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.
23 #include <linux/kvm_host.h>
24 #include <linux/module.h>
25 #include <linux/kernel.h>
27 #include <linux/highmem.h>
28 #include <linux/sched.h>
29 #include <linux/moduleparam.h>
30 #include <linux/mod_devicetable.h>
31 #include <linux/trace_events.h>
32 #include <linux/slab.h>
33 #include <linux/tboot.h>
34 #include <linux/hrtimer.h>
35 #include "kvm_cache_regs.h"
42 #include <asm/virtext.h>
44 #include <asm/fpu/internal.h>
45 #include <asm/perf_event.h>
46 #include <asm/debugreg.h>
47 #include <asm/kexec.h>
49 #include <asm/irq_remapping.h>
54 #define __ex(x) __kvm_handle_fault_on_reboot(x)
55 #define __ex_clear(x, reg) \
56 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
58 MODULE_AUTHOR("Qumranet");
59 MODULE_LICENSE("GPL");
61 static const struct x86_cpu_id vmx_cpu_id[] = {
62 X86_FEATURE_MATCH(X86_FEATURE_VMX),
65 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
67 static bool __read_mostly enable_vpid = 1;
68 module_param_named(vpid, enable_vpid, bool, 0444);
70 static bool __read_mostly flexpriority_enabled = 1;
71 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
73 static bool __read_mostly enable_ept = 1;
74 module_param_named(ept, enable_ept, bool, S_IRUGO);
76 static bool __read_mostly enable_unrestricted_guest = 1;
77 module_param_named(unrestricted_guest,
78 enable_unrestricted_guest, bool, S_IRUGO);
80 static bool __read_mostly enable_ept_ad_bits = 1;
81 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
83 static bool __read_mostly emulate_invalid_guest_state = true;
84 module_param(emulate_invalid_guest_state, bool, S_IRUGO);
86 static bool __read_mostly vmm_exclusive = 1;
87 module_param(vmm_exclusive, bool, S_IRUGO);
89 static bool __read_mostly fasteoi = 1;
90 module_param(fasteoi, bool, S_IRUGO);
92 static bool __read_mostly enable_apicv = 1;
93 module_param(enable_apicv, bool, S_IRUGO);
95 static bool __read_mostly enable_shadow_vmcs = 1;
96 module_param_named(enable_shadow_vmcs, enable_shadow_vmcs, bool, S_IRUGO);
98 * If nested=1, nested virtualization is supported, i.e., guests may use
99 * VMX and be a hypervisor for its own guests. If nested=0, guests may not
100 * use VMX instructions.
102 static bool __read_mostly nested = 0;
103 module_param(nested, bool, S_IRUGO);
105 static u64 __read_mostly host_xss;
107 static bool __read_mostly enable_pml = 1;
108 module_param_named(pml, enable_pml, bool, S_IRUGO);
110 #define KVM_VMX_TSC_MULTIPLIER_MAX 0xffffffffffffffffULL
112 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
113 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
114 #define KVM_VM_CR0_ALWAYS_ON \
115 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
116 #define KVM_CR4_GUEST_OWNED_BITS \
117 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
118 | X86_CR4_OSXMMEXCPT | X86_CR4_TSD)
120 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
121 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
123 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
125 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
128 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
129 * ple_gap: upper bound on the amount of time between two successive
130 * executions of PAUSE in a loop. Also indicate if ple enabled.
131 * According to test, this time is usually smaller than 128 cycles.
132 * ple_window: upper bound on the amount of time a guest is allowed to execute
133 * in a PAUSE loop. Tests indicate that most spinlocks are held for
134 * less than 2^12 cycles
135 * Time is measured based on a counter that runs at the same rate as the TSC,
136 * refer SDM volume 3b section 21.6.13 & 22.1.3.
138 #define KVM_VMX_DEFAULT_PLE_GAP 128
139 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
140 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
141 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
142 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
143 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
145 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
146 module_param(ple_gap, int, S_IRUGO);
148 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
149 module_param(ple_window, int, S_IRUGO);
151 /* Default doubles per-vcpu window every exit. */
152 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
153 module_param(ple_window_grow, int, S_IRUGO);
155 /* Default resets per-vcpu window every exit to ple_window. */
156 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
157 module_param(ple_window_shrink, int, S_IRUGO);
159 /* Default is to compute the maximum so we can never overflow. */
160 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
161 static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
162 module_param(ple_window_max, int, S_IRUGO);
164 extern const ulong vmx_return;
166 #define NR_AUTOLOAD_MSRS 8
167 #define VMCS02_POOL_SIZE 1
176 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
177 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
178 * loaded on this CPU (so we can clear them if the CPU goes down).
184 struct list_head loaded_vmcss_on_cpu_link;
187 struct shared_msr_entry {
194 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
195 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
196 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
197 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
198 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
199 * More than one of these structures may exist, if L1 runs multiple L2 guests.
200 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
201 * underlying hardware which will be used to run L2.
202 * This structure is packed to ensure that its layout is identical across
203 * machines (necessary for live migration).
204 * If there are changes in this struct, VMCS12_REVISION must be changed.
206 typedef u64 natural_width;
207 struct __packed vmcs12 {
208 /* According to the Intel spec, a VMCS region must start with the
209 * following two fields. Then follow implementation-specific data.
214 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
215 u32 padding[7]; /* room for future expansion */
220 u64 vm_exit_msr_store_addr;
221 u64 vm_exit_msr_load_addr;
222 u64 vm_entry_msr_load_addr;
224 u64 virtual_apic_page_addr;
225 u64 apic_access_addr;
226 u64 posted_intr_desc_addr;
228 u64 eoi_exit_bitmap0;
229 u64 eoi_exit_bitmap1;
230 u64 eoi_exit_bitmap2;
231 u64 eoi_exit_bitmap3;
233 u64 guest_physical_address;
234 u64 vmcs_link_pointer;
235 u64 guest_ia32_debugctl;
238 u64 guest_ia32_perf_global_ctrl;
246 u64 host_ia32_perf_global_ctrl;
247 u64 padding64[8]; /* room for future expansion */
249 * To allow migration of L1 (complete with its L2 guests) between
250 * machines of different natural widths (32 or 64 bit), we cannot have
251 * unsigned long fields with no explict size. We use u64 (aliased
252 * natural_width) instead. Luckily, x86 is little-endian.
254 natural_width cr0_guest_host_mask;
255 natural_width cr4_guest_host_mask;
256 natural_width cr0_read_shadow;
257 natural_width cr4_read_shadow;
258 natural_width cr3_target_value0;
259 natural_width cr3_target_value1;
260 natural_width cr3_target_value2;
261 natural_width cr3_target_value3;
262 natural_width exit_qualification;
263 natural_width guest_linear_address;
264 natural_width guest_cr0;
265 natural_width guest_cr3;
266 natural_width guest_cr4;
267 natural_width guest_es_base;
268 natural_width guest_cs_base;
269 natural_width guest_ss_base;
270 natural_width guest_ds_base;
271 natural_width guest_fs_base;
272 natural_width guest_gs_base;
273 natural_width guest_ldtr_base;
274 natural_width guest_tr_base;
275 natural_width guest_gdtr_base;
276 natural_width guest_idtr_base;
277 natural_width guest_dr7;
278 natural_width guest_rsp;
279 natural_width guest_rip;
280 natural_width guest_rflags;
281 natural_width guest_pending_dbg_exceptions;
282 natural_width guest_sysenter_esp;
283 natural_width guest_sysenter_eip;
284 natural_width host_cr0;
285 natural_width host_cr3;
286 natural_width host_cr4;
287 natural_width host_fs_base;
288 natural_width host_gs_base;
289 natural_width host_tr_base;
290 natural_width host_gdtr_base;
291 natural_width host_idtr_base;
292 natural_width host_ia32_sysenter_esp;
293 natural_width host_ia32_sysenter_eip;
294 natural_width host_rsp;
295 natural_width host_rip;
296 natural_width paddingl[8]; /* room for future expansion */
297 u32 pin_based_vm_exec_control;
298 u32 cpu_based_vm_exec_control;
299 u32 exception_bitmap;
300 u32 page_fault_error_code_mask;
301 u32 page_fault_error_code_match;
302 u32 cr3_target_count;
303 u32 vm_exit_controls;
304 u32 vm_exit_msr_store_count;
305 u32 vm_exit_msr_load_count;
306 u32 vm_entry_controls;
307 u32 vm_entry_msr_load_count;
308 u32 vm_entry_intr_info_field;
309 u32 vm_entry_exception_error_code;
310 u32 vm_entry_instruction_len;
312 u32 secondary_vm_exec_control;
313 u32 vm_instruction_error;
315 u32 vm_exit_intr_info;
316 u32 vm_exit_intr_error_code;
317 u32 idt_vectoring_info_field;
318 u32 idt_vectoring_error_code;
319 u32 vm_exit_instruction_len;
320 u32 vmx_instruction_info;
327 u32 guest_ldtr_limit;
329 u32 guest_gdtr_limit;
330 u32 guest_idtr_limit;
331 u32 guest_es_ar_bytes;
332 u32 guest_cs_ar_bytes;
333 u32 guest_ss_ar_bytes;
334 u32 guest_ds_ar_bytes;
335 u32 guest_fs_ar_bytes;
336 u32 guest_gs_ar_bytes;
337 u32 guest_ldtr_ar_bytes;
338 u32 guest_tr_ar_bytes;
339 u32 guest_interruptibility_info;
340 u32 guest_activity_state;
341 u32 guest_sysenter_cs;
342 u32 host_ia32_sysenter_cs;
343 u32 vmx_preemption_timer_value;
344 u32 padding32[7]; /* room for future expansion */
345 u16 virtual_processor_id;
347 u16 guest_es_selector;
348 u16 guest_cs_selector;
349 u16 guest_ss_selector;
350 u16 guest_ds_selector;
351 u16 guest_fs_selector;
352 u16 guest_gs_selector;
353 u16 guest_ldtr_selector;
354 u16 guest_tr_selector;
355 u16 guest_intr_status;
356 u16 host_es_selector;
357 u16 host_cs_selector;
358 u16 host_ss_selector;
359 u16 host_ds_selector;
360 u16 host_fs_selector;
361 u16 host_gs_selector;
362 u16 host_tr_selector;
366 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
367 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
368 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
370 #define VMCS12_REVISION 0x11e57ed0
373 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
374 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
375 * current implementation, 4K are reserved to avoid future complications.
377 #define VMCS12_SIZE 0x1000
379 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
381 struct list_head list;
383 struct loaded_vmcs vmcs02;
387 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
388 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
391 /* Has the level1 guest done vmxon? */
395 /* The guest-physical address of the current VMCS L1 keeps for L2 */
397 /* The host-usable pointer to the above */
398 struct page *current_vmcs12_page;
399 struct vmcs12 *current_vmcs12;
400 struct vmcs *current_shadow_vmcs;
402 * Indicates if the shadow vmcs must be updated with the
403 * data hold by vmcs12
405 bool sync_shadow_vmcs;
407 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
408 struct list_head vmcs02_pool;
410 u64 vmcs01_tsc_offset;
411 bool change_vmcs01_virtual_x2apic_mode;
412 /* L2 must run next, and mustn't decide to exit to L1. */
413 bool nested_run_pending;
415 * Guest pages referred to in vmcs02 with host-physical pointers, so
416 * we must keep them pinned while L2 runs.
418 struct page *apic_access_page;
419 struct page *virtual_apic_page;
420 struct page *pi_desc_page;
421 struct pi_desc *pi_desc;
424 u64 msr_ia32_feature_control;
426 struct hrtimer preemption_timer;
427 bool preemption_timer_expired;
429 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
435 u32 nested_vmx_procbased_ctls_low;
436 u32 nested_vmx_procbased_ctls_high;
437 u32 nested_vmx_true_procbased_ctls_low;
438 u32 nested_vmx_secondary_ctls_low;
439 u32 nested_vmx_secondary_ctls_high;
440 u32 nested_vmx_pinbased_ctls_low;
441 u32 nested_vmx_pinbased_ctls_high;
442 u32 nested_vmx_exit_ctls_low;
443 u32 nested_vmx_exit_ctls_high;
444 u32 nested_vmx_true_exit_ctls_low;
445 u32 nested_vmx_entry_ctls_low;
446 u32 nested_vmx_entry_ctls_high;
447 u32 nested_vmx_true_entry_ctls_low;
448 u32 nested_vmx_misc_low;
449 u32 nested_vmx_misc_high;
450 u32 nested_vmx_ept_caps;
451 u32 nested_vmx_vpid_caps;
454 #define POSTED_INTR_ON 0
455 #define POSTED_INTR_SN 1
457 /* Posted-Interrupt Descriptor */
459 u32 pir[8]; /* Posted interrupt requested */
462 /* bit 256 - Outstanding Notification */
464 /* bit 257 - Suppress Notification */
466 /* bit 271:258 - Reserved */
468 /* bit 279:272 - Notification Vector */
470 /* bit 287:280 - Reserved */
472 /* bit 319:288 - Notification Destination */
480 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
482 return test_and_set_bit(POSTED_INTR_ON,
483 (unsigned long *)&pi_desc->control);
486 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
488 return test_and_clear_bit(POSTED_INTR_ON,
489 (unsigned long *)&pi_desc->control);
492 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
494 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
497 static inline void pi_clear_sn(struct pi_desc *pi_desc)
499 return clear_bit(POSTED_INTR_SN,
500 (unsigned long *)&pi_desc->control);
503 static inline void pi_set_sn(struct pi_desc *pi_desc)
505 return set_bit(POSTED_INTR_SN,
506 (unsigned long *)&pi_desc->control);
509 static inline int pi_test_on(struct pi_desc *pi_desc)
511 return test_bit(POSTED_INTR_ON,
512 (unsigned long *)&pi_desc->control);
515 static inline int pi_test_sn(struct pi_desc *pi_desc)
517 return test_bit(POSTED_INTR_SN,
518 (unsigned long *)&pi_desc->control);
522 struct kvm_vcpu vcpu;
523 unsigned long host_rsp;
525 bool nmi_known_unmasked;
527 u32 idt_vectoring_info;
529 struct shared_msr_entry *guest_msrs;
532 unsigned long host_idt_base;
534 u64 msr_host_kernel_gs_base;
535 u64 msr_guest_kernel_gs_base;
537 u32 vm_entry_controls_shadow;
538 u32 vm_exit_controls_shadow;
540 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
541 * non-nested (L1) guest, it always points to vmcs01. For a nested
542 * guest (L2), it points to a different VMCS.
544 struct loaded_vmcs vmcs01;
545 struct loaded_vmcs *loaded_vmcs;
546 bool __launched; /* temporary, used in vmx_vcpu_run */
547 struct msr_autoload {
549 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
550 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
554 u16 fs_sel, gs_sel, ldt_sel;
558 int gs_ldt_reload_needed;
559 int fs_reload_needed;
560 u64 msr_host_bndcfgs;
561 unsigned long vmcs_host_cr4; /* May not match real cr4 */
566 struct kvm_segment segs[8];
569 u32 bitmask; /* 4 bits per segment (1 bit per field) */
570 struct kvm_save_segment {
578 bool emulation_required;
580 /* Support for vnmi-less CPUs */
581 int soft_vnmi_blocked;
583 s64 vnmi_blocked_time;
586 /* Posted interrupt descriptor */
587 struct pi_desc pi_desc;
589 /* Support for a guest hypervisor (nested VMX) */
590 struct nested_vmx nested;
592 /* Dynamic PLE window. */
594 bool ple_window_dirty;
596 /* Support for PML */
597 #define PML_ENTITY_NUM 512
600 u64 current_tsc_ratio;
603 enum segment_cache_field {
612 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
614 return container_of(vcpu, struct vcpu_vmx, vcpu);
617 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
619 return &(to_vmx(vcpu)->pi_desc);
622 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
623 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
624 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
625 [number##_HIGH] = VMCS12_OFFSET(name)+4
628 static unsigned long shadow_read_only_fields[] = {
630 * We do NOT shadow fields that are modified when L0
631 * traps and emulates any vmx instruction (e.g. VMPTRLD,
632 * VMXON...) executed by L1.
633 * For example, VM_INSTRUCTION_ERROR is read
634 * by L1 if a vmx instruction fails (part of the error path).
635 * Note the code assumes this logic. If for some reason
636 * we start shadowing these fields then we need to
637 * force a shadow sync when L0 emulates vmx instructions
638 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
639 * by nested_vmx_failValid)
643 VM_EXIT_INSTRUCTION_LEN,
644 IDT_VECTORING_INFO_FIELD,
645 IDT_VECTORING_ERROR_CODE,
646 VM_EXIT_INTR_ERROR_CODE,
648 GUEST_LINEAR_ADDRESS,
649 GUEST_PHYSICAL_ADDRESS
651 static int max_shadow_read_only_fields =
652 ARRAY_SIZE(shadow_read_only_fields);
654 static unsigned long shadow_read_write_fields[] = {
661 GUEST_INTERRUPTIBILITY_INFO,
674 CPU_BASED_VM_EXEC_CONTROL,
675 VM_ENTRY_EXCEPTION_ERROR_CODE,
676 VM_ENTRY_INTR_INFO_FIELD,
677 VM_ENTRY_INSTRUCTION_LEN,
678 VM_ENTRY_EXCEPTION_ERROR_CODE,
684 static int max_shadow_read_write_fields =
685 ARRAY_SIZE(shadow_read_write_fields);
687 static const unsigned short vmcs_field_to_offset_table[] = {
688 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
689 FIELD(POSTED_INTR_NV, posted_intr_nv),
690 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
691 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
692 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
693 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
694 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
695 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
696 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
697 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
698 FIELD(GUEST_INTR_STATUS, guest_intr_status),
699 FIELD(HOST_ES_SELECTOR, host_es_selector),
700 FIELD(HOST_CS_SELECTOR, host_cs_selector),
701 FIELD(HOST_SS_SELECTOR, host_ss_selector),
702 FIELD(HOST_DS_SELECTOR, host_ds_selector),
703 FIELD(HOST_FS_SELECTOR, host_fs_selector),
704 FIELD(HOST_GS_SELECTOR, host_gs_selector),
705 FIELD(HOST_TR_SELECTOR, host_tr_selector),
706 FIELD64(IO_BITMAP_A, io_bitmap_a),
707 FIELD64(IO_BITMAP_B, io_bitmap_b),
708 FIELD64(MSR_BITMAP, msr_bitmap),
709 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
710 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
711 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
712 FIELD64(TSC_OFFSET, tsc_offset),
713 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
714 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
715 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
716 FIELD64(EPT_POINTER, ept_pointer),
717 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
718 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
719 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
720 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
721 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
722 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
723 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
724 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
725 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
726 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
727 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
728 FIELD64(GUEST_PDPTR0, guest_pdptr0),
729 FIELD64(GUEST_PDPTR1, guest_pdptr1),
730 FIELD64(GUEST_PDPTR2, guest_pdptr2),
731 FIELD64(GUEST_PDPTR3, guest_pdptr3),
732 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
733 FIELD64(HOST_IA32_PAT, host_ia32_pat),
734 FIELD64(HOST_IA32_EFER, host_ia32_efer),
735 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
736 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
737 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
738 FIELD(EXCEPTION_BITMAP, exception_bitmap),
739 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
740 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
741 FIELD(CR3_TARGET_COUNT, cr3_target_count),
742 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
743 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
744 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
745 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
746 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
747 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
748 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
749 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
750 FIELD(TPR_THRESHOLD, tpr_threshold),
751 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
752 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
753 FIELD(VM_EXIT_REASON, vm_exit_reason),
754 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
755 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
756 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
757 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
758 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
759 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
760 FIELD(GUEST_ES_LIMIT, guest_es_limit),
761 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
762 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
763 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
764 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
765 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
766 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
767 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
768 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
769 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
770 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
771 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
772 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
773 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
774 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
775 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
776 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
777 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
778 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
779 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
780 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
781 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
782 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
783 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
784 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
785 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
786 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
787 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
788 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
789 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
790 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
791 FIELD(EXIT_QUALIFICATION, exit_qualification),
792 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
793 FIELD(GUEST_CR0, guest_cr0),
794 FIELD(GUEST_CR3, guest_cr3),
795 FIELD(GUEST_CR4, guest_cr4),
796 FIELD(GUEST_ES_BASE, guest_es_base),
797 FIELD(GUEST_CS_BASE, guest_cs_base),
798 FIELD(GUEST_SS_BASE, guest_ss_base),
799 FIELD(GUEST_DS_BASE, guest_ds_base),
800 FIELD(GUEST_FS_BASE, guest_fs_base),
801 FIELD(GUEST_GS_BASE, guest_gs_base),
802 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
803 FIELD(GUEST_TR_BASE, guest_tr_base),
804 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
805 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
806 FIELD(GUEST_DR7, guest_dr7),
807 FIELD(GUEST_RSP, guest_rsp),
808 FIELD(GUEST_RIP, guest_rip),
809 FIELD(GUEST_RFLAGS, guest_rflags),
810 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
811 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
812 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
813 FIELD(HOST_CR0, host_cr0),
814 FIELD(HOST_CR3, host_cr3),
815 FIELD(HOST_CR4, host_cr4),
816 FIELD(HOST_FS_BASE, host_fs_base),
817 FIELD(HOST_GS_BASE, host_gs_base),
818 FIELD(HOST_TR_BASE, host_tr_base),
819 FIELD(HOST_GDTR_BASE, host_gdtr_base),
820 FIELD(HOST_IDTR_BASE, host_idtr_base),
821 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
822 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
823 FIELD(HOST_RSP, host_rsp),
824 FIELD(HOST_RIP, host_rip),
827 static inline short vmcs_field_to_offset(unsigned long field)
829 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
831 if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
832 vmcs_field_to_offset_table[field] == 0)
835 return vmcs_field_to_offset_table[field];
838 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
840 return to_vmx(vcpu)->nested.current_vmcs12;
843 static struct page *nested_get_page(struct kvm_vcpu *vcpu, gpa_t addr)
845 struct page *page = kvm_vcpu_gfn_to_page(vcpu, addr >> PAGE_SHIFT);
846 if (is_error_page(page))
852 static void nested_release_page(struct page *page)
854 kvm_release_page_dirty(page);
857 static void nested_release_page_clean(struct page *page)
859 kvm_release_page_clean(page);
862 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
863 static u64 construct_eptp(unsigned long root_hpa);
864 static void kvm_cpu_vmxon(u64 addr);
865 static void kvm_cpu_vmxoff(void);
866 static bool vmx_xsaves_supported(void);
867 static int vmx_cpu_uses_apicv(struct kvm_vcpu *vcpu);
868 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
869 static void vmx_set_segment(struct kvm_vcpu *vcpu,
870 struct kvm_segment *var, int seg);
871 static void vmx_get_segment(struct kvm_vcpu *vcpu,
872 struct kvm_segment *var, int seg);
873 static bool guest_state_valid(struct kvm_vcpu *vcpu);
874 static u32 vmx_segment_access_rights(struct kvm_segment *var);
875 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu);
876 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
877 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
878 static int alloc_identity_pagetable(struct kvm *kvm);
880 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
881 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
883 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
884 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
886 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
887 static DEFINE_PER_CPU(struct desc_ptr, host_gdt);
890 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
891 * can find which vCPU should be waken up.
893 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
894 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
896 static unsigned long *vmx_io_bitmap_a;
897 static unsigned long *vmx_io_bitmap_b;
898 static unsigned long *vmx_msr_bitmap_legacy;
899 static unsigned long *vmx_msr_bitmap_longmode;
900 static unsigned long *vmx_msr_bitmap_legacy_x2apic;
901 static unsigned long *vmx_msr_bitmap_longmode_x2apic;
902 static unsigned long *vmx_msr_bitmap_nested;
903 static unsigned long *vmx_vmread_bitmap;
904 static unsigned long *vmx_vmwrite_bitmap;
906 static bool cpu_has_load_ia32_efer;
907 static bool cpu_has_load_perf_global_ctrl;
909 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
910 static DEFINE_SPINLOCK(vmx_vpid_lock);
912 static struct vmcs_config {
916 u32 pin_based_exec_ctrl;
917 u32 cpu_based_exec_ctrl;
918 u32 cpu_based_2nd_exec_ctrl;
923 static struct vmx_capability {
928 #define VMX_SEGMENT_FIELD(seg) \
929 [VCPU_SREG_##seg] = { \
930 .selector = GUEST_##seg##_SELECTOR, \
931 .base = GUEST_##seg##_BASE, \
932 .limit = GUEST_##seg##_LIMIT, \
933 .ar_bytes = GUEST_##seg##_AR_BYTES, \
936 static const struct kvm_vmx_segment_field {
941 } kvm_vmx_segment_fields[] = {
942 VMX_SEGMENT_FIELD(CS),
943 VMX_SEGMENT_FIELD(DS),
944 VMX_SEGMENT_FIELD(ES),
945 VMX_SEGMENT_FIELD(FS),
946 VMX_SEGMENT_FIELD(GS),
947 VMX_SEGMENT_FIELD(SS),
948 VMX_SEGMENT_FIELD(TR),
949 VMX_SEGMENT_FIELD(LDTR),
952 static u64 host_efer;
954 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
957 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
958 * away by decrementing the array size.
960 static const u32 vmx_msr_index[] = {
962 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
964 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
967 static inline bool is_page_fault(u32 intr_info)
969 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
970 INTR_INFO_VALID_MASK)) ==
971 (INTR_TYPE_HARD_EXCEPTION | PF_VECTOR | INTR_INFO_VALID_MASK);
974 static inline bool is_no_device(u32 intr_info)
976 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
977 INTR_INFO_VALID_MASK)) ==
978 (INTR_TYPE_HARD_EXCEPTION | NM_VECTOR | INTR_INFO_VALID_MASK);
981 static inline bool is_invalid_opcode(u32 intr_info)
983 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
984 INTR_INFO_VALID_MASK)) ==
985 (INTR_TYPE_HARD_EXCEPTION | UD_VECTOR | INTR_INFO_VALID_MASK);
988 static inline bool is_external_interrupt(u32 intr_info)
990 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
991 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
994 static inline bool is_machine_check(u32 intr_info)
996 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
997 INTR_INFO_VALID_MASK)) ==
998 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1001 static inline bool cpu_has_vmx_msr_bitmap(void)
1003 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1006 static inline bool cpu_has_vmx_tpr_shadow(void)
1008 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1011 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1013 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1016 static inline bool cpu_has_secondary_exec_ctrls(void)
1018 return vmcs_config.cpu_based_exec_ctrl &
1019 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1022 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1024 return vmcs_config.cpu_based_2nd_exec_ctrl &
1025 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1028 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1030 return vmcs_config.cpu_based_2nd_exec_ctrl &
1031 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1034 static inline bool cpu_has_vmx_apic_register_virt(void)
1036 return vmcs_config.cpu_based_2nd_exec_ctrl &
1037 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1040 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1042 return vmcs_config.cpu_based_2nd_exec_ctrl &
1043 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1046 static inline bool cpu_has_vmx_posted_intr(void)
1048 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1049 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1052 static inline bool cpu_has_vmx_apicv(void)
1054 return cpu_has_vmx_apic_register_virt() &&
1055 cpu_has_vmx_virtual_intr_delivery() &&
1056 cpu_has_vmx_posted_intr();
1059 static inline bool cpu_has_vmx_flexpriority(void)
1061 return cpu_has_vmx_tpr_shadow() &&
1062 cpu_has_vmx_virtualize_apic_accesses();
1065 static inline bool cpu_has_vmx_ept_execute_only(void)
1067 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1070 static inline bool cpu_has_vmx_ept_2m_page(void)
1072 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1075 static inline bool cpu_has_vmx_ept_1g_page(void)
1077 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1080 static inline bool cpu_has_vmx_ept_4levels(void)
1082 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1085 static inline bool cpu_has_vmx_ept_ad_bits(void)
1087 return vmx_capability.ept & VMX_EPT_AD_BIT;
1090 static inline bool cpu_has_vmx_invept_context(void)
1092 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1095 static inline bool cpu_has_vmx_invept_global(void)
1097 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1100 static inline bool cpu_has_vmx_invvpid_single(void)
1102 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1105 static inline bool cpu_has_vmx_invvpid_global(void)
1107 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1110 static inline bool cpu_has_vmx_ept(void)
1112 return vmcs_config.cpu_based_2nd_exec_ctrl &
1113 SECONDARY_EXEC_ENABLE_EPT;
1116 static inline bool cpu_has_vmx_unrestricted_guest(void)
1118 return vmcs_config.cpu_based_2nd_exec_ctrl &
1119 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1122 static inline bool cpu_has_vmx_ple(void)
1124 return vmcs_config.cpu_based_2nd_exec_ctrl &
1125 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1128 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1130 return flexpriority_enabled && lapic_in_kernel(vcpu);
1133 static inline bool cpu_has_vmx_vpid(void)
1135 return vmcs_config.cpu_based_2nd_exec_ctrl &
1136 SECONDARY_EXEC_ENABLE_VPID;
1139 static inline bool cpu_has_vmx_rdtscp(void)
1141 return vmcs_config.cpu_based_2nd_exec_ctrl &
1142 SECONDARY_EXEC_RDTSCP;
1145 static inline bool cpu_has_vmx_invpcid(void)
1147 return vmcs_config.cpu_based_2nd_exec_ctrl &
1148 SECONDARY_EXEC_ENABLE_INVPCID;
1151 static inline bool cpu_has_virtual_nmis(void)
1153 return vmcs_config.pin_based_exec_ctrl & PIN_BASED_VIRTUAL_NMIS;
1156 static inline bool cpu_has_vmx_wbinvd_exit(void)
1158 return vmcs_config.cpu_based_2nd_exec_ctrl &
1159 SECONDARY_EXEC_WBINVD_EXITING;
1162 static inline bool cpu_has_vmx_shadow_vmcs(void)
1165 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1166 /* check if the cpu supports writing r/o exit information fields */
1167 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1170 return vmcs_config.cpu_based_2nd_exec_ctrl &
1171 SECONDARY_EXEC_SHADOW_VMCS;
1174 static inline bool cpu_has_vmx_pml(void)
1176 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1179 static inline bool cpu_has_vmx_tsc_scaling(void)
1181 return vmcs_config.cpu_based_2nd_exec_ctrl &
1182 SECONDARY_EXEC_TSC_SCALING;
1185 static inline bool report_flexpriority(void)
1187 return flexpriority_enabled;
1190 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1192 return vmcs12->cpu_based_vm_exec_control & bit;
1195 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1197 return (vmcs12->cpu_based_vm_exec_control &
1198 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1199 (vmcs12->secondary_vm_exec_control & bit);
1202 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1204 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1207 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1209 return vmcs12->pin_based_vm_exec_control &
1210 PIN_BASED_VMX_PREEMPTION_TIMER;
1213 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1215 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1218 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1220 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES) &&
1221 vmx_xsaves_supported();
1224 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1226 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1229 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1231 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1234 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1236 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1239 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1241 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1244 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1246 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1249 static inline bool is_nmi(u32 intr_info)
1251 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1252 == (INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK);
1255 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1257 unsigned long exit_qualification);
1258 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1259 struct vmcs12 *vmcs12,
1260 u32 reason, unsigned long qualification);
1262 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1266 for (i = 0; i < vmx->nmsrs; ++i)
1267 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1272 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1278 } operand = { vpid, 0, gva };
1280 asm volatile (__ex(ASM_VMX_INVVPID)
1281 /* CF==1 or ZF==1 --> rc = -1 */
1282 "; ja 1f ; ud2 ; 1:"
1283 : : "a"(&operand), "c"(ext) : "cc", "memory");
1286 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1290 } operand = {eptp, gpa};
1292 asm volatile (__ex(ASM_VMX_INVEPT)
1293 /* CF==1 or ZF==1 --> rc = -1 */
1294 "; ja 1f ; ud2 ; 1:\n"
1295 : : "a" (&operand), "c" (ext) : "cc", "memory");
1298 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1302 i = __find_msr_index(vmx, msr);
1304 return &vmx->guest_msrs[i];
1308 static void vmcs_clear(struct vmcs *vmcs)
1310 u64 phys_addr = __pa(vmcs);
1313 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1314 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1317 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1321 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1323 vmcs_clear(loaded_vmcs->vmcs);
1324 loaded_vmcs->cpu = -1;
1325 loaded_vmcs->launched = 0;
1328 static void vmcs_load(struct vmcs *vmcs)
1330 u64 phys_addr = __pa(vmcs);
1333 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1334 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1337 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1341 #ifdef CONFIG_KEXEC_CORE
1343 * This bitmap is used to indicate whether the vmclear
1344 * operation is enabled on all cpus. All disabled by
1347 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1349 static inline void crash_enable_local_vmclear(int cpu)
1351 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1354 static inline void crash_disable_local_vmclear(int cpu)
1356 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1359 static inline int crash_local_vmclear_enabled(int cpu)
1361 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1364 static void crash_vmclear_local_loaded_vmcss(void)
1366 int cpu = raw_smp_processor_id();
1367 struct loaded_vmcs *v;
1369 if (!crash_local_vmclear_enabled(cpu))
1372 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1373 loaded_vmcss_on_cpu_link)
1374 vmcs_clear(v->vmcs);
1377 static inline void crash_enable_local_vmclear(int cpu) { }
1378 static inline void crash_disable_local_vmclear(int cpu) { }
1379 #endif /* CONFIG_KEXEC_CORE */
1381 static void __loaded_vmcs_clear(void *arg)
1383 struct loaded_vmcs *loaded_vmcs = arg;
1384 int cpu = raw_smp_processor_id();
1386 if (loaded_vmcs->cpu != cpu)
1387 return; /* vcpu migration can race with cpu offline */
1388 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1389 per_cpu(current_vmcs, cpu) = NULL;
1390 crash_disable_local_vmclear(cpu);
1391 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1394 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1395 * is before setting loaded_vmcs->vcpu to -1 which is done in
1396 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1397 * then adds the vmcs into percpu list before it is deleted.
1401 loaded_vmcs_init(loaded_vmcs);
1402 crash_enable_local_vmclear(cpu);
1405 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1407 int cpu = loaded_vmcs->cpu;
1410 smp_call_function_single(cpu,
1411 __loaded_vmcs_clear, loaded_vmcs, 1);
1414 static inline void vpid_sync_vcpu_single(int vpid)
1419 if (cpu_has_vmx_invvpid_single())
1420 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1423 static inline void vpid_sync_vcpu_global(void)
1425 if (cpu_has_vmx_invvpid_global())
1426 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1429 static inline void vpid_sync_context(int vpid)
1431 if (cpu_has_vmx_invvpid_single())
1432 vpid_sync_vcpu_single(vpid);
1434 vpid_sync_vcpu_global();
1437 static inline void ept_sync_global(void)
1439 if (cpu_has_vmx_invept_global())
1440 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1443 static inline void ept_sync_context(u64 eptp)
1446 if (cpu_has_vmx_invept_context())
1447 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1453 static __always_inline unsigned long vmcs_readl(unsigned long field)
1455 unsigned long value;
1457 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1458 : "=a"(value) : "d"(field) : "cc");
1462 static __always_inline u16 vmcs_read16(unsigned long field)
1464 return vmcs_readl(field);
1467 static __always_inline u32 vmcs_read32(unsigned long field)
1469 return vmcs_readl(field);
1472 static __always_inline u64 vmcs_read64(unsigned long field)
1474 #ifdef CONFIG_X86_64
1475 return vmcs_readl(field);
1477 return vmcs_readl(field) | ((u64)vmcs_readl(field+1) << 32);
1481 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1483 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1484 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1488 static void vmcs_writel(unsigned long field, unsigned long value)
1492 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1493 : "=q"(error) : "a"(value), "d"(field) : "cc");
1494 if (unlikely(error))
1495 vmwrite_error(field, value);
1498 static void vmcs_write16(unsigned long field, u16 value)
1500 vmcs_writel(field, value);
1503 static void vmcs_write32(unsigned long field, u32 value)
1505 vmcs_writel(field, value);
1508 static void vmcs_write64(unsigned long field, u64 value)
1510 vmcs_writel(field, value);
1511 #ifndef CONFIG_X86_64
1513 vmcs_writel(field+1, value >> 32);
1517 static void vmcs_clear_bits(unsigned long field, u32 mask)
1519 vmcs_writel(field, vmcs_readl(field) & ~mask);
1522 static void vmcs_set_bits(unsigned long field, u32 mask)
1524 vmcs_writel(field, vmcs_readl(field) | mask);
1527 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1529 vmcs_write32(VM_ENTRY_CONTROLS, val);
1530 vmx->vm_entry_controls_shadow = val;
1533 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1535 if (vmx->vm_entry_controls_shadow != val)
1536 vm_entry_controls_init(vmx, val);
1539 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1541 return vmx->vm_entry_controls_shadow;
1545 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1547 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1550 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1552 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1555 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1557 vmcs_write32(VM_EXIT_CONTROLS, val);
1558 vmx->vm_exit_controls_shadow = val;
1561 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1563 if (vmx->vm_exit_controls_shadow != val)
1564 vm_exit_controls_init(vmx, val);
1567 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1569 return vmx->vm_exit_controls_shadow;
1573 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1575 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1578 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1580 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1583 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1585 vmx->segment_cache.bitmask = 0;
1588 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1592 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1594 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1595 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1596 vmx->segment_cache.bitmask = 0;
1598 ret = vmx->segment_cache.bitmask & mask;
1599 vmx->segment_cache.bitmask |= mask;
1603 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1605 u16 *p = &vmx->segment_cache.seg[seg].selector;
1607 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1608 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1612 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1614 ulong *p = &vmx->segment_cache.seg[seg].base;
1616 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1617 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1621 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1623 u32 *p = &vmx->segment_cache.seg[seg].limit;
1625 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1626 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1630 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1632 u32 *p = &vmx->segment_cache.seg[seg].ar;
1634 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1635 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1639 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1643 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1644 (1u << NM_VECTOR) | (1u << DB_VECTOR) | (1u << AC_VECTOR);
1645 if ((vcpu->guest_debug &
1646 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1647 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1648 eb |= 1u << BP_VECTOR;
1649 if (to_vmx(vcpu)->rmode.vm86_active)
1652 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1653 if (vcpu->fpu_active)
1654 eb &= ~(1u << NM_VECTOR);
1656 /* When we are running a nested L2 guest and L1 specified for it a
1657 * certain exception bitmap, we must trap the same exceptions and pass
1658 * them to L1. When running L2, we will only handle the exceptions
1659 * specified above if L1 did not want them.
1661 if (is_guest_mode(vcpu))
1662 eb |= get_vmcs12(vcpu)->exception_bitmap;
1664 vmcs_write32(EXCEPTION_BITMAP, eb);
1667 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1668 unsigned long entry, unsigned long exit)
1670 vm_entry_controls_clearbit(vmx, entry);
1671 vm_exit_controls_clearbit(vmx, exit);
1674 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1677 struct msr_autoload *m = &vmx->msr_autoload;
1681 if (cpu_has_load_ia32_efer) {
1682 clear_atomic_switch_msr_special(vmx,
1683 VM_ENTRY_LOAD_IA32_EFER,
1684 VM_EXIT_LOAD_IA32_EFER);
1688 case MSR_CORE_PERF_GLOBAL_CTRL:
1689 if (cpu_has_load_perf_global_ctrl) {
1690 clear_atomic_switch_msr_special(vmx,
1691 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1692 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1698 for (i = 0; i < m->nr; ++i)
1699 if (m->guest[i].index == msr)
1705 m->guest[i] = m->guest[m->nr];
1706 m->host[i] = m->host[m->nr];
1707 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1708 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1711 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1712 unsigned long entry, unsigned long exit,
1713 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1714 u64 guest_val, u64 host_val)
1716 vmcs_write64(guest_val_vmcs, guest_val);
1717 vmcs_write64(host_val_vmcs, host_val);
1718 vm_entry_controls_setbit(vmx, entry);
1719 vm_exit_controls_setbit(vmx, exit);
1722 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1723 u64 guest_val, u64 host_val)
1726 struct msr_autoload *m = &vmx->msr_autoload;
1730 if (cpu_has_load_ia32_efer) {
1731 add_atomic_switch_msr_special(vmx,
1732 VM_ENTRY_LOAD_IA32_EFER,
1733 VM_EXIT_LOAD_IA32_EFER,
1736 guest_val, host_val);
1740 case MSR_CORE_PERF_GLOBAL_CTRL:
1741 if (cpu_has_load_perf_global_ctrl) {
1742 add_atomic_switch_msr_special(vmx,
1743 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1744 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1745 GUEST_IA32_PERF_GLOBAL_CTRL,
1746 HOST_IA32_PERF_GLOBAL_CTRL,
1747 guest_val, host_val);
1751 case MSR_IA32_PEBS_ENABLE:
1752 /* PEBS needs a quiescent period after being disabled (to write
1753 * a record). Disabling PEBS through VMX MSR swapping doesn't
1754 * provide that period, so a CPU could write host's record into
1757 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1760 for (i = 0; i < m->nr; ++i)
1761 if (m->guest[i].index == msr)
1764 if (i == NR_AUTOLOAD_MSRS) {
1765 printk_once(KERN_WARNING "Not enough msr switch entries. "
1766 "Can't add msr %x\n", msr);
1768 } else if (i == m->nr) {
1770 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1771 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1774 m->guest[i].index = msr;
1775 m->guest[i].value = guest_val;
1776 m->host[i].index = msr;
1777 m->host[i].value = host_val;
1780 static void reload_tss(void)
1783 * VT restores TR but not its size. Useless.
1785 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1786 struct desc_struct *descs;
1788 descs = (void *)gdt->address;
1789 descs[GDT_ENTRY_TSS].type = 9; /* available TSS */
1793 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
1795 u64 guest_efer = vmx->vcpu.arch.efer;
1796 u64 ignore_bits = 0;
1800 * NX is needed to handle CR0.WP=1, CR4.SMEP=1. Testing
1801 * host CPUID is more efficient than testing guest CPUID
1802 * or CR4. Host SMEP is anyway a requirement for guest SMEP.
1804 if (boot_cpu_has(X86_FEATURE_SMEP))
1805 guest_efer |= EFER_NX;
1806 else if (!(guest_efer & EFER_NX))
1807 ignore_bits |= EFER_NX;
1811 * LMA and LME handled by hardware; SCE meaningless outside long mode.
1813 ignore_bits |= EFER_SCE;
1814 #ifdef CONFIG_X86_64
1815 ignore_bits |= EFER_LMA | EFER_LME;
1816 /* SCE is meaningful only in long mode on Intel */
1817 if (guest_efer & EFER_LMA)
1818 ignore_bits &= ~(u64)EFER_SCE;
1821 clear_atomic_switch_msr(vmx, MSR_EFER);
1824 * On EPT, we can't emulate NX, so we must switch EFER atomically.
1825 * On CPUs that support "load IA32_EFER", always switch EFER
1826 * atomically, since it's faster than switching it manually.
1828 if (cpu_has_load_ia32_efer ||
1829 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
1830 if (!(guest_efer & EFER_LMA))
1831 guest_efer &= ~EFER_LME;
1832 if (guest_efer != host_efer)
1833 add_atomic_switch_msr(vmx, MSR_EFER,
1834 guest_efer, host_efer);
1837 guest_efer &= ~ignore_bits;
1838 guest_efer |= host_efer & ignore_bits;
1840 vmx->guest_msrs[efer_offset].data = guest_efer;
1841 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
1847 static unsigned long segment_base(u16 selector)
1849 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
1850 struct desc_struct *d;
1851 unsigned long table_base;
1854 if (!(selector & ~3))
1857 table_base = gdt->address;
1859 if (selector & 4) { /* from ldt */
1860 u16 ldt_selector = kvm_read_ldt();
1862 if (!(ldt_selector & ~3))
1865 table_base = segment_base(ldt_selector);
1867 d = (struct desc_struct *)(table_base + (selector & ~7));
1868 v = get_desc_base(d);
1869 #ifdef CONFIG_X86_64
1870 if (d->s == 0 && (d->type == 2 || d->type == 9 || d->type == 11))
1871 v |= ((unsigned long)((struct ldttss_desc64 *)d)->base3) << 32;
1876 static inline unsigned long kvm_read_tr_base(void)
1879 asm("str %0" : "=g"(tr));
1880 return segment_base(tr);
1883 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
1885 struct vcpu_vmx *vmx = to_vmx(vcpu);
1888 if (vmx->host_state.loaded)
1891 vmx->host_state.loaded = 1;
1893 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
1894 * allow segment selectors with cpl > 0 or ti == 1.
1896 vmx->host_state.ldt_sel = kvm_read_ldt();
1897 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
1898 savesegment(fs, vmx->host_state.fs_sel);
1899 if (!(vmx->host_state.fs_sel & 7)) {
1900 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
1901 vmx->host_state.fs_reload_needed = 0;
1903 vmcs_write16(HOST_FS_SELECTOR, 0);
1904 vmx->host_state.fs_reload_needed = 1;
1906 savesegment(gs, vmx->host_state.gs_sel);
1907 if (!(vmx->host_state.gs_sel & 7))
1908 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
1910 vmcs_write16(HOST_GS_SELECTOR, 0);
1911 vmx->host_state.gs_ldt_reload_needed = 1;
1914 #ifdef CONFIG_X86_64
1915 savesegment(ds, vmx->host_state.ds_sel);
1916 savesegment(es, vmx->host_state.es_sel);
1919 #ifdef CONFIG_X86_64
1920 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
1921 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
1923 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
1924 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
1927 #ifdef CONFIG_X86_64
1928 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1929 if (is_long_mode(&vmx->vcpu))
1930 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1932 if (boot_cpu_has(X86_FEATURE_MPX))
1933 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1934 for (i = 0; i < vmx->save_nmsrs; ++i)
1935 kvm_set_shared_msr(vmx->guest_msrs[i].index,
1936 vmx->guest_msrs[i].data,
1937 vmx->guest_msrs[i].mask);
1940 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
1942 if (!vmx->host_state.loaded)
1945 ++vmx->vcpu.stat.host_state_reload;
1946 vmx->host_state.loaded = 0;
1947 #ifdef CONFIG_X86_64
1948 if (is_long_mode(&vmx->vcpu))
1949 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
1951 if (vmx->host_state.gs_ldt_reload_needed) {
1952 kvm_load_ldt(vmx->host_state.ldt_sel);
1953 #ifdef CONFIG_X86_64
1954 load_gs_index(vmx->host_state.gs_sel);
1956 loadsegment(gs, vmx->host_state.gs_sel);
1959 if (vmx->host_state.fs_reload_needed)
1960 loadsegment(fs, vmx->host_state.fs_sel);
1961 #ifdef CONFIG_X86_64
1962 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
1963 loadsegment(ds, vmx->host_state.ds_sel);
1964 loadsegment(es, vmx->host_state.es_sel);
1968 #ifdef CONFIG_X86_64
1969 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
1971 if (vmx->host_state.msr_host_bndcfgs)
1972 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
1974 * If the FPU is not active (through the host task or
1975 * the guest vcpu), then restore the cr0.TS bit.
1977 if (!fpregs_active() && !vmx->vcpu.guest_fpu_loaded)
1979 load_gdt(this_cpu_ptr(&host_gdt));
1982 static void vmx_load_host_state(struct vcpu_vmx *vmx)
1985 __vmx_load_host_state(vmx);
1989 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
1991 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
1992 struct pi_desc old, new;
1995 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
1996 !irq_remapping_cap(IRQ_POSTING_CAP))
2000 old.control = new.control = pi_desc->control;
2003 * If 'nv' field is POSTED_INTR_WAKEUP_VECTOR, there
2004 * are two possible cases:
2005 * 1. After running 'pre_block', context switch
2006 * happened. For this case, 'sn' was set in
2007 * vmx_vcpu_put(), so we need to clear it here.
2008 * 2. After running 'pre_block', we were blocked,
2009 * and woken up by some other guy. For this case,
2010 * we don't need to do anything, 'pi_post_block'
2011 * will do everything for us. However, we cannot
2012 * check whether it is case #1 or case #2 here
2013 * (maybe, not needed), so we also clear sn here,
2014 * I think it is not a big deal.
2016 if (pi_desc->nv != POSTED_INTR_WAKEUP_VECTOR) {
2017 if (vcpu->cpu != cpu) {
2018 dest = cpu_physical_id(cpu);
2020 if (x2apic_enabled())
2023 new.ndst = (dest << 8) & 0xFF00;
2026 /* set 'NV' to 'notification vector' */
2027 new.nv = POSTED_INTR_VECTOR;
2030 /* Allow posting non-urgent interrupts */
2032 } while (cmpxchg64(&pi_desc->control, old.control,
2033 new.control) != old.control);
2036 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2037 * vcpu mutex is already taken.
2039 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2041 struct vcpu_vmx *vmx = to_vmx(vcpu);
2042 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
2045 kvm_cpu_vmxon(phys_addr);
2046 else if (vmx->loaded_vmcs->cpu != cpu)
2047 loaded_vmcs_clear(vmx->loaded_vmcs);
2049 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2050 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2051 vmcs_load(vmx->loaded_vmcs->vmcs);
2054 if (vmx->loaded_vmcs->cpu != cpu) {
2055 struct desc_ptr *gdt = this_cpu_ptr(&host_gdt);
2056 unsigned long sysenter_esp;
2058 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2059 local_irq_disable();
2060 crash_disable_local_vmclear(cpu);
2063 * Read loaded_vmcs->cpu should be before fetching
2064 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2065 * See the comments in __loaded_vmcs_clear().
2069 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2070 &per_cpu(loaded_vmcss_on_cpu, cpu));
2071 crash_enable_local_vmclear(cpu);
2075 * Linux uses per-cpu TSS and GDT, so set these when switching
2078 vmcs_writel(HOST_TR_BASE, kvm_read_tr_base()); /* 22.2.4 */
2079 vmcs_writel(HOST_GDTR_BASE, gdt->address); /* 22.2.4 */
2081 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2082 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2084 vmx->loaded_vmcs->cpu = cpu;
2087 /* Setup TSC multiplier */
2088 if (kvm_has_tsc_control &&
2089 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio) {
2090 vmx->current_tsc_ratio = vcpu->arch.tsc_scaling_ratio;
2091 vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2094 vmx_vcpu_pi_load(vcpu, cpu);
2097 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2099 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2101 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2102 !irq_remapping_cap(IRQ_POSTING_CAP))
2105 /* Set SN when the vCPU is preempted */
2106 if (vcpu->preempted)
2110 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2112 vmx_vcpu_pi_put(vcpu);
2114 __vmx_load_host_state(to_vmx(vcpu));
2115 if (!vmm_exclusive) {
2116 __loaded_vmcs_clear(to_vmx(vcpu)->loaded_vmcs);
2122 static void vmx_fpu_activate(struct kvm_vcpu *vcpu)
2126 if (vcpu->fpu_active)
2128 vcpu->fpu_active = 1;
2129 cr0 = vmcs_readl(GUEST_CR0);
2130 cr0 &= ~(X86_CR0_TS | X86_CR0_MP);
2131 cr0 |= kvm_read_cr0_bits(vcpu, X86_CR0_TS | X86_CR0_MP);
2132 vmcs_writel(GUEST_CR0, cr0);
2133 update_exception_bitmap(vcpu);
2134 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
2135 if (is_guest_mode(vcpu))
2136 vcpu->arch.cr0_guest_owned_bits &=
2137 ~get_vmcs12(vcpu)->cr0_guest_host_mask;
2138 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2141 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2144 * Return the cr0 value that a nested guest would read. This is a combination
2145 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2146 * its hypervisor (cr0_read_shadow).
2148 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2150 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2151 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2153 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2155 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2156 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2159 static void vmx_fpu_deactivate(struct kvm_vcpu *vcpu)
2161 /* Note that there is no vcpu->fpu_active = 0 here. The caller must
2162 * set this *before* calling this function.
2164 vmx_decache_cr0_guest_bits(vcpu);
2165 vmcs_set_bits(GUEST_CR0, X86_CR0_TS | X86_CR0_MP);
2166 update_exception_bitmap(vcpu);
2167 vcpu->arch.cr0_guest_owned_bits = 0;
2168 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
2169 if (is_guest_mode(vcpu)) {
2171 * L1's specified read shadow might not contain the TS bit,
2172 * so now that we turned on shadowing of this bit, we need to
2173 * set this bit of the shadow. Like in nested_vmx_run we need
2174 * nested_read_cr0(vmcs12), but vmcs12->guest_cr0 is not yet
2175 * up-to-date here because we just decached cr0.TS (and we'll
2176 * only update vmcs12->guest_cr0 on nested exit).
2178 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2179 vmcs12->guest_cr0 = (vmcs12->guest_cr0 & ~X86_CR0_TS) |
2180 (vcpu->arch.cr0 & X86_CR0_TS);
2181 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
2183 vmcs_writel(CR0_READ_SHADOW, vcpu->arch.cr0);
2186 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2188 unsigned long rflags, save_rflags;
2190 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2191 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2192 rflags = vmcs_readl(GUEST_RFLAGS);
2193 if (to_vmx(vcpu)->rmode.vm86_active) {
2194 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2195 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2196 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2198 to_vmx(vcpu)->rflags = rflags;
2200 return to_vmx(vcpu)->rflags;
2203 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2205 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2206 to_vmx(vcpu)->rflags = rflags;
2207 if (to_vmx(vcpu)->rmode.vm86_active) {
2208 to_vmx(vcpu)->rmode.save_rflags = rflags;
2209 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2211 vmcs_writel(GUEST_RFLAGS, rflags);
2214 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2216 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2219 if (interruptibility & GUEST_INTR_STATE_STI)
2220 ret |= KVM_X86_SHADOW_INT_STI;
2221 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2222 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2227 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2229 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2230 u32 interruptibility = interruptibility_old;
2232 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2234 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2235 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2236 else if (mask & KVM_X86_SHADOW_INT_STI)
2237 interruptibility |= GUEST_INTR_STATE_STI;
2239 if ((interruptibility != interruptibility_old))
2240 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2243 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2247 rip = kvm_rip_read(vcpu);
2248 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2249 kvm_rip_write(vcpu, rip);
2251 /* skipping an emulated instruction also counts */
2252 vmx_set_interrupt_shadow(vcpu, 0);
2256 * KVM wants to inject page-faults which it got to the guest. This function
2257 * checks whether in a nested guest, we need to inject them to L1 or L2.
2259 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned nr)
2261 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2263 if (!(vmcs12->exception_bitmap & (1u << nr)))
2266 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
2267 vmcs_read32(VM_EXIT_INTR_INFO),
2268 vmcs_readl(EXIT_QUALIFICATION));
2272 static void vmx_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
2273 bool has_error_code, u32 error_code,
2276 struct vcpu_vmx *vmx = to_vmx(vcpu);
2277 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2279 if (!reinject && is_guest_mode(vcpu) &&
2280 nested_vmx_check_exception(vcpu, nr))
2283 if (has_error_code) {
2284 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2285 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2288 if (vmx->rmode.vm86_active) {
2290 if (kvm_exception_is_soft(nr))
2291 inc_eip = vcpu->arch.event_exit_inst_len;
2292 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2293 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2297 if (kvm_exception_is_soft(nr)) {
2298 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2299 vmx->vcpu.arch.event_exit_inst_len);
2300 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2302 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2304 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2307 static bool vmx_rdtscp_supported(void)
2309 return cpu_has_vmx_rdtscp();
2312 static bool vmx_invpcid_supported(void)
2314 return cpu_has_vmx_invpcid() && enable_ept;
2318 * Swap MSR entry in host/guest MSR entry array.
2320 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2322 struct shared_msr_entry tmp;
2324 tmp = vmx->guest_msrs[to];
2325 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2326 vmx->guest_msrs[from] = tmp;
2329 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2331 unsigned long *msr_bitmap;
2333 if (is_guest_mode(vcpu))
2334 msr_bitmap = vmx_msr_bitmap_nested;
2335 else if (vcpu->arch.apic_base & X2APIC_ENABLE) {
2336 if (is_long_mode(vcpu))
2337 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2339 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2341 if (is_long_mode(vcpu))
2342 msr_bitmap = vmx_msr_bitmap_longmode;
2344 msr_bitmap = vmx_msr_bitmap_legacy;
2347 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2351 * Set up the vmcs to automatically save and restore system
2352 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2353 * mode, as fiddling with msrs is very expensive.
2355 static void setup_msrs(struct vcpu_vmx *vmx)
2357 int save_nmsrs, index;
2360 #ifdef CONFIG_X86_64
2361 if (is_long_mode(&vmx->vcpu)) {
2362 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2364 move_msr_up(vmx, index, save_nmsrs++);
2365 index = __find_msr_index(vmx, MSR_LSTAR);
2367 move_msr_up(vmx, index, save_nmsrs++);
2368 index = __find_msr_index(vmx, MSR_CSTAR);
2370 move_msr_up(vmx, index, save_nmsrs++);
2371 index = __find_msr_index(vmx, MSR_TSC_AUX);
2372 if (index >= 0 && guest_cpuid_has_rdtscp(&vmx->vcpu))
2373 move_msr_up(vmx, index, save_nmsrs++);
2375 * MSR_STAR is only needed on long mode guests, and only
2376 * if efer.sce is enabled.
2378 index = __find_msr_index(vmx, MSR_STAR);
2379 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2380 move_msr_up(vmx, index, save_nmsrs++);
2383 index = __find_msr_index(vmx, MSR_EFER);
2384 if (index >= 0 && update_transition_efer(vmx, index))
2385 move_msr_up(vmx, index, save_nmsrs++);
2387 vmx->save_nmsrs = save_nmsrs;
2389 if (cpu_has_vmx_msr_bitmap())
2390 vmx_set_msr_bitmap(&vmx->vcpu);
2394 * reads and returns guest's timestamp counter "register"
2395 * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2396 * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2398 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2400 u64 host_tsc, tsc_offset;
2403 tsc_offset = vmcs_read64(TSC_OFFSET);
2404 return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2408 * Like guest_read_tsc, but always returns L1's notion of the timestamp
2409 * counter, even if a nested guest (L2) is currently running.
2411 static u64 vmx_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
2415 tsc_offset = is_guest_mode(vcpu) ?
2416 to_vmx(vcpu)->nested.vmcs01_tsc_offset :
2417 vmcs_read64(TSC_OFFSET);
2418 return host_tsc + tsc_offset;
2421 static u64 vmx_read_tsc_offset(struct kvm_vcpu *vcpu)
2423 return vmcs_read64(TSC_OFFSET);
2427 * writes 'offset' into guest's timestamp counter offset register
2429 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2431 if (is_guest_mode(vcpu)) {
2433 * We're here if L1 chose not to trap WRMSR to TSC. According
2434 * to the spec, this should set L1's TSC; The offset that L1
2435 * set for L2 remains unchanged, and still needs to be added
2436 * to the newly set TSC to get L2's TSC.
2438 struct vmcs12 *vmcs12;
2439 to_vmx(vcpu)->nested.vmcs01_tsc_offset = offset;
2440 /* recalculate vmcs02.TSC_OFFSET: */
2441 vmcs12 = get_vmcs12(vcpu);
2442 vmcs_write64(TSC_OFFSET, offset +
2443 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2444 vmcs12->tsc_offset : 0));
2446 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2447 vmcs_read64(TSC_OFFSET), offset);
2448 vmcs_write64(TSC_OFFSET, offset);
2452 static void vmx_adjust_tsc_offset_guest(struct kvm_vcpu *vcpu, s64 adjustment)
2454 u64 offset = vmcs_read64(TSC_OFFSET);
2456 vmcs_write64(TSC_OFFSET, offset + adjustment);
2457 if (is_guest_mode(vcpu)) {
2458 /* Even when running L2, the adjustment needs to apply to L1 */
2459 to_vmx(vcpu)->nested.vmcs01_tsc_offset += adjustment;
2461 trace_kvm_write_tsc_offset(vcpu->vcpu_id, offset,
2462 offset + adjustment);
2465 static bool guest_cpuid_has_vmx(struct kvm_vcpu *vcpu)
2467 struct kvm_cpuid_entry2 *best = kvm_find_cpuid_entry(vcpu, 1, 0);
2468 return best && (best->ecx & (1 << (X86_FEATURE_VMX & 31)));
2472 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2473 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2474 * all guests if the "nested" module option is off, and can also be disabled
2475 * for a single guest by disabling its VMX cpuid bit.
2477 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2479 return nested && guest_cpuid_has_vmx(vcpu);
2483 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2484 * returned for the various VMX controls MSRs when nested VMX is enabled.
2485 * The same values should also be used to verify that vmcs12 control fields are
2486 * valid during nested entry from L1 to L2.
2487 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2488 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2489 * bit in the high half is on if the corresponding bit in the control field
2490 * may be on. See also vmx_control_verify().
2492 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2495 * Note that as a general rule, the high half of the MSRs (bits in
2496 * the control fields which may be 1) should be initialized by the
2497 * intersection of the underlying hardware's MSR (i.e., features which
2498 * can be supported) and the list of features we want to expose -
2499 * because they are known to be properly supported in our code.
2500 * Also, usually, the low half of the MSRs (bits which must be 1) can
2501 * be set to 0, meaning that L1 may turn off any of these bits. The
2502 * reason is that if one of these bits is necessary, it will appear
2503 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2504 * fields of vmcs01 and vmcs02, will turn these bits off - and
2505 * nested_vmx_exit_handled() will not pass related exits to L1.
2506 * These rules have exceptions below.
2509 /* pin-based controls */
2510 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2511 vmx->nested.nested_vmx_pinbased_ctls_low,
2512 vmx->nested.nested_vmx_pinbased_ctls_high);
2513 vmx->nested.nested_vmx_pinbased_ctls_low |=
2514 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2515 vmx->nested.nested_vmx_pinbased_ctls_high &=
2516 PIN_BASED_EXT_INTR_MASK |
2517 PIN_BASED_NMI_EXITING |
2518 PIN_BASED_VIRTUAL_NMIS;
2519 vmx->nested.nested_vmx_pinbased_ctls_high |=
2520 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2521 PIN_BASED_VMX_PREEMPTION_TIMER;
2522 if (vmx_cpu_uses_apicv(&vmx->vcpu))
2523 vmx->nested.nested_vmx_pinbased_ctls_high |=
2524 PIN_BASED_POSTED_INTR;
2527 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2528 vmx->nested.nested_vmx_exit_ctls_low,
2529 vmx->nested.nested_vmx_exit_ctls_high);
2530 vmx->nested.nested_vmx_exit_ctls_low =
2531 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2533 vmx->nested.nested_vmx_exit_ctls_high &=
2534 #ifdef CONFIG_X86_64
2535 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2537 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2538 vmx->nested.nested_vmx_exit_ctls_high |=
2539 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2540 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2541 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2543 if (kvm_mpx_supported())
2544 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2546 /* We support free control of debug control saving. */
2547 vmx->nested.nested_vmx_true_exit_ctls_low =
2548 vmx->nested.nested_vmx_exit_ctls_low &
2549 ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2551 /* entry controls */
2552 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2553 vmx->nested.nested_vmx_entry_ctls_low,
2554 vmx->nested.nested_vmx_entry_ctls_high);
2555 vmx->nested.nested_vmx_entry_ctls_low =
2556 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2557 vmx->nested.nested_vmx_entry_ctls_high &=
2558 #ifdef CONFIG_X86_64
2559 VM_ENTRY_IA32E_MODE |
2561 VM_ENTRY_LOAD_IA32_PAT;
2562 vmx->nested.nested_vmx_entry_ctls_high |=
2563 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2564 if (kvm_mpx_supported())
2565 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2567 /* We support free control of debug control loading. */
2568 vmx->nested.nested_vmx_true_entry_ctls_low =
2569 vmx->nested.nested_vmx_entry_ctls_low &
2570 ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2572 /* cpu-based controls */
2573 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2574 vmx->nested.nested_vmx_procbased_ctls_low,
2575 vmx->nested.nested_vmx_procbased_ctls_high);
2576 vmx->nested.nested_vmx_procbased_ctls_low =
2577 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2578 vmx->nested.nested_vmx_procbased_ctls_high &=
2579 CPU_BASED_VIRTUAL_INTR_PENDING |
2580 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2581 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2582 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2583 CPU_BASED_CR3_STORE_EXITING |
2584 #ifdef CONFIG_X86_64
2585 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2587 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2588 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2589 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2590 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2591 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2593 * We can allow some features even when not supported by the
2594 * hardware. For example, L1 can specify an MSR bitmap - and we
2595 * can use it to avoid exits to L1 - even when L0 runs L2
2596 * without MSR bitmaps.
2598 vmx->nested.nested_vmx_procbased_ctls_high |=
2599 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2600 CPU_BASED_USE_MSR_BITMAPS;
2602 /* We support free control of CR3 access interception. */
2603 vmx->nested.nested_vmx_true_procbased_ctls_low =
2604 vmx->nested.nested_vmx_procbased_ctls_low &
2605 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2607 /* secondary cpu-based controls */
2608 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2609 vmx->nested.nested_vmx_secondary_ctls_low,
2610 vmx->nested.nested_vmx_secondary_ctls_high);
2611 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2612 vmx->nested.nested_vmx_secondary_ctls_high &=
2613 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2614 SECONDARY_EXEC_RDTSCP |
2615 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2616 SECONDARY_EXEC_ENABLE_VPID |
2617 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2618 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2619 SECONDARY_EXEC_WBINVD_EXITING |
2620 SECONDARY_EXEC_XSAVES |
2621 SECONDARY_EXEC_PCOMMIT;
2624 /* nested EPT: emulate EPT also to L1 */
2625 vmx->nested.nested_vmx_secondary_ctls_high |=
2626 SECONDARY_EXEC_ENABLE_EPT;
2627 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2628 VMX_EPTP_WB_BIT | VMX_EPT_2MB_PAGE_BIT |
2630 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2632 * For nested guests, we don't do anything specific
2633 * for single context invalidation. Hence, only advertise
2634 * support for global context invalidation.
2636 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT;
2638 vmx->nested.nested_vmx_ept_caps = 0;
2641 * Old versions of KVM use the single-context version without
2642 * checking for support, so declare that it is supported even
2643 * though it is treated as global context. The alternative is
2644 * not failing the single-context invvpid, and it is worse.
2647 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2648 VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT |
2649 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
2651 vmx->nested.nested_vmx_vpid_caps = 0;
2653 if (enable_unrestricted_guest)
2654 vmx->nested.nested_vmx_secondary_ctls_high |=
2655 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2657 /* miscellaneous data */
2658 rdmsr(MSR_IA32_VMX_MISC,
2659 vmx->nested.nested_vmx_misc_low,
2660 vmx->nested.nested_vmx_misc_high);
2661 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2662 vmx->nested.nested_vmx_misc_low |=
2663 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2664 VMX_MISC_ACTIVITY_HLT;
2665 vmx->nested.nested_vmx_misc_high = 0;
2668 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2671 * Bits 0 in high must be 0, and bits 1 in low must be 1.
2673 return ((control & high) | low) == control;
2676 static inline u64 vmx_control_msr(u32 low, u32 high)
2678 return low | ((u64)high << 32);
2681 /* Returns 0 on success, non-0 otherwise. */
2682 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
2684 struct vcpu_vmx *vmx = to_vmx(vcpu);
2686 switch (msr_index) {
2687 case MSR_IA32_VMX_BASIC:
2689 * This MSR reports some information about VMX support. We
2690 * should return information about the VMX we emulate for the
2691 * guest, and the VMCS structure we give it - not about the
2692 * VMX support of the underlying hardware.
2694 *pdata = VMCS12_REVISION | VMX_BASIC_TRUE_CTLS |
2695 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2696 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2698 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2699 case MSR_IA32_VMX_PINBASED_CTLS:
2700 *pdata = vmx_control_msr(
2701 vmx->nested.nested_vmx_pinbased_ctls_low,
2702 vmx->nested.nested_vmx_pinbased_ctls_high);
2704 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2705 *pdata = vmx_control_msr(
2706 vmx->nested.nested_vmx_true_procbased_ctls_low,
2707 vmx->nested.nested_vmx_procbased_ctls_high);
2709 case MSR_IA32_VMX_PROCBASED_CTLS:
2710 *pdata = vmx_control_msr(
2711 vmx->nested.nested_vmx_procbased_ctls_low,
2712 vmx->nested.nested_vmx_procbased_ctls_high);
2714 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2715 *pdata = vmx_control_msr(
2716 vmx->nested.nested_vmx_true_exit_ctls_low,
2717 vmx->nested.nested_vmx_exit_ctls_high);
2719 case MSR_IA32_VMX_EXIT_CTLS:
2720 *pdata = vmx_control_msr(
2721 vmx->nested.nested_vmx_exit_ctls_low,
2722 vmx->nested.nested_vmx_exit_ctls_high);
2724 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2725 *pdata = vmx_control_msr(
2726 vmx->nested.nested_vmx_true_entry_ctls_low,
2727 vmx->nested.nested_vmx_entry_ctls_high);
2729 case MSR_IA32_VMX_ENTRY_CTLS:
2730 *pdata = vmx_control_msr(
2731 vmx->nested.nested_vmx_entry_ctls_low,
2732 vmx->nested.nested_vmx_entry_ctls_high);
2734 case MSR_IA32_VMX_MISC:
2735 *pdata = vmx_control_msr(
2736 vmx->nested.nested_vmx_misc_low,
2737 vmx->nested.nested_vmx_misc_high);
2740 * These MSRs specify bits which the guest must keep fixed (on or off)
2741 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2742 * We picked the standard core2 setting.
2744 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2745 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2746 case MSR_IA32_VMX_CR0_FIXED0:
2747 *pdata = VMXON_CR0_ALWAYSON;
2749 case MSR_IA32_VMX_CR0_FIXED1:
2752 case MSR_IA32_VMX_CR4_FIXED0:
2753 *pdata = VMXON_CR4_ALWAYSON;
2755 case MSR_IA32_VMX_CR4_FIXED1:
2758 case MSR_IA32_VMX_VMCS_ENUM:
2759 *pdata = 0x2e; /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2761 case MSR_IA32_VMX_PROCBASED_CTLS2:
2762 *pdata = vmx_control_msr(
2763 vmx->nested.nested_vmx_secondary_ctls_low,
2764 vmx->nested.nested_vmx_secondary_ctls_high);
2766 case MSR_IA32_VMX_EPT_VPID_CAP:
2767 /* Currently, no nested vpid support */
2768 *pdata = vmx->nested.nested_vmx_ept_caps |
2769 ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
2779 * Reads an msr value (of 'msr_index') into 'pdata'.
2780 * Returns 0 on success, non-0 otherwise.
2781 * Assumes vcpu_load() was already called.
2783 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2785 struct shared_msr_entry *msr;
2787 switch (msr_info->index) {
2788 #ifdef CONFIG_X86_64
2790 msr_info->data = vmcs_readl(GUEST_FS_BASE);
2793 msr_info->data = vmcs_readl(GUEST_GS_BASE);
2795 case MSR_KERNEL_GS_BASE:
2796 vmx_load_host_state(to_vmx(vcpu));
2797 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
2801 return kvm_get_msr_common(vcpu, msr_info);
2803 msr_info->data = guest_read_tsc(vcpu);
2805 case MSR_IA32_SYSENTER_CS:
2806 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
2808 case MSR_IA32_SYSENTER_EIP:
2809 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
2811 case MSR_IA32_SYSENTER_ESP:
2812 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
2814 case MSR_IA32_BNDCFGS:
2815 if (!kvm_mpx_supported() ||
2816 (!msr_info->host_initiated && !guest_cpuid_has_mpx(vcpu)))
2818 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
2820 case MSR_IA32_FEATURE_CONTROL:
2821 if (!nested_vmx_allowed(vcpu))
2823 msr_info->data = to_vmx(vcpu)->nested.msr_ia32_feature_control;
2825 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2826 if (!nested_vmx_allowed(vcpu))
2828 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
2830 if (!vmx_xsaves_supported())
2832 msr_info->data = vcpu->arch.ia32_xss;
2835 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
2837 /* Otherwise falls through */
2839 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
2841 msr_info->data = msr->data;
2844 return kvm_get_msr_common(vcpu, msr_info);
2850 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
2853 * Writes msr value into into the appropriate "register".
2854 * Returns 0 on success, non-0 otherwise.
2855 * Assumes vcpu_load() was already called.
2857 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
2859 struct vcpu_vmx *vmx = to_vmx(vcpu);
2860 struct shared_msr_entry *msr;
2862 u32 msr_index = msr_info->index;
2863 u64 data = msr_info->data;
2865 switch (msr_index) {
2867 ret = kvm_set_msr_common(vcpu, msr_info);
2869 #ifdef CONFIG_X86_64
2871 vmx_segment_cache_clear(vmx);
2872 vmcs_writel(GUEST_FS_BASE, data);
2875 vmx_segment_cache_clear(vmx);
2876 vmcs_writel(GUEST_GS_BASE, data);
2878 case MSR_KERNEL_GS_BASE:
2879 vmx_load_host_state(vmx);
2880 vmx->msr_guest_kernel_gs_base = data;
2883 case MSR_IA32_SYSENTER_CS:
2884 vmcs_write32(GUEST_SYSENTER_CS, data);
2886 case MSR_IA32_SYSENTER_EIP:
2887 vmcs_writel(GUEST_SYSENTER_EIP, data);
2889 case MSR_IA32_SYSENTER_ESP:
2890 vmcs_writel(GUEST_SYSENTER_ESP, data);
2892 case MSR_IA32_BNDCFGS:
2893 if (!kvm_mpx_supported() ||
2894 (!msr_info->host_initiated && !guest_cpuid_has_mpx(vcpu)))
2896 if (is_noncanonical_address(data & PAGE_MASK) ||
2897 (data & MSR_IA32_BNDCFGS_RSVD))
2899 vmcs_write64(GUEST_BNDCFGS, data);
2902 kvm_write_tsc(vcpu, msr_info);
2904 case MSR_IA32_CR_PAT:
2905 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
2906 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
2908 vmcs_write64(GUEST_IA32_PAT, data);
2909 vcpu->arch.pat = data;
2912 ret = kvm_set_msr_common(vcpu, msr_info);
2914 case MSR_IA32_TSC_ADJUST:
2915 ret = kvm_set_msr_common(vcpu, msr_info);
2917 case MSR_IA32_FEATURE_CONTROL:
2918 if (!nested_vmx_allowed(vcpu) ||
2919 (to_vmx(vcpu)->nested.msr_ia32_feature_control &
2920 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
2922 vmx->nested.msr_ia32_feature_control = data;
2923 if (msr_info->host_initiated && data == 0)
2924 vmx_leave_nested(vcpu);
2926 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
2927 return 1; /* they are read-only */
2929 if (!vmx_xsaves_supported())
2932 * The only supported bit as of Skylake is bit 8, but
2933 * it is not supported on KVM.
2937 vcpu->arch.ia32_xss = data;
2938 if (vcpu->arch.ia32_xss != host_xss)
2939 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
2940 vcpu->arch.ia32_xss, host_xss);
2942 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
2945 if (!guest_cpuid_has_rdtscp(vcpu) && !msr_info->host_initiated)
2947 /* Check reserved bit, higher 32 bits should be zero */
2948 if ((data >> 32) != 0)
2950 /* Otherwise falls through */
2952 msr = find_msr_entry(vmx, msr_index);
2954 u64 old_msr_data = msr->data;
2956 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
2958 ret = kvm_set_shared_msr(msr->index, msr->data,
2962 msr->data = old_msr_data;
2966 ret = kvm_set_msr_common(vcpu, msr_info);
2972 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
2974 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
2977 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
2980 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
2982 case VCPU_EXREG_PDPTR:
2984 ept_save_pdptrs(vcpu);
2991 static __init int cpu_has_kvm_support(void)
2993 return cpu_has_vmx();
2996 static __init int vmx_disabled_by_bios(void)
3000 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3001 if (msr & FEATURE_CONTROL_LOCKED) {
3002 /* launched w/ TXT and VMX disabled */
3003 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3006 /* launched w/o TXT and VMX only enabled w/ TXT */
3007 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3008 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3009 && !tboot_enabled()) {
3010 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3011 "activate TXT before enabling KVM\n");
3014 /* launched w/o TXT and VMX disabled */
3015 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3016 && !tboot_enabled())
3023 static void kvm_cpu_vmxon(u64 addr)
3025 asm volatile (ASM_VMX_VMXON_RAX
3026 : : "a"(&addr), "m"(addr)
3030 static int hardware_enable(void)
3032 int cpu = raw_smp_processor_id();
3033 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3036 if (cr4_read_shadow() & X86_CR4_VMXE)
3039 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3040 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3041 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3044 * Now we can enable the vmclear operation in kdump
3045 * since the loaded_vmcss_on_cpu list on this cpu
3046 * has been initialized.
3048 * Though the cpu is not in VMX operation now, there
3049 * is no problem to enable the vmclear operation
3050 * for the loaded_vmcss_on_cpu list is empty!
3052 crash_enable_local_vmclear(cpu);
3054 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3056 test_bits = FEATURE_CONTROL_LOCKED;
3057 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3058 if (tboot_enabled())
3059 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3061 if ((old & test_bits) != test_bits) {
3062 /* enable and lock */
3063 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3065 cr4_set_bits(X86_CR4_VMXE);
3067 if (vmm_exclusive) {
3068 kvm_cpu_vmxon(phys_addr);
3072 native_store_gdt(this_cpu_ptr(&host_gdt));
3077 static void vmclear_local_loaded_vmcss(void)
3079 int cpu = raw_smp_processor_id();
3080 struct loaded_vmcs *v, *n;
3082 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3083 loaded_vmcss_on_cpu_link)
3084 __loaded_vmcs_clear(v);
3088 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3091 static void kvm_cpu_vmxoff(void)
3093 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3096 static void hardware_disable(void)
3098 if (vmm_exclusive) {
3099 vmclear_local_loaded_vmcss();
3102 cr4_clear_bits(X86_CR4_VMXE);
3105 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3106 u32 msr, u32 *result)
3108 u32 vmx_msr_low, vmx_msr_high;
3109 u32 ctl = ctl_min | ctl_opt;
3111 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3113 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3114 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3116 /* Ensure minimum (required) set of control bits are supported. */
3124 static __init bool allow_1_setting(u32 msr, u32 ctl)
3126 u32 vmx_msr_low, vmx_msr_high;
3128 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3129 return vmx_msr_high & ctl;
3132 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3134 u32 vmx_msr_low, vmx_msr_high;
3135 u32 min, opt, min2, opt2;
3136 u32 _pin_based_exec_control = 0;
3137 u32 _cpu_based_exec_control = 0;
3138 u32 _cpu_based_2nd_exec_control = 0;
3139 u32 _vmexit_control = 0;
3140 u32 _vmentry_control = 0;
3142 min = CPU_BASED_HLT_EXITING |
3143 #ifdef CONFIG_X86_64
3144 CPU_BASED_CR8_LOAD_EXITING |
3145 CPU_BASED_CR8_STORE_EXITING |
3147 CPU_BASED_CR3_LOAD_EXITING |
3148 CPU_BASED_CR3_STORE_EXITING |
3149 CPU_BASED_USE_IO_BITMAPS |
3150 CPU_BASED_MOV_DR_EXITING |
3151 CPU_BASED_USE_TSC_OFFSETING |
3152 CPU_BASED_MWAIT_EXITING |
3153 CPU_BASED_MONITOR_EXITING |
3154 CPU_BASED_INVLPG_EXITING |
3155 CPU_BASED_RDPMC_EXITING;
3157 opt = CPU_BASED_TPR_SHADOW |
3158 CPU_BASED_USE_MSR_BITMAPS |
3159 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3160 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3161 &_cpu_based_exec_control) < 0)
3163 #ifdef CONFIG_X86_64
3164 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3165 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3166 ~CPU_BASED_CR8_STORE_EXITING;
3168 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3170 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3171 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3172 SECONDARY_EXEC_WBINVD_EXITING |
3173 SECONDARY_EXEC_ENABLE_VPID |
3174 SECONDARY_EXEC_ENABLE_EPT |
3175 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3176 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3177 SECONDARY_EXEC_RDTSCP |
3178 SECONDARY_EXEC_ENABLE_INVPCID |
3179 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3180 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3181 SECONDARY_EXEC_SHADOW_VMCS |
3182 SECONDARY_EXEC_XSAVES |
3183 SECONDARY_EXEC_ENABLE_PML |
3184 SECONDARY_EXEC_PCOMMIT |
3185 SECONDARY_EXEC_TSC_SCALING;
3186 if (adjust_vmx_controls(min2, opt2,
3187 MSR_IA32_VMX_PROCBASED_CTLS2,
3188 &_cpu_based_2nd_exec_control) < 0)
3191 #ifndef CONFIG_X86_64
3192 if (!(_cpu_based_2nd_exec_control &
3193 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3194 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3197 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3198 _cpu_based_2nd_exec_control &= ~(
3199 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3200 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3201 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3203 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3204 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3206 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3207 CPU_BASED_CR3_STORE_EXITING |
3208 CPU_BASED_INVLPG_EXITING);
3209 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3210 vmx_capability.ept, vmx_capability.vpid);
3213 min = VM_EXIT_SAVE_DEBUG_CONTROLS;
3214 #ifdef CONFIG_X86_64
3215 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3217 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3218 VM_EXIT_ACK_INTR_ON_EXIT | VM_EXIT_CLEAR_BNDCFGS;
3219 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3220 &_vmexit_control) < 0)
3223 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING;
3224 opt = PIN_BASED_VIRTUAL_NMIS | PIN_BASED_POSTED_INTR;
3225 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3226 &_pin_based_exec_control) < 0)
3229 if (!(_cpu_based_2nd_exec_control &
3230 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) ||
3231 !(_vmexit_control & VM_EXIT_ACK_INTR_ON_EXIT))
3232 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3234 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3235 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3236 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3237 &_vmentry_control) < 0)
3240 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3242 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3243 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3246 #ifdef CONFIG_X86_64
3247 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3248 if (vmx_msr_high & (1u<<16))
3252 /* Require Write-Back (WB) memory type for VMCS accesses. */
3253 if (((vmx_msr_high >> 18) & 15) != 6)
3256 vmcs_conf->size = vmx_msr_high & 0x1fff;
3257 vmcs_conf->order = get_order(vmcs_config.size);
3258 vmcs_conf->revision_id = vmx_msr_low;
3260 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3261 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3262 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3263 vmcs_conf->vmexit_ctrl = _vmexit_control;
3264 vmcs_conf->vmentry_ctrl = _vmentry_control;
3266 cpu_has_load_ia32_efer =
3267 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3268 VM_ENTRY_LOAD_IA32_EFER)
3269 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3270 VM_EXIT_LOAD_IA32_EFER);
3272 cpu_has_load_perf_global_ctrl =
3273 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3274 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3275 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3276 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3279 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3280 * but due to arrata below it can't be used. Workaround is to use
3281 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3283 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3288 * BC86,AAY89,BD102 (model 44)
3292 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3293 switch (boot_cpu_data.x86_model) {
3299 cpu_has_load_perf_global_ctrl = false;
3300 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3301 "does not work properly. Using workaround\n");
3309 rdmsrl(MSR_IA32_XSS, host_xss);
3314 static struct vmcs *alloc_vmcs_cpu(int cpu)
3316 int node = cpu_to_node(cpu);
3320 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3323 vmcs = page_address(pages);
3324 memset(vmcs, 0, vmcs_config.size);
3325 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3329 static struct vmcs *alloc_vmcs(void)
3331 return alloc_vmcs_cpu(raw_smp_processor_id());
3334 static void free_vmcs(struct vmcs *vmcs)
3336 free_pages((unsigned long)vmcs, vmcs_config.order);
3340 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3342 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3344 if (!loaded_vmcs->vmcs)
3346 loaded_vmcs_clear(loaded_vmcs);
3347 free_vmcs(loaded_vmcs->vmcs);
3348 loaded_vmcs->vmcs = NULL;
3351 static void free_kvm_area(void)
3355 for_each_possible_cpu(cpu) {
3356 free_vmcs(per_cpu(vmxarea, cpu));
3357 per_cpu(vmxarea, cpu) = NULL;
3361 static void init_vmcs_shadow_fields(void)
3365 /* No checks for read only fields yet */
3367 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3368 switch (shadow_read_write_fields[i]) {
3370 if (!kvm_mpx_supported())
3378 shadow_read_write_fields[j] =
3379 shadow_read_write_fields[i];
3382 max_shadow_read_write_fields = j;
3384 /* shadowed fields guest access without vmexit */
3385 for (i = 0; i < max_shadow_read_write_fields; i++) {
3386 clear_bit(shadow_read_write_fields[i],
3387 vmx_vmwrite_bitmap);
3388 clear_bit(shadow_read_write_fields[i],
3391 for (i = 0; i < max_shadow_read_only_fields; i++)
3392 clear_bit(shadow_read_only_fields[i],
3396 static __init int alloc_kvm_area(void)
3400 for_each_possible_cpu(cpu) {
3403 vmcs = alloc_vmcs_cpu(cpu);
3409 per_cpu(vmxarea, cpu) = vmcs;
3414 static bool emulation_required(struct kvm_vcpu *vcpu)
3416 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
3419 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3420 struct kvm_segment *save)
3422 if (!emulate_invalid_guest_state) {
3424 * CS and SS RPL should be equal during guest entry according
3425 * to VMX spec, but in reality it is not always so. Since vcpu
3426 * is in the middle of the transition from real mode to
3427 * protected mode it is safe to assume that RPL 0 is a good
3430 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3431 save->selector &= ~SEGMENT_RPL_MASK;
3432 save->dpl = save->selector & SEGMENT_RPL_MASK;
3435 vmx_set_segment(vcpu, save, seg);
3438 static void enter_pmode(struct kvm_vcpu *vcpu)
3440 unsigned long flags;
3441 struct vcpu_vmx *vmx = to_vmx(vcpu);
3444 * Update real mode segment cache. It may be not up-to-date if sement
3445 * register was written while vcpu was in a guest mode.
3447 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3448 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3449 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3450 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3451 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3452 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3454 vmx->rmode.vm86_active = 0;
3456 vmx_segment_cache_clear(vmx);
3458 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3460 flags = vmcs_readl(GUEST_RFLAGS);
3461 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3462 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3463 vmcs_writel(GUEST_RFLAGS, flags);
3465 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3466 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3468 update_exception_bitmap(vcpu);
3470 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3471 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3472 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3473 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3474 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3475 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3478 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3480 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3481 struct kvm_segment var = *save;
3484 if (seg == VCPU_SREG_CS)
3487 if (!emulate_invalid_guest_state) {
3488 var.selector = var.base >> 4;
3489 var.base = var.base & 0xffff0;
3499 if (save->base & 0xf)
3500 printk_once(KERN_WARNING "kvm: segment base is not "
3501 "paragraph aligned when entering "
3502 "protected mode (seg=%d)", seg);
3505 vmcs_write16(sf->selector, var.selector);
3506 vmcs_writel(sf->base, var.base);
3507 vmcs_write32(sf->limit, var.limit);
3508 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
3511 static void enter_rmode(struct kvm_vcpu *vcpu)
3513 unsigned long flags;
3514 struct vcpu_vmx *vmx = to_vmx(vcpu);
3516 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3517 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3518 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3519 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3520 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3521 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3522 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3524 vmx->rmode.vm86_active = 1;
3527 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
3528 * vcpu. Warn the user that an update is overdue.
3530 if (!vcpu->kvm->arch.tss_addr)
3531 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
3532 "called before entering vcpu\n");
3534 vmx_segment_cache_clear(vmx);
3536 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
3537 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
3538 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
3540 flags = vmcs_readl(GUEST_RFLAGS);
3541 vmx->rmode.save_rflags = flags;
3543 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
3545 vmcs_writel(GUEST_RFLAGS, flags);
3546 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
3547 update_exception_bitmap(vcpu);
3549 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3550 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3551 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3552 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3553 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3554 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3556 kvm_mmu_reset_context(vcpu);
3559 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
3561 struct vcpu_vmx *vmx = to_vmx(vcpu);
3562 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
3568 * Force kernel_gs_base reloading before EFER changes, as control
3569 * of this msr depends on is_long_mode().
3571 vmx_load_host_state(to_vmx(vcpu));
3572 vcpu->arch.efer = efer;
3573 if (efer & EFER_LMA) {
3574 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3577 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3579 msr->data = efer & ~EFER_LME;
3584 #ifdef CONFIG_X86_64
3586 static void enter_lmode(struct kvm_vcpu *vcpu)
3590 vmx_segment_cache_clear(to_vmx(vcpu));
3592 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
3593 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
3594 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
3596 vmcs_write32(GUEST_TR_AR_BYTES,
3597 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
3598 | VMX_AR_TYPE_BUSY_64_TSS);
3600 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
3603 static void exit_lmode(struct kvm_vcpu *vcpu)
3605 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
3606 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
3611 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
3613 vpid_sync_context(vpid);
3615 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
3617 ept_sync_context(construct_eptp(vcpu->arch.mmu.root_hpa));
3621 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
3623 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
3626 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
3628 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
3630 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
3631 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
3634 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
3636 if (enable_ept && is_paging(vcpu))
3637 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
3638 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
3641 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
3643 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
3645 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
3646 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
3649 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
3651 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3653 if (!test_bit(VCPU_EXREG_PDPTR,
3654 (unsigned long *)&vcpu->arch.regs_dirty))
3657 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3658 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
3659 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
3660 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
3661 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
3665 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
3667 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
3669 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
3670 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
3671 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
3672 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
3673 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
3676 __set_bit(VCPU_EXREG_PDPTR,
3677 (unsigned long *)&vcpu->arch.regs_avail);
3678 __set_bit(VCPU_EXREG_PDPTR,
3679 (unsigned long *)&vcpu->arch.regs_dirty);
3682 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
3684 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
3686 struct kvm_vcpu *vcpu)
3688 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
3689 vmx_decache_cr3(vcpu);
3690 if (!(cr0 & X86_CR0_PG)) {
3691 /* From paging/starting to nonpaging */
3692 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3693 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
3694 (CPU_BASED_CR3_LOAD_EXITING |
3695 CPU_BASED_CR3_STORE_EXITING));
3696 vcpu->arch.cr0 = cr0;
3697 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3698 } else if (!is_paging(vcpu)) {
3699 /* From nonpaging to paging */
3700 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
3701 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
3702 ~(CPU_BASED_CR3_LOAD_EXITING |
3703 CPU_BASED_CR3_STORE_EXITING));
3704 vcpu->arch.cr0 = cr0;
3705 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
3708 if (!(cr0 & X86_CR0_WP))
3709 *hw_cr0 &= ~X86_CR0_WP;
3712 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
3714 struct vcpu_vmx *vmx = to_vmx(vcpu);
3715 unsigned long hw_cr0;
3717 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
3718 if (enable_unrestricted_guest)
3719 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
3721 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
3723 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
3726 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
3730 #ifdef CONFIG_X86_64
3731 if (vcpu->arch.efer & EFER_LME) {
3732 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
3734 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
3740 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
3742 if (!vcpu->fpu_active)
3743 hw_cr0 |= X86_CR0_TS | X86_CR0_MP;
3745 vmcs_writel(CR0_READ_SHADOW, cr0);
3746 vmcs_writel(GUEST_CR0, hw_cr0);
3747 vcpu->arch.cr0 = cr0;
3749 /* depends on vcpu->arch.cr0 to be set to a new value */
3750 vmx->emulation_required = emulation_required(vcpu);
3753 static u64 construct_eptp(unsigned long root_hpa)
3757 /* TODO write the value reading from MSR */
3758 eptp = VMX_EPT_DEFAULT_MT |
3759 VMX_EPT_DEFAULT_GAW << VMX_EPT_GAW_EPTP_SHIFT;
3760 if (enable_ept_ad_bits)
3761 eptp |= VMX_EPT_AD_ENABLE_BIT;
3762 eptp |= (root_hpa & PAGE_MASK);
3767 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
3769 unsigned long guest_cr3;
3774 eptp = construct_eptp(cr3);
3775 vmcs_write64(EPT_POINTER, eptp);
3776 if (is_paging(vcpu) || is_guest_mode(vcpu))
3777 guest_cr3 = kvm_read_cr3(vcpu);
3779 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
3780 ept_load_pdptrs(vcpu);
3783 vmx_flush_tlb(vcpu);
3784 vmcs_writel(GUEST_CR3, guest_cr3);
3787 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
3790 * Pass through host's Machine Check Enable value to hw_cr4, which
3791 * is in force while we are in guest mode. Do not let guests control
3792 * this bit, even if host CR4.MCE == 0.
3794 unsigned long hw_cr4 =
3795 (cr4_read_shadow() & X86_CR4_MCE) |
3796 (cr4 & ~X86_CR4_MCE) |
3797 (to_vmx(vcpu)->rmode.vm86_active ?
3798 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
3800 if (cr4 & X86_CR4_VMXE) {
3802 * To use VMXON (and later other VMX instructions), a guest
3803 * must first be able to turn on cr4.VMXE (see handle_vmon()).
3804 * So basically the check on whether to allow nested VMX
3807 if (!nested_vmx_allowed(vcpu))
3810 if (to_vmx(vcpu)->nested.vmxon &&
3811 ((cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON))
3814 vcpu->arch.cr4 = cr4;
3816 if (!is_paging(vcpu)) {
3817 hw_cr4 &= ~X86_CR4_PAE;
3818 hw_cr4 |= X86_CR4_PSE;
3819 } else if (!(cr4 & X86_CR4_PAE)) {
3820 hw_cr4 &= ~X86_CR4_PAE;
3824 if (!enable_unrestricted_guest && !is_paging(vcpu))
3826 * SMEP/SMAP is disabled if CPU is in non-paging mode in
3827 * hardware. However KVM always uses paging mode without
3828 * unrestricted guest.
3829 * To emulate this behavior, SMEP/SMAP needs to be manually
3830 * disabled when guest switches to non-paging mode.
3832 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP);
3834 vmcs_writel(CR4_READ_SHADOW, cr4);
3835 vmcs_writel(GUEST_CR4, hw_cr4);
3839 static void vmx_get_segment(struct kvm_vcpu *vcpu,
3840 struct kvm_segment *var, int seg)
3842 struct vcpu_vmx *vmx = to_vmx(vcpu);
3845 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3846 *var = vmx->rmode.segs[seg];
3847 if (seg == VCPU_SREG_TR
3848 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
3850 var->base = vmx_read_guest_seg_base(vmx, seg);
3851 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3854 var->base = vmx_read_guest_seg_base(vmx, seg);
3855 var->limit = vmx_read_guest_seg_limit(vmx, seg);
3856 var->selector = vmx_read_guest_seg_selector(vmx, seg);
3857 ar = vmx_read_guest_seg_ar(vmx, seg);
3858 var->unusable = (ar >> 16) & 1;
3859 var->type = ar & 15;
3860 var->s = (ar >> 4) & 1;
3861 var->dpl = (ar >> 5) & 3;
3863 * Some userspaces do not preserve unusable property. Since usable
3864 * segment has to be present according to VMX spec we can use present
3865 * property to amend userspace bug by making unusable segment always
3866 * nonpresent. vmx_segment_access_rights() already marks nonpresent
3867 * segment as unusable.
3869 var->present = !var->unusable;
3870 var->avl = (ar >> 12) & 1;
3871 var->l = (ar >> 13) & 1;
3872 var->db = (ar >> 14) & 1;
3873 var->g = (ar >> 15) & 1;
3876 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
3878 struct kvm_segment s;
3880 if (to_vmx(vcpu)->rmode.vm86_active) {
3881 vmx_get_segment(vcpu, &s, seg);
3884 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
3887 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
3889 struct vcpu_vmx *vmx = to_vmx(vcpu);
3891 if (unlikely(vmx->rmode.vm86_active))
3894 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
3895 return VMX_AR_DPL(ar);
3899 static u32 vmx_segment_access_rights(struct kvm_segment *var)
3903 if (var->unusable || !var->present)
3906 ar = var->type & 15;
3907 ar |= (var->s & 1) << 4;
3908 ar |= (var->dpl & 3) << 5;
3909 ar |= (var->present & 1) << 7;
3910 ar |= (var->avl & 1) << 12;
3911 ar |= (var->l & 1) << 13;
3912 ar |= (var->db & 1) << 14;
3913 ar |= (var->g & 1) << 15;
3919 static void vmx_set_segment(struct kvm_vcpu *vcpu,
3920 struct kvm_segment *var, int seg)
3922 struct vcpu_vmx *vmx = to_vmx(vcpu);
3923 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3925 vmx_segment_cache_clear(vmx);
3927 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
3928 vmx->rmode.segs[seg] = *var;
3929 if (seg == VCPU_SREG_TR)
3930 vmcs_write16(sf->selector, var->selector);
3932 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
3936 vmcs_writel(sf->base, var->base);
3937 vmcs_write32(sf->limit, var->limit);
3938 vmcs_write16(sf->selector, var->selector);
3941 * Fix the "Accessed" bit in AR field of segment registers for older
3943 * IA32 arch specifies that at the time of processor reset the
3944 * "Accessed" bit in the AR field of segment registers is 1. And qemu
3945 * is setting it to 0 in the userland code. This causes invalid guest
3946 * state vmexit when "unrestricted guest" mode is turned on.
3947 * Fix for this setup issue in cpu_reset is being pushed in the qemu
3948 * tree. Newer qemu binaries with that qemu fix would not need this
3951 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
3952 var->type |= 0x1; /* Accessed */
3954 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
3957 vmx->emulation_required = emulation_required(vcpu);
3960 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
3962 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
3964 *db = (ar >> 14) & 1;
3965 *l = (ar >> 13) & 1;
3968 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3970 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
3971 dt->address = vmcs_readl(GUEST_IDTR_BASE);
3974 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3976 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
3977 vmcs_writel(GUEST_IDTR_BASE, dt->address);
3980 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3982 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
3983 dt->address = vmcs_readl(GUEST_GDTR_BASE);
3986 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
3988 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
3989 vmcs_writel(GUEST_GDTR_BASE, dt->address);
3992 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
3994 struct kvm_segment var;
3997 vmx_get_segment(vcpu, &var, seg);
3999 if (seg == VCPU_SREG_CS)
4001 ar = vmx_segment_access_rights(&var);
4003 if (var.base != (var.selector << 4))
4005 if (var.limit != 0xffff)
4013 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4015 struct kvm_segment cs;
4016 unsigned int cs_rpl;
4018 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4019 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4023 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4027 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4028 if (cs.dpl > cs_rpl)
4031 if (cs.dpl != cs_rpl)
4037 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4041 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4043 struct kvm_segment ss;
4044 unsigned int ss_rpl;
4046 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4047 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4051 if (ss.type != 3 && ss.type != 7)
4055 if (ss.dpl != ss_rpl) /* DPL != RPL */
4063 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4065 struct kvm_segment var;
4068 vmx_get_segment(vcpu, &var, seg);
4069 rpl = var.selector & SEGMENT_RPL_MASK;
4077 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4078 if (var.dpl < rpl) /* DPL < RPL */
4082 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4088 static bool tr_valid(struct kvm_vcpu *vcpu)
4090 struct kvm_segment tr;
4092 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4096 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4098 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4106 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4108 struct kvm_segment ldtr;
4110 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4114 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4124 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4126 struct kvm_segment cs, ss;
4128 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4129 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4131 return ((cs.selector & SEGMENT_RPL_MASK) ==
4132 (ss.selector & SEGMENT_RPL_MASK));
4136 * Check if guest state is valid. Returns true if valid, false if
4138 * We assume that registers are always usable
4140 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4142 if (enable_unrestricted_guest)
4145 /* real mode guest state checks */
4146 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4147 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4149 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4151 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4153 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4155 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4157 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4160 /* protected mode guest state checks */
4161 if (!cs_ss_rpl_check(vcpu))
4163 if (!code_segment_valid(vcpu))
4165 if (!stack_segment_valid(vcpu))
4167 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4169 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4171 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4173 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4175 if (!tr_valid(vcpu))
4177 if (!ldtr_valid(vcpu))
4181 * - Add checks on RIP
4182 * - Add checks on RFLAGS
4188 static int init_rmode_tss(struct kvm *kvm)
4194 idx = srcu_read_lock(&kvm->srcu);
4195 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4196 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4199 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4200 r = kvm_write_guest_page(kvm, fn++, &data,
4201 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4204 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4207 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4211 r = kvm_write_guest_page(kvm, fn, &data,
4212 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4215 srcu_read_unlock(&kvm->srcu, idx);
4219 static int init_rmode_identity_map(struct kvm *kvm)
4222 pfn_t identity_map_pfn;
4228 /* Protect kvm->arch.ept_identity_pagetable_done. */
4229 mutex_lock(&kvm->slots_lock);
4231 if (likely(kvm->arch.ept_identity_pagetable_done))
4234 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4236 r = alloc_identity_pagetable(kvm);
4240 idx = srcu_read_lock(&kvm->srcu);
4241 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4244 /* Set up identity-mapping pagetable for EPT in real mode */
4245 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4246 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4247 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4248 r = kvm_write_guest_page(kvm, identity_map_pfn,
4249 &tmp, i * sizeof(tmp), sizeof(tmp));
4253 kvm->arch.ept_identity_pagetable_done = true;
4256 srcu_read_unlock(&kvm->srcu, idx);
4259 mutex_unlock(&kvm->slots_lock);
4263 static void seg_setup(int seg)
4265 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4268 vmcs_write16(sf->selector, 0);
4269 vmcs_writel(sf->base, 0);
4270 vmcs_write32(sf->limit, 0xffff);
4272 if (seg == VCPU_SREG_CS)
4273 ar |= 0x08; /* code segment */
4275 vmcs_write32(sf->ar_bytes, ar);
4278 static int alloc_apic_access_page(struct kvm *kvm)
4283 mutex_lock(&kvm->slots_lock);
4284 if (kvm->arch.apic_access_page_done)
4286 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4287 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4291 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4292 if (is_error_page(page)) {
4298 * Do not pin the page in memory, so that memory hot-unplug
4299 * is able to migrate it.
4302 kvm->arch.apic_access_page_done = true;
4304 mutex_unlock(&kvm->slots_lock);
4308 static int alloc_identity_pagetable(struct kvm *kvm)
4310 /* Called with kvm->slots_lock held. */
4314 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4316 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4317 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4322 static int allocate_vpid(void)
4328 spin_lock(&vmx_vpid_lock);
4329 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4330 if (vpid < VMX_NR_VPIDS)
4331 __set_bit(vpid, vmx_vpid_bitmap);
4334 spin_unlock(&vmx_vpid_lock);
4338 static void free_vpid(int vpid)
4340 if (!enable_vpid || vpid == 0)
4342 spin_lock(&vmx_vpid_lock);
4343 __clear_bit(vpid, vmx_vpid_bitmap);
4344 spin_unlock(&vmx_vpid_lock);
4347 #define MSR_TYPE_R 1
4348 #define MSR_TYPE_W 2
4349 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4352 int f = sizeof(unsigned long);
4354 if (!cpu_has_vmx_msr_bitmap())
4358 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4359 * have the write-low and read-high bitmap offsets the wrong way round.
4360 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4362 if (msr <= 0x1fff) {
4363 if (type & MSR_TYPE_R)
4365 __clear_bit(msr, msr_bitmap + 0x000 / f);
4367 if (type & MSR_TYPE_W)
4369 __clear_bit(msr, msr_bitmap + 0x800 / f);
4371 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4373 if (type & MSR_TYPE_R)
4375 __clear_bit(msr, msr_bitmap + 0x400 / f);
4377 if (type & MSR_TYPE_W)
4379 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4384 static void __vmx_enable_intercept_for_msr(unsigned long *msr_bitmap,
4387 int f = sizeof(unsigned long);
4389 if (!cpu_has_vmx_msr_bitmap())
4393 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4394 * have the write-low and read-high bitmap offsets the wrong way round.
4395 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4397 if (msr <= 0x1fff) {
4398 if (type & MSR_TYPE_R)
4400 __set_bit(msr, msr_bitmap + 0x000 / f);
4402 if (type & MSR_TYPE_W)
4404 __set_bit(msr, msr_bitmap + 0x800 / f);
4406 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4408 if (type & MSR_TYPE_R)
4410 __set_bit(msr, msr_bitmap + 0x400 / f);
4412 if (type & MSR_TYPE_W)
4414 __set_bit(msr, msr_bitmap + 0xc00 / f);
4420 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4421 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4423 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4424 unsigned long *msr_bitmap_nested,
4427 int f = sizeof(unsigned long);
4429 if (!cpu_has_vmx_msr_bitmap()) {
4435 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4436 * have the write-low and read-high bitmap offsets the wrong way round.
4437 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4439 if (msr <= 0x1fff) {
4440 if (type & MSR_TYPE_R &&
4441 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4443 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4445 if (type & MSR_TYPE_W &&
4446 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4448 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4450 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4452 if (type & MSR_TYPE_R &&
4453 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4455 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4457 if (type & MSR_TYPE_W &&
4458 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4460 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4465 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4468 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4469 msr, MSR_TYPE_R | MSR_TYPE_W);
4470 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4471 msr, MSR_TYPE_R | MSR_TYPE_W);
4474 static void vmx_enable_intercept_msr_read_x2apic(u32 msr)
4476 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4478 __vmx_enable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4482 static void vmx_disable_intercept_msr_read_x2apic(u32 msr)
4484 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4486 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4490 static void vmx_disable_intercept_msr_write_x2apic(u32 msr)
4492 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
4494 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
4498 static int vmx_cpu_uses_apicv(struct kvm_vcpu *vcpu)
4500 return enable_apicv && lapic_in_kernel(vcpu);
4503 static int vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
4505 struct vcpu_vmx *vmx = to_vmx(vcpu);
4510 if (vmx->nested.pi_desc &&
4511 vmx->nested.pi_pending) {
4512 vmx->nested.pi_pending = false;
4513 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
4516 max_irr = find_last_bit(
4517 (unsigned long *)vmx->nested.pi_desc->pir, 256);
4522 vapic_page = kmap(vmx->nested.virtual_apic_page);
4527 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
4528 kunmap(vmx->nested.virtual_apic_page);
4530 status = vmcs_read16(GUEST_INTR_STATUS);
4531 if ((u8)max_irr > ((u8)status & 0xff)) {
4533 status |= (u8)max_irr;
4534 vmcs_write16(GUEST_INTR_STATUS, status);
4540 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu)
4543 if (vcpu->mode == IN_GUEST_MODE) {
4545 * The vector of interrupt to be delivered to vcpu had
4546 * been set in PIR before this function.
4548 * Following cases will be reached in this block, and
4549 * we always send a notification event in all cases as
4552 * Case 1: vcpu keeps in non-root mode. Sending a
4553 * notification event posts the interrupt to vcpu.
4555 * Case 2: vcpu exits to root mode and is still
4556 * runnable. PIR will be synced to vIRR before the
4557 * next vcpu entry. Sending a notification event in
4558 * this case has no effect, as vcpu is not in root
4561 * Case 3: vcpu exits to root mode and is blocked.
4562 * vcpu_block() has already synced PIR to vIRR and
4563 * never blocks vcpu if vIRR is not cleared. Therefore,
4564 * a blocked vcpu here does not wait for any requested
4565 * interrupts in PIR, and sending a notification event
4566 * which has no effect is safe here.
4569 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu),
4570 POSTED_INTR_VECTOR);
4577 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
4580 struct vcpu_vmx *vmx = to_vmx(vcpu);
4582 if (is_guest_mode(vcpu) &&
4583 vector == vmx->nested.posted_intr_nv) {
4584 /* the PIR and ON have been set by L1. */
4585 kvm_vcpu_trigger_posted_interrupt(vcpu);
4587 * If a posted intr is not recognized by hardware,
4588 * we will accomplish it in the next vmentry.
4590 vmx->nested.pi_pending = true;
4591 kvm_make_request(KVM_REQ_EVENT, vcpu);
4597 * Send interrupt to vcpu via posted interrupt way.
4598 * 1. If target vcpu is running(non-root mode), send posted interrupt
4599 * notification to vcpu and hardware will sync PIR to vIRR atomically.
4600 * 2. If target vcpu isn't running(root mode), kick it to pick up the
4601 * interrupt from PIR in next vmentry.
4603 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
4605 struct vcpu_vmx *vmx = to_vmx(vcpu);
4608 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
4612 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
4615 r = pi_test_and_set_on(&vmx->pi_desc);
4616 kvm_make_request(KVM_REQ_EVENT, vcpu);
4617 if (r || !kvm_vcpu_trigger_posted_interrupt(vcpu))
4618 kvm_vcpu_kick(vcpu);
4621 static void vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
4623 struct vcpu_vmx *vmx = to_vmx(vcpu);
4625 if (!pi_test_and_clear_on(&vmx->pi_desc))
4628 kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
4631 static void vmx_sync_pir_to_irr_dummy(struct kvm_vcpu *vcpu)
4637 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
4638 * will not change in the lifetime of the guest.
4639 * Note that host-state that does change is set elsewhere. E.g., host-state
4640 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
4642 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
4649 vmcs_writel(HOST_CR0, read_cr0() & ~X86_CR0_TS); /* 22.2.3 */
4650 vmcs_writel(HOST_CR3, read_cr3()); /* 22.2.3 FIXME: shadow tables */
4652 /* Save the most likely value for this task's CR4 in the VMCS. */
4653 cr4 = cr4_read_shadow();
4654 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
4655 vmx->host_state.vmcs_host_cr4 = cr4;
4657 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
4658 #ifdef CONFIG_X86_64
4660 * Load null selectors, so we can avoid reloading them in
4661 * __vmx_load_host_state(), in case userspace uses the null selectors
4662 * too (the expected case).
4664 vmcs_write16(HOST_DS_SELECTOR, 0);
4665 vmcs_write16(HOST_ES_SELECTOR, 0);
4667 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4668 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4670 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
4671 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
4673 native_store_idt(&dt);
4674 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
4675 vmx->host_idt_base = dt.address;
4677 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
4679 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
4680 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
4681 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
4682 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
4684 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
4685 rdmsr(MSR_IA32_CR_PAT, low32, high32);
4686 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
4690 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
4692 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
4694 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
4695 if (is_guest_mode(&vmx->vcpu))
4696 vmx->vcpu.arch.cr4_guest_owned_bits &=
4697 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
4698 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
4701 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
4703 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
4705 if (!vmx_cpu_uses_apicv(&vmx->vcpu))
4706 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
4707 return pin_based_exec_ctrl;
4710 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
4712 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
4714 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
4715 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
4717 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
4718 exec_control &= ~CPU_BASED_TPR_SHADOW;
4719 #ifdef CONFIG_X86_64
4720 exec_control |= CPU_BASED_CR8_STORE_EXITING |
4721 CPU_BASED_CR8_LOAD_EXITING;
4725 exec_control |= CPU_BASED_CR3_STORE_EXITING |
4726 CPU_BASED_CR3_LOAD_EXITING |
4727 CPU_BASED_INVLPG_EXITING;
4728 return exec_control;
4731 static u32 vmx_secondary_exec_control(struct vcpu_vmx *vmx)
4733 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
4734 if (!cpu_need_virtualize_apic_accesses(&vmx->vcpu))
4735 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
4737 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
4739 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
4740 enable_unrestricted_guest = 0;
4741 /* Enable INVPCID for non-ept guests may cause performance regression. */
4742 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
4744 if (!enable_unrestricted_guest)
4745 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
4747 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
4748 if (!vmx_cpu_uses_apicv(&vmx->vcpu))
4749 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
4750 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
4751 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
4752 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
4754 We can NOT enable shadow_vmcs here because we don't have yet
4757 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
4760 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
4762 /* Currently, we allow L1 guest to directly run pcommit instruction. */
4763 exec_control &= ~SECONDARY_EXEC_PCOMMIT;
4765 return exec_control;
4768 static void ept_set_mmio_spte_mask(void)
4771 * EPT Misconfigurations can be generated if the value of bits 2:0
4772 * of an EPT paging-structure entry is 110b (write/execute).
4773 * Also, magic bits (0x3ull << 62) is set to quickly identify mmio
4776 kvm_mmu_set_mmio_spte_mask((0x3ull << 62) | 0x6ull);
4779 #define VMX_XSS_EXIT_BITMAP 0
4781 * Sets up the vmcs for emulated real mode.
4783 static int vmx_vcpu_setup(struct vcpu_vmx *vmx)
4785 #ifdef CONFIG_X86_64
4791 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
4792 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
4794 if (enable_shadow_vmcs) {
4795 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
4796 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
4798 if (cpu_has_vmx_msr_bitmap())
4799 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
4801 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
4804 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
4806 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
4808 if (cpu_has_secondary_exec_ctrls())
4809 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
4810 vmx_secondary_exec_control(vmx));
4812 if (vmx_cpu_uses_apicv(&vmx->vcpu)) {
4813 vmcs_write64(EOI_EXIT_BITMAP0, 0);
4814 vmcs_write64(EOI_EXIT_BITMAP1, 0);
4815 vmcs_write64(EOI_EXIT_BITMAP2, 0);
4816 vmcs_write64(EOI_EXIT_BITMAP3, 0);
4818 vmcs_write16(GUEST_INTR_STATUS, 0);
4820 vmcs_write64(POSTED_INTR_NV, POSTED_INTR_VECTOR);
4821 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
4825 vmcs_write32(PLE_GAP, ple_gap);
4826 vmx->ple_window = ple_window;
4827 vmx->ple_window_dirty = true;
4830 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
4831 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
4832 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
4834 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
4835 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
4836 vmx_set_constant_host_state(vmx);
4837 #ifdef CONFIG_X86_64
4838 rdmsrl(MSR_FS_BASE, a);
4839 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
4840 rdmsrl(MSR_GS_BASE, a);
4841 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
4843 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
4844 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
4847 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
4848 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
4849 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
4850 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
4851 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
4853 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
4854 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
4856 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
4857 u32 index = vmx_msr_index[i];
4858 u32 data_low, data_high;
4861 if (rdmsr_safe(index, &data_low, &data_high) < 0)
4863 if (wrmsr_safe(index, data_low, data_high) < 0)
4865 vmx->guest_msrs[j].index = i;
4866 vmx->guest_msrs[j].data = 0;
4867 vmx->guest_msrs[j].mask = -1ull;
4872 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
4874 /* 22.2.1, 20.8.1 */
4875 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
4877 vmcs_writel(CR0_GUEST_HOST_MASK, ~0UL);
4878 set_cr4_guest_host_mask(vmx);
4880 if (vmx_xsaves_supported())
4881 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
4884 ASSERT(vmx->pml_pg);
4885 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
4886 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
4892 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
4894 struct vcpu_vmx *vmx = to_vmx(vcpu);
4895 struct msr_data apic_base_msr;
4898 vmx->rmode.vm86_active = 0;
4900 vmx->soft_vnmi_blocked = 0;
4902 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
4903 kvm_set_cr8(vcpu, 0);
4906 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
4907 MSR_IA32_APICBASE_ENABLE;
4908 if (kvm_vcpu_is_reset_bsp(vcpu))
4909 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
4910 apic_base_msr.host_initiated = true;
4911 kvm_set_apic_base(vcpu, &apic_base_msr);
4914 vmx_segment_cache_clear(vmx);
4916 seg_setup(VCPU_SREG_CS);
4917 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
4918 vmcs_write32(GUEST_CS_BASE, 0xffff0000);
4920 seg_setup(VCPU_SREG_DS);
4921 seg_setup(VCPU_SREG_ES);
4922 seg_setup(VCPU_SREG_FS);
4923 seg_setup(VCPU_SREG_GS);
4924 seg_setup(VCPU_SREG_SS);
4926 vmcs_write16(GUEST_TR_SELECTOR, 0);
4927 vmcs_writel(GUEST_TR_BASE, 0);
4928 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
4929 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4931 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
4932 vmcs_writel(GUEST_LDTR_BASE, 0);
4933 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
4934 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
4937 vmcs_write32(GUEST_SYSENTER_CS, 0);
4938 vmcs_writel(GUEST_SYSENTER_ESP, 0);
4939 vmcs_writel(GUEST_SYSENTER_EIP, 0);
4940 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
4943 vmcs_writel(GUEST_RFLAGS, 0x02);
4944 kvm_rip_write(vcpu, 0xfff0);
4946 vmcs_writel(GUEST_GDTR_BASE, 0);
4947 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
4949 vmcs_writel(GUEST_IDTR_BASE, 0);
4950 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
4952 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
4953 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
4954 vmcs_write32(GUEST_PENDING_DBG_EXCEPTIONS, 0);
4958 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
4960 if (cpu_has_vmx_tpr_shadow() && !init_event) {
4961 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
4962 if (cpu_need_tpr_shadow(vcpu))
4963 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
4964 __pa(vcpu->arch.apic->regs));
4965 vmcs_write32(TPR_THRESHOLD, 0);
4968 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
4970 if (vmx_cpu_uses_apicv(vcpu))
4971 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
4974 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
4976 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
4977 vmx->vcpu.arch.cr0 = cr0;
4978 vmx_set_cr0(vcpu, cr0); /* enter rmode */
4979 vmx_set_cr4(vcpu, 0);
4980 vmx_set_efer(vcpu, 0);
4981 vmx_fpu_activate(vcpu);
4982 update_exception_bitmap(vcpu);
4984 vpid_sync_context(vmx->vpid);
4988 * In nested virtualization, check if L1 asked to exit on external interrupts.
4989 * For most existing hypervisors, this will always return true.
4991 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
4993 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
4994 PIN_BASED_EXT_INTR_MASK;
4998 * In nested virtualization, check if L1 has set
4999 * VM_EXIT_ACK_INTR_ON_EXIT
5001 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5003 return get_vmcs12(vcpu)->vm_exit_controls &
5004 VM_EXIT_ACK_INTR_ON_EXIT;
5007 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5009 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5010 PIN_BASED_NMI_EXITING;
5013 static void enable_irq_window(struct kvm_vcpu *vcpu)
5015 u32 cpu_based_vm_exec_control;
5017 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5018 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_INTR_PENDING;
5019 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5022 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5024 u32 cpu_based_vm_exec_control;
5026 if (!cpu_has_virtual_nmis() ||
5027 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5028 enable_irq_window(vcpu);
5032 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5033 cpu_based_vm_exec_control |= CPU_BASED_VIRTUAL_NMI_PENDING;
5034 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5037 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5039 struct vcpu_vmx *vmx = to_vmx(vcpu);
5041 int irq = vcpu->arch.interrupt.nr;
5043 trace_kvm_inj_virq(irq);
5045 ++vcpu->stat.irq_injections;
5046 if (vmx->rmode.vm86_active) {
5048 if (vcpu->arch.interrupt.soft)
5049 inc_eip = vcpu->arch.event_exit_inst_len;
5050 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5051 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5054 intr = irq | INTR_INFO_VALID_MASK;
5055 if (vcpu->arch.interrupt.soft) {
5056 intr |= INTR_TYPE_SOFT_INTR;
5057 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5058 vmx->vcpu.arch.event_exit_inst_len);
5060 intr |= INTR_TYPE_EXT_INTR;
5061 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5064 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5066 struct vcpu_vmx *vmx = to_vmx(vcpu);
5068 if (is_guest_mode(vcpu))
5071 if (!cpu_has_virtual_nmis()) {
5073 * Tracking the NMI-blocked state in software is built upon
5074 * finding the next open IRQ window. This, in turn, depends on
5075 * well-behaving guests: They have to keep IRQs disabled at
5076 * least as long as the NMI handler runs. Otherwise we may
5077 * cause NMI nesting, maybe breaking the guest. But as this is
5078 * highly unlikely, we can live with the residual risk.
5080 vmx->soft_vnmi_blocked = 1;
5081 vmx->vnmi_blocked_time = 0;
5084 ++vcpu->stat.nmi_injections;
5085 vmx->nmi_known_unmasked = false;
5086 if (vmx->rmode.vm86_active) {
5087 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5088 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5091 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5092 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5095 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5097 if (!cpu_has_virtual_nmis())
5098 return to_vmx(vcpu)->soft_vnmi_blocked;
5099 if (to_vmx(vcpu)->nmi_known_unmasked)
5101 return vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5104 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5106 struct vcpu_vmx *vmx = to_vmx(vcpu);
5108 if (!cpu_has_virtual_nmis()) {
5109 if (vmx->soft_vnmi_blocked != masked) {
5110 vmx->soft_vnmi_blocked = masked;
5111 vmx->vnmi_blocked_time = 0;
5114 vmx->nmi_known_unmasked = !masked;
5116 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5117 GUEST_INTR_STATE_NMI);
5119 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5120 GUEST_INTR_STATE_NMI);
5124 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5126 if (to_vmx(vcpu)->nested.nested_run_pending)
5129 if (!cpu_has_virtual_nmis() && to_vmx(vcpu)->soft_vnmi_blocked)
5132 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5133 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5134 | GUEST_INTR_STATE_NMI));
5137 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5139 return (!to_vmx(vcpu)->nested.nested_run_pending &&
5140 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5141 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5142 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5145 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5149 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5153 kvm->arch.tss_addr = addr;
5154 return init_rmode_tss(kvm);
5157 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5162 * Update instruction length as we may reinject the exception
5163 * from user space while in guest debugging mode.
5165 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5166 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5167 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5171 if (vcpu->guest_debug &
5172 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5189 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5190 int vec, u32 err_code)
5193 * Instruction with address size override prefix opcode 0x67
5194 * Cause the #SS fault with 0 error code in VM86 mode.
5196 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5197 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5198 if (vcpu->arch.halt_request) {
5199 vcpu->arch.halt_request = 0;
5200 return kvm_vcpu_halt(vcpu);
5208 * Forward all other exceptions that are valid in real mode.
5209 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5210 * the required debugging infrastructure rework.
5212 kvm_queue_exception(vcpu, vec);
5217 * Trigger machine check on the host. We assume all the MSRs are already set up
5218 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5219 * We pass a fake environment to the machine check handler because we want
5220 * the guest to be always treated like user space, no matter what context
5221 * it used internally.
5223 static void kvm_machine_check(void)
5225 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5226 struct pt_regs regs = {
5227 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5228 .flags = X86_EFLAGS_IF,
5231 do_machine_check(®s, 0);
5235 static int handle_machine_check(struct kvm_vcpu *vcpu)
5237 /* already handled by vcpu_run */
5241 static int handle_exception(struct kvm_vcpu *vcpu)
5243 struct vcpu_vmx *vmx = to_vmx(vcpu);
5244 struct kvm_run *kvm_run = vcpu->run;
5245 u32 intr_info, ex_no, error_code;
5246 unsigned long cr2, rip, dr6;
5248 enum emulation_result er;
5250 vect_info = vmx->idt_vectoring_info;
5251 intr_info = vmx->exit_intr_info;
5253 if (is_machine_check(intr_info))
5254 return handle_machine_check(vcpu);
5256 if (is_nmi(intr_info))
5257 return 1; /* already handled by vmx_vcpu_run() */
5259 if (is_no_device(intr_info)) {
5260 vmx_fpu_activate(vcpu);
5264 if (is_invalid_opcode(intr_info)) {
5265 if (is_guest_mode(vcpu)) {
5266 kvm_queue_exception(vcpu, UD_VECTOR);
5269 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5270 if (er == EMULATE_USER_EXIT)
5272 if (er != EMULATE_DONE)
5273 kvm_queue_exception(vcpu, UD_VECTOR);
5278 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5279 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5282 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5283 * MMIO, it is better to report an internal error.
5284 * See the comments in vmx_handle_exit.
5286 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5287 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5288 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5289 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5290 vcpu->run->internal.ndata = 3;
5291 vcpu->run->internal.data[0] = vect_info;
5292 vcpu->run->internal.data[1] = intr_info;
5293 vcpu->run->internal.data[2] = error_code;
5297 if (is_page_fault(intr_info)) {
5298 /* EPT won't cause page fault directly */
5300 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5301 trace_kvm_page_fault(cr2, error_code);
5303 if (kvm_event_needs_reinjection(vcpu))
5304 kvm_mmu_unprotect_page_virt(vcpu, cr2);
5305 return kvm_mmu_page_fault(vcpu, cr2, error_code, NULL, 0);
5308 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5310 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5311 return handle_rmode_exception(vcpu, ex_no, error_code);
5315 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5318 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5319 if (!(vcpu->guest_debug &
5320 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5321 vcpu->arch.dr6 &= ~15;
5322 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5323 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5324 skip_emulated_instruction(vcpu);
5326 kvm_queue_exception(vcpu, DB_VECTOR);
5329 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5330 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5334 * Update instruction length as we may reinject #BP from
5335 * user space while in guest debugging mode. Reading it for
5336 * #DB as well causes no harm, it is not used in that case.
5338 vmx->vcpu.arch.event_exit_inst_len =
5339 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5340 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5341 rip = kvm_rip_read(vcpu);
5342 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5343 kvm_run->debug.arch.exception = ex_no;
5346 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5347 kvm_run->ex.exception = ex_no;
5348 kvm_run->ex.error_code = error_code;
5354 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5356 ++vcpu->stat.irq_exits;
5360 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5362 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5366 static int handle_io(struct kvm_vcpu *vcpu)
5368 unsigned long exit_qualification;
5369 int size, in, string;
5372 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5373 string = (exit_qualification & 16) != 0;
5374 in = (exit_qualification & 8) != 0;
5376 ++vcpu->stat.io_exits;
5379 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5381 port = exit_qualification >> 16;
5382 size = (exit_qualification & 7) + 1;
5383 skip_emulated_instruction(vcpu);
5385 return kvm_fast_pio_out(vcpu, size, port);
5389 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
5392 * Patch in the VMCALL instruction:
5394 hypercall[0] = 0x0f;
5395 hypercall[1] = 0x01;
5396 hypercall[2] = 0xc1;
5399 static bool nested_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
5401 unsigned long always_on = VMXON_CR0_ALWAYSON;
5402 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5404 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
5405 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
5406 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
5407 always_on &= ~(X86_CR0_PE | X86_CR0_PG);
5408 return (val & always_on) == always_on;
5411 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
5412 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
5414 if (is_guest_mode(vcpu)) {
5415 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5416 unsigned long orig_val = val;
5419 * We get here when L2 changed cr0 in a way that did not change
5420 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
5421 * but did change L0 shadowed bits. So we first calculate the
5422 * effective cr0 value that L1 would like to write into the
5423 * hardware. It consists of the L2-owned bits from the new
5424 * value combined with the L1-owned bits from L1's guest_cr0.
5426 val = (val & ~vmcs12->cr0_guest_host_mask) |
5427 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
5429 if (!nested_cr0_valid(vcpu, val))
5432 if (kvm_set_cr0(vcpu, val))
5434 vmcs_writel(CR0_READ_SHADOW, orig_val);
5437 if (to_vmx(vcpu)->nested.vmxon &&
5438 ((val & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON))
5440 return kvm_set_cr0(vcpu, val);
5444 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
5446 if (is_guest_mode(vcpu)) {
5447 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5448 unsigned long orig_val = val;
5450 /* analogously to handle_set_cr0 */
5451 val = (val & ~vmcs12->cr4_guest_host_mask) |
5452 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
5453 if (kvm_set_cr4(vcpu, val))
5455 vmcs_writel(CR4_READ_SHADOW, orig_val);
5458 return kvm_set_cr4(vcpu, val);
5461 /* called to set cr0 as approriate for clts instruction exit. */
5462 static void handle_clts(struct kvm_vcpu *vcpu)
5464 if (is_guest_mode(vcpu)) {
5466 * We get here when L2 did CLTS, and L1 didn't shadow CR0.TS
5467 * but we did (!fpu_active). We need to keep GUEST_CR0.TS on,
5468 * just pretend it's off (also in arch.cr0 for fpu_activate).
5470 vmcs_writel(CR0_READ_SHADOW,
5471 vmcs_readl(CR0_READ_SHADOW) & ~X86_CR0_TS);
5472 vcpu->arch.cr0 &= ~X86_CR0_TS;
5474 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
5477 static int handle_cr(struct kvm_vcpu *vcpu)
5479 unsigned long exit_qualification, val;
5484 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5485 cr = exit_qualification & 15;
5486 reg = (exit_qualification >> 8) & 15;
5487 switch ((exit_qualification >> 4) & 3) {
5488 case 0: /* mov to cr */
5489 val = kvm_register_readl(vcpu, reg);
5490 trace_kvm_cr_write(cr, val);
5493 err = handle_set_cr0(vcpu, val);
5494 kvm_complete_insn_gp(vcpu, err);
5497 err = kvm_set_cr3(vcpu, val);
5498 kvm_complete_insn_gp(vcpu, err);
5501 err = handle_set_cr4(vcpu, val);
5502 kvm_complete_insn_gp(vcpu, err);
5505 u8 cr8_prev = kvm_get_cr8(vcpu);
5507 err = kvm_set_cr8(vcpu, cr8);
5508 kvm_complete_insn_gp(vcpu, err);
5509 if (lapic_in_kernel(vcpu))
5511 if (cr8_prev <= cr8)
5513 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
5520 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
5521 skip_emulated_instruction(vcpu);
5522 vmx_fpu_activate(vcpu);
5524 case 1: /*mov from cr*/
5527 val = kvm_read_cr3(vcpu);
5528 kvm_register_write(vcpu, reg, val);
5529 trace_kvm_cr_read(cr, val);
5530 skip_emulated_instruction(vcpu);
5533 val = kvm_get_cr8(vcpu);
5534 kvm_register_write(vcpu, reg, val);
5535 trace_kvm_cr_read(cr, val);
5536 skip_emulated_instruction(vcpu);
5541 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
5542 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
5543 kvm_lmsw(vcpu, val);
5545 skip_emulated_instruction(vcpu);
5550 vcpu->run->exit_reason = 0;
5551 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
5552 (int)(exit_qualification >> 4) & 3, cr);
5556 static int handle_dr(struct kvm_vcpu *vcpu)
5558 unsigned long exit_qualification;
5561 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5562 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
5564 /* First, if DR does not exist, trigger UD */
5565 if (!kvm_require_dr(vcpu, dr))
5568 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
5569 if (!kvm_require_cpl(vcpu, 0))
5571 dr7 = vmcs_readl(GUEST_DR7);
5574 * As the vm-exit takes precedence over the debug trap, we
5575 * need to emulate the latter, either for the host or the
5576 * guest debugging itself.
5578 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
5579 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
5580 vcpu->run->debug.arch.dr7 = dr7;
5581 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
5582 vcpu->run->debug.arch.exception = DB_VECTOR;
5583 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
5586 vcpu->arch.dr6 &= ~15;
5587 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
5588 kvm_queue_exception(vcpu, DB_VECTOR);
5593 if (vcpu->guest_debug == 0) {
5594 u32 cpu_based_vm_exec_control;
5596 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5597 cpu_based_vm_exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5598 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5601 * No more DR vmexits; force a reload of the debug registers
5602 * and reenter on this instruction. The next vmexit will
5603 * retrieve the full state of the debug registers.
5605 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
5609 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
5610 if (exit_qualification & TYPE_MOV_FROM_DR) {
5613 if (kvm_get_dr(vcpu, dr, &val))
5615 kvm_register_write(vcpu, reg, val);
5617 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
5620 skip_emulated_instruction(vcpu);
5624 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
5626 return vcpu->arch.dr6;
5629 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
5633 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
5635 u32 cpu_based_vm_exec_control;
5637 get_debugreg(vcpu->arch.db[0], 0);
5638 get_debugreg(vcpu->arch.db[1], 1);
5639 get_debugreg(vcpu->arch.db[2], 2);
5640 get_debugreg(vcpu->arch.db[3], 3);
5641 get_debugreg(vcpu->arch.dr6, 6);
5642 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
5644 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
5646 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5647 cpu_based_vm_exec_control |= CPU_BASED_MOV_DR_EXITING;
5648 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5651 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
5653 vmcs_writel(GUEST_DR7, val);
5656 static int handle_cpuid(struct kvm_vcpu *vcpu)
5658 kvm_emulate_cpuid(vcpu);
5662 static int handle_rdmsr(struct kvm_vcpu *vcpu)
5664 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5665 struct msr_data msr_info;
5667 msr_info.index = ecx;
5668 msr_info.host_initiated = false;
5669 if (vmx_get_msr(vcpu, &msr_info)) {
5670 trace_kvm_msr_read_ex(ecx);
5671 kvm_inject_gp(vcpu, 0);
5675 trace_kvm_msr_read(ecx, msr_info.data);
5677 /* FIXME: handling of bits 32:63 of rax, rdx */
5678 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
5679 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
5680 skip_emulated_instruction(vcpu);
5684 static int handle_wrmsr(struct kvm_vcpu *vcpu)
5686 struct msr_data msr;
5687 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
5688 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
5689 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
5693 msr.host_initiated = false;
5694 if (kvm_set_msr(vcpu, &msr) != 0) {
5695 trace_kvm_msr_write_ex(ecx, data);
5696 kvm_inject_gp(vcpu, 0);
5700 trace_kvm_msr_write(ecx, data);
5701 skip_emulated_instruction(vcpu);
5705 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
5707 kvm_make_request(KVM_REQ_EVENT, vcpu);
5711 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
5713 u32 cpu_based_vm_exec_control;
5715 /* clear pending irq */
5716 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5717 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
5718 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5720 kvm_make_request(KVM_REQ_EVENT, vcpu);
5722 ++vcpu->stat.irq_window_exits;
5726 static int handle_halt(struct kvm_vcpu *vcpu)
5728 return kvm_emulate_halt(vcpu);
5731 static int handle_vmcall(struct kvm_vcpu *vcpu)
5733 kvm_emulate_hypercall(vcpu);
5737 static int handle_invd(struct kvm_vcpu *vcpu)
5739 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5742 static int handle_invlpg(struct kvm_vcpu *vcpu)
5744 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5746 kvm_mmu_invlpg(vcpu, exit_qualification);
5747 skip_emulated_instruction(vcpu);
5751 static int handle_rdpmc(struct kvm_vcpu *vcpu)
5755 err = kvm_rdpmc(vcpu);
5756 kvm_complete_insn_gp(vcpu, err);
5761 static int handle_wbinvd(struct kvm_vcpu *vcpu)
5763 kvm_emulate_wbinvd(vcpu);
5767 static int handle_xsetbv(struct kvm_vcpu *vcpu)
5769 u64 new_bv = kvm_read_edx_eax(vcpu);
5770 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
5772 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
5773 skip_emulated_instruction(vcpu);
5777 static int handle_xsaves(struct kvm_vcpu *vcpu)
5779 skip_emulated_instruction(vcpu);
5780 WARN(1, "this should never happen\n");
5784 static int handle_xrstors(struct kvm_vcpu *vcpu)
5786 skip_emulated_instruction(vcpu);
5787 WARN(1, "this should never happen\n");
5791 static int handle_apic_access(struct kvm_vcpu *vcpu)
5793 if (likely(fasteoi)) {
5794 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5795 int access_type, offset;
5797 access_type = exit_qualification & APIC_ACCESS_TYPE;
5798 offset = exit_qualification & APIC_ACCESS_OFFSET;
5800 * Sane guest uses MOV to write EOI, with written value
5801 * not cared. So make a short-circuit here by avoiding
5802 * heavy instruction emulation.
5804 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
5805 (offset == APIC_EOI)) {
5806 kvm_lapic_set_eoi(vcpu);
5807 skip_emulated_instruction(vcpu);
5811 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5814 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
5816 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5817 int vector = exit_qualification & 0xff;
5819 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
5820 kvm_apic_set_eoi_accelerated(vcpu, vector);
5824 static int handle_apic_write(struct kvm_vcpu *vcpu)
5826 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5827 u32 offset = exit_qualification & 0xfff;
5829 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
5830 kvm_apic_write_nodecode(vcpu, offset);
5834 static int handle_task_switch(struct kvm_vcpu *vcpu)
5836 struct vcpu_vmx *vmx = to_vmx(vcpu);
5837 unsigned long exit_qualification;
5838 bool has_error_code = false;
5841 int reason, type, idt_v, idt_index;
5843 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
5844 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
5845 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
5847 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5849 reason = (u32)exit_qualification >> 30;
5850 if (reason == TASK_SWITCH_GATE && idt_v) {
5852 case INTR_TYPE_NMI_INTR:
5853 vcpu->arch.nmi_injected = false;
5854 vmx_set_nmi_mask(vcpu, true);
5856 case INTR_TYPE_EXT_INTR:
5857 case INTR_TYPE_SOFT_INTR:
5858 kvm_clear_interrupt_queue(vcpu);
5860 case INTR_TYPE_HARD_EXCEPTION:
5861 if (vmx->idt_vectoring_info &
5862 VECTORING_INFO_DELIVER_CODE_MASK) {
5863 has_error_code = true;
5865 vmcs_read32(IDT_VECTORING_ERROR_CODE);
5868 case INTR_TYPE_SOFT_EXCEPTION:
5869 kvm_clear_exception_queue(vcpu);
5875 tss_selector = exit_qualification;
5877 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
5878 type != INTR_TYPE_EXT_INTR &&
5879 type != INTR_TYPE_NMI_INTR))
5880 skip_emulated_instruction(vcpu);
5882 if (kvm_task_switch(vcpu, tss_selector,
5883 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
5884 has_error_code, error_code) == EMULATE_FAIL) {
5885 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5886 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
5887 vcpu->run->internal.ndata = 0;
5892 * TODO: What about debug traps on tss switch?
5893 * Are we supposed to inject them and update dr6?
5899 static int handle_ept_violation(struct kvm_vcpu *vcpu)
5901 unsigned long exit_qualification;
5906 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5908 gla_validity = (exit_qualification >> 7) & 0x3;
5909 if (gla_validity != 0x3 && gla_validity != 0x1 && gla_validity != 0) {
5910 printk(KERN_ERR "EPT: Handling EPT violation failed!\n");
5911 printk(KERN_ERR "EPT: GPA: 0x%lx, GVA: 0x%lx\n",
5912 (long unsigned int)vmcs_read64(GUEST_PHYSICAL_ADDRESS),
5913 vmcs_readl(GUEST_LINEAR_ADDRESS));
5914 printk(KERN_ERR "EPT: Exit qualification is 0x%lx\n",
5915 (long unsigned int)exit_qualification);
5916 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5917 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_VIOLATION;
5922 * EPT violation happened while executing iret from NMI,
5923 * "blocked by NMI" bit has to be set before next VM entry.
5924 * There are errata that may cause this bit to not be set:
5927 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
5928 cpu_has_virtual_nmis() &&
5929 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
5930 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
5932 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5933 trace_kvm_page_fault(gpa, exit_qualification);
5935 /* It is a write fault? */
5936 error_code = exit_qualification & PFERR_WRITE_MASK;
5937 /* It is a fetch fault? */
5938 error_code |= (exit_qualification << 2) & PFERR_FETCH_MASK;
5939 /* ept page table is present? */
5940 error_code |= (exit_qualification >> 3) & PFERR_PRESENT_MASK;
5942 vcpu->arch.exit_qualification = exit_qualification;
5944 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
5947 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
5952 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
5953 if (!kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
5954 skip_emulated_instruction(vcpu);
5955 trace_kvm_fast_mmio(gpa);
5959 ret = handle_mmio_page_fault(vcpu, gpa, true);
5960 if (likely(ret == RET_MMIO_PF_EMULATE))
5961 return x86_emulate_instruction(vcpu, gpa, 0, NULL, 0) ==
5964 if (unlikely(ret == RET_MMIO_PF_INVALID))
5965 return kvm_mmu_page_fault(vcpu, gpa, 0, NULL, 0);
5967 if (unlikely(ret == RET_MMIO_PF_RETRY))
5970 /* It is the real ept misconfig */
5973 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
5974 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
5979 static int handle_nmi_window(struct kvm_vcpu *vcpu)
5981 u32 cpu_based_vm_exec_control;
5983 /* clear pending NMI */
5984 cpu_based_vm_exec_control = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
5985 cpu_based_vm_exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
5986 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control);
5987 ++vcpu->stat.nmi_window_exits;
5988 kvm_make_request(KVM_REQ_EVENT, vcpu);
5993 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
5995 struct vcpu_vmx *vmx = to_vmx(vcpu);
5996 enum emulation_result err = EMULATE_DONE;
5999 bool intr_window_requested;
6000 unsigned count = 130;
6002 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6003 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6005 while (vmx->emulation_required && count-- != 0) {
6006 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6007 return handle_interrupt_window(&vmx->vcpu);
6009 if (test_bit(KVM_REQ_EVENT, &vcpu->requests))
6012 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
6014 if (err == EMULATE_USER_EXIT) {
6015 ++vcpu->stat.mmio_exits;
6020 if (err != EMULATE_DONE) {
6021 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6022 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6023 vcpu->run->internal.ndata = 0;
6027 if (vcpu->arch.halt_request) {
6028 vcpu->arch.halt_request = 0;
6029 ret = kvm_vcpu_halt(vcpu);
6033 if (signal_pending(current))
6043 static int __grow_ple_window(int val)
6045 if (ple_window_grow < 1)
6048 val = min(val, ple_window_actual_max);
6050 if (ple_window_grow < ple_window)
6051 val *= ple_window_grow;
6053 val += ple_window_grow;
6058 static int __shrink_ple_window(int val, int modifier, int minimum)
6063 if (modifier < ple_window)
6068 return max(val, minimum);
6071 static void grow_ple_window(struct kvm_vcpu *vcpu)
6073 struct vcpu_vmx *vmx = to_vmx(vcpu);
6074 int old = vmx->ple_window;
6076 vmx->ple_window = __grow_ple_window(old);
6078 if (vmx->ple_window != old)
6079 vmx->ple_window_dirty = true;
6081 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6084 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6086 struct vcpu_vmx *vmx = to_vmx(vcpu);
6087 int old = vmx->ple_window;
6089 vmx->ple_window = __shrink_ple_window(old,
6090 ple_window_shrink, ple_window);
6092 if (vmx->ple_window != old)
6093 vmx->ple_window_dirty = true;
6095 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6099 * ple_window_actual_max is computed to be one grow_ple_window() below
6100 * ple_window_max. (See __grow_ple_window for the reason.)
6101 * This prevents overflows, because ple_window_max is int.
6102 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6104 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6106 static void update_ple_window_actual_max(void)
6108 ple_window_actual_max =
6109 __shrink_ple_window(max(ple_window_max, ple_window),
6110 ple_window_grow, INT_MIN);
6114 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6116 static void wakeup_handler(void)
6118 struct kvm_vcpu *vcpu;
6119 int cpu = smp_processor_id();
6121 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6122 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6123 blocked_vcpu_list) {
6124 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6126 if (pi_test_on(pi_desc) == 1)
6127 kvm_vcpu_kick(vcpu);
6129 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6132 static __init int hardware_setup(void)
6134 int r = -ENOMEM, i, msr;
6136 rdmsrl_safe(MSR_EFER, &host_efer);
6138 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6139 kvm_define_shared_msr(i, vmx_msr_index[i]);
6141 vmx_io_bitmap_a = (unsigned long *)__get_free_page(GFP_KERNEL);
6142 if (!vmx_io_bitmap_a)
6145 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6146 if (!vmx_io_bitmap_b)
6149 vmx_msr_bitmap_legacy = (unsigned long *)__get_free_page(GFP_KERNEL);
6150 if (!vmx_msr_bitmap_legacy)
6153 vmx_msr_bitmap_legacy_x2apic =
6154 (unsigned long *)__get_free_page(GFP_KERNEL);
6155 if (!vmx_msr_bitmap_legacy_x2apic)
6158 vmx_msr_bitmap_longmode = (unsigned long *)__get_free_page(GFP_KERNEL);
6159 if (!vmx_msr_bitmap_longmode)
6162 vmx_msr_bitmap_longmode_x2apic =
6163 (unsigned long *)__get_free_page(GFP_KERNEL);
6164 if (!vmx_msr_bitmap_longmode_x2apic)
6168 vmx_msr_bitmap_nested =
6169 (unsigned long *)__get_free_page(GFP_KERNEL);
6170 if (!vmx_msr_bitmap_nested)
6174 vmx_vmread_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6175 if (!vmx_vmread_bitmap)
6178 vmx_vmwrite_bitmap = (unsigned long *)__get_free_page(GFP_KERNEL);
6179 if (!vmx_vmwrite_bitmap)
6182 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6183 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6185 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6187 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6189 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6190 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6192 memset(vmx_msr_bitmap_nested, 0xff, PAGE_SIZE);
6194 if (setup_vmcs_config(&vmcs_config) < 0) {
6199 if (boot_cpu_has(X86_FEATURE_NX))
6200 kvm_enable_efer_bits(EFER_NX);
6202 if (!cpu_has_vmx_vpid())
6204 if (!cpu_has_vmx_shadow_vmcs())
6205 enable_shadow_vmcs = 0;
6206 if (enable_shadow_vmcs)
6207 init_vmcs_shadow_fields();
6209 if (!cpu_has_vmx_ept() ||
6210 !cpu_has_vmx_ept_4levels()) {
6212 enable_unrestricted_guest = 0;
6213 enable_ept_ad_bits = 0;
6216 if (!cpu_has_vmx_ept_ad_bits())
6217 enable_ept_ad_bits = 0;
6219 if (!cpu_has_vmx_unrestricted_guest())
6220 enable_unrestricted_guest = 0;
6222 if (!cpu_has_vmx_flexpriority())
6223 flexpriority_enabled = 0;
6226 * set_apic_access_page_addr() is used to reload apic access
6227 * page upon invalidation. No need to do anything if not
6228 * using the APIC_ACCESS_ADDR VMCS field.
6230 if (!flexpriority_enabled)
6231 kvm_x86_ops->set_apic_access_page_addr = NULL;
6233 if (!cpu_has_vmx_tpr_shadow())
6234 kvm_x86_ops->update_cr8_intercept = NULL;
6236 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6237 kvm_disable_largepages();
6239 if (!cpu_has_vmx_ple())
6242 if (!cpu_has_vmx_apicv())
6245 if (cpu_has_vmx_tsc_scaling()) {
6246 kvm_has_tsc_control = true;
6247 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6248 kvm_tsc_scaling_ratio_frac_bits = 48;
6252 kvm_x86_ops->update_cr8_intercept = NULL;
6254 kvm_x86_ops->hwapic_irr_update = NULL;
6255 kvm_x86_ops->hwapic_isr_update = NULL;
6256 kvm_x86_ops->deliver_posted_interrupt = NULL;
6257 kvm_x86_ops->sync_pir_to_irr = vmx_sync_pir_to_irr_dummy;
6260 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6261 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6262 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6263 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6264 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6265 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6267 memcpy(vmx_msr_bitmap_legacy_x2apic,
6268 vmx_msr_bitmap_legacy, PAGE_SIZE);
6269 memcpy(vmx_msr_bitmap_longmode_x2apic,
6270 vmx_msr_bitmap_longmode, PAGE_SIZE);
6272 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6275 for (msr = 0x800; msr <= 0x8ff; msr++)
6276 vmx_disable_intercept_msr_read_x2apic(msr);
6278 /* According SDM, in x2apic mode, the whole id reg is used.
6279 * But in KVM, it only use the highest eight bits. Need to
6281 vmx_enable_intercept_msr_read_x2apic(0x802);
6283 vmx_enable_intercept_msr_read_x2apic(0x839);
6285 vmx_disable_intercept_msr_write_x2apic(0x808);
6287 vmx_disable_intercept_msr_write_x2apic(0x80b);
6289 vmx_disable_intercept_msr_write_x2apic(0x83f);
6293 kvm_mmu_set_mask_ptes(0ull,
6294 (enable_ept_ad_bits) ? VMX_EPT_ACCESS_BIT : 0ull,
6295 (enable_ept_ad_bits) ? VMX_EPT_DIRTY_BIT : 0ull,
6296 0ull, VMX_EPT_EXECUTABLE_MASK);
6297 ept_set_mmio_spte_mask();
6302 update_ple_window_actual_max();
6305 * Only enable PML when hardware supports PML feature, and both EPT
6306 * and EPT A/D bit features are enabled -- PML depends on them to work.
6308 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6312 kvm_x86_ops->slot_enable_log_dirty = NULL;
6313 kvm_x86_ops->slot_disable_log_dirty = NULL;
6314 kvm_x86_ops->flush_log_dirty = NULL;
6315 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6318 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6320 return alloc_kvm_area();
6323 free_page((unsigned long)vmx_vmwrite_bitmap);
6325 free_page((unsigned long)vmx_vmread_bitmap);
6328 free_page((unsigned long)vmx_msr_bitmap_nested);
6330 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6332 free_page((unsigned long)vmx_msr_bitmap_longmode);
6334 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6336 free_page((unsigned long)vmx_msr_bitmap_legacy);
6338 free_page((unsigned long)vmx_io_bitmap_b);
6340 free_page((unsigned long)vmx_io_bitmap_a);
6345 static __exit void hardware_unsetup(void)
6347 free_page((unsigned long)vmx_msr_bitmap_legacy_x2apic);
6348 free_page((unsigned long)vmx_msr_bitmap_longmode_x2apic);
6349 free_page((unsigned long)vmx_msr_bitmap_legacy);
6350 free_page((unsigned long)vmx_msr_bitmap_longmode);
6351 free_page((unsigned long)vmx_io_bitmap_b);
6352 free_page((unsigned long)vmx_io_bitmap_a);
6353 free_page((unsigned long)vmx_vmwrite_bitmap);
6354 free_page((unsigned long)vmx_vmread_bitmap);
6356 free_page((unsigned long)vmx_msr_bitmap_nested);
6362 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6363 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6365 static int handle_pause(struct kvm_vcpu *vcpu)
6368 grow_ple_window(vcpu);
6370 skip_emulated_instruction(vcpu);
6371 kvm_vcpu_on_spin(vcpu);
6376 static int handle_nop(struct kvm_vcpu *vcpu)
6378 skip_emulated_instruction(vcpu);
6382 static int handle_mwait(struct kvm_vcpu *vcpu)
6384 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6385 return handle_nop(vcpu);
6388 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6393 static int handle_monitor(struct kvm_vcpu *vcpu)
6395 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6396 return handle_nop(vcpu);
6400 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6401 * We could reuse a single VMCS for all the L2 guests, but we also want the
6402 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6403 * allows keeping them loaded on the processor, and in the future will allow
6404 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6405 * every entry if they never change.
6406 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6407 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6409 * The following functions allocate and free a vmcs02 in this pool.
6412 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6413 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6415 struct vmcs02_list *item;
6416 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6417 if (item->vmptr == vmx->nested.current_vmptr) {
6418 list_move(&item->list, &vmx->nested.vmcs02_pool);
6419 return &item->vmcs02;
6422 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6423 /* Recycle the least recently used VMCS. */
6424 item = list_entry(vmx->nested.vmcs02_pool.prev,
6425 struct vmcs02_list, list);
6426 item->vmptr = vmx->nested.current_vmptr;
6427 list_move(&item->list, &vmx->nested.vmcs02_pool);
6428 return &item->vmcs02;
6431 /* Create a new VMCS */
6432 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6435 item->vmcs02.vmcs = alloc_vmcs();
6436 if (!item->vmcs02.vmcs) {
6440 loaded_vmcs_init(&item->vmcs02);
6441 item->vmptr = vmx->nested.current_vmptr;
6442 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6443 vmx->nested.vmcs02_num++;
6444 return &item->vmcs02;
6447 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6448 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6450 struct vmcs02_list *item;
6451 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6452 if (item->vmptr == vmptr) {
6453 free_loaded_vmcs(&item->vmcs02);
6454 list_del(&item->list);
6456 vmx->nested.vmcs02_num--;
6462 * Free all VMCSs saved for this vcpu, except the one pointed by
6463 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
6464 * must be &vmx->vmcs01.
6466 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
6468 struct vmcs02_list *item, *n;
6470 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
6471 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
6473 * Something will leak if the above WARN triggers. Better than
6476 if (vmx->loaded_vmcs == &item->vmcs02)
6479 free_loaded_vmcs(&item->vmcs02);
6480 list_del(&item->list);
6482 vmx->nested.vmcs02_num--;
6487 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
6488 * set the success or error code of an emulated VMX instruction, as specified
6489 * by Vol 2B, VMX Instruction Reference, "Conventions".
6491 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
6493 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
6494 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6495 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
6498 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
6500 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6501 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
6502 X86_EFLAGS_SF | X86_EFLAGS_OF))
6506 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
6507 u32 vm_instruction_error)
6509 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
6511 * failValid writes the error number to the current VMCS, which
6512 * can't be done there isn't a current VMCS.
6514 nested_vmx_failInvalid(vcpu);
6517 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
6518 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
6519 X86_EFLAGS_SF | X86_EFLAGS_OF))
6521 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
6523 * We don't need to force a shadow sync because
6524 * VM_INSTRUCTION_ERROR is not shadowed
6528 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
6530 /* TODO: not to reset guest simply here. */
6531 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6532 pr_warn("kvm: nested vmx abort, indicator %d\n", indicator);
6535 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
6537 struct vcpu_vmx *vmx =
6538 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
6540 vmx->nested.preemption_timer_expired = true;
6541 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
6542 kvm_vcpu_kick(&vmx->vcpu);
6544 return HRTIMER_NORESTART;
6548 * Decode the memory-address operand of a vmx instruction, as recorded on an
6549 * exit caused by such an instruction (run by a guest hypervisor).
6550 * On success, returns 0. When the operand is invalid, returns 1 and throws
6553 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
6554 unsigned long exit_qualification,
6555 u32 vmx_instruction_info, bool wr, gva_t *ret)
6559 struct kvm_segment s;
6562 * According to Vol. 3B, "Information for VM Exits Due to Instruction
6563 * Execution", on an exit, vmx_instruction_info holds most of the
6564 * addressing components of the operand. Only the displacement part
6565 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
6566 * For how an actual address is calculated from all these components,
6567 * refer to Vol. 1, "Operand Addressing".
6569 int scaling = vmx_instruction_info & 3;
6570 int addr_size = (vmx_instruction_info >> 7) & 7;
6571 bool is_reg = vmx_instruction_info & (1u << 10);
6572 int seg_reg = (vmx_instruction_info >> 15) & 7;
6573 int index_reg = (vmx_instruction_info >> 18) & 0xf;
6574 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
6575 int base_reg = (vmx_instruction_info >> 23) & 0xf;
6576 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
6579 kvm_queue_exception(vcpu, UD_VECTOR);
6583 /* Addr = segment_base + offset */
6584 /* offset = base + [index * scale] + displacement */
6585 off = exit_qualification; /* holds the displacement */
6587 off += kvm_register_read(vcpu, base_reg);
6589 off += kvm_register_read(vcpu, index_reg)<<scaling;
6590 vmx_get_segment(vcpu, &s, seg_reg);
6591 *ret = s.base + off;
6593 if (addr_size == 1) /* 32 bit */
6596 /* Checks for #GP/#SS exceptions. */
6598 if (is_long_mode(vcpu)) {
6599 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
6600 * non-canonical form. This is the only check on the memory
6601 * destination for long mode!
6603 exn = is_noncanonical_address(*ret);
6604 } else if (is_protmode(vcpu)) {
6605 /* Protected mode: apply checks for segment validity in the
6607 * - segment type check (#GP(0) may be thrown)
6608 * - usability check (#GP(0)/#SS(0))
6609 * - limit check (#GP(0)/#SS(0))
6612 /* #GP(0) if the destination operand is located in a
6613 * read-only data segment or any code segment.
6615 exn = ((s.type & 0xa) == 0 || (s.type & 8));
6617 /* #GP(0) if the source operand is located in an
6618 * execute-only code segment
6620 exn = ((s.type & 0xa) == 8);
6622 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
6625 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
6627 exn = (s.unusable != 0);
6628 /* Protected mode: #GP(0)/#SS(0) if the memory
6629 * operand is outside the segment limit.
6631 exn = exn || (off + sizeof(u64) > s.limit);
6634 kvm_queue_exception_e(vcpu,
6635 seg_reg == VCPU_SREG_SS ?
6636 SS_VECTOR : GP_VECTOR,
6645 * This function performs the various checks including
6646 * - if it's 4KB aligned
6647 * - No bits beyond the physical address width are set
6648 * - Returns 0 on success or else 1
6649 * (Intel SDM Section 30.3)
6651 static int nested_vmx_check_vmptr(struct kvm_vcpu *vcpu, int exit_reason,
6656 struct x86_exception e;
6658 struct vcpu_vmx *vmx = to_vmx(vcpu);
6659 int maxphyaddr = cpuid_maxphyaddr(vcpu);
6661 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
6662 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
6665 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vmptr,
6666 sizeof(vmptr), &e)) {
6667 kvm_inject_page_fault(vcpu, &e);
6671 switch (exit_reason) {
6672 case EXIT_REASON_VMON:
6675 * The first 4 bytes of VMXON region contain the supported
6676 * VMCS revision identifier
6678 * Note - IA32_VMX_BASIC[48] will never be 1
6679 * for the nested case;
6680 * which replaces physical address width with 32
6683 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6684 nested_vmx_failInvalid(vcpu);
6685 skip_emulated_instruction(vcpu);
6689 page = nested_get_page(vcpu, vmptr);
6691 nested_vmx_failInvalid(vcpu);
6692 skip_emulated_instruction(vcpu);
6695 if (*(u32 *)kmap(page) != VMCS12_REVISION) {
6697 nested_release_page_clean(page);
6698 nested_vmx_failInvalid(vcpu);
6699 skip_emulated_instruction(vcpu);
6703 nested_release_page_clean(page);
6704 vmx->nested.vmxon_ptr = vmptr;
6706 case EXIT_REASON_VMCLEAR:
6707 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6708 nested_vmx_failValid(vcpu,
6709 VMXERR_VMCLEAR_INVALID_ADDRESS);
6710 skip_emulated_instruction(vcpu);
6714 if (vmptr == vmx->nested.vmxon_ptr) {
6715 nested_vmx_failValid(vcpu,
6716 VMXERR_VMCLEAR_VMXON_POINTER);
6717 skip_emulated_instruction(vcpu);
6721 case EXIT_REASON_VMPTRLD:
6722 if (!PAGE_ALIGNED(vmptr) || (vmptr >> maxphyaddr)) {
6723 nested_vmx_failValid(vcpu,
6724 VMXERR_VMPTRLD_INVALID_ADDRESS);
6725 skip_emulated_instruction(vcpu);
6729 if (vmptr == vmx->nested.vmxon_ptr) {
6730 nested_vmx_failValid(vcpu,
6731 VMXERR_VMCLEAR_VMXON_POINTER);
6732 skip_emulated_instruction(vcpu);
6737 return 1; /* shouldn't happen */
6746 * Emulate the VMXON instruction.
6747 * Currently, we just remember that VMX is active, and do not save or even
6748 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
6749 * do not currently need to store anything in that guest-allocated memory
6750 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
6751 * argument is different from the VMXON pointer (which the spec says they do).
6753 static int handle_vmon(struct kvm_vcpu *vcpu)
6755 struct kvm_segment cs;
6756 struct vcpu_vmx *vmx = to_vmx(vcpu);
6757 struct vmcs *shadow_vmcs;
6758 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
6759 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
6761 /* The Intel VMX Instruction Reference lists a bunch of bits that
6762 * are prerequisite to running VMXON, most notably cr4.VMXE must be
6763 * set to 1 (see vmx_set_cr4() for when we allow the guest to set this).
6764 * Otherwise, we should fail with #UD. We test these now:
6766 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE) ||
6767 !kvm_read_cr0_bits(vcpu, X86_CR0_PE) ||
6768 (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
6769 kvm_queue_exception(vcpu, UD_VECTOR);
6773 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6774 if (is_long_mode(vcpu) && !cs.l) {
6775 kvm_queue_exception(vcpu, UD_VECTOR);
6779 if (vmx_get_cpl(vcpu)) {
6780 kvm_inject_gp(vcpu, 0);
6784 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMON, NULL))
6787 if (vmx->nested.vmxon) {
6788 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
6789 skip_emulated_instruction(vcpu);
6793 if ((vmx->nested.msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
6794 != VMXON_NEEDED_FEATURES) {
6795 kvm_inject_gp(vcpu, 0);
6799 if (enable_shadow_vmcs) {
6800 shadow_vmcs = alloc_vmcs();
6803 /* mark vmcs as shadow */
6804 shadow_vmcs->revision_id |= (1u << 31);
6805 /* init shadow vmcs */
6806 vmcs_clear(shadow_vmcs);
6807 vmx->nested.current_shadow_vmcs = shadow_vmcs;
6810 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
6811 vmx->nested.vmcs02_num = 0;
6813 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
6815 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
6817 vmx->nested.vmxon = true;
6819 skip_emulated_instruction(vcpu);
6820 nested_vmx_succeed(vcpu);
6825 * Intel's VMX Instruction Reference specifies a common set of prerequisites
6826 * for running VMX instructions (except VMXON, whose prerequisites are
6827 * slightly different). It also specifies what exception to inject otherwise.
6829 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
6831 struct kvm_segment cs;
6832 struct vcpu_vmx *vmx = to_vmx(vcpu);
6834 if (!vmx->nested.vmxon) {
6835 kvm_queue_exception(vcpu, UD_VECTOR);
6839 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
6840 if ((vmx_get_rflags(vcpu) & X86_EFLAGS_VM) ||
6841 (is_long_mode(vcpu) && !cs.l)) {
6842 kvm_queue_exception(vcpu, UD_VECTOR);
6846 if (vmx_get_cpl(vcpu)) {
6847 kvm_inject_gp(vcpu, 0);
6854 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
6856 if (vmx->nested.current_vmptr == -1ull)
6859 /* current_vmptr and current_vmcs12 are always set/reset together */
6860 if (WARN_ON(vmx->nested.current_vmcs12 == NULL))
6863 if (enable_shadow_vmcs) {
6864 /* copy to memory all shadowed fields in case
6865 they were modified */
6866 copy_shadow_to_vmcs12(vmx);
6867 vmx->nested.sync_shadow_vmcs = false;
6868 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
6869 SECONDARY_EXEC_SHADOW_VMCS);
6870 vmcs_write64(VMCS_LINK_POINTER, -1ull);
6872 vmx->nested.posted_intr_nv = -1;
6873 kunmap(vmx->nested.current_vmcs12_page);
6874 nested_release_page(vmx->nested.current_vmcs12_page);
6875 vmx->nested.current_vmptr = -1ull;
6876 vmx->nested.current_vmcs12 = NULL;
6880 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
6881 * just stops using VMX.
6883 static void free_nested(struct vcpu_vmx *vmx)
6885 if (!vmx->nested.vmxon)
6888 vmx->nested.vmxon = false;
6889 free_vpid(vmx->nested.vpid02);
6890 nested_release_vmcs12(vmx);
6891 if (enable_shadow_vmcs)
6892 free_vmcs(vmx->nested.current_shadow_vmcs);
6893 /* Unpin physical memory we referred to in current vmcs02 */
6894 if (vmx->nested.apic_access_page) {
6895 nested_release_page(vmx->nested.apic_access_page);
6896 vmx->nested.apic_access_page = NULL;
6898 if (vmx->nested.virtual_apic_page) {
6899 nested_release_page(vmx->nested.virtual_apic_page);
6900 vmx->nested.virtual_apic_page = NULL;
6902 if (vmx->nested.pi_desc_page) {
6903 kunmap(vmx->nested.pi_desc_page);
6904 nested_release_page(vmx->nested.pi_desc_page);
6905 vmx->nested.pi_desc_page = NULL;
6906 vmx->nested.pi_desc = NULL;
6909 nested_free_all_saved_vmcss(vmx);
6912 /* Emulate the VMXOFF instruction */
6913 static int handle_vmoff(struct kvm_vcpu *vcpu)
6915 if (!nested_vmx_check_permission(vcpu))
6917 free_nested(to_vmx(vcpu));
6918 skip_emulated_instruction(vcpu);
6919 nested_vmx_succeed(vcpu);
6923 /* Emulate the VMCLEAR instruction */
6924 static int handle_vmclear(struct kvm_vcpu *vcpu)
6926 struct vcpu_vmx *vmx = to_vmx(vcpu);
6928 struct vmcs12 *vmcs12;
6931 if (!nested_vmx_check_permission(vcpu))
6934 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMCLEAR, &vmptr))
6937 if (vmptr == vmx->nested.current_vmptr)
6938 nested_release_vmcs12(vmx);
6940 page = nested_get_page(vcpu, vmptr);
6943 * For accurate processor emulation, VMCLEAR beyond available
6944 * physical memory should do nothing at all. However, it is
6945 * possible that a nested vmx bug, not a guest hypervisor bug,
6946 * resulted in this case, so let's shut down before doing any
6949 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
6952 vmcs12 = kmap(page);
6953 vmcs12->launch_state = 0;
6955 nested_release_page(page);
6957 nested_free_vmcs02(vmx, vmptr);
6959 skip_emulated_instruction(vcpu);
6960 nested_vmx_succeed(vcpu);
6964 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
6966 /* Emulate the VMLAUNCH instruction */
6967 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
6969 return nested_vmx_run(vcpu, true);
6972 /* Emulate the VMRESUME instruction */
6973 static int handle_vmresume(struct kvm_vcpu *vcpu)
6976 return nested_vmx_run(vcpu, false);
6979 enum vmcs_field_type {
6980 VMCS_FIELD_TYPE_U16 = 0,
6981 VMCS_FIELD_TYPE_U64 = 1,
6982 VMCS_FIELD_TYPE_U32 = 2,
6983 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
6986 static inline int vmcs_field_type(unsigned long field)
6988 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
6989 return VMCS_FIELD_TYPE_U32;
6990 return (field >> 13) & 0x3 ;
6993 static inline int vmcs_field_readonly(unsigned long field)
6995 return (((field >> 10) & 0x3) == 1);
6999 * Read a vmcs12 field. Since these can have varying lengths and we return
7000 * one type, we chose the biggest type (u64) and zero-extend the return value
7001 * to that size. Note that the caller, handle_vmread, might need to use only
7002 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7003 * 64-bit fields are to be returned).
7005 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7006 unsigned long field, u64 *ret)
7008 short offset = vmcs_field_to_offset(field);
7014 p = ((char *)(get_vmcs12(vcpu))) + offset;
7016 switch (vmcs_field_type(field)) {
7017 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7018 *ret = *((natural_width *)p);
7020 case VMCS_FIELD_TYPE_U16:
7023 case VMCS_FIELD_TYPE_U32:
7026 case VMCS_FIELD_TYPE_U64:
7036 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7037 unsigned long field, u64 field_value){
7038 short offset = vmcs_field_to_offset(field);
7039 char *p = ((char *) get_vmcs12(vcpu)) + offset;
7043 switch (vmcs_field_type(field)) {
7044 case VMCS_FIELD_TYPE_U16:
7045 *(u16 *)p = field_value;
7047 case VMCS_FIELD_TYPE_U32:
7048 *(u32 *)p = field_value;
7050 case VMCS_FIELD_TYPE_U64:
7051 *(u64 *)p = field_value;
7053 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7054 *(natural_width *)p = field_value;
7063 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7066 unsigned long field;
7068 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7069 const unsigned long *fields = shadow_read_write_fields;
7070 const int num_fields = max_shadow_read_write_fields;
7074 vmcs_load(shadow_vmcs);
7076 for (i = 0; i < num_fields; i++) {
7078 switch (vmcs_field_type(field)) {
7079 case VMCS_FIELD_TYPE_U16:
7080 field_value = vmcs_read16(field);
7082 case VMCS_FIELD_TYPE_U32:
7083 field_value = vmcs_read32(field);
7085 case VMCS_FIELD_TYPE_U64:
7086 field_value = vmcs_read64(field);
7088 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7089 field_value = vmcs_readl(field);
7095 vmcs12_write_any(&vmx->vcpu, field, field_value);
7098 vmcs_clear(shadow_vmcs);
7099 vmcs_load(vmx->loaded_vmcs->vmcs);
7104 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7106 const unsigned long *fields[] = {
7107 shadow_read_write_fields,
7108 shadow_read_only_fields
7110 const int max_fields[] = {
7111 max_shadow_read_write_fields,
7112 max_shadow_read_only_fields
7115 unsigned long field;
7116 u64 field_value = 0;
7117 struct vmcs *shadow_vmcs = vmx->nested.current_shadow_vmcs;
7119 vmcs_load(shadow_vmcs);
7121 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7122 for (i = 0; i < max_fields[q]; i++) {
7123 field = fields[q][i];
7124 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7126 switch (vmcs_field_type(field)) {
7127 case VMCS_FIELD_TYPE_U16:
7128 vmcs_write16(field, (u16)field_value);
7130 case VMCS_FIELD_TYPE_U32:
7131 vmcs_write32(field, (u32)field_value);
7133 case VMCS_FIELD_TYPE_U64:
7134 vmcs_write64(field, (u64)field_value);
7136 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7137 vmcs_writel(field, (long)field_value);
7146 vmcs_clear(shadow_vmcs);
7147 vmcs_load(vmx->loaded_vmcs->vmcs);
7151 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7152 * used before) all generate the same failure when it is missing.
7154 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7156 struct vcpu_vmx *vmx = to_vmx(vcpu);
7157 if (vmx->nested.current_vmptr == -1ull) {
7158 nested_vmx_failInvalid(vcpu);
7159 skip_emulated_instruction(vcpu);
7165 static int handle_vmread(struct kvm_vcpu *vcpu)
7167 unsigned long field;
7169 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7170 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7173 if (!nested_vmx_check_permission(vcpu) ||
7174 !nested_vmx_check_vmcs12(vcpu))
7177 /* Decode instruction info and find the field to read */
7178 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7179 /* Read the field, zero-extended to a u64 field_value */
7180 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7181 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7182 skip_emulated_instruction(vcpu);
7186 * Now copy part of this value to register or memory, as requested.
7187 * Note that the number of bits actually copied is 32 or 64 depending
7188 * on the guest's mode (32 or 64 bit), not on the given field's length.
7190 if (vmx_instruction_info & (1u << 10)) {
7191 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7194 if (get_vmx_mem_address(vcpu, exit_qualification,
7195 vmx_instruction_info, true, &gva))
7197 /* _system ok, as nested_vmx_check_permission verified cpl=0 */
7198 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7199 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7202 nested_vmx_succeed(vcpu);
7203 skip_emulated_instruction(vcpu);
7208 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7210 unsigned long field;
7212 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7213 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7214 /* The value to write might be 32 or 64 bits, depending on L1's long
7215 * mode, and eventually we need to write that into a field of several
7216 * possible lengths. The code below first zero-extends the value to 64
7217 * bit (field_value), and then copies only the approriate number of
7218 * bits into the vmcs12 field.
7220 u64 field_value = 0;
7221 struct x86_exception e;
7223 if (!nested_vmx_check_permission(vcpu) ||
7224 !nested_vmx_check_vmcs12(vcpu))
7227 if (vmx_instruction_info & (1u << 10))
7228 field_value = kvm_register_readl(vcpu,
7229 (((vmx_instruction_info) >> 3) & 0xf));
7231 if (get_vmx_mem_address(vcpu, exit_qualification,
7232 vmx_instruction_info, false, &gva))
7234 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7235 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7236 kvm_inject_page_fault(vcpu, &e);
7242 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7243 if (vmcs_field_readonly(field)) {
7244 nested_vmx_failValid(vcpu,
7245 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7246 skip_emulated_instruction(vcpu);
7250 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7251 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7252 skip_emulated_instruction(vcpu);
7256 nested_vmx_succeed(vcpu);
7257 skip_emulated_instruction(vcpu);
7261 /* Emulate the VMPTRLD instruction */
7262 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7264 struct vcpu_vmx *vmx = to_vmx(vcpu);
7267 if (!nested_vmx_check_permission(vcpu))
7270 if (nested_vmx_check_vmptr(vcpu, EXIT_REASON_VMPTRLD, &vmptr))
7273 if (vmx->nested.current_vmptr != vmptr) {
7274 struct vmcs12 *new_vmcs12;
7276 page = nested_get_page(vcpu, vmptr);
7278 nested_vmx_failInvalid(vcpu);
7279 skip_emulated_instruction(vcpu);
7282 new_vmcs12 = kmap(page);
7283 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7285 nested_release_page_clean(page);
7286 nested_vmx_failValid(vcpu,
7287 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7288 skip_emulated_instruction(vcpu);
7292 nested_release_vmcs12(vmx);
7293 vmx->nested.current_vmptr = vmptr;
7294 vmx->nested.current_vmcs12 = new_vmcs12;
7295 vmx->nested.current_vmcs12_page = page;
7296 if (enable_shadow_vmcs) {
7297 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7298 SECONDARY_EXEC_SHADOW_VMCS);
7299 vmcs_write64(VMCS_LINK_POINTER,
7300 __pa(vmx->nested.current_shadow_vmcs));
7301 vmx->nested.sync_shadow_vmcs = true;
7305 nested_vmx_succeed(vcpu);
7306 skip_emulated_instruction(vcpu);
7310 /* Emulate the VMPTRST instruction */
7311 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7313 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7314 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7316 struct x86_exception e;
7318 if (!nested_vmx_check_permission(vcpu))
7321 if (get_vmx_mem_address(vcpu, exit_qualification,
7322 vmx_instruction_info, true, &vmcs_gva))
7324 /* ok to use *_system, as nested_vmx_check_permission verified cpl=0 */
7325 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7326 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7328 kvm_inject_page_fault(vcpu, &e);
7331 nested_vmx_succeed(vcpu);
7332 skip_emulated_instruction(vcpu);
7336 /* Emulate the INVEPT instruction */
7337 static int handle_invept(struct kvm_vcpu *vcpu)
7339 struct vcpu_vmx *vmx = to_vmx(vcpu);
7340 u32 vmx_instruction_info, types;
7343 struct x86_exception e;
7348 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7349 SECONDARY_EXEC_ENABLE_EPT) ||
7350 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7351 kvm_queue_exception(vcpu, UD_VECTOR);
7355 if (!nested_vmx_check_permission(vcpu))
7358 if (!kvm_read_cr0_bits(vcpu, X86_CR0_PE)) {
7359 kvm_queue_exception(vcpu, UD_VECTOR);
7363 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7364 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7366 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7368 if (!(types & (1UL << type))) {
7369 nested_vmx_failValid(vcpu,
7370 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7371 skip_emulated_instruction(vcpu);
7375 /* According to the Intel VMX instruction reference, the memory
7376 * operand is read even if it isn't needed (e.g., for type==global)
7378 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7379 vmx_instruction_info, false, &gva))
7381 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7382 sizeof(operand), &e)) {
7383 kvm_inject_page_fault(vcpu, &e);
7388 case VMX_EPT_EXTENT_GLOBAL:
7389 kvm_mmu_sync_roots(vcpu);
7390 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7391 nested_vmx_succeed(vcpu);
7394 /* Trap single context invalidation invept calls */
7399 skip_emulated_instruction(vcpu);
7403 static int handle_invvpid(struct kvm_vcpu *vcpu)
7405 struct vcpu_vmx *vmx = to_vmx(vcpu);
7406 u32 vmx_instruction_info;
7407 unsigned long type, types;
7409 struct x86_exception e;
7412 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7413 SECONDARY_EXEC_ENABLE_VPID) ||
7414 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7415 kvm_queue_exception(vcpu, UD_VECTOR);
7419 if (!nested_vmx_check_permission(vcpu))
7422 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7423 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7425 types = (vmx->nested.nested_vmx_vpid_caps >> 8) & 0x7;
7427 if (!(types & (1UL << type))) {
7428 nested_vmx_failValid(vcpu,
7429 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7430 skip_emulated_instruction(vcpu);
7434 /* according to the intel vmx instruction reference, the memory
7435 * operand is read even if it isn't needed (e.g., for type==global)
7437 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7438 vmx_instruction_info, false, &gva))
7440 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &vpid,
7442 kvm_inject_page_fault(vcpu, &e);
7447 case VMX_VPID_EXTENT_SINGLE_CONTEXT:
7449 * Old versions of KVM use the single-context version so we
7450 * have to support it; just treat it the same as all-context.
7452 case VMX_VPID_EXTENT_ALL_CONTEXT:
7453 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
7454 nested_vmx_succeed(vcpu);
7457 /* Trap individual address invalidation invvpid calls */
7462 skip_emulated_instruction(vcpu);
7466 static int handle_pml_full(struct kvm_vcpu *vcpu)
7468 unsigned long exit_qualification;
7470 trace_kvm_pml_full(vcpu->vcpu_id);
7472 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7475 * PML buffer FULL happened while executing iret from NMI,
7476 * "blocked by NMI" bit has to be set before next VM entry.
7478 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7479 cpu_has_virtual_nmis() &&
7480 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7481 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7482 GUEST_INTR_STATE_NMI);
7485 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7486 * here.., and there's no userspace involvement needed for PML.
7491 static int handle_pcommit(struct kvm_vcpu *vcpu)
7493 /* we never catch pcommit instruct for L1 guest. */
7499 * The exit handlers return 1 if the exit was handled fully and guest execution
7500 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
7501 * to be done to userspace and return 0.
7503 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
7504 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
7505 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
7506 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
7507 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
7508 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
7509 [EXIT_REASON_CR_ACCESS] = handle_cr,
7510 [EXIT_REASON_DR_ACCESS] = handle_dr,
7511 [EXIT_REASON_CPUID] = handle_cpuid,
7512 [EXIT_REASON_MSR_READ] = handle_rdmsr,
7513 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
7514 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
7515 [EXIT_REASON_HLT] = handle_halt,
7516 [EXIT_REASON_INVD] = handle_invd,
7517 [EXIT_REASON_INVLPG] = handle_invlpg,
7518 [EXIT_REASON_RDPMC] = handle_rdpmc,
7519 [EXIT_REASON_VMCALL] = handle_vmcall,
7520 [EXIT_REASON_VMCLEAR] = handle_vmclear,
7521 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
7522 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
7523 [EXIT_REASON_VMPTRST] = handle_vmptrst,
7524 [EXIT_REASON_VMREAD] = handle_vmread,
7525 [EXIT_REASON_VMRESUME] = handle_vmresume,
7526 [EXIT_REASON_VMWRITE] = handle_vmwrite,
7527 [EXIT_REASON_VMOFF] = handle_vmoff,
7528 [EXIT_REASON_VMON] = handle_vmon,
7529 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
7530 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
7531 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
7532 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
7533 [EXIT_REASON_WBINVD] = handle_wbinvd,
7534 [EXIT_REASON_XSETBV] = handle_xsetbv,
7535 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
7536 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
7537 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
7538 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
7539 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
7540 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
7541 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
7542 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
7543 [EXIT_REASON_INVEPT] = handle_invept,
7544 [EXIT_REASON_INVVPID] = handle_invvpid,
7545 [EXIT_REASON_XSAVES] = handle_xsaves,
7546 [EXIT_REASON_XRSTORS] = handle_xrstors,
7547 [EXIT_REASON_PML_FULL] = handle_pml_full,
7548 [EXIT_REASON_PCOMMIT] = handle_pcommit,
7551 static const int kvm_vmx_max_exit_handlers =
7552 ARRAY_SIZE(kvm_vmx_exit_handlers);
7554 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
7555 struct vmcs12 *vmcs12)
7557 unsigned long exit_qualification;
7558 gpa_t bitmap, last_bitmap;
7563 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
7564 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
7566 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7568 port = exit_qualification >> 16;
7569 size = (exit_qualification & 7) + 1;
7571 last_bitmap = (gpa_t)-1;
7576 bitmap = vmcs12->io_bitmap_a;
7577 else if (port < 0x10000)
7578 bitmap = vmcs12->io_bitmap_b;
7581 bitmap += (port & 0x7fff) / 8;
7583 if (last_bitmap != bitmap)
7584 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
7586 if (b & (1 << (port & 7)))
7591 last_bitmap = bitmap;
7598 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
7599 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
7600 * disinterest in the current event (read or write a specific MSR) by using an
7601 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
7603 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
7604 struct vmcs12 *vmcs12, u32 exit_reason)
7606 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
7609 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
7613 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
7614 * for the four combinations of read/write and low/high MSR numbers.
7615 * First we need to figure out which of the four to use:
7617 bitmap = vmcs12->msr_bitmap;
7618 if (exit_reason == EXIT_REASON_MSR_WRITE)
7620 if (msr_index >= 0xc0000000) {
7621 msr_index -= 0xc0000000;
7625 /* Then read the msr_index'th bit from this bitmap: */
7626 if (msr_index < 1024*8) {
7628 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
7630 return 1 & (b >> (msr_index & 7));
7632 return true; /* let L1 handle the wrong parameter */
7636 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
7637 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
7638 * intercept (via guest_host_mask etc.) the current event.
7640 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
7641 struct vmcs12 *vmcs12)
7643 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7644 int cr = exit_qualification & 15;
7645 int reg = (exit_qualification >> 8) & 15;
7646 unsigned long val = kvm_register_readl(vcpu, reg);
7648 switch ((exit_qualification >> 4) & 3) {
7649 case 0: /* mov to cr */
7652 if (vmcs12->cr0_guest_host_mask &
7653 (val ^ vmcs12->cr0_read_shadow))
7657 if ((vmcs12->cr3_target_count >= 1 &&
7658 vmcs12->cr3_target_value0 == val) ||
7659 (vmcs12->cr3_target_count >= 2 &&
7660 vmcs12->cr3_target_value1 == val) ||
7661 (vmcs12->cr3_target_count >= 3 &&
7662 vmcs12->cr3_target_value2 == val) ||
7663 (vmcs12->cr3_target_count >= 4 &&
7664 vmcs12->cr3_target_value3 == val))
7666 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
7670 if (vmcs12->cr4_guest_host_mask &
7671 (vmcs12->cr4_read_shadow ^ val))
7675 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
7681 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
7682 (vmcs12->cr0_read_shadow & X86_CR0_TS))
7685 case 1: /* mov from cr */
7688 if (vmcs12->cpu_based_vm_exec_control &
7689 CPU_BASED_CR3_STORE_EXITING)
7693 if (vmcs12->cpu_based_vm_exec_control &
7694 CPU_BASED_CR8_STORE_EXITING)
7701 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
7702 * cr0. Other attempted changes are ignored, with no exit.
7704 if (vmcs12->cr0_guest_host_mask & 0xe &
7705 (val ^ vmcs12->cr0_read_shadow))
7707 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
7708 !(vmcs12->cr0_read_shadow & 0x1) &&
7717 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
7718 * should handle it ourselves in L0 (and then continue L2). Only call this
7719 * when in is_guest_mode (L2).
7721 static bool nested_vmx_exit_handled(struct kvm_vcpu *vcpu)
7723 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
7724 struct vcpu_vmx *vmx = to_vmx(vcpu);
7725 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
7726 u32 exit_reason = vmx->exit_reason;
7728 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
7729 vmcs_readl(EXIT_QUALIFICATION),
7730 vmx->idt_vectoring_info,
7732 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
7735 if (vmx->nested.nested_run_pending)
7738 if (unlikely(vmx->fail)) {
7739 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
7740 vmcs_read32(VM_INSTRUCTION_ERROR));
7744 switch (exit_reason) {
7745 case EXIT_REASON_EXCEPTION_NMI:
7746 if (is_nmi(intr_info))
7748 else if (is_page_fault(intr_info))
7750 else if (is_no_device(intr_info) &&
7751 !(vmcs12->guest_cr0 & X86_CR0_TS))
7753 return vmcs12->exception_bitmap &
7754 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
7755 case EXIT_REASON_EXTERNAL_INTERRUPT:
7757 case EXIT_REASON_TRIPLE_FAULT:
7759 case EXIT_REASON_PENDING_INTERRUPT:
7760 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
7761 case EXIT_REASON_NMI_WINDOW:
7762 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
7763 case EXIT_REASON_TASK_SWITCH:
7765 case EXIT_REASON_CPUID:
7767 case EXIT_REASON_HLT:
7768 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
7769 case EXIT_REASON_INVD:
7771 case EXIT_REASON_INVLPG:
7772 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
7773 case EXIT_REASON_RDPMC:
7774 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
7775 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
7776 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
7777 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
7778 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
7779 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
7780 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
7781 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
7782 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
7784 * VMX instructions trap unconditionally. This allows L1 to
7785 * emulate them for its L2 guest, i.e., allows 3-level nesting!
7788 case EXIT_REASON_CR_ACCESS:
7789 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
7790 case EXIT_REASON_DR_ACCESS:
7791 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
7792 case EXIT_REASON_IO_INSTRUCTION:
7793 return nested_vmx_exit_handled_io(vcpu, vmcs12);
7794 case EXIT_REASON_MSR_READ:
7795 case EXIT_REASON_MSR_WRITE:
7796 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
7797 case EXIT_REASON_INVALID_STATE:
7799 case EXIT_REASON_MWAIT_INSTRUCTION:
7800 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
7801 case EXIT_REASON_MONITOR_TRAP_FLAG:
7802 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
7803 case EXIT_REASON_MONITOR_INSTRUCTION:
7804 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
7805 case EXIT_REASON_PAUSE_INSTRUCTION:
7806 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
7807 nested_cpu_has2(vmcs12,
7808 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
7809 case EXIT_REASON_MCE_DURING_VMENTRY:
7811 case EXIT_REASON_TPR_BELOW_THRESHOLD:
7812 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
7813 case EXIT_REASON_APIC_ACCESS:
7814 return nested_cpu_has2(vmcs12,
7815 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
7816 case EXIT_REASON_APIC_WRITE:
7817 case EXIT_REASON_EOI_INDUCED:
7818 /* apic_write and eoi_induced should exit unconditionally. */
7820 case EXIT_REASON_EPT_VIOLATION:
7822 * L0 always deals with the EPT violation. If nested EPT is
7823 * used, and the nested mmu code discovers that the address is
7824 * missing in the guest EPT table (EPT12), the EPT violation
7825 * will be injected with nested_ept_inject_page_fault()
7828 case EXIT_REASON_EPT_MISCONFIG:
7830 * L2 never uses directly L1's EPT, but rather L0's own EPT
7831 * table (shadow on EPT) or a merged EPT table that L0 built
7832 * (EPT on EPT). So any problems with the structure of the
7833 * table is L0's fault.
7836 case EXIT_REASON_WBINVD:
7837 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
7838 case EXIT_REASON_XSETBV:
7840 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
7842 * This should never happen, since it is not possible to
7843 * set XSS to a non-zero value---neither in L1 nor in L2.
7844 * If if it were, XSS would have to be checked against
7845 * the XSS exit bitmap in vmcs12.
7847 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
7848 case EXIT_REASON_PCOMMIT:
7849 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_PCOMMIT);
7850 case EXIT_REASON_PML_FULL:
7851 /* We don't expose PML support to L1. */
7858 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
7860 *info1 = vmcs_readl(EXIT_QUALIFICATION);
7861 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
7864 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
7867 __free_page(vmx->pml_pg);
7872 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
7874 struct vcpu_vmx *vmx = to_vmx(vcpu);
7878 pml_idx = vmcs_read16(GUEST_PML_INDEX);
7880 /* Do nothing if PML buffer is empty */
7881 if (pml_idx == (PML_ENTITY_NUM - 1))
7884 /* PML index always points to next available PML buffer entity */
7885 if (pml_idx >= PML_ENTITY_NUM)
7890 pml_buf = page_address(vmx->pml_pg);
7891 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
7894 gpa = pml_buf[pml_idx];
7895 WARN_ON(gpa & (PAGE_SIZE - 1));
7896 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
7899 /* reset PML index */
7900 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
7904 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
7905 * Called before reporting dirty_bitmap to userspace.
7907 static void kvm_flush_pml_buffers(struct kvm *kvm)
7910 struct kvm_vcpu *vcpu;
7912 * We only need to kick vcpu out of guest mode here, as PML buffer
7913 * is flushed at beginning of all VMEXITs, and it's obvious that only
7914 * vcpus running in guest are possible to have unflushed GPAs in PML
7917 kvm_for_each_vcpu(i, vcpu, kvm)
7918 kvm_vcpu_kick(vcpu);
7921 static void vmx_dump_sel(char *name, uint32_t sel)
7923 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
7924 name, vmcs_read16(sel),
7925 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
7926 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
7927 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
7930 static void vmx_dump_dtsel(char *name, uint32_t limit)
7932 pr_err("%s limit=0x%08x, base=0x%016lx\n",
7933 name, vmcs_read32(limit),
7934 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
7937 static void dump_vmcs(void)
7939 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
7940 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
7941 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
7942 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
7943 u32 secondary_exec_control = 0;
7944 unsigned long cr4 = vmcs_readl(GUEST_CR4);
7945 u64 efer = vmcs_readl(GUEST_IA32_EFER);
7948 if (cpu_has_secondary_exec_ctrls())
7949 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
7951 pr_err("*** Guest State ***\n");
7952 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
7953 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
7954 vmcs_readl(CR0_GUEST_HOST_MASK));
7955 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
7956 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
7957 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
7958 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
7959 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
7961 pr_err("PDPTR0 = 0x%016lx PDPTR1 = 0x%016lx\n",
7962 vmcs_readl(GUEST_PDPTR0), vmcs_readl(GUEST_PDPTR1));
7963 pr_err("PDPTR2 = 0x%016lx PDPTR3 = 0x%016lx\n",
7964 vmcs_readl(GUEST_PDPTR2), vmcs_readl(GUEST_PDPTR3));
7966 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
7967 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
7968 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
7969 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
7970 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
7971 vmcs_readl(GUEST_SYSENTER_ESP),
7972 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
7973 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
7974 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
7975 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
7976 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
7977 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
7978 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
7979 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
7980 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
7981 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
7982 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
7983 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
7984 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
7985 pr_err("EFER = 0x%016llx PAT = 0x%016lx\n",
7986 efer, vmcs_readl(GUEST_IA32_PAT));
7987 pr_err("DebugCtl = 0x%016lx DebugExceptions = 0x%016lx\n",
7988 vmcs_readl(GUEST_IA32_DEBUGCTL),
7989 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
7990 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
7991 pr_err("PerfGlobCtl = 0x%016lx\n",
7992 vmcs_readl(GUEST_IA32_PERF_GLOBAL_CTRL));
7993 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
7994 pr_err("BndCfgS = 0x%016lx\n", vmcs_readl(GUEST_BNDCFGS));
7995 pr_err("Interruptibility = %08x ActivityState = %08x\n",
7996 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
7997 vmcs_read32(GUEST_ACTIVITY_STATE));
7998 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
7999 pr_err("InterruptStatus = %04x\n",
8000 vmcs_read16(GUEST_INTR_STATUS));
8002 pr_err("*** Host State ***\n");
8003 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
8004 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8005 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8006 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8007 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8008 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8009 vmcs_read16(HOST_TR_SELECTOR));
8010 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8011 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8012 vmcs_readl(HOST_TR_BASE));
8013 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8014 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8015 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8016 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8017 vmcs_readl(HOST_CR4));
8018 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8019 vmcs_readl(HOST_IA32_SYSENTER_ESP),
8020 vmcs_read32(HOST_IA32_SYSENTER_CS),
8021 vmcs_readl(HOST_IA32_SYSENTER_EIP));
8022 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8023 pr_err("EFER = 0x%016lx PAT = 0x%016lx\n",
8024 vmcs_readl(HOST_IA32_EFER), vmcs_readl(HOST_IA32_PAT));
8025 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8026 pr_err("PerfGlobCtl = 0x%016lx\n",
8027 vmcs_readl(HOST_IA32_PERF_GLOBAL_CTRL));
8029 pr_err("*** Control State ***\n");
8030 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8031 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8032 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8033 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8034 vmcs_read32(EXCEPTION_BITMAP),
8035 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8036 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8037 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8038 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8039 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8040 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8041 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8042 vmcs_read32(VM_EXIT_INTR_INFO),
8043 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8044 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8045 pr_err(" reason=%08x qualification=%016lx\n",
8046 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8047 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8048 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8049 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8050 pr_err("TSC Offset = 0x%016lx\n", vmcs_readl(TSC_OFFSET));
8051 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8052 pr_err("TSC Multiplier = 0x%016lx\n",
8053 vmcs_readl(TSC_MULTIPLIER));
8054 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8055 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8056 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8057 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8058 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8059 pr_err("EPT pointer = 0x%016lx\n", vmcs_readl(EPT_POINTER));
8060 n = vmcs_read32(CR3_TARGET_COUNT);
8061 for (i = 0; i + 1 < n; i += 4)
8062 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8063 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8064 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8066 pr_err("CR3 target%u=%016lx\n",
8067 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8068 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8069 pr_err("PLE Gap=%08x Window=%08x\n",
8070 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8071 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8072 pr_err("Virtual processor ID = 0x%04x\n",
8073 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8077 * The guest has exited. See if we can fix it or if we need userspace
8080 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8082 struct vcpu_vmx *vmx = to_vmx(vcpu);
8083 u32 exit_reason = vmx->exit_reason;
8084 u32 vectoring_info = vmx->idt_vectoring_info;
8086 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8089 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8090 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8091 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8092 * mode as if vcpus is in root mode, the PML buffer must has been
8096 vmx_flush_pml_buffer(vcpu);
8098 /* If guest state is invalid, start emulating */
8099 if (vmx->emulation_required)
8100 return handle_invalid_guest_state(vcpu);
8102 if (is_guest_mode(vcpu) && nested_vmx_exit_handled(vcpu)) {
8103 nested_vmx_vmexit(vcpu, exit_reason,
8104 vmcs_read32(VM_EXIT_INTR_INFO),
8105 vmcs_readl(EXIT_QUALIFICATION));
8109 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8111 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8112 vcpu->run->fail_entry.hardware_entry_failure_reason
8117 if (unlikely(vmx->fail)) {
8118 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8119 vcpu->run->fail_entry.hardware_entry_failure_reason
8120 = vmcs_read32(VM_INSTRUCTION_ERROR);
8126 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8127 * delivery event since it indicates guest is accessing MMIO.
8128 * The vm-exit can be triggered again after return to guest that
8129 * will cause infinite loop.
8131 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8132 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8133 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8134 exit_reason != EXIT_REASON_PML_FULL &&
8135 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8136 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8137 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8138 vcpu->run->internal.ndata = 2;
8139 vcpu->run->internal.data[0] = vectoring_info;
8140 vcpu->run->internal.data[1] = exit_reason;
8144 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked &&
8145 !(is_guest_mode(vcpu) && nested_cpu_has_virtual_nmis(
8146 get_vmcs12(vcpu))))) {
8147 if (vmx_interrupt_allowed(vcpu)) {
8148 vmx->soft_vnmi_blocked = 0;
8149 } else if (vmx->vnmi_blocked_time > 1000000000LL &&
8150 vcpu->arch.nmi_pending) {
8152 * This CPU don't support us in finding the end of an
8153 * NMI-blocked window if the guest runs with IRQs
8154 * disabled. So we pull the trigger after 1 s of
8155 * futile waiting, but inform the user about this.
8157 printk(KERN_WARNING "%s: Breaking out of NMI-blocked "
8158 "state on VCPU %d after 1 s timeout\n",
8159 __func__, vcpu->vcpu_id);
8160 vmx->soft_vnmi_blocked = 0;
8164 if (exit_reason < kvm_vmx_max_exit_handlers
8165 && kvm_vmx_exit_handlers[exit_reason])
8166 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8168 WARN_ONCE(1, "vmx: unexpected exit reason 0x%x\n", exit_reason);
8169 kvm_queue_exception(vcpu, UD_VECTOR);
8174 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8176 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8178 if (is_guest_mode(vcpu) &&
8179 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8182 if (irr == -1 || tpr < irr) {
8183 vmcs_write32(TPR_THRESHOLD, 0);
8187 vmcs_write32(TPR_THRESHOLD, irr);
8190 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8192 u32 sec_exec_control;
8194 /* Postpone execution until vmcs01 is the current VMCS. */
8195 if (is_guest_mode(vcpu)) {
8196 to_vmx(vcpu)->nested.change_vmcs01_virtual_x2apic_mode = true;
8201 * There is not point to enable virtualize x2apic without enable
8204 if (!cpu_has_vmx_virtualize_x2apic_mode() ||
8205 !vmx_cpu_uses_apicv(vcpu))
8208 if (!cpu_need_tpr_shadow(vcpu))
8211 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8214 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8215 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8217 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8218 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8220 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8222 vmx_set_msr_bitmap(vcpu);
8225 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8227 struct vcpu_vmx *vmx = to_vmx(vcpu);
8230 * Currently we do not handle the nested case where L2 has an
8231 * APIC access page of its own; that page is still pinned.
8232 * Hence, we skip the case where the VCPU is in guest mode _and_
8233 * L1 prepared an APIC access page for L2.
8235 * For the case where L1 and L2 share the same APIC access page
8236 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8237 * in the vmcs12), this function will only update either the vmcs01
8238 * or the vmcs02. If the former, the vmcs02 will be updated by
8239 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8240 * the next L2->L1 exit.
8242 if (!is_guest_mode(vcpu) ||
8243 !nested_cpu_has2(vmx->nested.current_vmcs12,
8244 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
8245 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8248 static void vmx_hwapic_isr_update(struct kvm *kvm, int isr)
8256 status = vmcs_read16(GUEST_INTR_STATUS);
8261 vmcs_write16(GUEST_INTR_STATUS, status);
8265 static void vmx_set_rvi(int vector)
8273 status = vmcs_read16(GUEST_INTR_STATUS);
8274 old = (u8)status & 0xff;
8275 if ((u8)vector != old) {
8277 status |= (u8)vector;
8278 vmcs_write16(GUEST_INTR_STATUS, status);
8282 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8284 if (!is_guest_mode(vcpu)) {
8285 vmx_set_rvi(max_irr);
8293 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8296 if (nested_exit_on_intr(vcpu))
8300 * Else, fall back to pre-APICv interrupt injection since L2
8301 * is run without virtual interrupt delivery.
8303 if (!kvm_event_needs_reinjection(vcpu) &&
8304 vmx_interrupt_allowed(vcpu)) {
8305 kvm_queue_interrupt(vcpu, max_irr, false);
8306 vmx_inject_irq(vcpu);
8310 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu)
8312 u64 *eoi_exit_bitmap = vcpu->arch.eoi_exit_bitmap;
8313 if (!vmx_cpu_uses_apicv(vcpu))
8316 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
8317 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
8318 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
8319 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
8322 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
8326 if (!(vmx->exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
8327 || vmx->exit_reason == EXIT_REASON_EXCEPTION_NMI))
8330 vmx->exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8331 exit_intr_info = vmx->exit_intr_info;
8333 /* Handle machine checks before interrupts are enabled */
8334 if (is_machine_check(exit_intr_info))
8335 kvm_machine_check();
8337 /* We need to handle NMIs before interrupts are enabled */
8338 if (is_nmi(exit_intr_info)) {
8339 kvm_before_handle_nmi(&vmx->vcpu);
8341 kvm_after_handle_nmi(&vmx->vcpu);
8345 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
8347 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8350 * If external interrupt exists, IF bit is set in rflags/eflags on the
8351 * interrupt stack frame, and interrupt will be enabled on a return
8352 * from interrupt handler.
8354 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
8355 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
8356 unsigned int vector;
8357 unsigned long entry;
8359 struct vcpu_vmx *vmx = to_vmx(vcpu);
8360 #ifdef CONFIG_X86_64
8364 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8365 desc = (gate_desc *)vmx->host_idt_base + vector;
8366 entry = gate_offset(*desc);
8368 #ifdef CONFIG_X86_64
8369 "mov %%" _ASM_SP ", %[sp]\n\t"
8370 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
8375 "orl $0x200, (%%" _ASM_SP ")\n\t"
8376 __ASM_SIZE(push) " $%c[cs]\n\t"
8377 "call *%[entry]\n\t"
8379 #ifdef CONFIG_X86_64
8384 [ss]"i"(__KERNEL_DS),
8385 [cs]"i"(__KERNEL_CS)
8391 static bool vmx_has_high_real_mode_segbase(void)
8393 return enable_unrestricted_guest || emulate_invalid_guest_state;
8396 static bool vmx_mpx_supported(void)
8398 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
8399 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
8402 static bool vmx_xsaves_supported(void)
8404 return vmcs_config.cpu_based_2nd_exec_ctrl &
8405 SECONDARY_EXEC_XSAVES;
8408 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
8413 bool idtv_info_valid;
8415 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8417 if (cpu_has_virtual_nmis()) {
8418 if (vmx->nmi_known_unmasked)
8421 * Can't use vmx->exit_intr_info since we're not sure what
8422 * the exit reason is.
8424 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8425 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
8426 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
8428 * SDM 3: 27.7.1.2 (September 2008)
8429 * Re-set bit "block by NMI" before VM entry if vmexit caused by
8430 * a guest IRET fault.
8431 * SDM 3: 23.2.2 (September 2008)
8432 * Bit 12 is undefined in any of the following cases:
8433 * If the VM exit sets the valid bit in the IDT-vectoring
8434 * information field.
8435 * If the VM exit is due to a double fault.
8437 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
8438 vector != DF_VECTOR && !idtv_info_valid)
8439 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
8440 GUEST_INTR_STATE_NMI);
8442 vmx->nmi_known_unmasked =
8443 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
8444 & GUEST_INTR_STATE_NMI);
8445 } else if (unlikely(vmx->soft_vnmi_blocked))
8446 vmx->vnmi_blocked_time +=
8447 ktime_to_ns(ktime_sub(ktime_get(), vmx->entry_time));
8450 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
8451 u32 idt_vectoring_info,
8452 int instr_len_field,
8453 int error_code_field)
8457 bool idtv_info_valid;
8459 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
8461 vcpu->arch.nmi_injected = false;
8462 kvm_clear_exception_queue(vcpu);
8463 kvm_clear_interrupt_queue(vcpu);
8465 if (!idtv_info_valid)
8468 kvm_make_request(KVM_REQ_EVENT, vcpu);
8470 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
8471 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
8474 case INTR_TYPE_NMI_INTR:
8475 vcpu->arch.nmi_injected = true;
8477 * SDM 3: 27.7.1.2 (September 2008)
8478 * Clear bit "block by NMI" before VM entry if a NMI
8481 vmx_set_nmi_mask(vcpu, false);
8483 case INTR_TYPE_SOFT_EXCEPTION:
8484 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8486 case INTR_TYPE_HARD_EXCEPTION:
8487 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
8488 u32 err = vmcs_read32(error_code_field);
8489 kvm_requeue_exception_e(vcpu, vector, err);
8491 kvm_requeue_exception(vcpu, vector);
8493 case INTR_TYPE_SOFT_INTR:
8494 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
8496 case INTR_TYPE_EXT_INTR:
8497 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
8504 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
8506 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
8507 VM_EXIT_INSTRUCTION_LEN,
8508 IDT_VECTORING_ERROR_CODE);
8511 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
8513 __vmx_complete_interrupts(vcpu,
8514 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8515 VM_ENTRY_INSTRUCTION_LEN,
8516 VM_ENTRY_EXCEPTION_ERROR_CODE);
8518 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
8521 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
8524 struct perf_guest_switch_msr *msrs;
8526 msrs = perf_guest_get_msrs(&nr_msrs);
8531 for (i = 0; i < nr_msrs; i++)
8532 if (msrs[i].host == msrs[i].guest)
8533 clear_atomic_switch_msr(vmx, msrs[i].msr);
8535 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
8539 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
8541 struct vcpu_vmx *vmx = to_vmx(vcpu);
8542 unsigned long debugctlmsr, cr4;
8544 /* Record the guest's net vcpu time for enforced NMI injections. */
8545 if (unlikely(!cpu_has_virtual_nmis() && vmx->soft_vnmi_blocked))
8546 vmx->entry_time = ktime_get();
8548 /* Don't enter VMX if guest state is invalid, let the exit handler
8549 start emulation until we arrive back to a valid state */
8550 if (vmx->emulation_required)
8553 if (vmx->ple_window_dirty) {
8554 vmx->ple_window_dirty = false;
8555 vmcs_write32(PLE_WINDOW, vmx->ple_window);
8558 if (vmx->nested.sync_shadow_vmcs) {
8559 copy_vmcs12_to_shadow(vmx);
8560 vmx->nested.sync_shadow_vmcs = false;
8563 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
8564 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
8565 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
8566 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
8568 cr4 = cr4_read_shadow();
8569 if (unlikely(cr4 != vmx->host_state.vmcs_host_cr4)) {
8570 vmcs_writel(HOST_CR4, cr4);
8571 vmx->host_state.vmcs_host_cr4 = cr4;
8574 /* When single-stepping over STI and MOV SS, we must clear the
8575 * corresponding interruptibility bits in the guest state. Otherwise
8576 * vmentry fails as it then expects bit 14 (BS) in pending debug
8577 * exceptions being set, but that's not correct for the guest debugging
8579 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
8580 vmx_set_interrupt_shadow(vcpu, 0);
8582 atomic_switch_perf_msrs(vmx);
8583 debugctlmsr = get_debugctlmsr();
8585 vmx->__launched = vmx->loaded_vmcs->launched;
8587 /* Store host registers */
8588 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
8589 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
8590 "push %%" _ASM_CX " \n\t"
8591 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8593 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
8594 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
8596 /* Reload cr2 if changed */
8597 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
8598 "mov %%cr2, %%" _ASM_DX " \n\t"
8599 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
8601 "mov %%" _ASM_AX", %%cr2 \n\t"
8603 /* Check if vmlaunch of vmresume is needed */
8604 "cmpl $0, %c[launched](%0) \n\t"
8605 /* Load guest registers. Don't clobber flags. */
8606 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
8607 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
8608 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
8609 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
8610 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
8611 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
8612 #ifdef CONFIG_X86_64
8613 "mov %c[r8](%0), %%r8 \n\t"
8614 "mov %c[r9](%0), %%r9 \n\t"
8615 "mov %c[r10](%0), %%r10 \n\t"
8616 "mov %c[r11](%0), %%r11 \n\t"
8617 "mov %c[r12](%0), %%r12 \n\t"
8618 "mov %c[r13](%0), %%r13 \n\t"
8619 "mov %c[r14](%0), %%r14 \n\t"
8620 "mov %c[r15](%0), %%r15 \n\t"
8622 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
8624 /* Enter guest mode */
8626 __ex(ASM_VMX_VMLAUNCH) "\n\t"
8628 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
8630 /* Save guest registers, load host registers, keep flags */
8631 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
8633 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
8634 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
8635 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
8636 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
8637 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
8638 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
8639 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
8640 #ifdef CONFIG_X86_64
8641 "mov %%r8, %c[r8](%0) \n\t"
8642 "mov %%r9, %c[r9](%0) \n\t"
8643 "mov %%r10, %c[r10](%0) \n\t"
8644 "mov %%r11, %c[r11](%0) \n\t"
8645 "mov %%r12, %c[r12](%0) \n\t"
8646 "mov %%r13, %c[r13](%0) \n\t"
8647 "mov %%r14, %c[r14](%0) \n\t"
8648 "mov %%r15, %c[r15](%0) \n\t"
8650 "mov %%cr2, %%" _ASM_AX " \n\t"
8651 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
8653 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
8654 "setbe %c[fail](%0) \n\t"
8655 ".pushsection .rodata \n\t"
8656 ".global vmx_return \n\t"
8657 "vmx_return: " _ASM_PTR " 2b \n\t"
8659 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
8660 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
8661 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
8662 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
8663 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
8664 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
8665 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
8666 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
8667 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
8668 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
8669 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
8670 #ifdef CONFIG_X86_64
8671 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
8672 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
8673 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
8674 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
8675 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
8676 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
8677 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
8678 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
8680 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
8681 [wordsize]"i"(sizeof(ulong))
8683 #ifdef CONFIG_X86_64
8684 , "rax", "rbx", "rdi", "rsi"
8685 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
8687 , "eax", "ebx", "edi", "esi"
8691 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
8693 update_debugctlmsr(debugctlmsr);
8695 #ifndef CONFIG_X86_64
8697 * The sysexit path does not restore ds/es, so we must set them to
8698 * a reasonable value ourselves.
8700 * We can't defer this to vmx_load_host_state() since that function
8701 * may be executed in interrupt context, which saves and restore segments
8702 * around it, nullifying its effect.
8704 loadsegment(ds, __USER_DS);
8705 loadsegment(es, __USER_DS);
8708 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
8709 | (1 << VCPU_EXREG_RFLAGS)
8710 | (1 << VCPU_EXREG_PDPTR)
8711 | (1 << VCPU_EXREG_SEGMENTS)
8712 | (1 << VCPU_EXREG_CR3));
8713 vcpu->arch.regs_dirty = 0;
8715 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
8717 vmx->loaded_vmcs->launched = 1;
8719 vmx->exit_reason = vmcs_read32(VM_EXIT_REASON);
8722 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
8723 * we did not inject a still-pending event to L1 now because of
8724 * nested_run_pending, we need to re-enable this bit.
8726 if (vmx->nested.nested_run_pending)
8727 kvm_make_request(KVM_REQ_EVENT, vcpu);
8729 vmx->nested.nested_run_pending = 0;
8731 vmx_complete_atomic_exit(vmx);
8732 vmx_recover_nmi_blocking(vmx);
8733 vmx_complete_interrupts(vmx);
8736 static void vmx_load_vmcs01(struct kvm_vcpu *vcpu)
8738 struct vcpu_vmx *vmx = to_vmx(vcpu);
8741 if (vmx->loaded_vmcs == &vmx->vmcs01)
8745 vmx->loaded_vmcs = &vmx->vmcs01;
8747 vmx_vcpu_load(vcpu, cpu);
8753 * Ensure that the current vmcs of the logical processor is the
8754 * vmcs01 of the vcpu before calling free_nested().
8756 static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
8758 struct vcpu_vmx *vmx = to_vmx(vcpu);
8761 r = vcpu_load(vcpu);
8763 vmx_load_vmcs01(vcpu);
8768 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
8770 struct vcpu_vmx *vmx = to_vmx(vcpu);
8773 vmx_destroy_pml_buffer(vmx);
8774 free_vpid(vmx->vpid);
8775 leave_guest_mode(vcpu);
8776 vmx_free_vcpu_nested(vcpu);
8777 free_loaded_vmcs(vmx->loaded_vmcs);
8778 kfree(vmx->guest_msrs);
8779 kvm_vcpu_uninit(vcpu);
8780 kmem_cache_free(kvm_vcpu_cache, vmx);
8783 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
8786 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
8790 return ERR_PTR(-ENOMEM);
8792 vmx->vpid = allocate_vpid();
8794 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
8801 * If PML is turned on, failure on enabling PML just results in failure
8802 * of creating the vcpu, therefore we can simplify PML logic (by
8803 * avoiding dealing with cases, such as enabling PML partially on vcpus
8804 * for the guest, etc.
8807 vmx->pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
8812 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
8813 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
8816 if (!vmx->guest_msrs)
8819 vmx->loaded_vmcs = &vmx->vmcs01;
8820 vmx->loaded_vmcs->vmcs = alloc_vmcs();
8821 if (!vmx->loaded_vmcs->vmcs)
8824 kvm_cpu_vmxon(__pa(per_cpu(vmxarea, raw_smp_processor_id())));
8825 loaded_vmcs_init(vmx->loaded_vmcs);
8830 vmx_vcpu_load(&vmx->vcpu, cpu);
8831 vmx->vcpu.cpu = cpu;
8832 err = vmx_vcpu_setup(vmx);
8833 vmx_vcpu_put(&vmx->vcpu);
8837 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
8838 err = alloc_apic_access_page(kvm);
8844 if (!kvm->arch.ept_identity_map_addr)
8845 kvm->arch.ept_identity_map_addr =
8846 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
8847 err = init_rmode_identity_map(kvm);
8853 nested_vmx_setup_ctls_msrs(vmx);
8854 vmx->nested.vpid02 = allocate_vpid();
8857 vmx->nested.posted_intr_nv = -1;
8858 vmx->nested.current_vmptr = -1ull;
8859 vmx->nested.current_vmcs12 = NULL;
8864 free_vpid(vmx->nested.vpid02);
8865 free_loaded_vmcs(vmx->loaded_vmcs);
8867 kfree(vmx->guest_msrs);
8869 vmx_destroy_pml_buffer(vmx);
8871 kvm_vcpu_uninit(&vmx->vcpu);
8873 free_vpid(vmx->vpid);
8874 kmem_cache_free(kvm_vcpu_cache, vmx);
8875 return ERR_PTR(err);
8878 static void __init vmx_check_processor_compat(void *rtn)
8880 struct vmcs_config vmcs_conf;
8883 if (setup_vmcs_config(&vmcs_conf) < 0)
8885 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
8886 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
8887 smp_processor_id());
8892 static int get_ept_level(void)
8894 return VMX_EPT_DEFAULT_GAW + 1;
8897 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
8902 /* For VT-d and EPT combination
8903 * 1. MMIO: always map as UC
8905 * a. VT-d without snooping control feature: can't guarantee the
8906 * result, try to trust guest.
8907 * b. VT-d with snooping control feature: snooping control feature of
8908 * VT-d engine can guarantee the cache correctness. Just set it
8909 * to WB to keep consistent with host. So the same as item 3.
8910 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
8911 * consistent with host MTRR
8914 cache = MTRR_TYPE_UNCACHABLE;
8918 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
8919 ipat = VMX_EPT_IPAT_BIT;
8920 cache = MTRR_TYPE_WRBACK;
8924 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
8925 ipat = VMX_EPT_IPAT_BIT;
8926 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
8927 cache = MTRR_TYPE_WRBACK;
8929 cache = MTRR_TYPE_UNCACHABLE;
8933 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
8936 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
8939 static int vmx_get_lpage_level(void)
8941 if (enable_ept && !cpu_has_vmx_ept_1g_page())
8942 return PT_DIRECTORY_LEVEL;
8944 /* For shadow and EPT supported 1GB page */
8945 return PT_PDPE_LEVEL;
8948 static void vmcs_set_secondary_exec_control(u32 new_ctl)
8951 * These bits in the secondary execution controls field
8952 * are dynamic, the others are mostly based on the hypervisor
8953 * architecture and the guest's CPUID. Do not touch the
8957 SECONDARY_EXEC_SHADOW_VMCS |
8958 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
8959 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8961 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8963 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
8964 (new_ctl & ~mask) | (cur_ctl & mask));
8967 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
8969 struct kvm_cpuid_entry2 *best;
8970 struct vcpu_vmx *vmx = to_vmx(vcpu);
8971 u32 secondary_exec_ctl = vmx_secondary_exec_control(vmx);
8973 if (vmx_rdtscp_supported()) {
8974 bool rdtscp_enabled = guest_cpuid_has_rdtscp(vcpu);
8975 if (!rdtscp_enabled)
8976 secondary_exec_ctl &= ~SECONDARY_EXEC_RDTSCP;
8980 vmx->nested.nested_vmx_secondary_ctls_high |=
8981 SECONDARY_EXEC_RDTSCP;
8983 vmx->nested.nested_vmx_secondary_ctls_high &=
8984 ~SECONDARY_EXEC_RDTSCP;
8988 /* Exposing INVPCID only when PCID is exposed */
8989 best = kvm_find_cpuid_entry(vcpu, 0x7, 0);
8990 if (vmx_invpcid_supported() &&
8991 (!best || !(best->ebx & bit(X86_FEATURE_INVPCID)) ||
8992 !guest_cpuid_has_pcid(vcpu))) {
8993 secondary_exec_ctl &= ~SECONDARY_EXEC_ENABLE_INVPCID;
8996 best->ebx &= ~bit(X86_FEATURE_INVPCID);
8999 if (cpu_has_secondary_exec_ctrls())
9000 vmcs_set_secondary_exec_control(secondary_exec_ctl);
9002 if (static_cpu_has(X86_FEATURE_PCOMMIT) && nested) {
9003 if (guest_cpuid_has_pcommit(vcpu))
9004 vmx->nested.nested_vmx_secondary_ctls_high |=
9005 SECONDARY_EXEC_PCOMMIT;
9007 vmx->nested.nested_vmx_secondary_ctls_high &=
9008 ~SECONDARY_EXEC_PCOMMIT;
9012 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9014 if (func == 1 && nested)
9015 entry->ecx |= bit(X86_FEATURE_VMX);
9018 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9019 struct x86_exception *fault)
9021 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9024 if (fault->error_code & PFERR_RSVD_MASK)
9025 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9027 exit_reason = EXIT_REASON_EPT_VIOLATION;
9028 nested_vmx_vmexit(vcpu, exit_reason, 0, vcpu->arch.exit_qualification);
9029 vmcs12->guest_physical_address = fault->address;
9032 /* Callbacks for nested_ept_init_mmu_context: */
9034 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9036 /* return the page table to be shadowed - in our case, EPT12 */
9037 return get_vmcs12(vcpu)->ept_pointer;
9040 static void nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9042 WARN_ON(mmu_is_nested(vcpu));
9043 kvm_init_shadow_ept_mmu(vcpu,
9044 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
9045 VMX_EPT_EXECUTE_ONLY_BIT);
9046 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
9047 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
9048 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
9050 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
9053 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
9055 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
9058 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9061 bool inequality, bit;
9063 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9065 (error_code & vmcs12->page_fault_error_code_mask) !=
9066 vmcs12->page_fault_error_code_match;
9067 return inequality ^ bit;
9070 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9071 struct x86_exception *fault)
9073 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9075 WARN_ON(!is_guest_mode(vcpu));
9077 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code))
9078 nested_vmx_vmexit(vcpu, to_vmx(vcpu)->exit_reason,
9079 vmcs_read32(VM_EXIT_INTR_INFO),
9080 vmcs_readl(EXIT_QUALIFICATION));
9082 kvm_inject_page_fault(vcpu, fault);
9085 static bool nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9086 struct vmcs12 *vmcs12)
9088 struct vcpu_vmx *vmx = to_vmx(vcpu);
9089 int maxphyaddr = cpuid_maxphyaddr(vcpu);
9091 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9092 if (!PAGE_ALIGNED(vmcs12->apic_access_addr) ||
9093 vmcs12->apic_access_addr >> maxphyaddr)
9097 * Translate L1 physical address to host physical
9098 * address for vmcs02. Keep the page pinned, so this
9099 * physical address remains valid. We keep a reference
9100 * to it so we can release it later.
9102 if (vmx->nested.apic_access_page) /* shouldn't happen */
9103 nested_release_page(vmx->nested.apic_access_page);
9104 vmx->nested.apic_access_page =
9105 nested_get_page(vcpu, vmcs12->apic_access_addr);
9108 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9109 if (!PAGE_ALIGNED(vmcs12->virtual_apic_page_addr) ||
9110 vmcs12->virtual_apic_page_addr >> maxphyaddr)
9113 if (vmx->nested.virtual_apic_page) /* shouldn't happen */
9114 nested_release_page(vmx->nested.virtual_apic_page);
9115 vmx->nested.virtual_apic_page =
9116 nested_get_page(vcpu, vmcs12->virtual_apic_page_addr);
9119 * Failing the vm entry is _not_ what the processor does
9120 * but it's basically the only possibility we have.
9121 * We could still enter the guest if CR8 load exits are
9122 * enabled, CR8 store exits are enabled, and virtualize APIC
9123 * access is disabled; in this case the processor would never
9124 * use the TPR shadow and we could simply clear the bit from
9125 * the execution control. But such a configuration is useless,
9126 * so let's keep the code simple.
9128 if (!vmx->nested.virtual_apic_page)
9132 if (nested_cpu_has_posted_intr(vmcs12)) {
9133 if (!IS_ALIGNED(vmcs12->posted_intr_desc_addr, 64) ||
9134 vmcs12->posted_intr_desc_addr >> maxphyaddr)
9137 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9138 kunmap(vmx->nested.pi_desc_page);
9139 nested_release_page(vmx->nested.pi_desc_page);
9141 vmx->nested.pi_desc_page =
9142 nested_get_page(vcpu, vmcs12->posted_intr_desc_addr);
9143 if (!vmx->nested.pi_desc_page)
9146 vmx->nested.pi_desc =
9147 (struct pi_desc *)kmap(vmx->nested.pi_desc_page);
9148 if (!vmx->nested.pi_desc) {
9149 nested_release_page_clean(vmx->nested.pi_desc_page);
9152 vmx->nested.pi_desc =
9153 (struct pi_desc *)((void *)vmx->nested.pi_desc +
9154 (unsigned long)(vmcs12->posted_intr_desc_addr &
9161 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9163 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9164 struct vcpu_vmx *vmx = to_vmx(vcpu);
9166 if (vcpu->arch.virtual_tsc_khz == 0)
9169 /* Make sure short timeouts reliably trigger an immediate vmexit.
9170 * hrtimer_start does not guarantee this. */
9171 if (preemption_timeout <= 1) {
9172 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9176 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9177 preemption_timeout *= 1000000;
9178 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9179 hrtimer_start(&vmx->nested.preemption_timer,
9180 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9183 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9184 struct vmcs12 *vmcs12)
9189 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9192 if (vmcs12_read_any(vcpu, MSR_BITMAP, &addr)) {
9196 maxphyaddr = cpuid_maxphyaddr(vcpu);
9198 if (!PAGE_ALIGNED(vmcs12->msr_bitmap) ||
9199 ((addr + PAGE_SIZE) >> maxphyaddr))
9206 * Merge L0's and L1's MSR bitmap, return false to indicate that
9207 * we do not use the hardware.
9209 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9210 struct vmcs12 *vmcs12)
9214 unsigned long *msr_bitmap;
9216 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
9219 page = nested_get_page(vcpu, vmcs12->msr_bitmap);
9224 msr_bitmap = (unsigned long *)kmap(page);
9226 nested_release_page_clean(page);
9231 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
9232 if (nested_cpu_has_apic_reg_virt(vmcs12))
9233 for (msr = 0x800; msr <= 0x8ff; msr++)
9234 nested_vmx_disable_intercept_for_msr(
9236 vmx_msr_bitmap_nested,
9238 /* TPR is allowed */
9239 nested_vmx_disable_intercept_for_msr(msr_bitmap,
9240 vmx_msr_bitmap_nested,
9241 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9242 MSR_TYPE_R | MSR_TYPE_W);
9243 if (nested_cpu_has_vid(vmcs12)) {
9244 /* EOI and self-IPI are allowed */
9245 nested_vmx_disable_intercept_for_msr(
9247 vmx_msr_bitmap_nested,
9248 APIC_BASE_MSR + (APIC_EOI >> 4),
9250 nested_vmx_disable_intercept_for_msr(
9252 vmx_msr_bitmap_nested,
9253 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9258 * Enable reading intercept of all the x2apic
9259 * MSRs. We should not rely on vmcs12 to do any
9260 * optimizations here, it may have been modified
9263 for (msr = 0x800; msr <= 0x8ff; msr++)
9264 __vmx_enable_intercept_for_msr(
9265 vmx_msr_bitmap_nested,
9269 __vmx_enable_intercept_for_msr(
9270 vmx_msr_bitmap_nested,
9271 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
9273 __vmx_enable_intercept_for_msr(
9274 vmx_msr_bitmap_nested,
9275 APIC_BASE_MSR + (APIC_EOI >> 4),
9277 __vmx_enable_intercept_for_msr(
9278 vmx_msr_bitmap_nested,
9279 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
9283 nested_release_page_clean(page);
9288 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
9289 struct vmcs12 *vmcs12)
9291 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9292 !nested_cpu_has_apic_reg_virt(vmcs12) &&
9293 !nested_cpu_has_vid(vmcs12) &&
9294 !nested_cpu_has_posted_intr(vmcs12))
9298 * If virtualize x2apic mode is enabled,
9299 * virtualize apic access must be disabled.
9301 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
9302 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
9306 * If virtual interrupt delivery is enabled,
9307 * we must exit on external interrupts.
9309 if (nested_cpu_has_vid(vmcs12) &&
9310 !nested_exit_on_intr(vcpu))
9314 * bits 15:8 should be zero in posted_intr_nv,
9315 * the descriptor address has been already checked
9316 * in nested_get_vmcs12_pages.
9318 if (nested_cpu_has_posted_intr(vmcs12) &&
9319 (!nested_cpu_has_vid(vmcs12) ||
9320 !nested_exit_intr_ack_set(vcpu) ||
9321 vmcs12->posted_intr_nv & 0xff00))
9324 /* tpr shadow is needed by all apicv features. */
9325 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
9331 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
9332 unsigned long count_field,
9333 unsigned long addr_field)
9338 if (vmcs12_read_any(vcpu, count_field, &count) ||
9339 vmcs12_read_any(vcpu, addr_field, &addr)) {
9345 maxphyaddr = cpuid_maxphyaddr(vcpu);
9346 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
9347 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
9348 pr_warn_ratelimited(
9349 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
9350 addr_field, maxphyaddr, count, addr);
9356 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
9357 struct vmcs12 *vmcs12)
9359 if (vmcs12->vm_exit_msr_load_count == 0 &&
9360 vmcs12->vm_exit_msr_store_count == 0 &&
9361 vmcs12->vm_entry_msr_load_count == 0)
9362 return 0; /* Fast path */
9363 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
9364 VM_EXIT_MSR_LOAD_ADDR) ||
9365 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
9366 VM_EXIT_MSR_STORE_ADDR) ||
9367 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
9368 VM_ENTRY_MSR_LOAD_ADDR))
9373 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
9374 struct vmx_msr_entry *e)
9376 /* x2APIC MSR accesses are not allowed */
9377 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
9379 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
9380 e->index == MSR_IA32_UCODE_REV)
9382 if (e->reserved != 0)
9387 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
9388 struct vmx_msr_entry *e)
9390 if (e->index == MSR_FS_BASE ||
9391 e->index == MSR_GS_BASE ||
9392 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
9393 nested_vmx_msr_check_common(vcpu, e))
9398 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
9399 struct vmx_msr_entry *e)
9401 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
9402 nested_vmx_msr_check_common(vcpu, e))
9408 * Load guest's/host's msr at nested entry/exit.
9409 * return 0 for success, entry index for failure.
9411 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9414 struct vmx_msr_entry e;
9415 struct msr_data msr;
9417 msr.host_initiated = false;
9418 for (i = 0; i < count; i++) {
9419 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
9421 pr_warn_ratelimited(
9422 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9423 __func__, i, gpa + i * sizeof(e));
9426 if (nested_vmx_load_msr_check(vcpu, &e)) {
9427 pr_warn_ratelimited(
9428 "%s check failed (%u, 0x%x, 0x%x)\n",
9429 __func__, i, e.index, e.reserved);
9432 msr.index = e.index;
9434 if (kvm_set_msr(vcpu, &msr)) {
9435 pr_warn_ratelimited(
9436 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9437 __func__, i, e.index, e.value);
9446 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
9449 struct vmx_msr_entry e;
9451 for (i = 0; i < count; i++) {
9452 struct msr_data msr_info;
9453 if (kvm_vcpu_read_guest(vcpu,
9454 gpa + i * sizeof(e),
9455 &e, 2 * sizeof(u32))) {
9456 pr_warn_ratelimited(
9457 "%s cannot read MSR entry (%u, 0x%08llx)\n",
9458 __func__, i, gpa + i * sizeof(e));
9461 if (nested_vmx_store_msr_check(vcpu, &e)) {
9462 pr_warn_ratelimited(
9463 "%s check failed (%u, 0x%x, 0x%x)\n",
9464 __func__, i, e.index, e.reserved);
9467 msr_info.host_initiated = false;
9468 msr_info.index = e.index;
9469 if (kvm_get_msr(vcpu, &msr_info)) {
9470 pr_warn_ratelimited(
9471 "%s cannot read MSR (%u, 0x%x)\n",
9472 __func__, i, e.index);
9475 if (kvm_vcpu_write_guest(vcpu,
9476 gpa + i * sizeof(e) +
9477 offsetof(struct vmx_msr_entry, value),
9478 &msr_info.data, sizeof(msr_info.data))) {
9479 pr_warn_ratelimited(
9480 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
9481 __func__, i, e.index, msr_info.data);
9489 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
9490 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
9491 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
9492 * guest in a way that will both be appropriate to L1's requests, and our
9493 * needs. In addition to modifying the active vmcs (which is vmcs02), this
9494 * function also has additional necessary side-effects, like setting various
9495 * vcpu->arch fields.
9497 static void prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
9499 struct vcpu_vmx *vmx = to_vmx(vcpu);
9502 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
9503 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
9504 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
9505 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
9506 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
9507 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
9508 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
9509 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
9510 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
9511 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
9512 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
9513 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
9514 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
9515 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
9516 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
9517 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
9518 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
9519 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
9520 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
9521 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
9522 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
9523 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
9524 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
9525 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
9526 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
9527 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
9528 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
9529 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
9530 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
9531 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
9532 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
9533 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
9534 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
9535 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
9536 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
9537 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
9539 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS) {
9540 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
9541 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
9543 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
9544 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
9546 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
9547 vmcs12->vm_entry_intr_info_field);
9548 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
9549 vmcs12->vm_entry_exception_error_code);
9550 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
9551 vmcs12->vm_entry_instruction_len);
9552 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
9553 vmcs12->guest_interruptibility_info);
9554 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
9555 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
9556 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
9557 vmcs12->guest_pending_dbg_exceptions);
9558 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
9559 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
9561 if (nested_cpu_has_xsaves(vmcs12))
9562 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
9563 vmcs_write64(VMCS_LINK_POINTER, -1ull);
9565 exec_control = vmcs12->pin_based_vm_exec_control;
9566 exec_control |= vmcs_config.pin_based_exec_ctrl;
9567 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
9569 if (nested_cpu_has_posted_intr(vmcs12)) {
9571 * Note that we use L0's vector here and in
9572 * vmx_deliver_nested_posted_interrupt.
9574 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
9575 vmx->nested.pi_pending = false;
9576 vmcs_write64(POSTED_INTR_NV, POSTED_INTR_VECTOR);
9577 vmcs_write64(POSTED_INTR_DESC_ADDR,
9578 page_to_phys(vmx->nested.pi_desc_page) +
9579 (unsigned long)(vmcs12->posted_intr_desc_addr &
9582 exec_control &= ~PIN_BASED_POSTED_INTR;
9584 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
9586 vmx->nested.preemption_timer_expired = false;
9587 if (nested_cpu_has_preemption_timer(vmcs12))
9588 vmx_start_preemption_timer(vcpu);
9591 * Whether page-faults are trapped is determined by a combination of
9592 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
9593 * If enable_ept, L0 doesn't care about page faults and we should
9594 * set all of these to L1's desires. However, if !enable_ept, L0 does
9595 * care about (at least some) page faults, and because it is not easy
9596 * (if at all possible?) to merge L0 and L1's desires, we simply ask
9597 * to exit on each and every L2 page fault. This is done by setting
9598 * MASK=MATCH=0 and (see below) EB.PF=1.
9599 * Note that below we don't need special code to set EB.PF beyond the
9600 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
9601 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
9602 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
9604 * A problem with this approach (when !enable_ept) is that L1 may be
9605 * injected with more page faults than it asked for. This could have
9606 * caused problems, but in practice existing hypervisors don't care.
9607 * To fix this, we will need to emulate the PFEC checking (on the L1
9608 * page tables), using walk_addr(), when injecting PFs to L1.
9610 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
9611 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
9612 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
9613 enable_ept ? vmcs12->page_fault_error_code_match : 0);
9615 if (cpu_has_secondary_exec_ctrls()) {
9616 exec_control = vmx_secondary_exec_control(vmx);
9618 /* Take the following fields only from vmcs12 */
9619 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
9620 SECONDARY_EXEC_RDTSCP |
9621 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
9622 SECONDARY_EXEC_APIC_REGISTER_VIRT |
9623 SECONDARY_EXEC_PCOMMIT);
9624 if (nested_cpu_has(vmcs12,
9625 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS))
9626 exec_control |= vmcs12->secondary_vm_exec_control;
9628 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES) {
9630 * If translation failed, no matter: This feature asks
9631 * to exit when accessing the given address, and if it
9632 * can never be accessed, this feature won't do
9635 if (!vmx->nested.apic_access_page)
9637 ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9639 vmcs_write64(APIC_ACCESS_ADDR,
9640 page_to_phys(vmx->nested.apic_access_page));
9641 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9642 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9644 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9645 kvm_vcpu_reload_apic_access_page(vcpu);
9648 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
9649 vmcs_write64(EOI_EXIT_BITMAP0,
9650 vmcs12->eoi_exit_bitmap0);
9651 vmcs_write64(EOI_EXIT_BITMAP1,
9652 vmcs12->eoi_exit_bitmap1);
9653 vmcs_write64(EOI_EXIT_BITMAP2,
9654 vmcs12->eoi_exit_bitmap2);
9655 vmcs_write64(EOI_EXIT_BITMAP3,
9656 vmcs12->eoi_exit_bitmap3);
9657 vmcs_write16(GUEST_INTR_STATUS,
9658 vmcs12->guest_intr_status);
9661 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
9666 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
9667 * Some constant fields are set here by vmx_set_constant_host_state().
9668 * Other fields are different per CPU, and will be set later when
9669 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
9671 vmx_set_constant_host_state(vmx);
9674 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
9675 * entry, but only if the current (host) sp changed from the value
9676 * we wrote last (vmx->host_rsp). This cache is no longer relevant
9677 * if we switch vmcs, and rather than hold a separate cache per vmcs,
9678 * here we just force the write to happen on entry.
9682 exec_control = vmx_exec_control(vmx); /* L0's desires */
9683 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
9684 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
9685 exec_control &= ~CPU_BASED_TPR_SHADOW;
9686 exec_control |= vmcs12->cpu_based_vm_exec_control;
9688 if (exec_control & CPU_BASED_TPR_SHADOW) {
9689 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
9690 page_to_phys(vmx->nested.virtual_apic_page));
9691 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
9693 #ifdef CONFIG_X86_64
9694 exec_control |= CPU_BASED_CR8_LOAD_EXITING |
9695 CPU_BASED_CR8_STORE_EXITING;
9699 if (cpu_has_vmx_msr_bitmap() &&
9700 exec_control & CPU_BASED_USE_MSR_BITMAPS) {
9701 nested_vmx_merge_msr_bitmap(vcpu, vmcs12);
9702 /* MSR_BITMAP will be set by following vmx_set_efer. */
9704 exec_control &= ~CPU_BASED_USE_MSR_BITMAPS;
9707 * Merging of IO bitmap not currently supported.
9708 * Rather, exit every time.
9710 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
9711 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
9713 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
9715 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
9716 * bitwise-or of what L1 wants to trap for L2, and what we want to
9717 * trap. Note that CR0.TS also needs updating - we do this later.
9719 update_exception_bitmap(vcpu);
9720 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
9721 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
9723 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
9724 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
9725 * bits are further modified by vmx_set_efer() below.
9727 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
9729 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
9730 * emulated by vmx_set_efer(), below.
9732 vm_entry_controls_init(vmx,
9733 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
9734 ~VM_ENTRY_IA32E_MODE) |
9735 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
9737 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT) {
9738 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
9739 vcpu->arch.pat = vmcs12->guest_ia32_pat;
9740 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
9741 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
9744 set_cr4_guest_host_mask(vmx);
9746 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
9747 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
9749 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
9750 vmcs_write64(TSC_OFFSET,
9751 vmx->nested.vmcs01_tsc_offset + vmcs12->tsc_offset);
9753 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
9757 * There is no direct mapping between vpid02 and vpid12, the
9758 * vpid02 is per-vCPU for L0 and reused while the value of
9759 * vpid12 is changed w/ one invvpid during nested vmentry.
9760 * The vpid12 is allocated by L1 for L2, so it will not
9761 * influence global bitmap(for vpid01 and vpid02 allocation)
9762 * even if spawn a lot of nested vCPUs.
9764 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
9765 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
9766 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
9767 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
9768 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
9771 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
9772 vmx_flush_tlb(vcpu);
9779 * Conceptually we want to copy the PML address and index from
9780 * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
9781 * since we always flush the log on each vmexit, this happens
9782 * to be equivalent to simply resetting the fields in vmcs02.
9784 ASSERT(vmx->pml_pg);
9785 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
9786 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
9789 if (nested_cpu_has_ept(vmcs12)) {
9790 kvm_mmu_unload(vcpu);
9791 nested_ept_init_mmu_context(vcpu);
9794 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)
9795 vcpu->arch.efer = vmcs12->guest_ia32_efer;
9796 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
9797 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
9799 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
9800 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
9801 vmx_set_efer(vcpu, vcpu->arch.efer);
9804 * This sets GUEST_CR0 to vmcs12->guest_cr0, with possibly a modified
9805 * TS bit (for lazy fpu) and bits which we consider mandatory enabled.
9806 * The CR0_READ_SHADOW is what L2 should have expected to read given
9807 * the specifications by L1; It's not enough to take
9808 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
9809 * have more bits than L1 expected.
9811 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
9812 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
9814 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
9815 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
9817 /* shadow page tables on either EPT or shadow page tables */
9818 kvm_set_cr3(vcpu, vmcs12->guest_cr3);
9819 kvm_mmu_reset_context(vcpu);
9822 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
9825 * L1 may access the L2's PDPTR, so save them to construct vmcs12
9828 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
9829 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
9830 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
9831 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
9834 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
9835 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
9839 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
9840 * for running an L2 nested guest.
9842 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
9844 struct vmcs12 *vmcs12;
9845 struct vcpu_vmx *vmx = to_vmx(vcpu);
9847 struct loaded_vmcs *vmcs02;
9851 if (!nested_vmx_check_permission(vcpu) ||
9852 !nested_vmx_check_vmcs12(vcpu))
9855 skip_emulated_instruction(vcpu);
9856 vmcs12 = get_vmcs12(vcpu);
9858 if (enable_shadow_vmcs)
9859 copy_shadow_to_vmcs12(vmx);
9862 * The nested entry process starts with enforcing various prerequisites
9863 * on vmcs12 as required by the Intel SDM, and act appropriately when
9864 * they fail: As the SDM explains, some conditions should cause the
9865 * instruction to fail, while others will cause the instruction to seem
9866 * to succeed, but return an EXIT_REASON_INVALID_STATE.
9867 * To speed up the normal (success) code path, we should avoid checking
9868 * for misconfigurations which will anyway be caught by the processor
9869 * when using the merged vmcs02.
9871 if (vmcs12->launch_state == launch) {
9872 nested_vmx_failValid(vcpu,
9873 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
9874 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
9878 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
9879 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT) {
9880 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9884 if (!nested_get_vmcs12_pages(vcpu, vmcs12)) {
9885 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9889 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12)) {
9890 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9894 if (nested_vmx_check_apicv_controls(vcpu, vmcs12)) {
9895 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9899 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12)) {
9900 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9904 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
9905 vmx->nested.nested_vmx_true_procbased_ctls_low,
9906 vmx->nested.nested_vmx_procbased_ctls_high) ||
9907 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
9908 vmx->nested.nested_vmx_secondary_ctls_low,
9909 vmx->nested.nested_vmx_secondary_ctls_high) ||
9910 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
9911 vmx->nested.nested_vmx_pinbased_ctls_low,
9912 vmx->nested.nested_vmx_pinbased_ctls_high) ||
9913 !vmx_control_verify(vmcs12->vm_exit_controls,
9914 vmx->nested.nested_vmx_true_exit_ctls_low,
9915 vmx->nested.nested_vmx_exit_ctls_high) ||
9916 !vmx_control_verify(vmcs12->vm_entry_controls,
9917 vmx->nested.nested_vmx_true_entry_ctls_low,
9918 vmx->nested.nested_vmx_entry_ctls_high))
9920 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
9924 if (((vmcs12->host_cr0 & VMXON_CR0_ALWAYSON) != VMXON_CR0_ALWAYSON) ||
9925 ((vmcs12->host_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
9926 nested_vmx_failValid(vcpu,
9927 VMXERR_ENTRY_INVALID_HOST_STATE_FIELD);
9931 if (!nested_cr0_valid(vcpu, vmcs12->guest_cr0) ||
9932 ((vmcs12->guest_cr4 & VMXON_CR4_ALWAYSON) != VMXON_CR4_ALWAYSON)) {
9933 nested_vmx_entry_failure(vcpu, vmcs12,
9934 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9937 if (vmcs12->vmcs_link_pointer != -1ull) {
9938 nested_vmx_entry_failure(vcpu, vmcs12,
9939 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_VMCS_LINK_PTR);
9944 * If the load IA32_EFER VM-entry control is 1, the following checks
9945 * are performed on the field for the IA32_EFER MSR:
9946 * - Bits reserved in the IA32_EFER MSR must be 0.
9947 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
9948 * the IA-32e mode guest VM-exit control. It must also be identical
9949 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
9952 if (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER) {
9953 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
9954 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
9955 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
9956 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
9957 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME))) {
9958 nested_vmx_entry_failure(vcpu, vmcs12,
9959 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9965 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
9966 * IA32_EFER MSR must be 0 in the field for that register. In addition,
9967 * the values of the LMA and LME bits in the field must each be that of
9968 * the host address-space size VM-exit control.
9970 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
9971 ia32e = (vmcs12->vm_exit_controls &
9972 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
9973 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
9974 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
9975 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME)) {
9976 nested_vmx_entry_failure(vcpu, vmcs12,
9977 EXIT_REASON_INVALID_STATE, ENTRY_FAIL_DEFAULT);
9983 * We're finally done with prerequisite checking, and can start with
9987 vmcs02 = nested_get_current_vmcs02(vmx);
9991 enter_guest_mode(vcpu);
9993 vmx->nested.vmcs01_tsc_offset = vmcs_read64(TSC_OFFSET);
9995 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
9996 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
9999 vmx->loaded_vmcs = vmcs02;
10000 vmx_vcpu_put(vcpu);
10001 vmx_vcpu_load(vcpu, cpu);
10005 vmx_segment_cache_clear(vmx);
10007 prepare_vmcs02(vcpu, vmcs12);
10009 msr_entry_idx = nested_vmx_load_msr(vcpu,
10010 vmcs12->vm_entry_msr_load_addr,
10011 vmcs12->vm_entry_msr_load_count);
10012 if (msr_entry_idx) {
10013 leave_guest_mode(vcpu);
10014 vmx_load_vmcs01(vcpu);
10015 nested_vmx_entry_failure(vcpu, vmcs12,
10016 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
10020 vmcs12->launch_state = 1;
10022 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
10023 return kvm_vcpu_halt(vcpu);
10025 vmx->nested.nested_run_pending = 1;
10028 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10029 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10030 * returned as far as L1 is concerned. It will only return (and set
10031 * the success flag) when L2 exits (see nested_vmx_vmexit()).
10037 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
10038 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
10039 * This function returns the new value we should put in vmcs12.guest_cr0.
10040 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
10041 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
10042 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
10043 * didn't trap the bit, because if L1 did, so would L0).
10044 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
10045 * been modified by L2, and L1 knows it. So just leave the old value of
10046 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
10047 * isn't relevant, because if L0 traps this bit it can set it to anything.
10048 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
10049 * changed these bits, and therefore they need to be updated, but L0
10050 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
10051 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
10053 static inline unsigned long
10054 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10057 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
10058 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
10059 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
10060 vcpu->arch.cr0_guest_owned_bits));
10063 static inline unsigned long
10064 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10067 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
10068 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
10069 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
10070 vcpu->arch.cr4_guest_owned_bits));
10073 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
10074 struct vmcs12 *vmcs12)
10079 if (vcpu->arch.exception.pending && vcpu->arch.exception.reinject) {
10080 nr = vcpu->arch.exception.nr;
10081 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10083 if (kvm_exception_is_soft(nr)) {
10084 vmcs12->vm_exit_instruction_len =
10085 vcpu->arch.event_exit_inst_len;
10086 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
10088 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
10090 if (vcpu->arch.exception.has_error_code) {
10091 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
10092 vmcs12->idt_vectoring_error_code =
10093 vcpu->arch.exception.error_code;
10096 vmcs12->idt_vectoring_info_field = idt_vectoring;
10097 } else if (vcpu->arch.nmi_injected) {
10098 vmcs12->idt_vectoring_info_field =
10099 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
10100 } else if (vcpu->arch.interrupt.pending) {
10101 nr = vcpu->arch.interrupt.nr;
10102 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
10104 if (vcpu->arch.interrupt.soft) {
10105 idt_vectoring |= INTR_TYPE_SOFT_INTR;
10106 vmcs12->vm_entry_instruction_len =
10107 vcpu->arch.event_exit_inst_len;
10109 idt_vectoring |= INTR_TYPE_EXT_INTR;
10111 vmcs12->idt_vectoring_info_field = idt_vectoring;
10115 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
10117 struct vcpu_vmx *vmx = to_vmx(vcpu);
10119 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
10120 vmx->nested.preemption_timer_expired) {
10121 if (vmx->nested.nested_run_pending)
10123 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
10127 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
10128 if (vmx->nested.nested_run_pending ||
10129 vcpu->arch.interrupt.pending)
10131 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
10132 NMI_VECTOR | INTR_TYPE_NMI_INTR |
10133 INTR_INFO_VALID_MASK, 0);
10135 * The NMI-triggered VM exit counts as injection:
10136 * clear this one and block further NMIs.
10138 vcpu->arch.nmi_pending = 0;
10139 vmx_set_nmi_mask(vcpu, true);
10143 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
10144 nested_exit_on_intr(vcpu)) {
10145 if (vmx->nested.nested_run_pending)
10147 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
10151 return vmx_complete_nested_posted_interrupt(vcpu);
10154 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
10156 ktime_t remaining =
10157 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
10160 if (ktime_to_ns(remaining) <= 0)
10163 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
10164 do_div(value, 1000000);
10165 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
10169 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
10170 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
10171 * and this function updates it to reflect the changes to the guest state while
10172 * L2 was running (and perhaps made some exits which were handled directly by L0
10173 * without going back to L1), and to reflect the exit reason.
10174 * Note that we do not have to copy here all VMCS fields, just those that
10175 * could have changed by the L2 guest or the exit - i.e., the guest-state and
10176 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
10177 * which already writes to vmcs12 directly.
10179 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10180 u32 exit_reason, u32 exit_intr_info,
10181 unsigned long exit_qualification)
10183 /* update guest state fields: */
10184 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
10185 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
10187 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
10188 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
10189 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
10191 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
10192 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
10193 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
10194 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
10195 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
10196 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
10197 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
10198 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
10199 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
10200 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
10201 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
10202 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
10203 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
10204 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
10205 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
10206 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
10207 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
10208 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
10209 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
10210 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
10211 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
10212 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
10213 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
10214 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
10215 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
10216 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
10217 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
10218 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
10219 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
10220 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
10221 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
10222 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
10223 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
10224 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
10225 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
10226 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
10228 vmcs12->guest_interruptibility_info =
10229 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
10230 vmcs12->guest_pending_dbg_exceptions =
10231 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
10232 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
10233 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
10235 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
10237 if (nested_cpu_has_preemption_timer(vmcs12)) {
10238 if (vmcs12->vm_exit_controls &
10239 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
10240 vmcs12->vmx_preemption_timer_value =
10241 vmx_get_preemption_timer_value(vcpu);
10242 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
10246 * In some cases (usually, nested EPT), L2 is allowed to change its
10247 * own CR3 without exiting. If it has changed it, we must keep it.
10248 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
10249 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
10251 * Additionally, restore L2's PDPTR to vmcs12.
10254 vmcs12->guest_cr3 = vmcs_read64(GUEST_CR3);
10255 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
10256 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
10257 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
10258 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
10261 if (nested_cpu_has_vid(vmcs12))
10262 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
10264 vmcs12->vm_entry_controls =
10265 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
10266 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
10268 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
10269 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
10270 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10273 /* TODO: These cannot have changed unless we have MSR bitmaps and
10274 * the relevant bit asks not to trap the change */
10275 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
10276 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
10277 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
10278 vmcs12->guest_ia32_efer = vcpu->arch.efer;
10279 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
10280 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
10281 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
10282 if (kvm_mpx_supported())
10283 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
10284 if (nested_cpu_has_xsaves(vmcs12))
10285 vmcs12->xss_exit_bitmap = vmcs_read64(XSS_EXIT_BITMAP);
10287 /* update exit information fields: */
10289 vmcs12->vm_exit_reason = exit_reason;
10290 vmcs12->exit_qualification = exit_qualification;
10292 vmcs12->vm_exit_intr_info = exit_intr_info;
10293 if ((vmcs12->vm_exit_intr_info &
10294 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
10295 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK))
10296 vmcs12->vm_exit_intr_error_code =
10297 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
10298 vmcs12->idt_vectoring_info_field = 0;
10299 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
10300 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
10302 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
10303 /* vm_entry_intr_info_field is cleared on exit. Emulate this
10304 * instead of reading the real value. */
10305 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
10308 * Transfer the event that L0 or L1 may wanted to inject into
10309 * L2 to IDT_VECTORING_INFO_FIELD.
10311 vmcs12_save_pending_event(vcpu, vmcs12);
10315 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
10316 * preserved above and would only end up incorrectly in L1.
10318 vcpu->arch.nmi_injected = false;
10319 kvm_clear_exception_queue(vcpu);
10320 kvm_clear_interrupt_queue(vcpu);
10324 * A part of what we need to when the nested L2 guest exits and we want to
10325 * run its L1 parent, is to reset L1's guest state to the host state specified
10327 * This function is to be called not only on normal nested exit, but also on
10328 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
10329 * Failures During or After Loading Guest State").
10330 * This function should be called when the active VMCS is L1's (vmcs01).
10332 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
10333 struct vmcs12 *vmcs12)
10335 struct kvm_segment seg;
10337 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
10338 vcpu->arch.efer = vmcs12->host_ia32_efer;
10339 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10340 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10342 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10343 vmx_set_efer(vcpu, vcpu->arch.efer);
10345 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
10346 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
10347 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
10349 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
10350 * actually changed, because it depends on the current state of
10351 * fpu_active (which may have changed).
10352 * Note that vmx_set_cr0 refers to efer set above.
10354 vmx_set_cr0(vcpu, vmcs12->host_cr0);
10356 * If we did fpu_activate()/fpu_deactivate() during L2's run, we need
10357 * to apply the same changes to L1's vmcs. We just set cr0 correctly,
10358 * but we also need to update cr0_guest_host_mask and exception_bitmap.
10360 update_exception_bitmap(vcpu);
10361 vcpu->arch.cr0_guest_owned_bits = (vcpu->fpu_active ? X86_CR0_TS : 0);
10362 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10365 * Note that CR4_GUEST_HOST_MASK is already set in the original vmcs01
10366 * (KVM doesn't change it)- no reason to call set_cr4_guest_host_mask();
10368 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
10369 vmx_set_cr4(vcpu, vmcs12->host_cr4);
10371 nested_ept_uninit_mmu_context(vcpu);
10373 kvm_set_cr3(vcpu, vmcs12->host_cr3);
10374 kvm_mmu_reset_context(vcpu);
10377 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
10381 * Trivially support vpid by letting L2s share their parent
10382 * L1's vpid. TODO: move to a more elaborate solution, giving
10383 * each L2 its own vpid and exposing the vpid feature to L1.
10385 vmx_flush_tlb(vcpu);
10389 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
10390 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
10391 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
10392 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
10393 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
10394 vmcs_write32(GUEST_IDTR_LIMIT, 0xFFFF);
10395 vmcs_write32(GUEST_GDTR_LIMIT, 0xFFFF);
10397 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
10398 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
10399 vmcs_write64(GUEST_BNDCFGS, 0);
10401 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
10402 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
10403 vcpu->arch.pat = vmcs12->host_ia32_pat;
10405 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
10406 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
10407 vmcs12->host_ia32_perf_global_ctrl);
10409 /* Set L1 segment info according to Intel SDM
10410 27.5.2 Loading Host Segment and Descriptor-Table Registers */
10411 seg = (struct kvm_segment) {
10413 .limit = 0xFFFFFFFF,
10414 .selector = vmcs12->host_cs_selector,
10420 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
10424 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
10425 seg = (struct kvm_segment) {
10427 .limit = 0xFFFFFFFF,
10434 seg.selector = vmcs12->host_ds_selector;
10435 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
10436 seg.selector = vmcs12->host_es_selector;
10437 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
10438 seg.selector = vmcs12->host_ss_selector;
10439 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
10440 seg.selector = vmcs12->host_fs_selector;
10441 seg.base = vmcs12->host_fs_base;
10442 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
10443 seg.selector = vmcs12->host_gs_selector;
10444 seg.base = vmcs12->host_gs_base;
10445 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
10446 seg = (struct kvm_segment) {
10447 .base = vmcs12->host_tr_base,
10449 .selector = vmcs12->host_tr_selector,
10453 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
10455 kvm_set_dr(vcpu, 7, 0x400);
10456 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
10458 if (cpu_has_vmx_msr_bitmap())
10459 vmx_set_msr_bitmap(vcpu);
10461 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
10462 vmcs12->vm_exit_msr_load_count))
10463 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
10467 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
10468 * and modify vmcs12 to make it see what it would expect to see there if
10469 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
10471 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
10472 u32 exit_intr_info,
10473 unsigned long exit_qualification)
10475 struct vcpu_vmx *vmx = to_vmx(vcpu);
10476 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10478 /* trying to cancel vmlaunch/vmresume is a bug */
10479 WARN_ON_ONCE(vmx->nested.nested_run_pending);
10481 leave_guest_mode(vcpu);
10482 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
10483 exit_qualification);
10485 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
10486 vmcs12->vm_exit_msr_store_count))
10487 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
10489 vmx_load_vmcs01(vcpu);
10491 if ((exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT)
10492 && nested_exit_intr_ack_set(vcpu)) {
10493 int irq = kvm_cpu_get_interrupt(vcpu);
10495 vmcs12->vm_exit_intr_info = irq |
10496 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
10499 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
10500 vmcs12->exit_qualification,
10501 vmcs12->idt_vectoring_info_field,
10502 vmcs12->vm_exit_intr_info,
10503 vmcs12->vm_exit_intr_error_code,
10506 vm_entry_controls_init(vmx, vmcs_read32(VM_ENTRY_CONTROLS));
10507 vm_exit_controls_init(vmx, vmcs_read32(VM_EXIT_CONTROLS));
10508 vmx_segment_cache_clear(vmx);
10510 /* if no vmcs02 cache requested, remove the one we used */
10511 if (VMCS02_POOL_SIZE == 0)
10512 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
10514 load_vmcs12_host_state(vcpu, vmcs12);
10516 /* Update TSC_OFFSET if TSC was changed while L2 ran */
10517 vmcs_write64(TSC_OFFSET, vmx->nested.vmcs01_tsc_offset);
10519 if (vmx->nested.change_vmcs01_virtual_x2apic_mode) {
10520 vmx->nested.change_vmcs01_virtual_x2apic_mode = false;
10521 vmx_set_virtual_x2apic_mode(vcpu,
10522 vcpu->arch.apic_base & X2APIC_ENABLE);
10525 /* This is needed for same reason as it was needed in prepare_vmcs02 */
10528 /* Unpin physical memory we referred to in vmcs02 */
10529 if (vmx->nested.apic_access_page) {
10530 nested_release_page(vmx->nested.apic_access_page);
10531 vmx->nested.apic_access_page = NULL;
10533 if (vmx->nested.virtual_apic_page) {
10534 nested_release_page(vmx->nested.virtual_apic_page);
10535 vmx->nested.virtual_apic_page = NULL;
10537 if (vmx->nested.pi_desc_page) {
10538 kunmap(vmx->nested.pi_desc_page);
10539 nested_release_page(vmx->nested.pi_desc_page);
10540 vmx->nested.pi_desc_page = NULL;
10541 vmx->nested.pi_desc = NULL;
10545 * We are now running in L2, mmu_notifier will force to reload the
10546 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
10548 kvm_vcpu_reload_apic_access_page(vcpu);
10551 * Exiting from L2 to L1, we're now back to L1 which thinks it just
10552 * finished a VMLAUNCH or VMRESUME instruction, so we need to set the
10553 * success or failure flag accordingly.
10555 if (unlikely(vmx->fail)) {
10557 nested_vmx_failValid(vcpu, vmcs_read32(VM_INSTRUCTION_ERROR));
10559 nested_vmx_succeed(vcpu);
10560 if (enable_shadow_vmcs)
10561 vmx->nested.sync_shadow_vmcs = true;
10563 /* in case we halted in L2 */
10564 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
10568 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
10570 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
10572 if (is_guest_mode(vcpu))
10573 nested_vmx_vmexit(vcpu, -1, 0, 0);
10574 free_nested(to_vmx(vcpu));
10578 * L1's failure to enter L2 is a subset of a normal exit, as explained in
10579 * 23.7 "VM-entry failures during or after loading guest state" (this also
10580 * lists the acceptable exit-reason and exit-qualification parameters).
10581 * It should only be called before L2 actually succeeded to run, and when
10582 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
10584 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
10585 struct vmcs12 *vmcs12,
10586 u32 reason, unsigned long qualification)
10588 load_vmcs12_host_state(vcpu, vmcs12);
10589 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
10590 vmcs12->exit_qualification = qualification;
10591 nested_vmx_succeed(vcpu);
10592 if (enable_shadow_vmcs)
10593 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
10596 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
10597 struct x86_instruction_info *info,
10598 enum x86_intercept_stage stage)
10600 return X86EMUL_CONTINUE;
10603 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
10606 shrink_ple_window(vcpu);
10609 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
10610 struct kvm_memory_slot *slot)
10612 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
10613 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
10616 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
10617 struct kvm_memory_slot *slot)
10619 kvm_mmu_slot_set_dirty(kvm, slot);
10622 static void vmx_flush_log_dirty(struct kvm *kvm)
10624 kvm_flush_pml_buffers(kvm);
10627 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
10628 struct kvm_memory_slot *memslot,
10629 gfn_t offset, unsigned long mask)
10631 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
10635 * This routine does the following things for vCPU which is going
10636 * to be blocked if VT-d PI is enabled.
10637 * - Store the vCPU to the wakeup list, so when interrupts happen
10638 * we can find the right vCPU to wake up.
10639 * - Change the Posted-interrupt descriptor as below:
10640 * 'NDST' <-- vcpu->pre_pcpu
10641 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
10642 * - If 'ON' is set during this process, which means at least one
10643 * interrupt is posted for this vCPU, we cannot block it, in
10644 * this case, return 1, otherwise, return 0.
10647 static int vmx_pre_block(struct kvm_vcpu *vcpu)
10649 unsigned long flags;
10651 struct pi_desc old, new;
10652 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10654 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10655 !irq_remapping_cap(IRQ_POSTING_CAP))
10658 vcpu->pre_pcpu = vcpu->cpu;
10659 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10660 vcpu->pre_pcpu), flags);
10661 list_add_tail(&vcpu->blocked_vcpu_list,
10662 &per_cpu(blocked_vcpu_on_cpu,
10664 spin_unlock_irqrestore(&per_cpu(blocked_vcpu_on_cpu_lock,
10665 vcpu->pre_pcpu), flags);
10668 old.control = new.control = pi_desc->control;
10671 * We should not block the vCPU if
10672 * an interrupt is posted for it.
10674 if (pi_test_on(pi_desc) == 1) {
10675 spin_lock_irqsave(&per_cpu(blocked_vcpu_on_cpu_lock,
10676 vcpu->pre_pcpu), flags);
10677 list_del(&vcpu->blocked_vcpu_list);
10678 spin_unlock_irqrestore(
10679 &per_cpu(blocked_vcpu_on_cpu_lock,
10680 vcpu->pre_pcpu), flags);
10681 vcpu->pre_pcpu = -1;
10686 WARN((pi_desc->sn == 1),
10687 "Warning: SN field of posted-interrupts "
10688 "is set before blocking\n");
10691 * Since vCPU can be preempted during this process,
10692 * vcpu->cpu could be different with pre_pcpu, we
10693 * need to set pre_pcpu as the destination of wakeup
10694 * notification event, then we can find the right vCPU
10695 * to wakeup in wakeup handler if interrupts happen
10696 * when the vCPU is in blocked state.
10698 dest = cpu_physical_id(vcpu->pre_pcpu);
10700 if (x2apic_enabled())
10703 new.ndst = (dest << 8) & 0xFF00;
10705 /* set 'NV' to 'wakeup vector' */
10706 new.nv = POSTED_INTR_WAKEUP_VECTOR;
10707 } while (cmpxchg64(&pi_desc->control, old.control,
10708 new.control) != old.control);
10713 static void vmx_post_block(struct kvm_vcpu *vcpu)
10715 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
10716 struct pi_desc old, new;
10718 unsigned long flags;
10720 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
10721 !irq_remapping_cap(IRQ_POSTING_CAP))
10725 old.control = new.control = pi_desc->control;
10727 dest = cpu_physical_id(vcpu->cpu);
10729 if (x2apic_enabled())
10732 new.ndst = (dest << 8) & 0xFF00;
10734 /* Allow posting non-urgent interrupts */
10737 /* set 'NV' to 'notification vector' */
10738 new.nv = POSTED_INTR_VECTOR;
10739 } while (cmpxchg64(&pi_desc->control, old.control,
10740 new.control) != old.control);
10742 if(vcpu->pre_pcpu != -1) {
10744 &per_cpu(blocked_vcpu_on_cpu_lock,
10745 vcpu->pre_pcpu), flags);
10746 list_del(&vcpu->blocked_vcpu_list);
10747 spin_unlock_irqrestore(
10748 &per_cpu(blocked_vcpu_on_cpu_lock,
10749 vcpu->pre_pcpu), flags);
10750 vcpu->pre_pcpu = -1;
10755 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
10758 * @host_irq: host irq of the interrupt
10759 * @guest_irq: gsi of the interrupt
10760 * @set: set or unset PI
10761 * returns 0 on success, < 0 on failure
10763 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
10764 uint32_t guest_irq, bool set)
10766 struct kvm_kernel_irq_routing_entry *e;
10767 struct kvm_irq_routing_table *irq_rt;
10768 struct kvm_lapic_irq irq;
10769 struct kvm_vcpu *vcpu;
10770 struct vcpu_data vcpu_info;
10773 if (!kvm_arch_has_assigned_device(kvm) ||
10774 !irq_remapping_cap(IRQ_POSTING_CAP))
10777 idx = srcu_read_lock(&kvm->irq_srcu);
10778 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
10779 if (guest_irq >= irq_rt->nr_rt_entries ||
10780 hlist_empty(&irq_rt->map[guest_irq])) {
10781 pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
10782 guest_irq, irq_rt->nr_rt_entries);
10786 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
10787 if (e->type != KVM_IRQ_ROUTING_MSI)
10790 * VT-d PI cannot support posting multicast/broadcast
10791 * interrupts to a vCPU, we still use interrupt remapping
10792 * for these kind of interrupts.
10794 * For lowest-priority interrupts, we only support
10795 * those with single CPU as the destination, e.g. user
10796 * configures the interrupts via /proc/irq or uses
10797 * irqbalance to make the interrupts single-CPU.
10799 * We will support full lowest-priority interrupt later.
10802 kvm_set_msi_irq(e, &irq);
10803 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu))
10806 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
10807 vcpu_info.vector = irq.vector;
10809 trace_kvm_pi_irte_update(vcpu->vcpu_id, e->gsi,
10810 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
10813 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
10815 ret = irq_set_vcpu_affinity(host_irq, NULL);
10818 printk(KERN_INFO "%s: failed to update PI IRTE\n",
10826 srcu_read_unlock(&kvm->irq_srcu, idx);
10830 static struct kvm_x86_ops vmx_x86_ops = {
10831 .cpu_has_kvm_support = cpu_has_kvm_support,
10832 .disabled_by_bios = vmx_disabled_by_bios,
10833 .hardware_setup = hardware_setup,
10834 .hardware_unsetup = hardware_unsetup,
10835 .check_processor_compatibility = vmx_check_processor_compat,
10836 .hardware_enable = hardware_enable,
10837 .hardware_disable = hardware_disable,
10838 .cpu_has_accelerated_tpr = report_flexpriority,
10839 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
10841 .vcpu_create = vmx_create_vcpu,
10842 .vcpu_free = vmx_free_vcpu,
10843 .vcpu_reset = vmx_vcpu_reset,
10845 .prepare_guest_switch = vmx_save_host_state,
10846 .vcpu_load = vmx_vcpu_load,
10847 .vcpu_put = vmx_vcpu_put,
10849 .update_bp_intercept = update_exception_bitmap,
10850 .get_msr = vmx_get_msr,
10851 .set_msr = vmx_set_msr,
10852 .get_segment_base = vmx_get_segment_base,
10853 .get_segment = vmx_get_segment,
10854 .set_segment = vmx_set_segment,
10855 .get_cpl = vmx_get_cpl,
10856 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
10857 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
10858 .decache_cr3 = vmx_decache_cr3,
10859 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
10860 .set_cr0 = vmx_set_cr0,
10861 .set_cr3 = vmx_set_cr3,
10862 .set_cr4 = vmx_set_cr4,
10863 .set_efer = vmx_set_efer,
10864 .get_idt = vmx_get_idt,
10865 .set_idt = vmx_set_idt,
10866 .get_gdt = vmx_get_gdt,
10867 .set_gdt = vmx_set_gdt,
10868 .get_dr6 = vmx_get_dr6,
10869 .set_dr6 = vmx_set_dr6,
10870 .set_dr7 = vmx_set_dr7,
10871 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
10872 .cache_reg = vmx_cache_reg,
10873 .get_rflags = vmx_get_rflags,
10874 .set_rflags = vmx_set_rflags,
10875 .fpu_activate = vmx_fpu_activate,
10876 .fpu_deactivate = vmx_fpu_deactivate,
10878 .tlb_flush = vmx_flush_tlb,
10880 .run = vmx_vcpu_run,
10881 .handle_exit = vmx_handle_exit,
10882 .skip_emulated_instruction = skip_emulated_instruction,
10883 .set_interrupt_shadow = vmx_set_interrupt_shadow,
10884 .get_interrupt_shadow = vmx_get_interrupt_shadow,
10885 .patch_hypercall = vmx_patch_hypercall,
10886 .set_irq = vmx_inject_irq,
10887 .set_nmi = vmx_inject_nmi,
10888 .queue_exception = vmx_queue_exception,
10889 .cancel_injection = vmx_cancel_injection,
10890 .interrupt_allowed = vmx_interrupt_allowed,
10891 .nmi_allowed = vmx_nmi_allowed,
10892 .get_nmi_mask = vmx_get_nmi_mask,
10893 .set_nmi_mask = vmx_set_nmi_mask,
10894 .enable_nmi_window = enable_nmi_window,
10895 .enable_irq_window = enable_irq_window,
10896 .update_cr8_intercept = update_cr8_intercept,
10897 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
10898 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
10899 .cpu_uses_apicv = vmx_cpu_uses_apicv,
10900 .load_eoi_exitmap = vmx_load_eoi_exitmap,
10901 .hwapic_irr_update = vmx_hwapic_irr_update,
10902 .hwapic_isr_update = vmx_hwapic_isr_update,
10903 .sync_pir_to_irr = vmx_sync_pir_to_irr,
10904 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
10906 .set_tss_addr = vmx_set_tss_addr,
10907 .get_tdp_level = get_ept_level,
10908 .get_mt_mask = vmx_get_mt_mask,
10910 .get_exit_info = vmx_get_exit_info,
10912 .get_lpage_level = vmx_get_lpage_level,
10914 .cpuid_update = vmx_cpuid_update,
10916 .rdtscp_supported = vmx_rdtscp_supported,
10917 .invpcid_supported = vmx_invpcid_supported,
10919 .set_supported_cpuid = vmx_set_supported_cpuid,
10921 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
10923 .read_tsc_offset = vmx_read_tsc_offset,
10924 .write_tsc_offset = vmx_write_tsc_offset,
10925 .adjust_tsc_offset_guest = vmx_adjust_tsc_offset_guest,
10926 .read_l1_tsc = vmx_read_l1_tsc,
10928 .set_tdp_cr3 = vmx_set_cr3,
10930 .check_intercept = vmx_check_intercept,
10931 .handle_external_intr = vmx_handle_external_intr,
10932 .mpx_supported = vmx_mpx_supported,
10933 .xsaves_supported = vmx_xsaves_supported,
10935 .check_nested_events = vmx_check_nested_events,
10937 .sched_in = vmx_sched_in,
10939 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
10940 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
10941 .flush_log_dirty = vmx_flush_log_dirty,
10942 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
10944 .pre_block = vmx_pre_block,
10945 .post_block = vmx_post_block,
10947 .pmu_ops = &intel_pmu_ops,
10949 .update_pi_irte = vmx_update_pi_irte,
10952 static int __init vmx_init(void)
10954 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
10955 __alignof__(struct vcpu_vmx), THIS_MODULE);
10959 #ifdef CONFIG_KEXEC_CORE
10960 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
10961 crash_vmclear_local_loaded_vmcss);
10967 static void __exit vmx_exit(void)
10969 #ifdef CONFIG_KEXEC_CORE
10970 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
10977 module_init(vmx_init)
10978 module_exit(vmx_exit)