM entry with
11419 * invalid control field(s)." Anything else indicates a
11422 WARN_ON_ONCE(vmx->fail && (vmcs_read32(VM_INSTRUCTION_ERROR) !=
11423 VMXERR_ENTRY_INVALID_CONTROL_FIELD));
11425 leave_guest_mode(vcpu);
11427 if (likely(!vmx->fail)) {
11428 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
11429 exit_qualification);
11431 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
11432 vmcs12->vm_exit_msr_store_count))
11433 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
11436 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
11437 vm_entry_controls_reset_shadow(vmx);
11438 vm_exit_controls_reset_shadow(vmx);
11439 vmx_segment_cache_clear(vmx);
11441 /* if no vmcs02 cache requested, remove the one we used */
11442 if (VMCS02_POOL_SIZE == 0)
11443 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
11445 /* Update any VMCS fields that might have changed while L2 ran */
11446 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11447 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11448 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
11449 if (vmx->hv_deadline_tsc == -1)
11450 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11451 PIN_BASED_VMX_PREEMPTION_TIMER);
11453 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11454 PIN_BASED_VMX_PREEMPTION_TIMER);
11455 if (kvm_has_tsc_control)
11456 decache_tsc_multiplier(vmx);
11458 if (vmx->nested.change_vmcs01_virtual_x2apic_mode) {
11459 vmx->nested.change_vmcs01_virtual_x2apic_mode = false;
11460 vmx_set_virtual_x2apic_mode(vcpu,
11461 vcpu->arch.apic_base & X2APIC_ENABLE);
11462 } else if (!nested_cpu_has_ept(vmcs12) &&
11463 nested_cpu_has2(vmcs12,
11464 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
11465 vmx_flush_tlb_ept_only(vcpu);
11468 /* This is needed for same reason as it was needed in prepare_vmcs02 */
11471 /* Unpin physical memory we referred to in vmcs02 */
11472 if (vmx->nested.apic_access_page) {
11473 kvm_release_page_dirty(vmx->nested.apic_access_page);
11474 vmx->nested.apic_access_page = NULL;
11476 if (vmx->nested.virtual_apic_page) {
11477 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
11478 vmx->nested.virtual_apic_page = NULL;
11480 if (vmx->nested.pi_desc_page) {
11481 kunmap(vmx->nested.pi_desc_page);
11482 kvm_release_page_dirty(vmx->nested.pi_desc_page);
11483 vmx->nested.pi_desc_page = NULL;
11484 vmx->nested.pi_desc = NULL;
11488 * We are now running in L2, mmu_notifier will force to reload the
11489 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
11491 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
11493 if (enable_shadow_vmcs)
11494 vmx->nested.sync_shadow_vmcs = true;
11496 /* in case we halted in L2 */
11497 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11499 if (likely(!vmx->fail)) {
11501 * TODO: SDM says that with acknowledge interrupt on
11502 * exit, bit 31 of the VM-exit interrupt information
11503 * (valid interrupt) is always set to 1 on
11504 * EXIT_REASON_EXTERNAL_INTERRUPT, so we shouldn't
11505 * need kvm_cpu_has_interrupt(). See the commit
11506 * message for details.
11508 if (nested_exit_intr_ack_set(vcpu) &&
11509 exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
11510 kvm_cpu_has_interrupt(vcpu)) {
11511 int irq = kvm_cpu_get_interrupt(vcpu);
11513 vmcs12->vm_exit_intr_info = irq |
11514 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
11517 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
11518 vmcs12->exit_qualification,
11519 vmcs12->idt_vectoring_info_field,
11520 vmcs12->vm_exit_intr_info,
11521 vmcs12->vm_exit_intr_error_code,
11524 load_vmcs12_host_state(vcpu, vmcs12);
11530 * After an early L2 VM-entry failure, we're now back
11531 * in L1 which thinks it just finished a VMLAUNCH or
11532 * VMRESUME instruction, so we need to set the failure
11533 * flag and the VM-instruction error field of the VMCS
11536 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
11538 * The emulated instruction was already skipped in
11539 * nested_vmx_run, but the updated RIP was never
11540 * written back to the vmcs01.
11542 skip_emulated_instruction(vcpu);
11547 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
11549 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
11551 if (is_guest_mode(vcpu)) {
11552 to_vmx(vcpu)->nested.nested_run_pending = 0;
11553 nested_vmx_vmexit(vcpu, -1, 0, 0);
11555 free_nested(to_vmx(vcpu));
11559 * L1's failure to enter L2 is a subset of a normal exit, as explained in
11560 * 23.7 "VM-entry failures during or after loading guest state" (this also
11561 * lists the acceptable exit-reason and exit-qualification parameters).
11562 * It should only be called before L2 actually succeeded to run, and when
11563 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
11565 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
11566 struct vmcs12 *vmcs12,
11567 u32 reason, unsigned long qualification)
11569 load_vmcs12_host_state(vcpu, vmcs12);
11570 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
11571 vmcs12->exit_qualification = qualification;
11572 nested_vmx_succeed(vcpu);
11573 if (enable_shadow_vmcs)
11574 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
11577 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
11578 struct x86_instruction_info *info,
11579 enum x86_intercept_stage stage)
11581 return X86EMUL_CONTINUE;
11584 #ifdef CONFIG_X86_64
11585 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
11586 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
11587 u64 divisor, u64 *result)
11589 u64 low = a << shift, high = a >> (64 - shift);
11591 /* To avoid the overflow on divq */
11592 if (high >= divisor)
11595 /* Low hold the result, high hold rem which is discarded */
11596 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
11597 "rm" (divisor), "0" (low), "1" (high));
11603 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
11605 struct vcpu_vmx *vmx = to_vmx(vcpu);
11606 u64 tscl = rdtsc();
11607 u64 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
11608 u64 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
11610 /* Convert to host delta tsc if tsc scaling is enabled */
11611 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
11612 u64_shl_div_u64(delta_tsc,
11613 kvm_tsc_scaling_ratio_frac_bits,
11614 vcpu->arch.tsc_scaling_ratio,
11619 * If the delta tsc can't fit in the 32 bit after the multi shift,
11620 * we can't use the preemption timer.
11621 * It's possible that it fits on later vmentries, but checking
11622 * on every vmentry is costly so we just use an hrtimer.
11624 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
11627 vmx->hv_deadline_tsc = tscl + delta_tsc;
11628 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11629 PIN_BASED_VMX_PREEMPTION_TIMER);
11631 return delta_tsc == 0;
11634 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
11636 struct vcpu_vmx *vmx = to_vmx(vcpu);
11637 vmx->hv_deadline_tsc = -1;
11638 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11639 PIN_BASED_VMX_PREEMPTION_TIMER);
11643 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
11646 shrink_ple_window(vcpu);
11649 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
11650 struct kvm_memory_slot *slot)
11652 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
11653 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
11656 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
11657 struct kvm_memory_slot *slot)
11659 kvm_mmu_slot_set_dirty(kvm, slot);
11662 static void vmx_flush_log_dirty(struct kvm *kvm)
11664 kvm_flush_pml_buffers(kvm);
11667 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
11669 struct vmcs12 *vmcs12;
11670 struct vcpu_vmx *vmx = to_vmx(vcpu);
11672 struct page *page = NULL;
11675 if (is_guest_mode(vcpu)) {
11676 WARN_ON_ONCE(vmx->nested.pml_full);
11679 * Check if PML is enabled for the nested guest.
11680 * Whether eptp bit 6 is set is already checked
11681 * as part of A/D emulation.
11683 vmcs12 = get_vmcs12(vcpu);
11684 if (!nested_cpu_has_pml(vmcs12))
11687 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
11688 vmx->nested.pml_full = true;
11692 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
11694 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->pml_address);
11695 if (is_error_page(page))
11698 pml_address = kmap(page);
11699 pml_address[vmcs12->guest_pml_index--] = gpa;
11701 kvm_release_page_clean(page);
11707 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
11708 struct kvm_memory_slot *memslot,
11709 gfn_t offset, unsigned long mask)
11711 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
11714 static void __pi_post_block(struct kvm_vcpu *vcpu)
11716 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11717 struct pi_desc old, new;
11721 old.control = new.control = pi_desc->control;
11722 WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
11723 "Wakeup handler not enabled while the VCPU is blocked\n");
11725 dest = cpu_physical_id(vcpu->cpu);
11727 if (x2apic_enabled())
11730 new.ndst = (dest << 8) & 0xFF00;
11732 /* set 'NV' to 'notification vector' */
11733 new.nv = POSTED_INTR_VECTOR;
11734 } while (cmpxchg64(&pi_desc->control, old.control,
11735 new.control) != old.control);
11737 if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
11738 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
11739 list_del(&vcpu->blocked_vcpu_list);
11740 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
11741 vcpu->pre_pcpu = -1;
11746 * This routine does the following things for vCPU which is going
11747 * to be blocked if VT-d PI is enabled.
11748 * - Store the vCPU to the wakeup list, so when interrupts happen
11749 * we can find the right vCPU to wake up.
11750 * - Change the Posted-interrupt descriptor as below:
11751 * 'NDST' <-- vcpu->pre_pcpu
11752 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
11753 * - If 'ON' is set during this process, which means at least one
11754 * interrupt is posted for this vCPU, we cannot block it, in
11755 * this case, return 1, otherwise, return 0.
11758 static int pi_pre_block(struct kvm_vcpu *vcpu)
11761 struct pi_desc old, new;
11762 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11764 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11765 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11766 !kvm_vcpu_apicv_active(vcpu))
11769 WARN_ON(irqs_disabled());
11770 local_irq_disable();
11771 if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
11772 vcpu->pre_pcpu = vcpu->cpu;
11773 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
11774 list_add_tail(&vcpu->blocked_vcpu_list,
11775 &per_cpu(blocked_vcpu_on_cpu,
11777 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
11781 old.control = new.control = pi_desc->control;
11783 WARN((pi_desc->sn == 1),
11784 "Warning: SN field of posted-interrupts "
11785 "is set before blocking\n");
11788 * Since vCPU can be preempted during this process,
11789 * vcpu->cpu could be different with pre_pcpu, we
11790 * need to set pre_pcpu as the destination of wakeup
11791 * notification event, then we can find the right vCPU
11792 * to wakeup in wakeup handler if interrupts happen
11793 * when the vCPU is in blocked state.
11795 dest = cpu_physical_id(vcpu->pre_pcpu);
11797 if (x2apic_enabled())
11800 new.ndst = (dest << 8) & 0xFF00;
11802 /* set 'NV' to 'wakeup vector' */
11803 new.nv = POSTED_INTR_WAKEUP_VECTOR;
11804 } while (cmpxchg64(&pi_desc->control, old.control,
11805 new.control) != old.control);
11807 /* We should not block the vCPU if an interrupt is posted for it. */
11808 if (pi_test_on(pi_desc) == 1)
11809 __pi_post_block(vcpu);
11811 local_irq_enable();
11812 return (vcpu->pre_pcpu == -1);
11815 static int vmx_pre_block(struct kvm_vcpu *vcpu)
11817 if (pi_pre_block(vcpu))
11820 if (kvm_lapic_hv_timer_in_use(vcpu))
11821 kvm_lapic_switch_to_sw_timer(vcpu);
11826 static void pi_post_block(struct kvm_vcpu *vcpu)
11828 if (vcpu->pre_pcpu == -1)
11831 WARN_ON(irqs_disabled());
11832 local_irq_disable();
11833 __pi_post_block(vcpu);
11834 local_irq_enable();
11837 static void vmx_post_block(struct kvm_vcpu *vcpu)
11839 if (kvm_x86_ops->set_hv_timer)
11840 kvm_lapic_switch_to_hv_timer(vcpu);
11842 pi_post_block(vcpu);
11846 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
11849 * @host_irq: host irq of the interrupt
11850 * @guest_irq: gsi of the interrupt
11851 * @set: set or unset PI
11852 * returns 0 on success, < 0 on failure
11854 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
11855 uint32_t guest_irq, bool set)
11857 struct kvm_kernel_irq_routing_entry *e;
11858 struct kvm_irq_routing_table *irq_rt;
11859 struct kvm_lapic_irq irq;
11860 struct kvm_vcpu *vcpu;
11861 struct vcpu_data vcpu_info;
11864 if (!kvm_arch_has_assigned_device(kvm) ||
11865 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11866 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
11869 idx = srcu_read_lock(&kvm->irq_srcu);
11870 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
11871 if (guest_irq >= irq_rt->nr_rt_entries ||
11872 hlist_empty(&irq_rt->map[guest_irq])) {
11873 pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
11874 guest_irq, irq_rt->nr_rt_entries);
11878 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
11879 if (e->type != KVM_IRQ_ROUTING_MSI)
11882 * VT-d PI cannot support posting multicast/broadcast
11883 * interrupts to a vCPU, we still use interrupt remapping
11884 * for these kind of interrupts.
11886 * For lowest-priority interrupts, we only support
11887 * those with single CPU as the destination, e.g. user
11888 * configures the interrupts via /proc/irq or uses
11889 * irqbalance to make the interrupts single-CPU.
11891 * We will support full lowest-priority interrupt later.
11894 kvm_set_msi_irq(kvm, e, &irq);
11895 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
11897 * Make sure the IRTE is in remapped mode if
11898 * we don't handle it in posted mode.
11900 ret = irq_set_vcpu_affinity(host_irq, NULL);
11903 "failed to back to remapped mode, irq: %u\n",
11911 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
11912 vcpu_info.vector = irq.vector;
11914 trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
11915 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
11918 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
11920 ret = irq_set_vcpu_affinity(host_irq, NULL);
11923 printk(KERN_INFO "%s: failed to update PI IRTE\n",
11931 srcu_read_unlock(&kvm->irq_srcu, idx);
11935 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
11937 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
11938 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
11939 FEATURE_CONTROL_LMCE;
11941 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
11942 ~FEATURE_CONTROL_LMCE;
11945 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
11946 .cpu_has_kvm_support = cpu_has_kvm_support,
11947 .disabled_by_bios = vmx_disabled_by_bios,
11948 .hardware_setup = hardware_setup,
11949 .hardware_unsetup = hardware_unsetup,
11950 .check_processor_compatibility = vmx_check_processor_compat,
11951 .hardware_enable = hardware_enable,
11952 .hardware_disable = hardware_disable,
11953 .cpu_has_accelerated_tpr = report_flexpriority,
11954 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
11956 .vcpu_create = vmx_create_vcpu,
11957 .vcpu_free = vmx_free_vcpu,
11958 .vcpu_reset = vmx_vcpu_reset,
11960 .prepare_guest_switch = vmx_save_host_state,
11961 .vcpu_load = vmx_vcpu_load,
11962 .vcpu_put = vmx_vcpu_put,
11964 .update_bp_intercept = update_exception_bitmap,
11965 .get_msr = vmx_get_msr,
11966 .set_msr = vmx_set_msr,
11967 .get_segment_base = vmx_get_segment_base,
11968 .get_segment = vmx_get_segment,
11969 .set_segment = vmx_set_segment,
11970 .get_cpl = vmx_get_cpl,
11971 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
11972 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
11973 .decache_cr3 = vmx_decache_cr3,
11974 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
11975 .set_cr0 = vmx_set_cr0,
11976 .set_cr3 = vmx_set_cr3,
11977 .set_cr4 = vmx_set_cr4,
11978 .set_efer = vmx_set_efer,
11979 .get_idt = vmx_get_idt,
11980 .set_idt = vmx_set_idt,
11981 .get_gdt = vmx_get_gdt,
11982 .set_gdt = vmx_set_gdt,
11983 .get_dr6 = vmx_get_dr6,
11984 .set_dr6 = vmx_set_dr6,
11985 .set_dr7 = vmx_set_dr7,
11986 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
11987 .cache_reg = vmx_cache_reg,
11988 .get_rflags = vmx_get_rflags,
11989 .set_rflags = vmx_set_rflags,
11991 .tlb_flush = vmx_flush_tlb,
11993 .run = vmx_vcpu_run,
11994 .handle_exit = vmx_handle_exit,
11995 .skip_emulated_instruction = skip_emulated_instruction,
11996 .set_interrupt_shadow = vmx_set_interrupt_shadow,
11997 .get_interrupt_shadow = vmx_get_interrupt_shadow,
11998 .patch_hypercall = vmx_patch_hypercall,
11999 .set_irq = vmx_inject_irq,
12000 .set_nmi = vmx_inject_nmi,
12001 .queue_exception = vmx_queue_exception,
12002 .cancel_injection = vmx_cancel_injection,
12003 .interrupt_allowed = vmx_interrupt_allowed,
12004 .nmi_allowed = vmx_nmi_allowed,
12005 .get_nmi_mask = vmx_get_nmi_mask,
12006 .set_nmi_mask = vmx_set_nmi_mask,
12007 .enable_nmi_window = enable_nmi_window,
12008 .enable_irq_window = enable_irq_window,
12009 .update_cr8_intercept = update_cr8_intercept,
12010 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
12011 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
12012 .get_enable_apicv = vmx_get_enable_apicv,
12013 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
12014 .load_eoi_exitmap = vmx_load_eoi_exitmap,
12015 .apicv_post_state_restore = vmx_apicv_post_state_restore,
12016 .hwapic_irr_update = vmx_hwapic_irr_update,
12017 .hwapic_isr_update = vmx_hwapic_isr_update,
12018 .sync_pir_to_irr = vmx_sync_pir_to_irr,
12019 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
12021 .set_tss_addr = vmx_set_tss_addr,
12022 .get_tdp_level = get_ept_level,
12023 .get_mt_mask = vmx_get_mt_mask,
12025 .get_exit_info = vmx_get_exit_info,
12027 .get_lpage_level = vmx_get_lpage_level,
12029 .cpuid_update = vmx_cpuid_update,
12031 .rdtscp_supported = vmx_rdtscp_supported,
12032 .invpcid_supported = vmx_invpcid_supported,
12034 .set_supported_cpuid = vmx_set_supported_cpuid,
12036 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
12038 .write_tsc_offset = vmx_write_tsc_offset,
12040 .set_tdp_cr3 = vmx_set_cr3,
12042 .check_intercept = vmx_check_intercept,
12043 .handle_external_intr = vmx_handle_external_intr,
12044 .mpx_supported = vmx_mpx_supported,
12045 .xsaves_supported = vmx_xsaves_supported,
12047 .check_nested_events = vmx_check_nested_events,
12049 .sched_in = vmx_sched_in,
12051 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
12052 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
12053 .flush_log_dirty = vmx_flush_log_dirty,
12054 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
12055 .write_log_dirty = vmx_write_pml_buffer,
12057 .pre_block = vmx_pre_block,
12058 .post_block = vmx_post_block,
12060 .pmu_ops = &intel_pmu_ops,
12062 .update_pi_irte = vmx_update_pi_irte,
12064 #ifdef CONFIG_X86_64
12065 .set_hv_timer = vmx_set_hv_timer,
12066 .cancel_hv_timer = vmx_cancel_hv_timer,
12069 .setup_mce = vmx_setup_mce,
12072 static int __init vmx_init(void)
12074 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
12075 __alignof__(struct vcpu_vmx), THIS_MODULE);
12079 #ifdef CONFIG_KEXEC_CORE
12080 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
12081 crash_vmclear_local_loaded_vmcss);
12087 static void __exit vmx_exit(void)
12089 #ifdef CONFIG_KEXEC_CORE
12090 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
12097 module_init(vmx_init)
12098 module_exit(vmx_exit)