2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
24 #include <linux/kvm_host.h>
25 #include <linux/module.h>
26 #include <linux/kernel.h>
28 #include <linux/highmem.h>
29 #include <linux/sched.h>
30 #include <linux/moduleparam.h>
31 #include <linux/mod_devicetable.h>
32 #include <linux/trace_events.h>
33 #include <linux/slab.h>
34 #include <linux/tboot.h>
35 #include <linux/hrtimer.h>
36 #include <linux/frame.h>
37 #include "kvm_cache_regs.h"
44 #include <asm/virtext.h>
46 #include <asm/fpu/internal.h>
47 #include <asm/perf_event.h>
48 #include <asm/debugreg.h>
49 #include <asm/kexec.h>
51 #include <asm/irq_remapping.h>
52 #include <asm/mmu_context.h>
57 #define __ex(x) __kvm_handle_fault_on_reboot(x)
58 #define __ex_clear(x, reg) \
59 ____kvm_handle_fault_on_reboot(x, "xor " reg " , " reg)
61 MODULE_AUTHOR("Qumranet");
62 MODULE_LICENSE("GPL");
64 static const struct x86_cpu_id vmx_cpu_id[] = {
65 X86_FEATURE_MATCH(X86_FEATURE_VMX),
68 MODULE_DEVICE_TABLE(x86cpu, vmx_cpu_id);
70 static bool __read_mostly enable_vpid = 1;
71 module_param_named(vpid, enable_vpid, bool, 0444);
73 static bool __read_mostly flexpriority_enabled = 1;
74 module_param_named(flexpriority, flexpriority_enabled, bool, S_IRUGO);
76 static bool __read_mostly enable_ept = 1;
77 module_param_named(ept, enable_ept, bool, S_IRUGO);
79 static bool __read_mostly enable_unrestricted_guest = 1;
80 module_param_named(unrestricted_guest,
81 enable_unrestricted_guest, bool, S_IRUGO);
83 static bool __read_mostly enable_ept_ad_bits = 1;
84 module_param_named(eptad, enable_ept_ad_bits, bool, S_IRUGO);
86 static bool __read_mostly emulate_invalid_guest_state = true;
87 module_param(emulate_invalid_guest_state, 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 /* Guest_tsc -> host_tsc conversion requires 64-bit division. */
113 static int __read_mostly cpu_preemption_timer_multi;
114 static bool __read_mostly enable_preemption_timer = 1;
116 module_param_named(preemption_timer, enable_preemption_timer, bool, S_IRUGO);
119 #define KVM_GUEST_CR0_MASK (X86_CR0_NW | X86_CR0_CD)
120 #define KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST (X86_CR0_WP | X86_CR0_NE)
121 #define KVM_VM_CR0_ALWAYS_ON \
122 (KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST | X86_CR0_PG | X86_CR0_PE)
123 #define KVM_CR4_GUEST_OWNED_BITS \
124 (X86_CR4_PVI | X86_CR4_DE | X86_CR4_PCE | X86_CR4_OSFXSR \
125 | X86_CR4_OSXMMEXCPT | X86_CR4_LA57 | X86_CR4_TSD)
127 #define KVM_PMODE_VM_CR4_ALWAYS_ON (X86_CR4_PAE | X86_CR4_VMXE)
128 #define KVM_RMODE_VM_CR4_ALWAYS_ON (X86_CR4_VME | X86_CR4_PAE | X86_CR4_VMXE)
130 #define RMODE_GUEST_OWNED_EFLAGS_BITS (~(X86_EFLAGS_IOPL | X86_EFLAGS_VM))
132 #define VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE 5
135 * Hyper-V requires all of these, so mark them as supported even though
136 * they are just treated the same as all-context.
138 #define VMX_VPID_EXTENT_SUPPORTED_MASK \
139 (VMX_VPID_EXTENT_INDIVIDUAL_ADDR_BIT | \
140 VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT | \
141 VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT | \
142 VMX_VPID_EXTENT_SINGLE_NON_GLOBAL_BIT)
145 * These 2 parameters are used to config the controls for Pause-Loop Exiting:
146 * ple_gap: upper bound on the amount of time between two successive
147 * executions of PAUSE in a loop. Also indicate if ple enabled.
148 * According to test, this time is usually smaller than 128 cycles.
149 * ple_window: upper bound on the amount of time a guest is allowed to execute
150 * in a PAUSE loop. Tests indicate that most spinlocks are held for
151 * less than 2^12 cycles
152 * Time is measured based on a counter that runs at the same rate as the TSC,
153 * refer SDM volume 3b section 21.6.13 & 22.1.3.
155 #define KVM_VMX_DEFAULT_PLE_GAP 128
156 #define KVM_VMX_DEFAULT_PLE_WINDOW 4096
157 #define KVM_VMX_DEFAULT_PLE_WINDOW_GROW 2
158 #define KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK 0
159 #define KVM_VMX_DEFAULT_PLE_WINDOW_MAX \
160 INT_MAX / KVM_VMX_DEFAULT_PLE_WINDOW_GROW
162 static int ple_gap = KVM_VMX_DEFAULT_PLE_GAP;
163 module_param(ple_gap, int, S_IRUGO);
165 static int ple_window = KVM_VMX_DEFAULT_PLE_WINDOW;
166 module_param(ple_window, int, S_IRUGO);
168 /* Default doubles per-vcpu window every exit. */
169 static int ple_window_grow = KVM_VMX_DEFAULT_PLE_WINDOW_GROW;
170 module_param(ple_window_grow, int, S_IRUGO);
172 /* Default resets per-vcpu window every exit to ple_window. */
173 static int ple_window_shrink = KVM_VMX_DEFAULT_PLE_WINDOW_SHRINK;
174 module_param(ple_window_shrink, int, S_IRUGO);
176 /* Default is to compute the maximum so we can never overflow. */
177 static int ple_window_actual_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
178 static int ple_window_max = KVM_VMX_DEFAULT_PLE_WINDOW_MAX;
179 module_param(ple_window_max, int, S_IRUGO);
181 extern const ulong vmx_return;
183 #define NR_AUTOLOAD_MSRS 8
184 #define VMCS02_POOL_SIZE 1
193 * Track a VMCS that may be loaded on a certain CPU. If it is (cpu!=-1), also
194 * remember whether it was VMLAUNCHed, and maintain a linked list of all VMCSs
195 * loaded on this CPU (so we can clear them if the CPU goes down).
199 struct vmcs *shadow_vmcs;
202 bool nmi_known_unmasked;
203 unsigned long vmcs_host_cr3; /* May not match real cr3 */
204 unsigned long vmcs_host_cr4; /* May not match real cr4 */
205 struct list_head loaded_vmcss_on_cpu_link;
208 struct shared_msr_entry {
215 * struct vmcs12 describes the state that our guest hypervisor (L1) keeps for a
216 * single nested guest (L2), hence the name vmcs12. Any VMX implementation has
217 * a VMCS structure, and vmcs12 is our emulated VMX's VMCS. This structure is
218 * stored in guest memory specified by VMPTRLD, but is opaque to the guest,
219 * which must access it using VMREAD/VMWRITE/VMCLEAR instructions.
220 * More than one of these structures may exist, if L1 runs multiple L2 guests.
221 * nested_vmx_run() will use the data here to build a vmcs02: a VMCS for the
222 * underlying hardware which will be used to run L2.
223 * This structure is packed to ensure that its layout is identical across
224 * machines (necessary for live migration).
225 * If there are changes in this struct, VMCS12_REVISION must be changed.
227 typedef u64 natural_width;
228 struct __packed vmcs12 {
229 /* According to the Intel spec, a VMCS region must start with the
230 * following two fields. Then follow implementation-specific data.
235 u32 launch_state; /* set to 0 by VMCLEAR, to 1 by VMLAUNCH */
236 u32 padding[7]; /* room for future expansion */
241 u64 vm_exit_msr_store_addr;
242 u64 vm_exit_msr_load_addr;
243 u64 vm_entry_msr_load_addr;
245 u64 virtual_apic_page_addr;
246 u64 apic_access_addr;
247 u64 posted_intr_desc_addr;
248 u64 vm_function_control;
250 u64 eoi_exit_bitmap0;
251 u64 eoi_exit_bitmap1;
252 u64 eoi_exit_bitmap2;
253 u64 eoi_exit_bitmap3;
254 u64 eptp_list_address;
256 u64 guest_physical_address;
257 u64 vmcs_link_pointer;
259 u64 guest_ia32_debugctl;
262 u64 guest_ia32_perf_global_ctrl;
270 u64 host_ia32_perf_global_ctrl;
271 u64 padding64[8]; /* room for future expansion */
273 * To allow migration of L1 (complete with its L2 guests) between
274 * machines of different natural widths (32 or 64 bit), we cannot have
275 * unsigned long fields with no explict size. We use u64 (aliased
276 * natural_width) instead. Luckily, x86 is little-endian.
278 natural_width cr0_guest_host_mask;
279 natural_width cr4_guest_host_mask;
280 natural_width cr0_read_shadow;
281 natural_width cr4_read_shadow;
282 natural_width cr3_target_value0;
283 natural_width cr3_target_value1;
284 natural_width cr3_target_value2;
285 natural_width cr3_target_value3;
286 natural_width exit_qualification;
287 natural_width guest_linear_address;
288 natural_width guest_cr0;
289 natural_width guest_cr3;
290 natural_width guest_cr4;
291 natural_width guest_es_base;
292 natural_width guest_cs_base;
293 natural_width guest_ss_base;
294 natural_width guest_ds_base;
295 natural_width guest_fs_base;
296 natural_width guest_gs_base;
297 natural_width guest_ldtr_base;
298 natural_width guest_tr_base;
299 natural_width guest_gdtr_base;
300 natural_width guest_idtr_base;
301 natural_width guest_dr7;
302 natural_width guest_rsp;
303 natural_width guest_rip;
304 natural_width guest_rflags;
305 natural_width guest_pending_dbg_exceptions;
306 natural_width guest_sysenter_esp;
307 natural_width guest_sysenter_eip;
308 natural_width host_cr0;
309 natural_width host_cr3;
310 natural_width host_cr4;
311 natural_width host_fs_base;
312 natural_width host_gs_base;
313 natural_width host_tr_base;
314 natural_width host_gdtr_base;
315 natural_width host_idtr_base;
316 natural_width host_ia32_sysenter_esp;
317 natural_width host_ia32_sysenter_eip;
318 natural_width host_rsp;
319 natural_width host_rip;
320 natural_width paddingl[8]; /* room for future expansion */
321 u32 pin_based_vm_exec_control;
322 u32 cpu_based_vm_exec_control;
323 u32 exception_bitmap;
324 u32 page_fault_error_code_mask;
325 u32 page_fault_error_code_match;
326 u32 cr3_target_count;
327 u32 vm_exit_controls;
328 u32 vm_exit_msr_store_count;
329 u32 vm_exit_msr_load_count;
330 u32 vm_entry_controls;
331 u32 vm_entry_msr_load_count;
332 u32 vm_entry_intr_info_field;
333 u32 vm_entry_exception_error_code;
334 u32 vm_entry_instruction_len;
336 u32 secondary_vm_exec_control;
337 u32 vm_instruction_error;
339 u32 vm_exit_intr_info;
340 u32 vm_exit_intr_error_code;
341 u32 idt_vectoring_info_field;
342 u32 idt_vectoring_error_code;
343 u32 vm_exit_instruction_len;
344 u32 vmx_instruction_info;
351 u32 guest_ldtr_limit;
353 u32 guest_gdtr_limit;
354 u32 guest_idtr_limit;
355 u32 guest_es_ar_bytes;
356 u32 guest_cs_ar_bytes;
357 u32 guest_ss_ar_bytes;
358 u32 guest_ds_ar_bytes;
359 u32 guest_fs_ar_bytes;
360 u32 guest_gs_ar_bytes;
361 u32 guest_ldtr_ar_bytes;
362 u32 guest_tr_ar_bytes;
363 u32 guest_interruptibility_info;
364 u32 guest_activity_state;
365 u32 guest_sysenter_cs;
366 u32 host_ia32_sysenter_cs;
367 u32 vmx_preemption_timer_value;
368 u32 padding32[7]; /* room for future expansion */
369 u16 virtual_processor_id;
371 u16 guest_es_selector;
372 u16 guest_cs_selector;
373 u16 guest_ss_selector;
374 u16 guest_ds_selector;
375 u16 guest_fs_selector;
376 u16 guest_gs_selector;
377 u16 guest_ldtr_selector;
378 u16 guest_tr_selector;
379 u16 guest_intr_status;
381 u16 host_es_selector;
382 u16 host_cs_selector;
383 u16 host_ss_selector;
384 u16 host_ds_selector;
385 u16 host_fs_selector;
386 u16 host_gs_selector;
387 u16 host_tr_selector;
391 * VMCS12_REVISION is an arbitrary id that should be changed if the content or
392 * layout of struct vmcs12 is changed. MSR_IA32_VMX_BASIC returns this id, and
393 * VMPTRLD verifies that the VMCS region that L1 is loading contains this id.
395 #define VMCS12_REVISION 0x11e57ed0
398 * VMCS12_SIZE is the number of bytes L1 should allocate for the VMXON region
399 * and any VMCS region. Although only sizeof(struct vmcs12) are used by the
400 * current implementation, 4K are reserved to avoid future complications.
402 #define VMCS12_SIZE 0x1000
404 /* Used to remember the last vmcs02 used for some recently used vmcs12s */
406 struct list_head list;
408 struct loaded_vmcs vmcs02;
412 * The nested_vmx structure is part of vcpu_vmx, and holds information we need
413 * for correct emulation of VMX (i.e., nested VMX) on this vcpu.
416 /* Has the level1 guest done vmxon? */
421 /* The guest-physical address of the current VMCS L1 keeps for L2 */
424 * Cache of the guest's VMCS, existing outside of guest memory.
425 * Loaded from guest memory during VMPTRLD. Flushed to guest
426 * memory during VMCLEAR and VMPTRLD.
428 struct vmcs12 *cached_vmcs12;
430 * Indicates if the shadow vmcs must be updated with the
431 * data hold by vmcs12
433 bool sync_shadow_vmcs;
435 /* vmcs02_list cache of VMCSs recently used to run L2 guests */
436 struct list_head vmcs02_pool;
438 bool change_vmcs01_virtual_x2apic_mode;
439 /* L2 must run next, and mustn't decide to exit to L1. */
440 bool nested_run_pending;
442 * Guest pages referred to in vmcs02 with host-physical pointers, so
443 * we must keep them pinned while L2 runs.
445 struct page *apic_access_page;
446 struct page *virtual_apic_page;
447 struct page *pi_desc_page;
448 struct pi_desc *pi_desc;
452 unsigned long *msr_bitmap;
454 struct hrtimer preemption_timer;
455 bool preemption_timer_expired;
457 /* to migrate it to L2 if VM_ENTRY_LOAD_DEBUG_CONTROLS is off */
464 * We only store the "true" versions of the VMX capability MSRs. We
465 * generate the "non-true" versions by setting the must-be-1 bits
466 * according to the SDM.
468 u32 nested_vmx_procbased_ctls_low;
469 u32 nested_vmx_procbased_ctls_high;
470 u32 nested_vmx_secondary_ctls_low;
471 u32 nested_vmx_secondary_ctls_high;
472 u32 nested_vmx_pinbased_ctls_low;
473 u32 nested_vmx_pinbased_ctls_high;
474 u32 nested_vmx_exit_ctls_low;
475 u32 nested_vmx_exit_ctls_high;
476 u32 nested_vmx_entry_ctls_low;
477 u32 nested_vmx_entry_ctls_high;
478 u32 nested_vmx_misc_low;
479 u32 nested_vmx_misc_high;
480 u32 nested_vmx_ept_caps;
481 u32 nested_vmx_vpid_caps;
482 u64 nested_vmx_basic;
483 u64 nested_vmx_cr0_fixed0;
484 u64 nested_vmx_cr0_fixed1;
485 u64 nested_vmx_cr4_fixed0;
486 u64 nested_vmx_cr4_fixed1;
487 u64 nested_vmx_vmcs_enum;
488 u64 nested_vmx_vmfunc_controls;
491 #define POSTED_INTR_ON 0
492 #define POSTED_INTR_SN 1
494 /* Posted-Interrupt Descriptor */
496 u32 pir[8]; /* Posted interrupt requested */
499 /* bit 256 - Outstanding Notification */
501 /* bit 257 - Suppress Notification */
503 /* bit 271:258 - Reserved */
505 /* bit 279:272 - Notification Vector */
507 /* bit 287:280 - Reserved */
509 /* bit 319:288 - Notification Destination */
517 static bool pi_test_and_set_on(struct pi_desc *pi_desc)
519 return test_and_set_bit(POSTED_INTR_ON,
520 (unsigned long *)&pi_desc->control);
523 static bool pi_test_and_clear_on(struct pi_desc *pi_desc)
525 return test_and_clear_bit(POSTED_INTR_ON,
526 (unsigned long *)&pi_desc->control);
529 static int pi_test_and_set_pir(int vector, struct pi_desc *pi_desc)
531 return test_and_set_bit(vector, (unsigned long *)pi_desc->pir);
534 static inline void pi_clear_sn(struct pi_desc *pi_desc)
536 return clear_bit(POSTED_INTR_SN,
537 (unsigned long *)&pi_desc->control);
540 static inline void pi_set_sn(struct pi_desc *pi_desc)
542 return set_bit(POSTED_INTR_SN,
543 (unsigned long *)&pi_desc->control);
546 static inline void pi_clear_on(struct pi_desc *pi_desc)
548 clear_bit(POSTED_INTR_ON,
549 (unsigned long *)&pi_desc->control);
552 static inline int pi_test_on(struct pi_desc *pi_desc)
554 return test_bit(POSTED_INTR_ON,
555 (unsigned long *)&pi_desc->control);
558 static inline int pi_test_sn(struct pi_desc *pi_desc)
560 return test_bit(POSTED_INTR_SN,
561 (unsigned long *)&pi_desc->control);
565 struct kvm_vcpu vcpu;
566 unsigned long host_rsp;
569 u32 idt_vectoring_info;
571 struct shared_msr_entry *guest_msrs;
574 unsigned long host_idt_base;
576 u64 msr_host_kernel_gs_base;
577 u64 msr_guest_kernel_gs_base;
579 u32 vm_entry_controls_shadow;
580 u32 vm_exit_controls_shadow;
581 u32 secondary_exec_control;
584 * loaded_vmcs points to the VMCS currently used in this vcpu. For a
585 * non-nested (L1) guest, it always points to vmcs01. For a nested
586 * guest (L2), it points to a different VMCS.
588 struct loaded_vmcs vmcs01;
589 struct loaded_vmcs *loaded_vmcs;
590 bool __launched; /* temporary, used in vmx_vcpu_run */
591 struct msr_autoload {
593 struct vmx_msr_entry guest[NR_AUTOLOAD_MSRS];
594 struct vmx_msr_entry host[NR_AUTOLOAD_MSRS];
598 u16 fs_sel, gs_sel, ldt_sel;
602 int gs_ldt_reload_needed;
603 int fs_reload_needed;
604 u64 msr_host_bndcfgs;
609 struct kvm_segment segs[8];
612 u32 bitmask; /* 4 bits per segment (1 bit per field) */
613 struct kvm_save_segment {
621 bool emulation_required;
625 /* Posted interrupt descriptor */
626 struct pi_desc pi_desc;
628 /* Support for a guest hypervisor (nested VMX) */
629 struct nested_vmx nested;
631 /* Dynamic PLE window. */
633 bool ple_window_dirty;
635 /* Support for PML */
636 #define PML_ENTITY_NUM 512
639 /* apic deadline value in host tsc */
642 u64 current_tsc_ratio;
647 * Only bits masked by msr_ia32_feature_control_valid_bits can be set in
648 * msr_ia32_feature_control. FEATURE_CONTROL_LOCKED is always included
649 * in msr_ia32_feature_control_valid_bits.
651 u64 msr_ia32_feature_control;
652 u64 msr_ia32_feature_control_valid_bits;
655 enum segment_cache_field {
664 static inline struct vcpu_vmx *to_vmx(struct kvm_vcpu *vcpu)
666 return container_of(vcpu, struct vcpu_vmx, vcpu);
669 static struct pi_desc *vcpu_to_pi_desc(struct kvm_vcpu *vcpu)
671 return &(to_vmx(vcpu)->pi_desc);
674 #define VMCS12_OFFSET(x) offsetof(struct vmcs12, x)
675 #define FIELD(number, name) [number] = VMCS12_OFFSET(name)
676 #define FIELD64(number, name) [number] = VMCS12_OFFSET(name), \
677 [number##_HIGH] = VMCS12_OFFSET(name)+4
680 static unsigned long shadow_read_only_fields[] = {
682 * We do NOT shadow fields that are modified when L0
683 * traps and emulates any vmx instruction (e.g. VMPTRLD,
684 * VMXON...) executed by L1.
685 * For example, VM_INSTRUCTION_ERROR is read
686 * by L1 if a vmx instruction fails (part of the error path).
687 * Note the code assumes this logic. If for some reason
688 * we start shadowing these fields then we need to
689 * force a shadow sync when L0 emulates vmx instructions
690 * (e.g. force a sync if VM_INSTRUCTION_ERROR is modified
691 * by nested_vmx_failValid)
695 VM_EXIT_INSTRUCTION_LEN,
696 IDT_VECTORING_INFO_FIELD,
697 IDT_VECTORING_ERROR_CODE,
698 VM_EXIT_INTR_ERROR_CODE,
700 GUEST_LINEAR_ADDRESS,
701 GUEST_PHYSICAL_ADDRESS
703 static int max_shadow_read_only_fields =
704 ARRAY_SIZE(shadow_read_only_fields);
706 static unsigned long shadow_read_write_fields[] = {
713 GUEST_INTERRUPTIBILITY_INFO,
726 CPU_BASED_VM_EXEC_CONTROL,
727 VM_ENTRY_EXCEPTION_ERROR_CODE,
728 VM_ENTRY_INTR_INFO_FIELD,
729 VM_ENTRY_INSTRUCTION_LEN,
730 VM_ENTRY_EXCEPTION_ERROR_CODE,
736 static int max_shadow_read_write_fields =
737 ARRAY_SIZE(shadow_read_write_fields);
739 static const unsigned short vmcs_field_to_offset_table[] = {
740 FIELD(VIRTUAL_PROCESSOR_ID, virtual_processor_id),
741 FIELD(POSTED_INTR_NV, posted_intr_nv),
742 FIELD(GUEST_ES_SELECTOR, guest_es_selector),
743 FIELD(GUEST_CS_SELECTOR, guest_cs_selector),
744 FIELD(GUEST_SS_SELECTOR, guest_ss_selector),
745 FIELD(GUEST_DS_SELECTOR, guest_ds_selector),
746 FIELD(GUEST_FS_SELECTOR, guest_fs_selector),
747 FIELD(GUEST_GS_SELECTOR, guest_gs_selector),
748 FIELD(GUEST_LDTR_SELECTOR, guest_ldtr_selector),
749 FIELD(GUEST_TR_SELECTOR, guest_tr_selector),
750 FIELD(GUEST_INTR_STATUS, guest_intr_status),
751 FIELD(GUEST_PML_INDEX, guest_pml_index),
752 FIELD(HOST_ES_SELECTOR, host_es_selector),
753 FIELD(HOST_CS_SELECTOR, host_cs_selector),
754 FIELD(HOST_SS_SELECTOR, host_ss_selector),
755 FIELD(HOST_DS_SELECTOR, host_ds_selector),
756 FIELD(HOST_FS_SELECTOR, host_fs_selector),
757 FIELD(HOST_GS_SELECTOR, host_gs_selector),
758 FIELD(HOST_TR_SELECTOR, host_tr_selector),
759 FIELD64(IO_BITMAP_A, io_bitmap_a),
760 FIELD64(IO_BITMAP_B, io_bitmap_b),
761 FIELD64(MSR_BITMAP, msr_bitmap),
762 FIELD64(VM_EXIT_MSR_STORE_ADDR, vm_exit_msr_store_addr),
763 FIELD64(VM_EXIT_MSR_LOAD_ADDR, vm_exit_msr_load_addr),
764 FIELD64(VM_ENTRY_MSR_LOAD_ADDR, vm_entry_msr_load_addr),
765 FIELD64(TSC_OFFSET, tsc_offset),
766 FIELD64(VIRTUAL_APIC_PAGE_ADDR, virtual_apic_page_addr),
767 FIELD64(APIC_ACCESS_ADDR, apic_access_addr),
768 FIELD64(POSTED_INTR_DESC_ADDR, posted_intr_desc_addr),
769 FIELD64(VM_FUNCTION_CONTROL, vm_function_control),
770 FIELD64(EPT_POINTER, ept_pointer),
771 FIELD64(EOI_EXIT_BITMAP0, eoi_exit_bitmap0),
772 FIELD64(EOI_EXIT_BITMAP1, eoi_exit_bitmap1),
773 FIELD64(EOI_EXIT_BITMAP2, eoi_exit_bitmap2),
774 FIELD64(EOI_EXIT_BITMAP3, eoi_exit_bitmap3),
775 FIELD64(EPTP_LIST_ADDRESS, eptp_list_address),
776 FIELD64(XSS_EXIT_BITMAP, xss_exit_bitmap),
777 FIELD64(GUEST_PHYSICAL_ADDRESS, guest_physical_address),
778 FIELD64(VMCS_LINK_POINTER, vmcs_link_pointer),
779 FIELD64(PML_ADDRESS, pml_address),
780 FIELD64(GUEST_IA32_DEBUGCTL, guest_ia32_debugctl),
781 FIELD64(GUEST_IA32_PAT, guest_ia32_pat),
782 FIELD64(GUEST_IA32_EFER, guest_ia32_efer),
783 FIELD64(GUEST_IA32_PERF_GLOBAL_CTRL, guest_ia32_perf_global_ctrl),
784 FIELD64(GUEST_PDPTR0, guest_pdptr0),
785 FIELD64(GUEST_PDPTR1, guest_pdptr1),
786 FIELD64(GUEST_PDPTR2, guest_pdptr2),
787 FIELD64(GUEST_PDPTR3, guest_pdptr3),
788 FIELD64(GUEST_BNDCFGS, guest_bndcfgs),
789 FIELD64(HOST_IA32_PAT, host_ia32_pat),
790 FIELD64(HOST_IA32_EFER, host_ia32_efer),
791 FIELD64(HOST_IA32_PERF_GLOBAL_CTRL, host_ia32_perf_global_ctrl),
792 FIELD(PIN_BASED_VM_EXEC_CONTROL, pin_based_vm_exec_control),
793 FIELD(CPU_BASED_VM_EXEC_CONTROL, cpu_based_vm_exec_control),
794 FIELD(EXCEPTION_BITMAP, exception_bitmap),
795 FIELD(PAGE_FAULT_ERROR_CODE_MASK, page_fault_error_code_mask),
796 FIELD(PAGE_FAULT_ERROR_CODE_MATCH, page_fault_error_code_match),
797 FIELD(CR3_TARGET_COUNT, cr3_target_count),
798 FIELD(VM_EXIT_CONTROLS, vm_exit_controls),
799 FIELD(VM_EXIT_MSR_STORE_COUNT, vm_exit_msr_store_count),
800 FIELD(VM_EXIT_MSR_LOAD_COUNT, vm_exit_msr_load_count),
801 FIELD(VM_ENTRY_CONTROLS, vm_entry_controls),
802 FIELD(VM_ENTRY_MSR_LOAD_COUNT, vm_entry_msr_load_count),
803 FIELD(VM_ENTRY_INTR_INFO_FIELD, vm_entry_intr_info_field),
804 FIELD(VM_ENTRY_EXCEPTION_ERROR_CODE, vm_entry_exception_error_code),
805 FIELD(VM_ENTRY_INSTRUCTION_LEN, vm_entry_instruction_len),
806 FIELD(TPR_THRESHOLD, tpr_threshold),
807 FIELD(SECONDARY_VM_EXEC_CONTROL, secondary_vm_exec_control),
808 FIELD(VM_INSTRUCTION_ERROR, vm_instruction_error),
809 FIELD(VM_EXIT_REASON, vm_exit_reason),
810 FIELD(VM_EXIT_INTR_INFO, vm_exit_intr_info),
811 FIELD(VM_EXIT_INTR_ERROR_CODE, vm_exit_intr_error_code),
812 FIELD(IDT_VECTORING_INFO_FIELD, idt_vectoring_info_field),
813 FIELD(IDT_VECTORING_ERROR_CODE, idt_vectoring_error_code),
814 FIELD(VM_EXIT_INSTRUCTION_LEN, vm_exit_instruction_len),
815 FIELD(VMX_INSTRUCTION_INFO, vmx_instruction_info),
816 FIELD(GUEST_ES_LIMIT, guest_es_limit),
817 FIELD(GUEST_CS_LIMIT, guest_cs_limit),
818 FIELD(GUEST_SS_LIMIT, guest_ss_limit),
819 FIELD(GUEST_DS_LIMIT, guest_ds_limit),
820 FIELD(GUEST_FS_LIMIT, guest_fs_limit),
821 FIELD(GUEST_GS_LIMIT, guest_gs_limit),
822 FIELD(GUEST_LDTR_LIMIT, guest_ldtr_limit),
823 FIELD(GUEST_TR_LIMIT, guest_tr_limit),
824 FIELD(GUEST_GDTR_LIMIT, guest_gdtr_limit),
825 FIELD(GUEST_IDTR_LIMIT, guest_idtr_limit),
826 FIELD(GUEST_ES_AR_BYTES, guest_es_ar_bytes),
827 FIELD(GUEST_CS_AR_BYTES, guest_cs_ar_bytes),
828 FIELD(GUEST_SS_AR_BYTES, guest_ss_ar_bytes),
829 FIELD(GUEST_DS_AR_BYTES, guest_ds_ar_bytes),
830 FIELD(GUEST_FS_AR_BYTES, guest_fs_ar_bytes),
831 FIELD(GUEST_GS_AR_BYTES, guest_gs_ar_bytes),
832 FIELD(GUEST_LDTR_AR_BYTES, guest_ldtr_ar_bytes),
833 FIELD(GUEST_TR_AR_BYTES, guest_tr_ar_bytes),
834 FIELD(GUEST_INTERRUPTIBILITY_INFO, guest_interruptibility_info),
835 FIELD(GUEST_ACTIVITY_STATE, guest_activity_state),
836 FIELD(GUEST_SYSENTER_CS, guest_sysenter_cs),
837 FIELD(HOST_IA32_SYSENTER_CS, host_ia32_sysenter_cs),
838 FIELD(VMX_PREEMPTION_TIMER_VALUE, vmx_preemption_timer_value),
839 FIELD(CR0_GUEST_HOST_MASK, cr0_guest_host_mask),
840 FIELD(CR4_GUEST_HOST_MASK, cr4_guest_host_mask),
841 FIELD(CR0_READ_SHADOW, cr0_read_shadow),
842 FIELD(CR4_READ_SHADOW, cr4_read_shadow),
843 FIELD(CR3_TARGET_VALUE0, cr3_target_value0),
844 FIELD(CR3_TARGET_VALUE1, cr3_target_value1),
845 FIELD(CR3_TARGET_VALUE2, cr3_target_value2),
846 FIELD(CR3_TARGET_VALUE3, cr3_target_value3),
847 FIELD(EXIT_QUALIFICATION, exit_qualification),
848 FIELD(GUEST_LINEAR_ADDRESS, guest_linear_address),
849 FIELD(GUEST_CR0, guest_cr0),
850 FIELD(GUEST_CR3, guest_cr3),
851 FIELD(GUEST_CR4, guest_cr4),
852 FIELD(GUEST_ES_BASE, guest_es_base),
853 FIELD(GUEST_CS_BASE, guest_cs_base),
854 FIELD(GUEST_SS_BASE, guest_ss_base),
855 FIELD(GUEST_DS_BASE, guest_ds_base),
856 FIELD(GUEST_FS_BASE, guest_fs_base),
857 FIELD(GUEST_GS_BASE, guest_gs_base),
858 FIELD(GUEST_LDTR_BASE, guest_ldtr_base),
859 FIELD(GUEST_TR_BASE, guest_tr_base),
860 FIELD(GUEST_GDTR_BASE, guest_gdtr_base),
861 FIELD(GUEST_IDTR_BASE, guest_idtr_base),
862 FIELD(GUEST_DR7, guest_dr7),
863 FIELD(GUEST_RSP, guest_rsp),
864 FIELD(GUEST_RIP, guest_rip),
865 FIELD(GUEST_RFLAGS, guest_rflags),
866 FIELD(GUEST_PENDING_DBG_EXCEPTIONS, guest_pending_dbg_exceptions),
867 FIELD(GUEST_SYSENTER_ESP, guest_sysenter_esp),
868 FIELD(GUEST_SYSENTER_EIP, guest_sysenter_eip),
869 FIELD(HOST_CR0, host_cr0),
870 FIELD(HOST_CR3, host_cr3),
871 FIELD(HOST_CR4, host_cr4),
872 FIELD(HOST_FS_BASE, host_fs_base),
873 FIELD(HOST_GS_BASE, host_gs_base),
874 FIELD(HOST_TR_BASE, host_tr_base),
875 FIELD(HOST_GDTR_BASE, host_gdtr_base),
876 FIELD(HOST_IDTR_BASE, host_idtr_base),
877 FIELD(HOST_IA32_SYSENTER_ESP, host_ia32_sysenter_esp),
878 FIELD(HOST_IA32_SYSENTER_EIP, host_ia32_sysenter_eip),
879 FIELD(HOST_RSP, host_rsp),
880 FIELD(HOST_RIP, host_rip),
883 static inline short vmcs_field_to_offset(unsigned long field)
885 BUILD_BUG_ON(ARRAY_SIZE(vmcs_field_to_offset_table) > SHRT_MAX);
887 if (field >= ARRAY_SIZE(vmcs_field_to_offset_table) ||
888 vmcs_field_to_offset_table[field] == 0)
891 return vmcs_field_to_offset_table[field];
894 static inline struct vmcs12 *get_vmcs12(struct kvm_vcpu *vcpu)
896 return to_vmx(vcpu)->nested.cached_vmcs12;
899 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu);
900 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu);
901 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa);
902 static bool vmx_xsaves_supported(void);
903 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr);
904 static void vmx_set_segment(struct kvm_vcpu *vcpu,
905 struct kvm_segment *var, int seg);
906 static void vmx_get_segment(struct kvm_vcpu *vcpu,
907 struct kvm_segment *var, int seg);
908 static bool guest_state_valid(struct kvm_vcpu *vcpu);
909 static u32 vmx_segment_access_rights(struct kvm_segment *var);
910 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx);
911 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx);
912 static int alloc_identity_pagetable(struct kvm *kvm);
913 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu);
914 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked);
915 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
918 static DEFINE_PER_CPU(struct vmcs *, vmxarea);
919 static DEFINE_PER_CPU(struct vmcs *, current_vmcs);
921 * We maintain a per-CPU linked-list of VMCS loaded on that CPU. This is needed
922 * when a CPU is brought down, and we need to VMCLEAR all VMCSs loaded on it.
924 static DEFINE_PER_CPU(struct list_head, loaded_vmcss_on_cpu);
927 * We maintian a per-CPU linked-list of vCPU, so in wakeup_handler() we
928 * can find which vCPU should be waken up.
930 static DEFINE_PER_CPU(struct list_head, blocked_vcpu_on_cpu);
931 static DEFINE_PER_CPU(spinlock_t, blocked_vcpu_on_cpu_lock);
936 VMX_MSR_BITMAP_LEGACY,
937 VMX_MSR_BITMAP_LONGMODE,
938 VMX_MSR_BITMAP_LEGACY_X2APIC_APICV,
939 VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV,
940 VMX_MSR_BITMAP_LEGACY_X2APIC,
941 VMX_MSR_BITMAP_LONGMODE_X2APIC,
947 static unsigned long *vmx_bitmap[VMX_BITMAP_NR];
949 #define vmx_io_bitmap_a (vmx_bitmap[VMX_IO_BITMAP_A])
950 #define vmx_io_bitmap_b (vmx_bitmap[VMX_IO_BITMAP_B])
951 #define vmx_msr_bitmap_legacy (vmx_bitmap[VMX_MSR_BITMAP_LEGACY])
952 #define vmx_msr_bitmap_longmode (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE])
953 #define vmx_msr_bitmap_legacy_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC_APICV])
954 #define vmx_msr_bitmap_longmode_x2apic_apicv (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC_APICV])
955 #define vmx_msr_bitmap_legacy_x2apic (vmx_bitmap[VMX_MSR_BITMAP_LEGACY_X2APIC])
956 #define vmx_msr_bitmap_longmode_x2apic (vmx_bitmap[VMX_MSR_BITMAP_LONGMODE_X2APIC])
957 #define vmx_vmread_bitmap (vmx_bitmap[VMX_VMREAD_BITMAP])
958 #define vmx_vmwrite_bitmap (vmx_bitmap[VMX_VMWRITE_BITMAP])
960 static bool cpu_has_load_ia32_efer;
961 static bool cpu_has_load_perf_global_ctrl;
963 static DECLARE_BITMAP(vmx_vpid_bitmap, VMX_NR_VPIDS);
964 static DEFINE_SPINLOCK(vmx_vpid_lock);
966 static struct vmcs_config {
971 u32 pin_based_exec_ctrl;
972 u32 cpu_based_exec_ctrl;
973 u32 cpu_based_2nd_exec_ctrl;
978 static struct vmx_capability {
983 #define VMX_SEGMENT_FIELD(seg) \
984 [VCPU_SREG_##seg] = { \
985 .selector = GUEST_##seg##_SELECTOR, \
986 .base = GUEST_##seg##_BASE, \
987 .limit = GUEST_##seg##_LIMIT, \
988 .ar_bytes = GUEST_##seg##_AR_BYTES, \
991 static const struct kvm_vmx_segment_field {
996 } kvm_vmx_segment_fields[] = {
997 VMX_SEGMENT_FIELD(CS),
998 VMX_SEGMENT_FIELD(DS),
999 VMX_SEGMENT_FIELD(ES),
1000 VMX_SEGMENT_FIELD(FS),
1001 VMX_SEGMENT_FIELD(GS),
1002 VMX_SEGMENT_FIELD(SS),
1003 VMX_SEGMENT_FIELD(TR),
1004 VMX_SEGMENT_FIELD(LDTR),
1007 static u64 host_efer;
1009 static void ept_save_pdptrs(struct kvm_vcpu *vcpu);
1012 * Keep MSR_STAR at the end, as setup_msrs() will try to optimize it
1013 * away by decrementing the array size.
1015 static const u32 vmx_msr_index[] = {
1016 #ifdef CONFIG_X86_64
1017 MSR_SYSCALL_MASK, MSR_LSTAR, MSR_CSTAR,
1019 MSR_EFER, MSR_TSC_AUX, MSR_STAR,
1022 static inline bool is_exception_n(u32 intr_info, u8 vector)
1024 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1025 INTR_INFO_VALID_MASK)) ==
1026 (INTR_TYPE_HARD_EXCEPTION | vector | INTR_INFO_VALID_MASK);
1029 static inline bool is_debug(u32 intr_info)
1031 return is_exception_n(intr_info, DB_VECTOR);
1034 static inline bool is_breakpoint(u32 intr_info)
1036 return is_exception_n(intr_info, BP_VECTOR);
1039 static inline bool is_page_fault(u32 intr_info)
1041 return is_exception_n(intr_info, PF_VECTOR);
1044 static inline bool is_no_device(u32 intr_info)
1046 return is_exception_n(intr_info, NM_VECTOR);
1049 static inline bool is_invalid_opcode(u32 intr_info)
1051 return is_exception_n(intr_info, UD_VECTOR);
1054 static inline bool is_external_interrupt(u32 intr_info)
1056 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1057 == (INTR_TYPE_EXT_INTR | INTR_INFO_VALID_MASK);
1060 static inline bool is_machine_check(u32 intr_info)
1062 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VECTOR_MASK |
1063 INTR_INFO_VALID_MASK)) ==
1064 (INTR_TYPE_HARD_EXCEPTION | MC_VECTOR | INTR_INFO_VALID_MASK);
1067 static inline bool cpu_has_vmx_msr_bitmap(void)
1069 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_USE_MSR_BITMAPS;
1072 static inline bool cpu_has_vmx_tpr_shadow(void)
1074 return vmcs_config.cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW;
1077 static inline bool cpu_need_tpr_shadow(struct kvm_vcpu *vcpu)
1079 return cpu_has_vmx_tpr_shadow() && lapic_in_kernel(vcpu);
1082 static inline bool cpu_has_secondary_exec_ctrls(void)
1084 return vmcs_config.cpu_based_exec_ctrl &
1085 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
1088 static inline bool cpu_has_vmx_virtualize_apic_accesses(void)
1090 return vmcs_config.cpu_based_2nd_exec_ctrl &
1091 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
1094 static inline bool cpu_has_vmx_virtualize_x2apic_mode(void)
1096 return vmcs_config.cpu_based_2nd_exec_ctrl &
1097 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
1100 static inline bool cpu_has_vmx_apic_register_virt(void)
1102 return vmcs_config.cpu_based_2nd_exec_ctrl &
1103 SECONDARY_EXEC_APIC_REGISTER_VIRT;
1106 static inline bool cpu_has_vmx_virtual_intr_delivery(void)
1108 return vmcs_config.cpu_based_2nd_exec_ctrl &
1109 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY;
1113 * Comment's format: document - errata name - stepping - processor name.
1115 * https://www.virtualbox.org/svn/vbox/trunk/src/VBox/VMM/VMMR0/HMR0.cpp
1117 static u32 vmx_preemption_cpu_tfms[] = {
1118 /* 323344.pdf - BA86 - D0 - Xeon 7500 Series */
1120 /* 323056.pdf - AAX65 - C2 - Xeon L3406 */
1121 /* 322814.pdf - AAT59 - C2 - i7-600, i5-500, i5-400 and i3-300 Mobile */
1122 /* 322911.pdf - AAU65 - C2 - i5-600, i3-500 Desktop and Pentium G6950 */
1124 /* 322911.pdf - AAU65 - K0 - i5-600, i3-500 Desktop and Pentium G6950 */
1126 /* 322373.pdf - AAO95 - B1 - Xeon 3400 Series */
1127 /* 322166.pdf - AAN92 - B1 - i7-800 and i5-700 Desktop */
1129 * 320767.pdf - AAP86 - B1 -
1130 * i7-900 Mobile Extreme, i7-800 and i7-700 Mobile
1133 /* 321333.pdf - AAM126 - C0 - Xeon 3500 */
1135 /* 321333.pdf - AAM126 - C1 - Xeon 3500 */
1137 /* 320836.pdf - AAJ124 - C0 - i7-900 Desktop Extreme and i7-900 Desktop */
1139 /* 321333.pdf - AAM126 - D0 - Xeon 3500 */
1140 /* 321324.pdf - AAK139 - D0 - Xeon 5500 */
1141 /* 320836.pdf - AAJ124 - D0 - i7-900 Extreme and i7-900 Desktop */
1145 static inline bool cpu_has_broken_vmx_preemption_timer(void)
1147 u32 eax = cpuid_eax(0x00000001), i;
1149 /* Clear the reserved bits */
1150 eax &= ~(0x3U << 14 | 0xfU << 28);
1151 for (i = 0; i < ARRAY_SIZE(vmx_preemption_cpu_tfms); i++)
1152 if (eax == vmx_preemption_cpu_tfms[i])
1158 static inline bool cpu_has_vmx_preemption_timer(void)
1160 return vmcs_config.pin_based_exec_ctrl &
1161 PIN_BASED_VMX_PREEMPTION_TIMER;
1164 static inline bool cpu_has_vmx_posted_intr(void)
1166 return IS_ENABLED(CONFIG_X86_LOCAL_APIC) &&
1167 vmcs_config.pin_based_exec_ctrl & PIN_BASED_POSTED_INTR;
1170 static inline bool cpu_has_vmx_apicv(void)
1172 return cpu_has_vmx_apic_register_virt() &&
1173 cpu_has_vmx_virtual_intr_delivery() &&
1174 cpu_has_vmx_posted_intr();
1177 static inline bool cpu_has_vmx_flexpriority(void)
1179 return cpu_has_vmx_tpr_shadow() &&
1180 cpu_has_vmx_virtualize_apic_accesses();
1183 static inline bool cpu_has_vmx_ept_execute_only(void)
1185 return vmx_capability.ept & VMX_EPT_EXECUTE_ONLY_BIT;
1188 static inline bool cpu_has_vmx_ept_2m_page(void)
1190 return vmx_capability.ept & VMX_EPT_2MB_PAGE_BIT;
1193 static inline bool cpu_has_vmx_ept_1g_page(void)
1195 return vmx_capability.ept & VMX_EPT_1GB_PAGE_BIT;
1198 static inline bool cpu_has_vmx_ept_4levels(void)
1200 return vmx_capability.ept & VMX_EPT_PAGE_WALK_4_BIT;
1203 static inline bool cpu_has_vmx_ept_mt_wb(void)
1205 return vmx_capability.ept & VMX_EPTP_WB_BIT;
1208 static inline bool cpu_has_vmx_ept_5levels(void)
1210 return vmx_capability.ept & VMX_EPT_PAGE_WALK_5_BIT;
1213 static inline bool cpu_has_vmx_ept_ad_bits(void)
1215 return vmx_capability.ept & VMX_EPT_AD_BIT;
1218 static inline bool cpu_has_vmx_invept_context(void)
1220 return vmx_capability.ept & VMX_EPT_EXTENT_CONTEXT_BIT;
1223 static inline bool cpu_has_vmx_invept_global(void)
1225 return vmx_capability.ept & VMX_EPT_EXTENT_GLOBAL_BIT;
1228 static inline bool cpu_has_vmx_invvpid_single(void)
1230 return vmx_capability.vpid & VMX_VPID_EXTENT_SINGLE_CONTEXT_BIT;
1233 static inline bool cpu_has_vmx_invvpid_global(void)
1235 return vmx_capability.vpid & VMX_VPID_EXTENT_GLOBAL_CONTEXT_BIT;
1238 static inline bool cpu_has_vmx_invvpid(void)
1240 return vmx_capability.vpid & VMX_VPID_INVVPID_BIT;
1243 static inline bool cpu_has_vmx_ept(void)
1245 return vmcs_config.cpu_based_2nd_exec_ctrl &
1246 SECONDARY_EXEC_ENABLE_EPT;
1249 static inline bool cpu_has_vmx_unrestricted_guest(void)
1251 return vmcs_config.cpu_based_2nd_exec_ctrl &
1252 SECONDARY_EXEC_UNRESTRICTED_GUEST;
1255 static inline bool cpu_has_vmx_ple(void)
1257 return vmcs_config.cpu_based_2nd_exec_ctrl &
1258 SECONDARY_EXEC_PAUSE_LOOP_EXITING;
1261 static inline bool cpu_has_vmx_basic_inout(void)
1263 return (((u64)vmcs_config.basic_cap << 32) & VMX_BASIC_INOUT);
1266 static inline bool cpu_need_virtualize_apic_accesses(struct kvm_vcpu *vcpu)
1268 return flexpriority_enabled && lapic_in_kernel(vcpu);
1271 static inline bool cpu_has_vmx_vpid(void)
1273 return vmcs_config.cpu_based_2nd_exec_ctrl &
1274 SECONDARY_EXEC_ENABLE_VPID;
1277 static inline bool cpu_has_vmx_rdtscp(void)
1279 return vmcs_config.cpu_based_2nd_exec_ctrl &
1280 SECONDARY_EXEC_RDTSCP;
1283 static inline bool cpu_has_vmx_invpcid(void)
1285 return vmcs_config.cpu_based_2nd_exec_ctrl &
1286 SECONDARY_EXEC_ENABLE_INVPCID;
1289 static inline bool cpu_has_vmx_wbinvd_exit(void)
1291 return vmcs_config.cpu_based_2nd_exec_ctrl &
1292 SECONDARY_EXEC_WBINVD_EXITING;
1295 static inline bool cpu_has_vmx_shadow_vmcs(void)
1298 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
1299 /* check if the cpu supports writing r/o exit information fields */
1300 if (!(vmx_msr & MSR_IA32_VMX_MISC_VMWRITE_SHADOW_RO_FIELDS))
1303 return vmcs_config.cpu_based_2nd_exec_ctrl &
1304 SECONDARY_EXEC_SHADOW_VMCS;
1307 static inline bool cpu_has_vmx_pml(void)
1309 return vmcs_config.cpu_based_2nd_exec_ctrl & SECONDARY_EXEC_ENABLE_PML;
1312 static inline bool cpu_has_vmx_tsc_scaling(void)
1314 return vmcs_config.cpu_based_2nd_exec_ctrl &
1315 SECONDARY_EXEC_TSC_SCALING;
1318 static inline bool cpu_has_vmx_vmfunc(void)
1320 return vmcs_config.cpu_based_2nd_exec_ctrl &
1321 SECONDARY_EXEC_ENABLE_VMFUNC;
1324 static inline bool report_flexpriority(void)
1326 return flexpriority_enabled;
1329 static inline unsigned nested_cpu_vmx_misc_cr3_count(struct kvm_vcpu *vcpu)
1331 return vmx_misc_cr3_count(to_vmx(vcpu)->nested.nested_vmx_misc_low);
1334 static inline bool nested_cpu_has(struct vmcs12 *vmcs12, u32 bit)
1336 return vmcs12->cpu_based_vm_exec_control & bit;
1339 static inline bool nested_cpu_has2(struct vmcs12 *vmcs12, u32 bit)
1341 return (vmcs12->cpu_based_vm_exec_control &
1342 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
1343 (vmcs12->secondary_vm_exec_control & bit);
1346 static inline bool nested_cpu_has_virtual_nmis(struct vmcs12 *vmcs12)
1348 return vmcs12->pin_based_vm_exec_control & PIN_BASED_VIRTUAL_NMIS;
1351 static inline bool nested_cpu_has_preemption_timer(struct vmcs12 *vmcs12)
1353 return vmcs12->pin_based_vm_exec_control &
1354 PIN_BASED_VMX_PREEMPTION_TIMER;
1357 static inline int nested_cpu_has_ept(struct vmcs12 *vmcs12)
1359 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_EPT);
1362 static inline bool nested_cpu_has_xsaves(struct vmcs12 *vmcs12)
1364 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
1367 static inline bool nested_cpu_has_pml(struct vmcs12 *vmcs12)
1369 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML);
1372 static inline bool nested_cpu_has_virt_x2apic_mode(struct vmcs12 *vmcs12)
1374 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE);
1377 static inline bool nested_cpu_has_vpid(struct vmcs12 *vmcs12)
1379 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VPID);
1382 static inline bool nested_cpu_has_apic_reg_virt(struct vmcs12 *vmcs12)
1384 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_APIC_REGISTER_VIRT);
1387 static inline bool nested_cpu_has_vid(struct vmcs12 *vmcs12)
1389 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
1392 static inline bool nested_cpu_has_posted_intr(struct vmcs12 *vmcs12)
1394 return vmcs12->pin_based_vm_exec_control & PIN_BASED_POSTED_INTR;
1397 static inline bool nested_cpu_has_vmfunc(struct vmcs12 *vmcs12)
1399 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_VMFUNC);
1402 static inline bool nested_cpu_has_eptp_switching(struct vmcs12 *vmcs12)
1404 return nested_cpu_has_vmfunc(vmcs12) &&
1405 (vmcs12->vm_function_control &
1406 VMX_VMFUNC_EPTP_SWITCHING);
1409 static inline bool is_nmi(u32 intr_info)
1411 return (intr_info & (INTR_INFO_INTR_TYPE_MASK | INTR_INFO_VALID_MASK))
1412 == (INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK);
1415 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
1417 unsigned long exit_qualification);
1418 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
1419 struct vmcs12 *vmcs12,
1420 u32 reason, unsigned long qualification);
1422 static int __find_msr_index(struct vcpu_vmx *vmx, u32 msr)
1426 for (i = 0; i < vmx->nmsrs; ++i)
1427 if (vmx_msr_index[vmx->guest_msrs[i].index] == msr)
1432 static inline void __invvpid(int ext, u16 vpid, gva_t gva)
1438 } operand = { vpid, 0, gva };
1440 asm volatile (__ex(ASM_VMX_INVVPID)
1441 /* CF==1 or ZF==1 --> rc = -1 */
1442 "; ja 1f ; ud2 ; 1:"
1443 : : "a"(&operand), "c"(ext) : "cc", "memory");
1446 static inline void __invept(int ext, u64 eptp, gpa_t gpa)
1450 } operand = {eptp, gpa};
1452 asm volatile (__ex(ASM_VMX_INVEPT)
1453 /* CF==1 or ZF==1 --> rc = -1 */
1454 "; ja 1f ; ud2 ; 1:\n"
1455 : : "a" (&operand), "c" (ext) : "cc", "memory");
1458 static struct shared_msr_entry *find_msr_entry(struct vcpu_vmx *vmx, u32 msr)
1462 i = __find_msr_index(vmx, msr);
1464 return &vmx->guest_msrs[i];
1468 static void vmcs_clear(struct vmcs *vmcs)
1470 u64 phys_addr = __pa(vmcs);
1473 asm volatile (__ex(ASM_VMX_VMCLEAR_RAX) "; setna %0"
1474 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1477 printk(KERN_ERR "kvm: vmclear fail: %p/%llx\n",
1481 static inline void loaded_vmcs_init(struct loaded_vmcs *loaded_vmcs)
1483 vmcs_clear(loaded_vmcs->vmcs);
1484 if (loaded_vmcs->shadow_vmcs && loaded_vmcs->launched)
1485 vmcs_clear(loaded_vmcs->shadow_vmcs);
1486 loaded_vmcs->cpu = -1;
1487 loaded_vmcs->launched = 0;
1490 static void vmcs_load(struct vmcs *vmcs)
1492 u64 phys_addr = __pa(vmcs);
1495 asm volatile (__ex(ASM_VMX_VMPTRLD_RAX) "; setna %0"
1496 : "=qm"(error) : "a"(&phys_addr), "m"(phys_addr)
1499 printk(KERN_ERR "kvm: vmptrld %p/%llx failed\n",
1503 #ifdef CONFIG_KEXEC_CORE
1505 * This bitmap is used to indicate whether the vmclear
1506 * operation is enabled on all cpus. All disabled by
1509 static cpumask_t crash_vmclear_enabled_bitmap = CPU_MASK_NONE;
1511 static inline void crash_enable_local_vmclear(int cpu)
1513 cpumask_set_cpu(cpu, &crash_vmclear_enabled_bitmap);
1516 static inline void crash_disable_local_vmclear(int cpu)
1518 cpumask_clear_cpu(cpu, &crash_vmclear_enabled_bitmap);
1521 static inline int crash_local_vmclear_enabled(int cpu)
1523 return cpumask_test_cpu(cpu, &crash_vmclear_enabled_bitmap);
1526 static void crash_vmclear_local_loaded_vmcss(void)
1528 int cpu = raw_smp_processor_id();
1529 struct loaded_vmcs *v;
1531 if (!crash_local_vmclear_enabled(cpu))
1534 list_for_each_entry(v, &per_cpu(loaded_vmcss_on_cpu, cpu),
1535 loaded_vmcss_on_cpu_link)
1536 vmcs_clear(v->vmcs);
1539 static inline void crash_enable_local_vmclear(int cpu) { }
1540 static inline void crash_disable_local_vmclear(int cpu) { }
1541 #endif /* CONFIG_KEXEC_CORE */
1543 static void __loaded_vmcs_clear(void *arg)
1545 struct loaded_vmcs *loaded_vmcs = arg;
1546 int cpu = raw_smp_processor_id();
1548 if (loaded_vmcs->cpu != cpu)
1549 return; /* vcpu migration can race with cpu offline */
1550 if (per_cpu(current_vmcs, cpu) == loaded_vmcs->vmcs)
1551 per_cpu(current_vmcs, cpu) = NULL;
1552 crash_disable_local_vmclear(cpu);
1553 list_del(&loaded_vmcs->loaded_vmcss_on_cpu_link);
1556 * we should ensure updating loaded_vmcs->loaded_vmcss_on_cpu_link
1557 * is before setting loaded_vmcs->vcpu to -1 which is done in
1558 * loaded_vmcs_init. Otherwise, other cpu can see vcpu = -1 fist
1559 * then adds the vmcs into percpu list before it is deleted.
1563 loaded_vmcs_init(loaded_vmcs);
1564 crash_enable_local_vmclear(cpu);
1567 static void loaded_vmcs_clear(struct loaded_vmcs *loaded_vmcs)
1569 int cpu = loaded_vmcs->cpu;
1572 smp_call_function_single(cpu,
1573 __loaded_vmcs_clear, loaded_vmcs, 1);
1576 static inline void vpid_sync_vcpu_single(int vpid)
1581 if (cpu_has_vmx_invvpid_single())
1582 __invvpid(VMX_VPID_EXTENT_SINGLE_CONTEXT, vpid, 0);
1585 static inline void vpid_sync_vcpu_global(void)
1587 if (cpu_has_vmx_invvpid_global())
1588 __invvpid(VMX_VPID_EXTENT_ALL_CONTEXT, 0, 0);
1591 static inline void vpid_sync_context(int vpid)
1593 if (cpu_has_vmx_invvpid_single())
1594 vpid_sync_vcpu_single(vpid);
1596 vpid_sync_vcpu_global();
1599 static inline void ept_sync_global(void)
1601 if (cpu_has_vmx_invept_global())
1602 __invept(VMX_EPT_EXTENT_GLOBAL, 0, 0);
1605 static inline void ept_sync_context(u64 eptp)
1607 if (cpu_has_vmx_invept_context())
1608 __invept(VMX_EPT_EXTENT_CONTEXT, eptp, 0);
1613 static __always_inline void vmcs_check16(unsigned long field)
1615 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1616 "16-bit accessor invalid for 64-bit field");
1617 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1618 "16-bit accessor invalid for 64-bit high field");
1619 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1620 "16-bit accessor invalid for 32-bit high field");
1621 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1622 "16-bit accessor invalid for natural width field");
1625 static __always_inline void vmcs_check32(unsigned long field)
1627 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1628 "32-bit accessor invalid for 16-bit field");
1629 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1630 "32-bit accessor invalid for natural width field");
1633 static __always_inline void vmcs_check64(unsigned long field)
1635 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1636 "64-bit accessor invalid for 16-bit field");
1637 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1638 "64-bit accessor invalid for 64-bit high field");
1639 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1640 "64-bit accessor invalid for 32-bit field");
1641 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x6000,
1642 "64-bit accessor invalid for natural width field");
1645 static __always_inline void vmcs_checkl(unsigned long field)
1647 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0,
1648 "Natural width accessor invalid for 16-bit field");
1649 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2000,
1650 "Natural width accessor invalid for 64-bit field");
1651 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6001) == 0x2001,
1652 "Natural width accessor invalid for 64-bit high field");
1653 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x4000,
1654 "Natural width accessor invalid for 32-bit field");
1657 static __always_inline unsigned long __vmcs_readl(unsigned long field)
1659 unsigned long value;
1661 asm volatile (__ex_clear(ASM_VMX_VMREAD_RDX_RAX, "%0")
1662 : "=a"(value) : "d"(field) : "cc");
1666 static __always_inline u16 vmcs_read16(unsigned long field)
1668 vmcs_check16(field);
1669 return __vmcs_readl(field);
1672 static __always_inline u32 vmcs_read32(unsigned long field)
1674 vmcs_check32(field);
1675 return __vmcs_readl(field);
1678 static __always_inline u64 vmcs_read64(unsigned long field)
1680 vmcs_check64(field);
1681 #ifdef CONFIG_X86_64
1682 return __vmcs_readl(field);
1684 return __vmcs_readl(field) | ((u64)__vmcs_readl(field+1) << 32);
1688 static __always_inline unsigned long vmcs_readl(unsigned long field)
1691 return __vmcs_readl(field);
1694 static noinline void vmwrite_error(unsigned long field, unsigned long value)
1696 printk(KERN_ERR "vmwrite error: reg %lx value %lx (err %d)\n",
1697 field, value, vmcs_read32(VM_INSTRUCTION_ERROR));
1701 static __always_inline void __vmcs_writel(unsigned long field, unsigned long value)
1705 asm volatile (__ex(ASM_VMX_VMWRITE_RAX_RDX) "; setna %0"
1706 : "=q"(error) : "a"(value), "d"(field) : "cc");
1707 if (unlikely(error))
1708 vmwrite_error(field, value);
1711 static __always_inline void vmcs_write16(unsigned long field, u16 value)
1713 vmcs_check16(field);
1714 __vmcs_writel(field, value);
1717 static __always_inline void vmcs_write32(unsigned long field, u32 value)
1719 vmcs_check32(field);
1720 __vmcs_writel(field, value);
1723 static __always_inline void vmcs_write64(unsigned long field, u64 value)
1725 vmcs_check64(field);
1726 __vmcs_writel(field, value);
1727 #ifndef CONFIG_X86_64
1729 __vmcs_writel(field+1, value >> 32);
1733 static __always_inline void vmcs_writel(unsigned long field, unsigned long value)
1736 __vmcs_writel(field, value);
1739 static __always_inline void vmcs_clear_bits(unsigned long field, u32 mask)
1741 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1742 "vmcs_clear_bits does not support 64-bit fields");
1743 __vmcs_writel(field, __vmcs_readl(field) & ~mask);
1746 static __always_inline void vmcs_set_bits(unsigned long field, u32 mask)
1748 BUILD_BUG_ON_MSG(__builtin_constant_p(field) && ((field) & 0x6000) == 0x2000,
1749 "vmcs_set_bits does not support 64-bit fields");
1750 __vmcs_writel(field, __vmcs_readl(field) | mask);
1753 static inline void vm_entry_controls_reset_shadow(struct vcpu_vmx *vmx)
1755 vmx->vm_entry_controls_shadow = vmcs_read32(VM_ENTRY_CONTROLS);
1758 static inline void vm_entry_controls_init(struct vcpu_vmx *vmx, u32 val)
1760 vmcs_write32(VM_ENTRY_CONTROLS, val);
1761 vmx->vm_entry_controls_shadow = val;
1764 static inline void vm_entry_controls_set(struct vcpu_vmx *vmx, u32 val)
1766 if (vmx->vm_entry_controls_shadow != val)
1767 vm_entry_controls_init(vmx, val);
1770 static inline u32 vm_entry_controls_get(struct vcpu_vmx *vmx)
1772 return vmx->vm_entry_controls_shadow;
1776 static inline void vm_entry_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1778 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) | val);
1781 static inline void vm_entry_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1783 vm_entry_controls_set(vmx, vm_entry_controls_get(vmx) & ~val);
1786 static inline void vm_exit_controls_reset_shadow(struct vcpu_vmx *vmx)
1788 vmx->vm_exit_controls_shadow = vmcs_read32(VM_EXIT_CONTROLS);
1791 static inline void vm_exit_controls_init(struct vcpu_vmx *vmx, u32 val)
1793 vmcs_write32(VM_EXIT_CONTROLS, val);
1794 vmx->vm_exit_controls_shadow = val;
1797 static inline void vm_exit_controls_set(struct vcpu_vmx *vmx, u32 val)
1799 if (vmx->vm_exit_controls_shadow != val)
1800 vm_exit_controls_init(vmx, val);
1803 static inline u32 vm_exit_controls_get(struct vcpu_vmx *vmx)
1805 return vmx->vm_exit_controls_shadow;
1809 static inline void vm_exit_controls_setbit(struct vcpu_vmx *vmx, u32 val)
1811 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) | val);
1814 static inline void vm_exit_controls_clearbit(struct vcpu_vmx *vmx, u32 val)
1816 vm_exit_controls_set(vmx, vm_exit_controls_get(vmx) & ~val);
1819 static void vmx_segment_cache_clear(struct vcpu_vmx *vmx)
1821 vmx->segment_cache.bitmask = 0;
1824 static bool vmx_segment_cache_test_set(struct vcpu_vmx *vmx, unsigned seg,
1828 u32 mask = 1 << (seg * SEG_FIELD_NR + field);
1830 if (!(vmx->vcpu.arch.regs_avail & (1 << VCPU_EXREG_SEGMENTS))) {
1831 vmx->vcpu.arch.regs_avail |= (1 << VCPU_EXREG_SEGMENTS);
1832 vmx->segment_cache.bitmask = 0;
1834 ret = vmx->segment_cache.bitmask & mask;
1835 vmx->segment_cache.bitmask |= mask;
1839 static u16 vmx_read_guest_seg_selector(struct vcpu_vmx *vmx, unsigned seg)
1841 u16 *p = &vmx->segment_cache.seg[seg].selector;
1843 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_SEL))
1844 *p = vmcs_read16(kvm_vmx_segment_fields[seg].selector);
1848 static ulong vmx_read_guest_seg_base(struct vcpu_vmx *vmx, unsigned seg)
1850 ulong *p = &vmx->segment_cache.seg[seg].base;
1852 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_BASE))
1853 *p = vmcs_readl(kvm_vmx_segment_fields[seg].base);
1857 static u32 vmx_read_guest_seg_limit(struct vcpu_vmx *vmx, unsigned seg)
1859 u32 *p = &vmx->segment_cache.seg[seg].limit;
1861 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_LIMIT))
1862 *p = vmcs_read32(kvm_vmx_segment_fields[seg].limit);
1866 static u32 vmx_read_guest_seg_ar(struct vcpu_vmx *vmx, unsigned seg)
1868 u32 *p = &vmx->segment_cache.seg[seg].ar;
1870 if (!vmx_segment_cache_test_set(vmx, seg, SEG_FIELD_AR))
1871 *p = vmcs_read32(kvm_vmx_segment_fields[seg].ar_bytes);
1875 static void update_exception_bitmap(struct kvm_vcpu *vcpu)
1879 eb = (1u << PF_VECTOR) | (1u << UD_VECTOR) | (1u << MC_VECTOR) |
1880 (1u << DB_VECTOR) | (1u << AC_VECTOR);
1881 if ((vcpu->guest_debug &
1882 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP)) ==
1883 (KVM_GUESTDBG_ENABLE | KVM_GUESTDBG_USE_SW_BP))
1884 eb |= 1u << BP_VECTOR;
1885 if (to_vmx(vcpu)->rmode.vm86_active)
1888 eb &= ~(1u << PF_VECTOR); /* bypass_guest_pf = 0 */
1890 /* When we are running a nested L2 guest and L1 specified for it a
1891 * certain exception bitmap, we must trap the same exceptions and pass
1892 * them to L1. When running L2, we will only handle the exceptions
1893 * specified above if L1 did not want them.
1895 if (is_guest_mode(vcpu))
1896 eb |= get_vmcs12(vcpu)->exception_bitmap;
1898 vmcs_write32(EXCEPTION_BITMAP, eb);
1901 static void clear_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1902 unsigned long entry, unsigned long exit)
1904 vm_entry_controls_clearbit(vmx, entry);
1905 vm_exit_controls_clearbit(vmx, exit);
1908 static void clear_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr)
1911 struct msr_autoload *m = &vmx->msr_autoload;
1915 if (cpu_has_load_ia32_efer) {
1916 clear_atomic_switch_msr_special(vmx,
1917 VM_ENTRY_LOAD_IA32_EFER,
1918 VM_EXIT_LOAD_IA32_EFER);
1922 case MSR_CORE_PERF_GLOBAL_CTRL:
1923 if (cpu_has_load_perf_global_ctrl) {
1924 clear_atomic_switch_msr_special(vmx,
1925 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1926 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
1932 for (i = 0; i < m->nr; ++i)
1933 if (m->guest[i].index == msr)
1939 m->guest[i] = m->guest[m->nr];
1940 m->host[i] = m->host[m->nr];
1941 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
1942 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
1945 static void add_atomic_switch_msr_special(struct vcpu_vmx *vmx,
1946 unsigned long entry, unsigned long exit,
1947 unsigned long guest_val_vmcs, unsigned long host_val_vmcs,
1948 u64 guest_val, u64 host_val)
1950 vmcs_write64(guest_val_vmcs, guest_val);
1951 vmcs_write64(host_val_vmcs, host_val);
1952 vm_entry_controls_setbit(vmx, entry);
1953 vm_exit_controls_setbit(vmx, exit);
1956 static void add_atomic_switch_msr(struct vcpu_vmx *vmx, unsigned msr,
1957 u64 guest_val, u64 host_val)
1960 struct msr_autoload *m = &vmx->msr_autoload;
1964 if (cpu_has_load_ia32_efer) {
1965 add_atomic_switch_msr_special(vmx,
1966 VM_ENTRY_LOAD_IA32_EFER,
1967 VM_EXIT_LOAD_IA32_EFER,
1970 guest_val, host_val);
1974 case MSR_CORE_PERF_GLOBAL_CTRL:
1975 if (cpu_has_load_perf_global_ctrl) {
1976 add_atomic_switch_msr_special(vmx,
1977 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL,
1978 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL,
1979 GUEST_IA32_PERF_GLOBAL_CTRL,
1980 HOST_IA32_PERF_GLOBAL_CTRL,
1981 guest_val, host_val);
1985 case MSR_IA32_PEBS_ENABLE:
1986 /* PEBS needs a quiescent period after being disabled (to write
1987 * a record). Disabling PEBS through VMX MSR swapping doesn't
1988 * provide that period, so a CPU could write host's record into
1991 wrmsrl(MSR_IA32_PEBS_ENABLE, 0);
1994 for (i = 0; i < m->nr; ++i)
1995 if (m->guest[i].index == msr)
1998 if (i == NR_AUTOLOAD_MSRS) {
1999 printk_once(KERN_WARNING "Not enough msr switch entries. "
2000 "Can't add msr %x\n", msr);
2002 } else if (i == m->nr) {
2004 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, m->nr);
2005 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, m->nr);
2008 m->guest[i].index = msr;
2009 m->guest[i].value = guest_val;
2010 m->host[i].index = msr;
2011 m->host[i].value = host_val;
2014 static bool update_transition_efer(struct vcpu_vmx *vmx, int efer_offset)
2016 u64 guest_efer = vmx->vcpu.arch.efer;
2017 u64 ignore_bits = 0;
2021 * NX is needed to handle CR0.WP=1, CR4.SMEP=1. Testing
2022 * host CPUID is more efficient than testing guest CPUID
2023 * or CR4. Host SMEP is anyway a requirement for guest SMEP.
2025 if (boot_cpu_has(X86_FEATURE_SMEP))
2026 guest_efer |= EFER_NX;
2027 else if (!(guest_efer & EFER_NX))
2028 ignore_bits |= EFER_NX;
2032 * LMA and LME handled by hardware; SCE meaningless outside long mode.
2034 ignore_bits |= EFER_SCE;
2035 #ifdef CONFIG_X86_64
2036 ignore_bits |= EFER_LMA | EFER_LME;
2037 /* SCE is meaningful only in long mode on Intel */
2038 if (guest_efer & EFER_LMA)
2039 ignore_bits &= ~(u64)EFER_SCE;
2042 clear_atomic_switch_msr(vmx, MSR_EFER);
2045 * On EPT, we can't emulate NX, so we must switch EFER atomically.
2046 * On CPUs that support "load IA32_EFER", always switch EFER
2047 * atomically, since it's faster than switching it manually.
2049 if (cpu_has_load_ia32_efer ||
2050 (enable_ept && ((vmx->vcpu.arch.efer ^ host_efer) & EFER_NX))) {
2051 if (!(guest_efer & EFER_LMA))
2052 guest_efer &= ~EFER_LME;
2053 if (guest_efer != host_efer)
2054 add_atomic_switch_msr(vmx, MSR_EFER,
2055 guest_efer, host_efer);
2058 guest_efer &= ~ignore_bits;
2059 guest_efer |= host_efer & ignore_bits;
2061 vmx->guest_msrs[efer_offset].data = guest_efer;
2062 vmx->guest_msrs[efer_offset].mask = ~ignore_bits;
2068 #ifdef CONFIG_X86_32
2070 * On 32-bit kernels, VM exits still load the FS and GS bases from the
2071 * VMCS rather than the segment table. KVM uses this helper to figure
2072 * out the current bases to poke them into the VMCS before entry.
2074 static unsigned long segment_base(u16 selector)
2076 struct desc_struct *table;
2079 if (!(selector & ~SEGMENT_RPL_MASK))
2082 table = get_current_gdt_ro();
2084 if ((selector & SEGMENT_TI_MASK) == SEGMENT_LDT) {
2085 u16 ldt_selector = kvm_read_ldt();
2087 if (!(ldt_selector & ~SEGMENT_RPL_MASK))
2090 table = (struct desc_struct *)segment_base(ldt_selector);
2092 v = get_desc_base(&table[selector >> 3]);
2097 static void vmx_save_host_state(struct kvm_vcpu *vcpu)
2099 struct vcpu_vmx *vmx = to_vmx(vcpu);
2102 if (vmx->host_state.loaded)
2105 vmx->host_state.loaded = 1;
2107 * Set host fs and gs selectors. Unfortunately, 22.2.3 does not
2108 * allow segment selectors with cpl > 0 or ti == 1.
2110 vmx->host_state.ldt_sel = kvm_read_ldt();
2111 vmx->host_state.gs_ldt_reload_needed = vmx->host_state.ldt_sel;
2112 savesegment(fs, vmx->host_state.fs_sel);
2113 if (!(vmx->host_state.fs_sel & 7)) {
2114 vmcs_write16(HOST_FS_SELECTOR, vmx->host_state.fs_sel);
2115 vmx->host_state.fs_reload_needed = 0;
2117 vmcs_write16(HOST_FS_SELECTOR, 0);
2118 vmx->host_state.fs_reload_needed = 1;
2120 savesegment(gs, vmx->host_state.gs_sel);
2121 if (!(vmx->host_state.gs_sel & 7))
2122 vmcs_write16(HOST_GS_SELECTOR, vmx->host_state.gs_sel);
2124 vmcs_write16(HOST_GS_SELECTOR, 0);
2125 vmx->host_state.gs_ldt_reload_needed = 1;
2128 #ifdef CONFIG_X86_64
2129 savesegment(ds, vmx->host_state.ds_sel);
2130 savesegment(es, vmx->host_state.es_sel);
2133 #ifdef CONFIG_X86_64
2134 vmcs_writel(HOST_FS_BASE, read_msr(MSR_FS_BASE));
2135 vmcs_writel(HOST_GS_BASE, read_msr(MSR_GS_BASE));
2137 vmcs_writel(HOST_FS_BASE, segment_base(vmx->host_state.fs_sel));
2138 vmcs_writel(HOST_GS_BASE, segment_base(vmx->host_state.gs_sel));
2141 #ifdef CONFIG_X86_64
2142 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2143 if (is_long_mode(&vmx->vcpu))
2144 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2146 if (boot_cpu_has(X86_FEATURE_MPX))
2147 rdmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2148 for (i = 0; i < vmx->save_nmsrs; ++i)
2149 kvm_set_shared_msr(vmx->guest_msrs[i].index,
2150 vmx->guest_msrs[i].data,
2151 vmx->guest_msrs[i].mask);
2154 static void __vmx_load_host_state(struct vcpu_vmx *vmx)
2156 if (!vmx->host_state.loaded)
2159 ++vmx->vcpu.stat.host_state_reload;
2160 vmx->host_state.loaded = 0;
2161 #ifdef CONFIG_X86_64
2162 if (is_long_mode(&vmx->vcpu))
2163 rdmsrl(MSR_KERNEL_GS_BASE, vmx->msr_guest_kernel_gs_base);
2165 if (vmx->host_state.gs_ldt_reload_needed) {
2166 kvm_load_ldt(vmx->host_state.ldt_sel);
2167 #ifdef CONFIG_X86_64
2168 load_gs_index(vmx->host_state.gs_sel);
2170 loadsegment(gs, vmx->host_state.gs_sel);
2173 if (vmx->host_state.fs_reload_needed)
2174 loadsegment(fs, vmx->host_state.fs_sel);
2175 #ifdef CONFIG_X86_64
2176 if (unlikely(vmx->host_state.ds_sel | vmx->host_state.es_sel)) {
2177 loadsegment(ds, vmx->host_state.ds_sel);
2178 loadsegment(es, vmx->host_state.es_sel);
2181 invalidate_tss_limit();
2182 #ifdef CONFIG_X86_64
2183 wrmsrl(MSR_KERNEL_GS_BASE, vmx->msr_host_kernel_gs_base);
2185 if (vmx->host_state.msr_host_bndcfgs)
2186 wrmsrl(MSR_IA32_BNDCFGS, vmx->host_state.msr_host_bndcfgs);
2187 load_fixmap_gdt(raw_smp_processor_id());
2190 static void vmx_load_host_state(struct vcpu_vmx *vmx)
2193 __vmx_load_host_state(vmx);
2197 static void vmx_vcpu_pi_load(struct kvm_vcpu *vcpu, int cpu)
2199 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2200 struct pi_desc old, new;
2204 * In case of hot-plug or hot-unplug, we may have to undo
2205 * vmx_vcpu_pi_put even if there is no assigned device. And we
2206 * always keep PI.NDST up to date for simplicity: it makes the
2207 * code easier, and CPU migration is not a fast path.
2209 if (!pi_test_sn(pi_desc) && vcpu->cpu == cpu)
2213 * First handle the simple case where no cmpxchg is necessary; just
2214 * allow posting non-urgent interrupts.
2216 * If the 'nv' field is POSTED_INTR_WAKEUP_VECTOR, do not change
2217 * PI.NDST: pi_post_block will do it for us and the wakeup_handler
2218 * expects the VCPU to be on the blocked_vcpu_list that matches
2221 if (pi_desc->nv == POSTED_INTR_WAKEUP_VECTOR ||
2223 pi_clear_sn(pi_desc);
2227 /* The full case. */
2229 old.control = new.control = pi_desc->control;
2231 dest = cpu_physical_id(cpu);
2233 if (x2apic_enabled())
2236 new.ndst = (dest << 8) & 0xFF00;
2239 } while (cmpxchg64(&pi_desc->control, old.control,
2240 new.control) != old.control);
2243 static void decache_tsc_multiplier(struct vcpu_vmx *vmx)
2245 vmx->current_tsc_ratio = vmx->vcpu.arch.tsc_scaling_ratio;
2246 vmcs_write64(TSC_MULTIPLIER, vmx->current_tsc_ratio);
2250 * Switches to specified vcpu, until a matching vcpu_put(), but assumes
2251 * vcpu mutex is already taken.
2253 static void vmx_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
2255 struct vcpu_vmx *vmx = to_vmx(vcpu);
2256 bool already_loaded = vmx->loaded_vmcs->cpu == cpu;
2258 if (!already_loaded) {
2259 loaded_vmcs_clear(vmx->loaded_vmcs);
2260 local_irq_disable();
2261 crash_disable_local_vmclear(cpu);
2264 * Read loaded_vmcs->cpu should be before fetching
2265 * loaded_vmcs->loaded_vmcss_on_cpu_link.
2266 * See the comments in __loaded_vmcs_clear().
2270 list_add(&vmx->loaded_vmcs->loaded_vmcss_on_cpu_link,
2271 &per_cpu(loaded_vmcss_on_cpu, cpu));
2272 crash_enable_local_vmclear(cpu);
2276 if (per_cpu(current_vmcs, cpu) != vmx->loaded_vmcs->vmcs) {
2277 per_cpu(current_vmcs, cpu) = vmx->loaded_vmcs->vmcs;
2278 vmcs_load(vmx->loaded_vmcs->vmcs);
2281 if (!already_loaded) {
2282 void *gdt = get_current_gdt_ro();
2283 unsigned long sysenter_esp;
2285 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
2288 * Linux uses per-cpu TSS and GDT, so set these when switching
2289 * processors. See 22.2.4.
2291 vmcs_writel(HOST_TR_BASE,
2292 (unsigned long)this_cpu_ptr(&cpu_tss));
2293 vmcs_writel(HOST_GDTR_BASE, (unsigned long)gdt); /* 22.2.4 */
2296 * VM exits change the host TR limit to 0x67 after a VM
2297 * exit. This is okay, since 0x67 covers everything except
2298 * the IO bitmap and have have code to handle the IO bitmap
2299 * being lost after a VM exit.
2301 BUILD_BUG_ON(IO_BITMAP_OFFSET - 1 != 0x67);
2303 rdmsrl(MSR_IA32_SYSENTER_ESP, sysenter_esp);
2304 vmcs_writel(HOST_IA32_SYSENTER_ESP, sysenter_esp); /* 22.2.3 */
2306 vmx->loaded_vmcs->cpu = cpu;
2309 /* Setup TSC multiplier */
2310 if (kvm_has_tsc_control &&
2311 vmx->current_tsc_ratio != vcpu->arch.tsc_scaling_ratio)
2312 decache_tsc_multiplier(vmx);
2314 vmx_vcpu_pi_load(vcpu, cpu);
2315 vmx->host_pkru = read_pkru();
2318 static void vmx_vcpu_pi_put(struct kvm_vcpu *vcpu)
2320 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
2322 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
2323 !irq_remapping_cap(IRQ_POSTING_CAP) ||
2324 !kvm_vcpu_apicv_active(vcpu))
2327 /* Set SN when the vCPU is preempted */
2328 if (vcpu->preempted)
2332 static void vmx_vcpu_put(struct kvm_vcpu *vcpu)
2334 vmx_vcpu_pi_put(vcpu);
2336 __vmx_load_host_state(to_vmx(vcpu));
2339 static bool emulation_required(struct kvm_vcpu *vcpu)
2341 return emulate_invalid_guest_state && !guest_state_valid(vcpu);
2344 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu);
2347 * Return the cr0 value that a nested guest would read. This is a combination
2348 * of the real cr0 used to run the guest (guest_cr0), and the bits shadowed by
2349 * its hypervisor (cr0_read_shadow).
2351 static inline unsigned long nested_read_cr0(struct vmcs12 *fields)
2353 return (fields->guest_cr0 & ~fields->cr0_guest_host_mask) |
2354 (fields->cr0_read_shadow & fields->cr0_guest_host_mask);
2356 static inline unsigned long nested_read_cr4(struct vmcs12 *fields)
2358 return (fields->guest_cr4 & ~fields->cr4_guest_host_mask) |
2359 (fields->cr4_read_shadow & fields->cr4_guest_host_mask);
2362 static unsigned long vmx_get_rflags(struct kvm_vcpu *vcpu)
2364 unsigned long rflags, save_rflags;
2366 if (!test_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail)) {
2367 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2368 rflags = vmcs_readl(GUEST_RFLAGS);
2369 if (to_vmx(vcpu)->rmode.vm86_active) {
2370 rflags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
2371 save_rflags = to_vmx(vcpu)->rmode.save_rflags;
2372 rflags |= save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
2374 to_vmx(vcpu)->rflags = rflags;
2376 return to_vmx(vcpu)->rflags;
2379 static void vmx_set_rflags(struct kvm_vcpu *vcpu, unsigned long rflags)
2381 unsigned long old_rflags = vmx_get_rflags(vcpu);
2383 __set_bit(VCPU_EXREG_RFLAGS, (ulong *)&vcpu->arch.regs_avail);
2384 to_vmx(vcpu)->rflags = rflags;
2385 if (to_vmx(vcpu)->rmode.vm86_active) {
2386 to_vmx(vcpu)->rmode.save_rflags = rflags;
2387 rflags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
2389 vmcs_writel(GUEST_RFLAGS, rflags);
2391 if ((old_rflags ^ to_vmx(vcpu)->rflags) & X86_EFLAGS_VM)
2392 to_vmx(vcpu)->emulation_required = emulation_required(vcpu);
2395 static u32 vmx_get_interrupt_shadow(struct kvm_vcpu *vcpu)
2397 u32 interruptibility = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2400 if (interruptibility & GUEST_INTR_STATE_STI)
2401 ret |= KVM_X86_SHADOW_INT_STI;
2402 if (interruptibility & GUEST_INTR_STATE_MOV_SS)
2403 ret |= KVM_X86_SHADOW_INT_MOV_SS;
2408 static void vmx_set_interrupt_shadow(struct kvm_vcpu *vcpu, int mask)
2410 u32 interruptibility_old = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
2411 u32 interruptibility = interruptibility_old;
2413 interruptibility &= ~(GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS);
2415 if (mask & KVM_X86_SHADOW_INT_MOV_SS)
2416 interruptibility |= GUEST_INTR_STATE_MOV_SS;
2417 else if (mask & KVM_X86_SHADOW_INT_STI)
2418 interruptibility |= GUEST_INTR_STATE_STI;
2420 if ((interruptibility != interruptibility_old))
2421 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, interruptibility);
2424 static void skip_emulated_instruction(struct kvm_vcpu *vcpu)
2428 rip = kvm_rip_read(vcpu);
2429 rip += vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
2430 kvm_rip_write(vcpu, rip);
2432 /* skipping an emulated instruction also counts */
2433 vmx_set_interrupt_shadow(vcpu, 0);
2436 static void nested_vmx_inject_exception_vmexit(struct kvm_vcpu *vcpu,
2437 unsigned long exit_qual)
2439 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2440 unsigned int nr = vcpu->arch.exception.nr;
2441 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2443 if (vcpu->arch.exception.has_error_code) {
2444 vmcs12->vm_exit_intr_error_code = vcpu->arch.exception.error_code;
2445 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2448 if (kvm_exception_is_soft(nr))
2449 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2451 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2453 if (!(vmcs12->idt_vectoring_info_field & VECTORING_INFO_VALID_MASK) &&
2454 vmx_get_nmi_mask(vcpu))
2455 intr_info |= INTR_INFO_UNBLOCK_NMI;
2457 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI, intr_info, exit_qual);
2461 * KVM wants to inject page-faults which it got to the guest. This function
2462 * checks whether in a nested guest, we need to inject them to L1 or L2.
2464 static int nested_vmx_check_exception(struct kvm_vcpu *vcpu, unsigned long *exit_qual)
2466 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
2467 unsigned int nr = vcpu->arch.exception.nr;
2469 if (nr == PF_VECTOR) {
2470 if (vcpu->arch.exception.nested_apf) {
2471 *exit_qual = vcpu->arch.apf.nested_apf_token;
2475 * FIXME: we must not write CR2 when L1 intercepts an L2 #PF exception.
2476 * The fix is to add the ancillary datum (CR2 or DR6) to structs
2477 * kvm_queued_exception and kvm_vcpu_events, so that CR2 and DR6
2478 * can be written only when inject_pending_event runs. This should be
2479 * conditional on a new capability---if the capability is disabled,
2480 * kvm_multiple_exception would write the ancillary information to
2481 * CR2 or DR6, for backwards ABI-compatibility.
2483 if (nested_vmx_is_page_fault_vmexit(vmcs12,
2484 vcpu->arch.exception.error_code)) {
2485 *exit_qual = vcpu->arch.cr2;
2489 if (vmcs12->exception_bitmap & (1u << nr)) {
2490 if (nr == DB_VECTOR)
2491 *exit_qual = vcpu->arch.dr6;
2501 static void vmx_queue_exception(struct kvm_vcpu *vcpu)
2503 struct vcpu_vmx *vmx = to_vmx(vcpu);
2504 unsigned nr = vcpu->arch.exception.nr;
2505 bool has_error_code = vcpu->arch.exception.has_error_code;
2506 u32 error_code = vcpu->arch.exception.error_code;
2507 u32 intr_info = nr | INTR_INFO_VALID_MASK;
2509 if (has_error_code) {
2510 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE, error_code);
2511 intr_info |= INTR_INFO_DELIVER_CODE_MASK;
2514 if (vmx->rmode.vm86_active) {
2516 if (kvm_exception_is_soft(nr))
2517 inc_eip = vcpu->arch.event_exit_inst_len;
2518 if (kvm_inject_realmode_interrupt(vcpu, nr, inc_eip) != EMULATE_DONE)
2519 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
2523 if (kvm_exception_is_soft(nr)) {
2524 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
2525 vmx->vcpu.arch.event_exit_inst_len);
2526 intr_info |= INTR_TYPE_SOFT_EXCEPTION;
2528 intr_info |= INTR_TYPE_HARD_EXCEPTION;
2530 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr_info);
2533 static bool vmx_rdtscp_supported(void)
2535 return cpu_has_vmx_rdtscp();
2538 static bool vmx_invpcid_supported(void)
2540 return cpu_has_vmx_invpcid() && enable_ept;
2544 * Swap MSR entry in host/guest MSR entry array.
2546 static void move_msr_up(struct vcpu_vmx *vmx, int from, int to)
2548 struct shared_msr_entry tmp;
2550 tmp = vmx->guest_msrs[to];
2551 vmx->guest_msrs[to] = vmx->guest_msrs[from];
2552 vmx->guest_msrs[from] = tmp;
2555 static void vmx_set_msr_bitmap(struct kvm_vcpu *vcpu)
2557 unsigned long *msr_bitmap;
2559 if (is_guest_mode(vcpu))
2560 msr_bitmap = to_vmx(vcpu)->nested.msr_bitmap;
2561 else if (cpu_has_secondary_exec_ctrls() &&
2562 (vmcs_read32(SECONDARY_VM_EXEC_CONTROL) &
2563 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE)) {
2564 if (enable_apicv && kvm_vcpu_apicv_active(vcpu)) {
2565 if (is_long_mode(vcpu))
2566 msr_bitmap = vmx_msr_bitmap_longmode_x2apic_apicv;
2568 msr_bitmap = vmx_msr_bitmap_legacy_x2apic_apicv;
2570 if (is_long_mode(vcpu))
2571 msr_bitmap = vmx_msr_bitmap_longmode_x2apic;
2573 msr_bitmap = vmx_msr_bitmap_legacy_x2apic;
2576 if (is_long_mode(vcpu))
2577 msr_bitmap = vmx_msr_bitmap_longmode;
2579 msr_bitmap = vmx_msr_bitmap_legacy;
2582 vmcs_write64(MSR_BITMAP, __pa(msr_bitmap));
2586 * Set up the vmcs to automatically save and restore system
2587 * msrs. Don't touch the 64-bit msrs if the guest is in legacy
2588 * mode, as fiddling with msrs is very expensive.
2590 static void setup_msrs(struct vcpu_vmx *vmx)
2592 int save_nmsrs, index;
2595 #ifdef CONFIG_X86_64
2596 if (is_long_mode(&vmx->vcpu)) {
2597 index = __find_msr_index(vmx, MSR_SYSCALL_MASK);
2599 move_msr_up(vmx, index, save_nmsrs++);
2600 index = __find_msr_index(vmx, MSR_LSTAR);
2602 move_msr_up(vmx, index, save_nmsrs++);
2603 index = __find_msr_index(vmx, MSR_CSTAR);
2605 move_msr_up(vmx, index, save_nmsrs++);
2606 index = __find_msr_index(vmx, MSR_TSC_AUX);
2607 if (index >= 0 && guest_cpuid_has(&vmx->vcpu, X86_FEATURE_RDTSCP))
2608 move_msr_up(vmx, index, save_nmsrs++);
2610 * MSR_STAR is only needed on long mode guests, and only
2611 * if efer.sce is enabled.
2613 index = __find_msr_index(vmx, MSR_STAR);
2614 if ((index >= 0) && (vmx->vcpu.arch.efer & EFER_SCE))
2615 move_msr_up(vmx, index, save_nmsrs++);
2618 index = __find_msr_index(vmx, MSR_EFER);
2619 if (index >= 0 && update_transition_efer(vmx, index))
2620 move_msr_up(vmx, index, save_nmsrs++);
2622 vmx->save_nmsrs = save_nmsrs;
2624 if (cpu_has_vmx_msr_bitmap())
2625 vmx_set_msr_bitmap(&vmx->vcpu);
2629 * reads and returns guest's timestamp counter "register"
2630 * guest_tsc = (host_tsc * tsc multiplier) >> 48 + tsc_offset
2631 * -- Intel TSC Scaling for Virtualization White Paper, sec 1.3
2633 static u64 guest_read_tsc(struct kvm_vcpu *vcpu)
2635 u64 host_tsc, tsc_offset;
2638 tsc_offset = vmcs_read64(TSC_OFFSET);
2639 return kvm_scale_tsc(vcpu, host_tsc) + tsc_offset;
2643 * writes 'offset' into guest's timestamp counter offset register
2645 static void vmx_write_tsc_offset(struct kvm_vcpu *vcpu, u64 offset)
2647 if (is_guest_mode(vcpu)) {
2649 * We're here if L1 chose not to trap WRMSR to TSC. According
2650 * to the spec, this should set L1's TSC; The offset that L1
2651 * set for L2 remains unchanged, and still needs to be added
2652 * to the newly set TSC to get L2's TSC.
2654 struct vmcs12 *vmcs12;
2655 /* recalculate vmcs02.TSC_OFFSET: */
2656 vmcs12 = get_vmcs12(vcpu);
2657 vmcs_write64(TSC_OFFSET, offset +
2658 (nested_cpu_has(vmcs12, CPU_BASED_USE_TSC_OFFSETING) ?
2659 vmcs12->tsc_offset : 0));
2661 trace_kvm_write_tsc_offset(vcpu->vcpu_id,
2662 vmcs_read64(TSC_OFFSET), offset);
2663 vmcs_write64(TSC_OFFSET, offset);
2668 * nested_vmx_allowed() checks whether a guest should be allowed to use VMX
2669 * instructions and MSRs (i.e., nested VMX). Nested VMX is disabled for
2670 * all guests if the "nested" module option is off, and can also be disabled
2671 * for a single guest by disabling its VMX cpuid bit.
2673 static inline bool nested_vmx_allowed(struct kvm_vcpu *vcpu)
2675 return nested && guest_cpuid_has(vcpu, X86_FEATURE_VMX);
2679 * nested_vmx_setup_ctls_msrs() sets up variables containing the values to be
2680 * returned for the various VMX controls MSRs when nested VMX is enabled.
2681 * The same values should also be used to verify that vmcs12 control fields are
2682 * valid during nested entry from L1 to L2.
2683 * Each of these control msrs has a low and high 32-bit half: A low bit is on
2684 * if the corresponding bit in the (32-bit) control field *must* be on, and a
2685 * bit in the high half is on if the corresponding bit in the control field
2686 * may be on. See also vmx_control_verify().
2688 static void nested_vmx_setup_ctls_msrs(struct vcpu_vmx *vmx)
2691 * Note that as a general rule, the high half of the MSRs (bits in
2692 * the control fields which may be 1) should be initialized by the
2693 * intersection of the underlying hardware's MSR (i.e., features which
2694 * can be supported) and the list of features we want to expose -
2695 * because they are known to be properly supported in our code.
2696 * Also, usually, the low half of the MSRs (bits which must be 1) can
2697 * be set to 0, meaning that L1 may turn off any of these bits. The
2698 * reason is that if one of these bits is necessary, it will appear
2699 * in vmcs01 and prepare_vmcs02, when it bitwise-or's the control
2700 * fields of vmcs01 and vmcs02, will turn these bits off - and
2701 * nested_vmx_exit_reflected() will not pass related exits to L1.
2702 * These rules have exceptions below.
2705 /* pin-based controls */
2706 rdmsr(MSR_IA32_VMX_PINBASED_CTLS,
2707 vmx->nested.nested_vmx_pinbased_ctls_low,
2708 vmx->nested.nested_vmx_pinbased_ctls_high);
2709 vmx->nested.nested_vmx_pinbased_ctls_low |=
2710 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2711 vmx->nested.nested_vmx_pinbased_ctls_high &=
2712 PIN_BASED_EXT_INTR_MASK |
2713 PIN_BASED_NMI_EXITING |
2714 PIN_BASED_VIRTUAL_NMIS;
2715 vmx->nested.nested_vmx_pinbased_ctls_high |=
2716 PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2717 PIN_BASED_VMX_PREEMPTION_TIMER;
2718 if (kvm_vcpu_apicv_active(&vmx->vcpu))
2719 vmx->nested.nested_vmx_pinbased_ctls_high |=
2720 PIN_BASED_POSTED_INTR;
2723 rdmsr(MSR_IA32_VMX_EXIT_CTLS,
2724 vmx->nested.nested_vmx_exit_ctls_low,
2725 vmx->nested.nested_vmx_exit_ctls_high);
2726 vmx->nested.nested_vmx_exit_ctls_low =
2727 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
2729 vmx->nested.nested_vmx_exit_ctls_high &=
2730 #ifdef CONFIG_X86_64
2731 VM_EXIT_HOST_ADDR_SPACE_SIZE |
2733 VM_EXIT_LOAD_IA32_PAT | VM_EXIT_SAVE_IA32_PAT;
2734 vmx->nested.nested_vmx_exit_ctls_high |=
2735 VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR |
2736 VM_EXIT_LOAD_IA32_EFER | VM_EXIT_SAVE_IA32_EFER |
2737 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER | VM_EXIT_ACK_INTR_ON_EXIT;
2739 if (kvm_mpx_supported())
2740 vmx->nested.nested_vmx_exit_ctls_high |= VM_EXIT_CLEAR_BNDCFGS;
2742 /* We support free control of debug control saving. */
2743 vmx->nested.nested_vmx_exit_ctls_low &= ~VM_EXIT_SAVE_DEBUG_CONTROLS;
2745 /* entry controls */
2746 rdmsr(MSR_IA32_VMX_ENTRY_CTLS,
2747 vmx->nested.nested_vmx_entry_ctls_low,
2748 vmx->nested.nested_vmx_entry_ctls_high);
2749 vmx->nested.nested_vmx_entry_ctls_low =
2750 VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
2751 vmx->nested.nested_vmx_entry_ctls_high &=
2752 #ifdef CONFIG_X86_64
2753 VM_ENTRY_IA32E_MODE |
2755 VM_ENTRY_LOAD_IA32_PAT;
2756 vmx->nested.nested_vmx_entry_ctls_high |=
2757 (VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR | VM_ENTRY_LOAD_IA32_EFER);
2758 if (kvm_mpx_supported())
2759 vmx->nested.nested_vmx_entry_ctls_high |= VM_ENTRY_LOAD_BNDCFGS;
2761 /* We support free control of debug control loading. */
2762 vmx->nested.nested_vmx_entry_ctls_low &= ~VM_ENTRY_LOAD_DEBUG_CONTROLS;
2764 /* cpu-based controls */
2765 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS,
2766 vmx->nested.nested_vmx_procbased_ctls_low,
2767 vmx->nested.nested_vmx_procbased_ctls_high);
2768 vmx->nested.nested_vmx_procbased_ctls_low =
2769 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
2770 vmx->nested.nested_vmx_procbased_ctls_high &=
2771 CPU_BASED_VIRTUAL_INTR_PENDING |
2772 CPU_BASED_VIRTUAL_NMI_PENDING | CPU_BASED_USE_TSC_OFFSETING |
2773 CPU_BASED_HLT_EXITING | CPU_BASED_INVLPG_EXITING |
2774 CPU_BASED_MWAIT_EXITING | CPU_BASED_CR3_LOAD_EXITING |
2775 CPU_BASED_CR3_STORE_EXITING |
2776 #ifdef CONFIG_X86_64
2777 CPU_BASED_CR8_LOAD_EXITING | CPU_BASED_CR8_STORE_EXITING |
2779 CPU_BASED_MOV_DR_EXITING | CPU_BASED_UNCOND_IO_EXITING |
2780 CPU_BASED_USE_IO_BITMAPS | CPU_BASED_MONITOR_TRAP_FLAG |
2781 CPU_BASED_MONITOR_EXITING | CPU_BASED_RDPMC_EXITING |
2782 CPU_BASED_RDTSC_EXITING | CPU_BASED_PAUSE_EXITING |
2783 CPU_BASED_TPR_SHADOW | CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
2785 * We can allow some features even when not supported by the
2786 * hardware. For example, L1 can specify an MSR bitmap - and we
2787 * can use it to avoid exits to L1 - even when L0 runs L2
2788 * without MSR bitmaps.
2790 vmx->nested.nested_vmx_procbased_ctls_high |=
2791 CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR |
2792 CPU_BASED_USE_MSR_BITMAPS;
2794 /* We support free control of CR3 access interception. */
2795 vmx->nested.nested_vmx_procbased_ctls_low &=
2796 ~(CPU_BASED_CR3_LOAD_EXITING | CPU_BASED_CR3_STORE_EXITING);
2799 * secondary cpu-based controls. Do not include those that
2800 * depend on CPUID bits, they are added later by vmx_cpuid_update.
2802 rdmsr(MSR_IA32_VMX_PROCBASED_CTLS2,
2803 vmx->nested.nested_vmx_secondary_ctls_low,
2804 vmx->nested.nested_vmx_secondary_ctls_high);
2805 vmx->nested.nested_vmx_secondary_ctls_low = 0;
2806 vmx->nested.nested_vmx_secondary_ctls_high &=
2807 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
2808 SECONDARY_EXEC_DESC |
2809 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
2810 SECONDARY_EXEC_APIC_REGISTER_VIRT |
2811 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
2812 SECONDARY_EXEC_WBINVD_EXITING;
2815 /* nested EPT: emulate EPT also to L1 */
2816 vmx->nested.nested_vmx_secondary_ctls_high |=
2817 SECONDARY_EXEC_ENABLE_EPT;
2818 vmx->nested.nested_vmx_ept_caps = VMX_EPT_PAGE_WALK_4_BIT |
2819 VMX_EPTP_WB_BIT | VMX_EPT_INVEPT_BIT;
2820 if (cpu_has_vmx_ept_execute_only())
2821 vmx->nested.nested_vmx_ept_caps |=
2822 VMX_EPT_EXECUTE_ONLY_BIT;
2823 vmx->nested.nested_vmx_ept_caps &= vmx_capability.ept;
2824 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_EXTENT_GLOBAL_BIT |
2825 VMX_EPT_EXTENT_CONTEXT_BIT | VMX_EPT_2MB_PAGE_BIT |
2826 VMX_EPT_1GB_PAGE_BIT;
2827 if (enable_ept_ad_bits) {
2828 vmx->nested.nested_vmx_secondary_ctls_high |=
2829 SECONDARY_EXEC_ENABLE_PML;
2830 vmx->nested.nested_vmx_ept_caps |= VMX_EPT_AD_BIT;
2833 vmx->nested.nested_vmx_ept_caps = 0;
2835 if (cpu_has_vmx_vmfunc()) {
2836 vmx->nested.nested_vmx_secondary_ctls_high |=
2837 SECONDARY_EXEC_ENABLE_VMFUNC;
2839 * Advertise EPTP switching unconditionally
2840 * since we emulate it
2842 vmx->nested.nested_vmx_vmfunc_controls =
2843 VMX_VMFUNC_EPTP_SWITCHING;
2847 * Old versions of KVM use the single-context version without
2848 * checking for support, so declare that it is supported even
2849 * though it is treated as global context. The alternative is
2850 * not failing the single-context invvpid, and it is worse.
2853 vmx->nested.nested_vmx_secondary_ctls_high |=
2854 SECONDARY_EXEC_ENABLE_VPID;
2855 vmx->nested.nested_vmx_vpid_caps = VMX_VPID_INVVPID_BIT |
2856 VMX_VPID_EXTENT_SUPPORTED_MASK;
2858 vmx->nested.nested_vmx_vpid_caps = 0;
2860 if (enable_unrestricted_guest)
2861 vmx->nested.nested_vmx_secondary_ctls_high |=
2862 SECONDARY_EXEC_UNRESTRICTED_GUEST;
2864 /* miscellaneous data */
2865 rdmsr(MSR_IA32_VMX_MISC,
2866 vmx->nested.nested_vmx_misc_low,
2867 vmx->nested.nested_vmx_misc_high);
2868 vmx->nested.nested_vmx_misc_low &= VMX_MISC_SAVE_EFER_LMA;
2869 vmx->nested.nested_vmx_misc_low |=
2870 VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE |
2871 VMX_MISC_ACTIVITY_HLT;
2872 vmx->nested.nested_vmx_misc_high = 0;
2875 * This MSR reports some information about VMX support. We
2876 * should return information about the VMX we emulate for the
2877 * guest, and the VMCS structure we give it - not about the
2878 * VMX support of the underlying hardware.
2880 vmx->nested.nested_vmx_basic =
2882 VMX_BASIC_TRUE_CTLS |
2883 ((u64)VMCS12_SIZE << VMX_BASIC_VMCS_SIZE_SHIFT) |
2884 (VMX_BASIC_MEM_TYPE_WB << VMX_BASIC_MEM_TYPE_SHIFT);
2886 if (cpu_has_vmx_basic_inout())
2887 vmx->nested.nested_vmx_basic |= VMX_BASIC_INOUT;
2890 * These MSRs specify bits which the guest must keep fixed on
2891 * while L1 is in VMXON mode (in L1's root mode, or running an L2).
2892 * We picked the standard core2 setting.
2894 #define VMXON_CR0_ALWAYSON (X86_CR0_PE | X86_CR0_PG | X86_CR0_NE)
2895 #define VMXON_CR4_ALWAYSON X86_CR4_VMXE
2896 vmx->nested.nested_vmx_cr0_fixed0 = VMXON_CR0_ALWAYSON;
2897 vmx->nested.nested_vmx_cr4_fixed0 = VMXON_CR4_ALWAYSON;
2899 /* These MSRs specify bits which the guest must keep fixed off. */
2900 rdmsrl(MSR_IA32_VMX_CR0_FIXED1, vmx->nested.nested_vmx_cr0_fixed1);
2901 rdmsrl(MSR_IA32_VMX_CR4_FIXED1, vmx->nested.nested_vmx_cr4_fixed1);
2903 /* highest index: VMX_PREEMPTION_TIMER_VALUE */
2904 vmx->nested.nested_vmx_vmcs_enum = 0x2e;
2908 * if fixed0[i] == 1: val[i] must be 1
2909 * if fixed1[i] == 0: val[i] must be 0
2911 static inline bool fixed_bits_valid(u64 val, u64 fixed0, u64 fixed1)
2913 return ((val & fixed1) | fixed0) == val;
2916 static inline bool vmx_control_verify(u32 control, u32 low, u32 high)
2918 return fixed_bits_valid(control, low, high);
2921 static inline u64 vmx_control_msr(u32 low, u32 high)
2923 return low | ((u64)high << 32);
2926 static bool is_bitwise_subset(u64 superset, u64 subset, u64 mask)
2931 return (superset | subset) == superset;
2934 static int vmx_restore_vmx_basic(struct vcpu_vmx *vmx, u64 data)
2936 const u64 feature_and_reserved =
2937 /* feature (except bit 48; see below) */
2938 BIT_ULL(49) | BIT_ULL(54) | BIT_ULL(55) |
2940 BIT_ULL(31) | GENMASK_ULL(47, 45) | GENMASK_ULL(63, 56);
2941 u64 vmx_basic = vmx->nested.nested_vmx_basic;
2943 if (!is_bitwise_subset(vmx_basic, data, feature_and_reserved))
2947 * KVM does not emulate a version of VMX that constrains physical
2948 * addresses of VMX structures (e.g. VMCS) to 32-bits.
2950 if (data & BIT_ULL(48))
2953 if (vmx_basic_vmcs_revision_id(vmx_basic) !=
2954 vmx_basic_vmcs_revision_id(data))
2957 if (vmx_basic_vmcs_size(vmx_basic) > vmx_basic_vmcs_size(data))
2960 vmx->nested.nested_vmx_basic = data;
2965 vmx_restore_control_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
2970 switch (msr_index) {
2971 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
2972 lowp = &vmx->nested.nested_vmx_pinbased_ctls_low;
2973 highp = &vmx->nested.nested_vmx_pinbased_ctls_high;
2975 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
2976 lowp = &vmx->nested.nested_vmx_procbased_ctls_low;
2977 highp = &vmx->nested.nested_vmx_procbased_ctls_high;
2979 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
2980 lowp = &vmx->nested.nested_vmx_exit_ctls_low;
2981 highp = &vmx->nested.nested_vmx_exit_ctls_high;
2983 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
2984 lowp = &vmx->nested.nested_vmx_entry_ctls_low;
2985 highp = &vmx->nested.nested_vmx_entry_ctls_high;
2987 case MSR_IA32_VMX_PROCBASED_CTLS2:
2988 lowp = &vmx->nested.nested_vmx_secondary_ctls_low;
2989 highp = &vmx->nested.nested_vmx_secondary_ctls_high;
2995 supported = vmx_control_msr(*lowp, *highp);
2997 /* Check must-be-1 bits are still 1. */
2998 if (!is_bitwise_subset(data, supported, GENMASK_ULL(31, 0)))
3001 /* Check must-be-0 bits are still 0. */
3002 if (!is_bitwise_subset(supported, data, GENMASK_ULL(63, 32)))
3006 *highp = data >> 32;
3010 static int vmx_restore_vmx_misc(struct vcpu_vmx *vmx, u64 data)
3012 const u64 feature_and_reserved_bits =
3014 BIT_ULL(5) | GENMASK_ULL(8, 6) | BIT_ULL(14) | BIT_ULL(15) |
3015 BIT_ULL(28) | BIT_ULL(29) | BIT_ULL(30) |
3017 GENMASK_ULL(13, 9) | BIT_ULL(31);
3020 vmx_misc = vmx_control_msr(vmx->nested.nested_vmx_misc_low,
3021 vmx->nested.nested_vmx_misc_high);
3023 if (!is_bitwise_subset(vmx_misc, data, feature_and_reserved_bits))
3026 if ((vmx->nested.nested_vmx_pinbased_ctls_high &
3027 PIN_BASED_VMX_PREEMPTION_TIMER) &&
3028 vmx_misc_preemption_timer_rate(data) !=
3029 vmx_misc_preemption_timer_rate(vmx_misc))
3032 if (vmx_misc_cr3_count(data) > vmx_misc_cr3_count(vmx_misc))
3035 if (vmx_misc_max_msr(data) > vmx_misc_max_msr(vmx_misc))
3038 if (vmx_misc_mseg_revid(data) != vmx_misc_mseg_revid(vmx_misc))
3041 vmx->nested.nested_vmx_misc_low = data;
3042 vmx->nested.nested_vmx_misc_high = data >> 32;
3046 static int vmx_restore_vmx_ept_vpid_cap(struct vcpu_vmx *vmx, u64 data)
3048 u64 vmx_ept_vpid_cap;
3050 vmx_ept_vpid_cap = vmx_control_msr(vmx->nested.nested_vmx_ept_caps,
3051 vmx->nested.nested_vmx_vpid_caps);
3053 /* Every bit is either reserved or a feature bit. */
3054 if (!is_bitwise_subset(vmx_ept_vpid_cap, data, -1ULL))
3057 vmx->nested.nested_vmx_ept_caps = data;
3058 vmx->nested.nested_vmx_vpid_caps = data >> 32;
3062 static int vmx_restore_fixed0_msr(struct vcpu_vmx *vmx, u32 msr_index, u64 data)
3066 switch (msr_index) {
3067 case MSR_IA32_VMX_CR0_FIXED0:
3068 msr = &vmx->nested.nested_vmx_cr0_fixed0;
3070 case MSR_IA32_VMX_CR4_FIXED0:
3071 msr = &vmx->nested.nested_vmx_cr4_fixed0;
3078 * 1 bits (which indicates bits which "must-be-1" during VMX operation)
3079 * must be 1 in the restored value.
3081 if (!is_bitwise_subset(data, *msr, -1ULL))
3089 * Called when userspace is restoring VMX MSRs.
3091 * Returns 0 on success, non-0 otherwise.
3093 static int vmx_set_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 data)
3095 struct vcpu_vmx *vmx = to_vmx(vcpu);
3097 switch (msr_index) {
3098 case MSR_IA32_VMX_BASIC:
3099 return vmx_restore_vmx_basic(vmx, data);
3100 case MSR_IA32_VMX_PINBASED_CTLS:
3101 case MSR_IA32_VMX_PROCBASED_CTLS:
3102 case MSR_IA32_VMX_EXIT_CTLS:
3103 case MSR_IA32_VMX_ENTRY_CTLS:
3105 * The "non-true" VMX capability MSRs are generated from the
3106 * "true" MSRs, so we do not support restoring them directly.
3108 * If userspace wants to emulate VMX_BASIC[55]=0, userspace
3109 * should restore the "true" MSRs with the must-be-1 bits
3110 * set according to the SDM Vol 3. A.2 "RESERVED CONTROLS AND
3111 * DEFAULT SETTINGS".
3114 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3115 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3116 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3117 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3118 case MSR_IA32_VMX_PROCBASED_CTLS2:
3119 return vmx_restore_control_msr(vmx, msr_index, data);
3120 case MSR_IA32_VMX_MISC:
3121 return vmx_restore_vmx_misc(vmx, data);
3122 case MSR_IA32_VMX_CR0_FIXED0:
3123 case MSR_IA32_VMX_CR4_FIXED0:
3124 return vmx_restore_fixed0_msr(vmx, msr_index, data);
3125 case MSR_IA32_VMX_CR0_FIXED1:
3126 case MSR_IA32_VMX_CR4_FIXED1:
3128 * These MSRs are generated based on the vCPU's CPUID, so we
3129 * do not support restoring them directly.
3132 case MSR_IA32_VMX_EPT_VPID_CAP:
3133 return vmx_restore_vmx_ept_vpid_cap(vmx, data);
3134 case MSR_IA32_VMX_VMCS_ENUM:
3135 vmx->nested.nested_vmx_vmcs_enum = data;
3139 * The rest of the VMX capability MSRs do not support restore.
3145 /* Returns 0 on success, non-0 otherwise. */
3146 static int vmx_get_vmx_msr(struct kvm_vcpu *vcpu, u32 msr_index, u64 *pdata)
3148 struct vcpu_vmx *vmx = to_vmx(vcpu);
3150 switch (msr_index) {
3151 case MSR_IA32_VMX_BASIC:
3152 *pdata = vmx->nested.nested_vmx_basic;
3154 case MSR_IA32_VMX_TRUE_PINBASED_CTLS:
3155 case MSR_IA32_VMX_PINBASED_CTLS:
3156 *pdata = vmx_control_msr(
3157 vmx->nested.nested_vmx_pinbased_ctls_low,
3158 vmx->nested.nested_vmx_pinbased_ctls_high);
3159 if (msr_index == MSR_IA32_VMX_PINBASED_CTLS)
3160 *pdata |= PIN_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3162 case MSR_IA32_VMX_TRUE_PROCBASED_CTLS:
3163 case MSR_IA32_VMX_PROCBASED_CTLS:
3164 *pdata = vmx_control_msr(
3165 vmx->nested.nested_vmx_procbased_ctls_low,
3166 vmx->nested.nested_vmx_procbased_ctls_high);
3167 if (msr_index == MSR_IA32_VMX_PROCBASED_CTLS)
3168 *pdata |= CPU_BASED_ALWAYSON_WITHOUT_TRUE_MSR;
3170 case MSR_IA32_VMX_TRUE_EXIT_CTLS:
3171 case MSR_IA32_VMX_EXIT_CTLS:
3172 *pdata = vmx_control_msr(
3173 vmx->nested.nested_vmx_exit_ctls_low,
3174 vmx->nested.nested_vmx_exit_ctls_high);
3175 if (msr_index == MSR_IA32_VMX_EXIT_CTLS)
3176 *pdata |= VM_EXIT_ALWAYSON_WITHOUT_TRUE_MSR;
3178 case MSR_IA32_VMX_TRUE_ENTRY_CTLS:
3179 case MSR_IA32_VMX_ENTRY_CTLS:
3180 *pdata = vmx_control_msr(
3181 vmx->nested.nested_vmx_entry_ctls_low,
3182 vmx->nested.nested_vmx_entry_ctls_high);
3183 if (msr_index == MSR_IA32_VMX_ENTRY_CTLS)
3184 *pdata |= VM_ENTRY_ALWAYSON_WITHOUT_TRUE_MSR;
3186 case MSR_IA32_VMX_MISC:
3187 *pdata = vmx_control_msr(
3188 vmx->nested.nested_vmx_misc_low,
3189 vmx->nested.nested_vmx_misc_high);
3191 case MSR_IA32_VMX_CR0_FIXED0:
3192 *pdata = vmx->nested.nested_vmx_cr0_fixed0;
3194 case MSR_IA32_VMX_CR0_FIXED1:
3195 *pdata = vmx->nested.nested_vmx_cr0_fixed1;
3197 case MSR_IA32_VMX_CR4_FIXED0:
3198 *pdata = vmx->nested.nested_vmx_cr4_fixed0;
3200 case MSR_IA32_VMX_CR4_FIXED1:
3201 *pdata = vmx->nested.nested_vmx_cr4_fixed1;
3203 case MSR_IA32_VMX_VMCS_ENUM:
3204 *pdata = vmx->nested.nested_vmx_vmcs_enum;
3206 case MSR_IA32_VMX_PROCBASED_CTLS2:
3207 *pdata = vmx_control_msr(
3208 vmx->nested.nested_vmx_secondary_ctls_low,
3209 vmx->nested.nested_vmx_secondary_ctls_high);
3211 case MSR_IA32_VMX_EPT_VPID_CAP:
3212 *pdata = vmx->nested.nested_vmx_ept_caps |
3213 ((u64)vmx->nested.nested_vmx_vpid_caps << 32);
3215 case MSR_IA32_VMX_VMFUNC:
3216 *pdata = vmx->nested.nested_vmx_vmfunc_controls;
3225 static inline bool vmx_feature_control_msr_valid(struct kvm_vcpu *vcpu,
3228 uint64_t valid_bits = to_vmx(vcpu)->msr_ia32_feature_control_valid_bits;
3230 return !(val & ~valid_bits);
3234 * Reads an msr value (of 'msr_index') into 'pdata'.
3235 * Returns 0 on success, non-0 otherwise.
3236 * Assumes vcpu_load() was already called.
3238 static int vmx_get_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3240 struct shared_msr_entry *msr;
3242 switch (msr_info->index) {
3243 #ifdef CONFIG_X86_64
3245 msr_info->data = vmcs_readl(GUEST_FS_BASE);
3248 msr_info->data = vmcs_readl(GUEST_GS_BASE);
3250 case MSR_KERNEL_GS_BASE:
3251 vmx_load_host_state(to_vmx(vcpu));
3252 msr_info->data = to_vmx(vcpu)->msr_guest_kernel_gs_base;
3256 return kvm_get_msr_common(vcpu, msr_info);
3258 msr_info->data = guest_read_tsc(vcpu);
3260 case MSR_IA32_SYSENTER_CS:
3261 msr_info->data = vmcs_read32(GUEST_SYSENTER_CS);
3263 case MSR_IA32_SYSENTER_EIP:
3264 msr_info->data = vmcs_readl(GUEST_SYSENTER_EIP);
3266 case MSR_IA32_SYSENTER_ESP:
3267 msr_info->data = vmcs_readl(GUEST_SYSENTER_ESP);
3269 case MSR_IA32_BNDCFGS:
3270 if (!kvm_mpx_supported() ||
3271 (!msr_info->host_initiated &&
3272 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
3274 msr_info->data = vmcs_read64(GUEST_BNDCFGS);
3276 case MSR_IA32_MCG_EXT_CTL:
3277 if (!msr_info->host_initiated &&
3278 !(to_vmx(vcpu)->msr_ia32_feature_control &
3279 FEATURE_CONTROL_LMCE))
3281 msr_info->data = vcpu->arch.mcg_ext_ctl;
3283 case MSR_IA32_FEATURE_CONTROL:
3284 msr_info->data = to_vmx(vcpu)->msr_ia32_feature_control;
3286 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3287 if (!nested_vmx_allowed(vcpu))
3289 return vmx_get_vmx_msr(vcpu, msr_info->index, &msr_info->data);
3291 if (!vmx_xsaves_supported())
3293 msr_info->data = vcpu->arch.ia32_xss;
3296 if (!msr_info->host_initiated &&
3297 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
3299 /* Otherwise falls through */
3301 msr = find_msr_entry(to_vmx(vcpu), msr_info->index);
3303 msr_info->data = msr->data;
3306 return kvm_get_msr_common(vcpu, msr_info);
3312 static void vmx_leave_nested(struct kvm_vcpu *vcpu);
3315 * Writes msr value into into the appropriate "register".
3316 * Returns 0 on success, non-0 otherwise.
3317 * Assumes vcpu_load() was already called.
3319 static int vmx_set_msr(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
3321 struct vcpu_vmx *vmx = to_vmx(vcpu);
3322 struct shared_msr_entry *msr;
3324 u32 msr_index = msr_info->index;
3325 u64 data = msr_info->data;
3327 switch (msr_index) {
3329 ret = kvm_set_msr_common(vcpu, msr_info);
3331 #ifdef CONFIG_X86_64
3333 vmx_segment_cache_clear(vmx);
3334 vmcs_writel(GUEST_FS_BASE, data);
3337 vmx_segment_cache_clear(vmx);
3338 vmcs_writel(GUEST_GS_BASE, data);
3340 case MSR_KERNEL_GS_BASE:
3341 vmx_load_host_state(vmx);
3342 vmx->msr_guest_kernel_gs_base = data;
3345 case MSR_IA32_SYSENTER_CS:
3346 vmcs_write32(GUEST_SYSENTER_CS, data);
3348 case MSR_IA32_SYSENTER_EIP:
3349 vmcs_writel(GUEST_SYSENTER_EIP, data);
3351 case MSR_IA32_SYSENTER_ESP:
3352 vmcs_writel(GUEST_SYSENTER_ESP, data);
3354 case MSR_IA32_BNDCFGS:
3355 if (!kvm_mpx_supported() ||
3356 (!msr_info->host_initiated &&
3357 !guest_cpuid_has(vcpu, X86_FEATURE_MPX)))
3359 if (is_noncanonical_address(data & PAGE_MASK, vcpu) ||
3360 (data & MSR_IA32_BNDCFGS_RSVD))
3362 vmcs_write64(GUEST_BNDCFGS, data);
3365 kvm_write_tsc(vcpu, msr_info);
3367 case MSR_IA32_CR_PAT:
3368 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
3369 if (!kvm_mtrr_valid(vcpu, MSR_IA32_CR_PAT, data))
3371 vmcs_write64(GUEST_IA32_PAT, data);
3372 vcpu->arch.pat = data;
3375 ret = kvm_set_msr_common(vcpu, msr_info);
3377 case MSR_IA32_TSC_ADJUST:
3378 ret = kvm_set_msr_common(vcpu, msr_info);
3380 case MSR_IA32_MCG_EXT_CTL:
3381 if ((!msr_info->host_initiated &&
3382 !(to_vmx(vcpu)->msr_ia32_feature_control &
3383 FEATURE_CONTROL_LMCE)) ||
3384 (data & ~MCG_EXT_CTL_LMCE_EN))
3386 vcpu->arch.mcg_ext_ctl = data;
3388 case MSR_IA32_FEATURE_CONTROL:
3389 if (!vmx_feature_control_msr_valid(vcpu, data) ||
3390 (to_vmx(vcpu)->msr_ia32_feature_control &
3391 FEATURE_CONTROL_LOCKED && !msr_info->host_initiated))
3393 vmx->msr_ia32_feature_control = data;
3394 if (msr_info->host_initiated && data == 0)
3395 vmx_leave_nested(vcpu);
3397 case MSR_IA32_VMX_BASIC ... MSR_IA32_VMX_VMFUNC:
3398 if (!msr_info->host_initiated)
3399 return 1; /* they are read-only */
3400 if (!nested_vmx_allowed(vcpu))
3402 return vmx_set_vmx_msr(vcpu, msr_index, data);
3404 if (!vmx_xsaves_supported())
3407 * The only supported bit as of Skylake is bit 8, but
3408 * it is not supported on KVM.
3412 vcpu->arch.ia32_xss = data;
3413 if (vcpu->arch.ia32_xss != host_xss)
3414 add_atomic_switch_msr(vmx, MSR_IA32_XSS,
3415 vcpu->arch.ia32_xss, host_xss);
3417 clear_atomic_switch_msr(vmx, MSR_IA32_XSS);
3420 if (!msr_info->host_initiated &&
3421 !guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP))
3423 /* Check reserved bit, higher 32 bits should be zero */
3424 if ((data >> 32) != 0)
3426 /* Otherwise falls through */
3428 msr = find_msr_entry(vmx, msr_index);
3430 u64 old_msr_data = msr->data;
3432 if (msr - vmx->guest_msrs < vmx->save_nmsrs) {
3434 ret = kvm_set_shared_msr(msr->index, msr->data,
3438 msr->data = old_msr_data;
3442 ret = kvm_set_msr_common(vcpu, msr_info);
3448 static void vmx_cache_reg(struct kvm_vcpu *vcpu, enum kvm_reg reg)
3450 __set_bit(reg, (unsigned long *)&vcpu->arch.regs_avail);
3453 vcpu->arch.regs[VCPU_REGS_RSP] = vmcs_readl(GUEST_RSP);
3456 vcpu->arch.regs[VCPU_REGS_RIP] = vmcs_readl(GUEST_RIP);
3458 case VCPU_EXREG_PDPTR:
3460 ept_save_pdptrs(vcpu);
3467 static __init int cpu_has_kvm_support(void)
3469 return cpu_has_vmx();
3472 static __init int vmx_disabled_by_bios(void)
3476 rdmsrl(MSR_IA32_FEATURE_CONTROL, msr);
3477 if (msr & FEATURE_CONTROL_LOCKED) {
3478 /* launched w/ TXT and VMX disabled */
3479 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3482 /* launched w/o TXT and VMX only enabled w/ TXT */
3483 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3484 && (msr & FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX)
3485 && !tboot_enabled()) {
3486 printk(KERN_WARNING "kvm: disable TXT in the BIOS or "
3487 "activate TXT before enabling KVM\n");
3490 /* launched w/o TXT and VMX disabled */
3491 if (!(msr & FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX)
3492 && !tboot_enabled())
3499 static void kvm_cpu_vmxon(u64 addr)
3501 cr4_set_bits(X86_CR4_VMXE);
3502 intel_pt_handle_vmx(1);
3504 asm volatile (ASM_VMX_VMXON_RAX
3505 : : "a"(&addr), "m"(addr)
3509 static int hardware_enable(void)
3511 int cpu = raw_smp_processor_id();
3512 u64 phys_addr = __pa(per_cpu(vmxarea, cpu));
3515 if (cr4_read_shadow() & X86_CR4_VMXE)
3518 INIT_LIST_HEAD(&per_cpu(loaded_vmcss_on_cpu, cpu));
3519 INIT_LIST_HEAD(&per_cpu(blocked_vcpu_on_cpu, cpu));
3520 spin_lock_init(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
3523 * Now we can enable the vmclear operation in kdump
3524 * since the loaded_vmcss_on_cpu list on this cpu
3525 * has been initialized.
3527 * Though the cpu is not in VMX operation now, there
3528 * is no problem to enable the vmclear operation
3529 * for the loaded_vmcss_on_cpu list is empty!
3531 crash_enable_local_vmclear(cpu);
3533 rdmsrl(MSR_IA32_FEATURE_CONTROL, old);
3535 test_bits = FEATURE_CONTROL_LOCKED;
3536 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
3537 if (tboot_enabled())
3538 test_bits |= FEATURE_CONTROL_VMXON_ENABLED_INSIDE_SMX;
3540 if ((old & test_bits) != test_bits) {
3541 /* enable and lock */
3542 wrmsrl(MSR_IA32_FEATURE_CONTROL, old | test_bits);
3544 kvm_cpu_vmxon(phys_addr);
3551 static void vmclear_local_loaded_vmcss(void)
3553 int cpu = raw_smp_processor_id();
3554 struct loaded_vmcs *v, *n;
3556 list_for_each_entry_safe(v, n, &per_cpu(loaded_vmcss_on_cpu, cpu),
3557 loaded_vmcss_on_cpu_link)
3558 __loaded_vmcs_clear(v);
3562 /* Just like cpu_vmxoff(), but with the __kvm_handle_fault_on_reboot()
3565 static void kvm_cpu_vmxoff(void)
3567 asm volatile (__ex(ASM_VMX_VMXOFF) : : : "cc");
3569 intel_pt_handle_vmx(0);
3570 cr4_clear_bits(X86_CR4_VMXE);
3573 static void hardware_disable(void)
3575 vmclear_local_loaded_vmcss();
3579 static __init int adjust_vmx_controls(u32 ctl_min, u32 ctl_opt,
3580 u32 msr, u32 *result)
3582 u32 vmx_msr_low, vmx_msr_high;
3583 u32 ctl = ctl_min | ctl_opt;
3585 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3587 ctl &= vmx_msr_high; /* bit == 0 in high word ==> must be zero */
3588 ctl |= vmx_msr_low; /* bit == 1 in low word ==> must be one */
3590 /* Ensure minimum (required) set of control bits are supported. */
3598 static __init bool allow_1_setting(u32 msr, u32 ctl)
3600 u32 vmx_msr_low, vmx_msr_high;
3602 rdmsr(msr, vmx_msr_low, vmx_msr_high);
3603 return vmx_msr_high & ctl;
3606 static __init int setup_vmcs_config(struct vmcs_config *vmcs_conf)
3608 u32 vmx_msr_low, vmx_msr_high;
3609 u32 min, opt, min2, opt2;
3610 u32 _pin_based_exec_control = 0;
3611 u32 _cpu_based_exec_control = 0;
3612 u32 _cpu_based_2nd_exec_control = 0;
3613 u32 _vmexit_control = 0;
3614 u32 _vmentry_control = 0;
3616 min = CPU_BASED_HLT_EXITING |
3617 #ifdef CONFIG_X86_64
3618 CPU_BASED_CR8_LOAD_EXITING |
3619 CPU_BASED_CR8_STORE_EXITING |
3621 CPU_BASED_CR3_LOAD_EXITING |
3622 CPU_BASED_CR3_STORE_EXITING |
3623 CPU_BASED_USE_IO_BITMAPS |
3624 CPU_BASED_MOV_DR_EXITING |
3625 CPU_BASED_USE_TSC_OFFSETING |
3626 CPU_BASED_INVLPG_EXITING |
3627 CPU_BASED_RDPMC_EXITING;
3629 if (!kvm_mwait_in_guest())
3630 min |= CPU_BASED_MWAIT_EXITING |
3631 CPU_BASED_MONITOR_EXITING;
3633 opt = CPU_BASED_TPR_SHADOW |
3634 CPU_BASED_USE_MSR_BITMAPS |
3635 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS;
3636 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PROCBASED_CTLS,
3637 &_cpu_based_exec_control) < 0)
3639 #ifdef CONFIG_X86_64
3640 if ((_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3641 _cpu_based_exec_control &= ~CPU_BASED_CR8_LOAD_EXITING &
3642 ~CPU_BASED_CR8_STORE_EXITING;
3644 if (_cpu_based_exec_control & CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) {
3646 opt2 = SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
3647 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3648 SECONDARY_EXEC_WBINVD_EXITING |
3649 SECONDARY_EXEC_ENABLE_VPID |
3650 SECONDARY_EXEC_ENABLE_EPT |
3651 SECONDARY_EXEC_UNRESTRICTED_GUEST |
3652 SECONDARY_EXEC_PAUSE_LOOP_EXITING |
3653 SECONDARY_EXEC_RDTSCP |
3654 SECONDARY_EXEC_ENABLE_INVPCID |
3655 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3656 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
3657 SECONDARY_EXEC_SHADOW_VMCS |
3658 SECONDARY_EXEC_XSAVES |
3659 SECONDARY_EXEC_RDSEED |
3660 SECONDARY_EXEC_RDRAND |
3661 SECONDARY_EXEC_ENABLE_PML |
3662 SECONDARY_EXEC_TSC_SCALING |
3663 SECONDARY_EXEC_ENABLE_VMFUNC;
3664 if (adjust_vmx_controls(min2, opt2,
3665 MSR_IA32_VMX_PROCBASED_CTLS2,
3666 &_cpu_based_2nd_exec_control) < 0)
3669 #ifndef CONFIG_X86_64
3670 if (!(_cpu_based_2nd_exec_control &
3671 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
3672 _cpu_based_exec_control &= ~CPU_BASED_TPR_SHADOW;
3675 if (!(_cpu_based_exec_control & CPU_BASED_TPR_SHADOW))
3676 _cpu_based_2nd_exec_control &= ~(
3677 SECONDARY_EXEC_APIC_REGISTER_VIRT |
3678 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
3679 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
3681 if (_cpu_based_2nd_exec_control & SECONDARY_EXEC_ENABLE_EPT) {
3682 /* CR3 accesses and invlpg don't need to cause VM Exits when EPT
3684 _cpu_based_exec_control &= ~(CPU_BASED_CR3_LOAD_EXITING |
3685 CPU_BASED_CR3_STORE_EXITING |
3686 CPU_BASED_INVLPG_EXITING);
3687 rdmsr(MSR_IA32_VMX_EPT_VPID_CAP,
3688 vmx_capability.ept, vmx_capability.vpid);
3691 min = VM_EXIT_SAVE_DEBUG_CONTROLS | VM_EXIT_ACK_INTR_ON_EXIT;
3692 #ifdef CONFIG_X86_64
3693 min |= VM_EXIT_HOST_ADDR_SPACE_SIZE;
3695 opt = VM_EXIT_SAVE_IA32_PAT | VM_EXIT_LOAD_IA32_PAT |
3696 VM_EXIT_CLEAR_BNDCFGS;
3697 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_EXIT_CTLS,
3698 &_vmexit_control) < 0)
3701 min = PIN_BASED_EXT_INTR_MASK | PIN_BASED_NMI_EXITING |
3702 PIN_BASED_VIRTUAL_NMIS;
3703 opt = PIN_BASED_POSTED_INTR | PIN_BASED_VMX_PREEMPTION_TIMER;
3704 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_PINBASED_CTLS,
3705 &_pin_based_exec_control) < 0)
3708 if (cpu_has_broken_vmx_preemption_timer())
3709 _pin_based_exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
3710 if (!(_cpu_based_2nd_exec_control &
3711 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY))
3712 _pin_based_exec_control &= ~PIN_BASED_POSTED_INTR;
3714 min = VM_ENTRY_LOAD_DEBUG_CONTROLS;
3715 opt = VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_BNDCFGS;
3716 if (adjust_vmx_controls(min, opt, MSR_IA32_VMX_ENTRY_CTLS,
3717 &_vmentry_control) < 0)
3720 rdmsr(MSR_IA32_VMX_BASIC, vmx_msr_low, vmx_msr_high);
3722 /* IA-32 SDM Vol 3B: VMCS size is never greater than 4kB. */
3723 if ((vmx_msr_high & 0x1fff) > PAGE_SIZE)
3726 #ifdef CONFIG_X86_64
3727 /* IA-32 SDM Vol 3B: 64-bit CPUs always have VMX_BASIC_MSR[48]==0. */
3728 if (vmx_msr_high & (1u<<16))
3732 /* Require Write-Back (WB) memory type for VMCS accesses. */
3733 if (((vmx_msr_high >> 18) & 15) != 6)
3736 vmcs_conf->size = vmx_msr_high & 0x1fff;
3737 vmcs_conf->order = get_order(vmcs_conf->size);
3738 vmcs_conf->basic_cap = vmx_msr_high & ~0x1fff;
3739 vmcs_conf->revision_id = vmx_msr_low;
3741 vmcs_conf->pin_based_exec_ctrl = _pin_based_exec_control;
3742 vmcs_conf->cpu_based_exec_ctrl = _cpu_based_exec_control;
3743 vmcs_conf->cpu_based_2nd_exec_ctrl = _cpu_based_2nd_exec_control;
3744 vmcs_conf->vmexit_ctrl = _vmexit_control;
3745 vmcs_conf->vmentry_ctrl = _vmentry_control;
3747 cpu_has_load_ia32_efer =
3748 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3749 VM_ENTRY_LOAD_IA32_EFER)
3750 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3751 VM_EXIT_LOAD_IA32_EFER);
3753 cpu_has_load_perf_global_ctrl =
3754 allow_1_setting(MSR_IA32_VMX_ENTRY_CTLS,
3755 VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
3756 && allow_1_setting(MSR_IA32_VMX_EXIT_CTLS,
3757 VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL);
3760 * Some cpus support VM_ENTRY_(LOAD|SAVE)_IA32_PERF_GLOBAL_CTRL
3761 * but due to errata below it can't be used. Workaround is to use
3762 * msr load mechanism to switch IA32_PERF_GLOBAL_CTRL.
3764 * VM Exit May Incorrectly Clear IA32_PERF_GLOBAL_CTRL [34:32]
3769 * BC86,AAY89,BD102 (model 44)
3773 if (cpu_has_load_perf_global_ctrl && boot_cpu_data.x86 == 0x6) {
3774 switch (boot_cpu_data.x86_model) {
3780 cpu_has_load_perf_global_ctrl = false;
3781 printk_once(KERN_WARNING"kvm: VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL "
3782 "does not work properly. Using workaround\n");
3789 if (boot_cpu_has(X86_FEATURE_XSAVES))
3790 rdmsrl(MSR_IA32_XSS, host_xss);
3795 static struct vmcs *alloc_vmcs_cpu(int cpu)
3797 int node = cpu_to_node(cpu);
3801 pages = __alloc_pages_node(node, GFP_KERNEL, vmcs_config.order);
3804 vmcs = page_address(pages);
3805 memset(vmcs, 0, vmcs_config.size);
3806 vmcs->revision_id = vmcs_config.revision_id; /* vmcs revision id */
3810 static struct vmcs *alloc_vmcs(void)
3812 return alloc_vmcs_cpu(raw_smp_processor_id());
3815 static void free_vmcs(struct vmcs *vmcs)
3817 free_pages((unsigned long)vmcs, vmcs_config.order);
3821 * Free a VMCS, but before that VMCLEAR it on the CPU where it was last loaded
3823 static void free_loaded_vmcs(struct loaded_vmcs *loaded_vmcs)
3825 if (!loaded_vmcs->vmcs)
3827 loaded_vmcs_clear(loaded_vmcs);
3828 free_vmcs(loaded_vmcs->vmcs);
3829 loaded_vmcs->vmcs = NULL;
3830 WARN_ON(loaded_vmcs->shadow_vmcs != NULL);
3833 static void free_kvm_area(void)
3837 for_each_possible_cpu(cpu) {
3838 free_vmcs(per_cpu(vmxarea, cpu));
3839 per_cpu(vmxarea, cpu) = NULL;
3843 enum vmcs_field_type {
3844 VMCS_FIELD_TYPE_U16 = 0,
3845 VMCS_FIELD_TYPE_U64 = 1,
3846 VMCS_FIELD_TYPE_U32 = 2,
3847 VMCS_FIELD_TYPE_NATURAL_WIDTH = 3
3850 static inline int vmcs_field_type(unsigned long field)
3852 if (0x1 & field) /* the *_HIGH fields are all 32 bit */
3853 return VMCS_FIELD_TYPE_U32;
3854 return (field >> 13) & 0x3 ;
3857 static inline int vmcs_field_readonly(unsigned long field)
3859 return (((field >> 10) & 0x3) == 1);
3862 static void init_vmcs_shadow_fields(void)
3866 /* No checks for read only fields yet */
3868 for (i = j = 0; i < max_shadow_read_write_fields; i++) {
3869 switch (shadow_read_write_fields[i]) {
3871 if (!kvm_mpx_supported())
3879 shadow_read_write_fields[j] =
3880 shadow_read_write_fields[i];
3883 max_shadow_read_write_fields = j;
3885 /* shadowed fields guest access without vmexit */
3886 for (i = 0; i < max_shadow_read_write_fields; i++) {
3887 unsigned long field = shadow_read_write_fields[i];
3889 clear_bit(field, vmx_vmwrite_bitmap);
3890 clear_bit(field, vmx_vmread_bitmap);
3891 if (vmcs_field_type(field) == VMCS_FIELD_TYPE_U64) {
3892 clear_bit(field + 1, vmx_vmwrite_bitmap);
3893 clear_bit(field + 1, vmx_vmread_bitmap);
3896 for (i = 0; i < max_shadow_read_only_fields; i++) {
3897 unsigned long field = shadow_read_only_fields[i];
3899 clear_bit(field, vmx_vmread_bitmap);
3900 if (vmcs_field_type(field) == VMCS_FIELD_TYPE_U64)
3901 clear_bit(field + 1, vmx_vmread_bitmap);
3905 static __init int alloc_kvm_area(void)
3909 for_each_possible_cpu(cpu) {
3912 vmcs = alloc_vmcs_cpu(cpu);
3918 per_cpu(vmxarea, cpu) = vmcs;
3923 static void fix_pmode_seg(struct kvm_vcpu *vcpu, int seg,
3924 struct kvm_segment *save)
3926 if (!emulate_invalid_guest_state) {
3928 * CS and SS RPL should be equal during guest entry according
3929 * to VMX spec, but in reality it is not always so. Since vcpu
3930 * is in the middle of the transition from real mode to
3931 * protected mode it is safe to assume that RPL 0 is a good
3934 if (seg == VCPU_SREG_CS || seg == VCPU_SREG_SS)
3935 save->selector &= ~SEGMENT_RPL_MASK;
3936 save->dpl = save->selector & SEGMENT_RPL_MASK;
3939 vmx_set_segment(vcpu, save, seg);
3942 static void enter_pmode(struct kvm_vcpu *vcpu)
3944 unsigned long flags;
3945 struct vcpu_vmx *vmx = to_vmx(vcpu);
3948 * Update real mode segment cache. It may be not up-to-date if sement
3949 * register was written while vcpu was in a guest mode.
3951 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
3952 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
3953 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
3954 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
3955 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
3956 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
3958 vmx->rmode.vm86_active = 0;
3960 vmx_segment_cache_clear(vmx);
3962 vmx_set_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
3964 flags = vmcs_readl(GUEST_RFLAGS);
3965 flags &= RMODE_GUEST_OWNED_EFLAGS_BITS;
3966 flags |= vmx->rmode.save_rflags & ~RMODE_GUEST_OWNED_EFLAGS_BITS;
3967 vmcs_writel(GUEST_RFLAGS, flags);
3969 vmcs_writel(GUEST_CR4, (vmcs_readl(GUEST_CR4) & ~X86_CR4_VME) |
3970 (vmcs_readl(CR4_READ_SHADOW) & X86_CR4_VME));
3972 update_exception_bitmap(vcpu);
3974 fix_pmode_seg(vcpu, VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
3975 fix_pmode_seg(vcpu, VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
3976 fix_pmode_seg(vcpu, VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
3977 fix_pmode_seg(vcpu, VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
3978 fix_pmode_seg(vcpu, VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
3979 fix_pmode_seg(vcpu, VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
3982 static void fix_rmode_seg(int seg, struct kvm_segment *save)
3984 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
3985 struct kvm_segment var = *save;
3988 if (seg == VCPU_SREG_CS)
3991 if (!emulate_invalid_guest_state) {
3992 var.selector = var.base >> 4;
3993 var.base = var.base & 0xffff0;
4003 if (save->base & 0xf)
4004 printk_once(KERN_WARNING "kvm: segment base is not "
4005 "paragraph aligned when entering "
4006 "protected mode (seg=%d)", seg);
4009 vmcs_write16(sf->selector, var.selector);
4010 vmcs_writel(sf->base, var.base);
4011 vmcs_write32(sf->limit, var.limit);
4012 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(&var));
4015 static void enter_rmode(struct kvm_vcpu *vcpu)
4017 unsigned long flags;
4018 struct vcpu_vmx *vmx = to_vmx(vcpu);
4020 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_TR], VCPU_SREG_TR);
4021 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_ES], VCPU_SREG_ES);
4022 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_DS], VCPU_SREG_DS);
4023 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_FS], VCPU_SREG_FS);
4024 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_GS], VCPU_SREG_GS);
4025 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_SS], VCPU_SREG_SS);
4026 vmx_get_segment(vcpu, &vmx->rmode.segs[VCPU_SREG_CS], VCPU_SREG_CS);
4028 vmx->rmode.vm86_active = 1;
4031 * Very old userspace does not call KVM_SET_TSS_ADDR before entering
4032 * vcpu. Warn the user that an update is overdue.
4034 if (!vcpu->kvm->arch.tss_addr)
4035 printk_once(KERN_WARNING "kvm: KVM_SET_TSS_ADDR need to be "
4036 "called before entering vcpu\n");
4038 vmx_segment_cache_clear(vmx);
4040 vmcs_writel(GUEST_TR_BASE, vcpu->kvm->arch.tss_addr);
4041 vmcs_write32(GUEST_TR_LIMIT, RMODE_TSS_SIZE - 1);
4042 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
4044 flags = vmcs_readl(GUEST_RFLAGS);
4045 vmx->rmode.save_rflags = flags;
4047 flags |= X86_EFLAGS_IOPL | X86_EFLAGS_VM;
4049 vmcs_writel(GUEST_RFLAGS, flags);
4050 vmcs_writel(GUEST_CR4, vmcs_readl(GUEST_CR4) | X86_CR4_VME);
4051 update_exception_bitmap(vcpu);
4053 fix_rmode_seg(VCPU_SREG_SS, &vmx->rmode.segs[VCPU_SREG_SS]);
4054 fix_rmode_seg(VCPU_SREG_CS, &vmx->rmode.segs[VCPU_SREG_CS]);
4055 fix_rmode_seg(VCPU_SREG_ES, &vmx->rmode.segs[VCPU_SREG_ES]);
4056 fix_rmode_seg(VCPU_SREG_DS, &vmx->rmode.segs[VCPU_SREG_DS]);
4057 fix_rmode_seg(VCPU_SREG_GS, &vmx->rmode.segs[VCPU_SREG_GS]);
4058 fix_rmode_seg(VCPU_SREG_FS, &vmx->rmode.segs[VCPU_SREG_FS]);
4060 kvm_mmu_reset_context(vcpu);
4063 static void vmx_set_efer(struct kvm_vcpu *vcpu, u64 efer)
4065 struct vcpu_vmx *vmx = to_vmx(vcpu);
4066 struct shared_msr_entry *msr = find_msr_entry(vmx, MSR_EFER);
4072 * Force kernel_gs_base reloading before EFER changes, as control
4073 * of this msr depends on is_long_mode().
4075 vmx_load_host_state(to_vmx(vcpu));
4076 vcpu->arch.efer = efer;
4077 if (efer & EFER_LMA) {
4078 vm_entry_controls_setbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4081 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4083 msr->data = efer & ~EFER_LME;
4088 #ifdef CONFIG_X86_64
4090 static void enter_lmode(struct kvm_vcpu *vcpu)
4094 vmx_segment_cache_clear(to_vmx(vcpu));
4096 guest_tr_ar = vmcs_read32(GUEST_TR_AR_BYTES);
4097 if ((guest_tr_ar & VMX_AR_TYPE_MASK) != VMX_AR_TYPE_BUSY_64_TSS) {
4098 pr_debug_ratelimited("%s: tss fixup for long mode. \n",
4100 vmcs_write32(GUEST_TR_AR_BYTES,
4101 (guest_tr_ar & ~VMX_AR_TYPE_MASK)
4102 | VMX_AR_TYPE_BUSY_64_TSS);
4104 vmx_set_efer(vcpu, vcpu->arch.efer | EFER_LMA);
4107 static void exit_lmode(struct kvm_vcpu *vcpu)
4109 vm_entry_controls_clearbit(to_vmx(vcpu), VM_ENTRY_IA32E_MODE);
4110 vmx_set_efer(vcpu, vcpu->arch.efer & ~EFER_LMA);
4115 static inline void __vmx_flush_tlb(struct kvm_vcpu *vcpu, int vpid)
4118 if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
4120 ept_sync_context(construct_eptp(vcpu, vcpu->arch.mmu.root_hpa));
4122 vpid_sync_context(vpid);
4126 static void vmx_flush_tlb(struct kvm_vcpu *vcpu)
4128 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->vpid);
4131 static void vmx_flush_tlb_ept_only(struct kvm_vcpu *vcpu)
4134 vmx_flush_tlb(vcpu);
4137 static void vmx_decache_cr0_guest_bits(struct kvm_vcpu *vcpu)
4139 ulong cr0_guest_owned_bits = vcpu->arch.cr0_guest_owned_bits;
4141 vcpu->arch.cr0 &= ~cr0_guest_owned_bits;
4142 vcpu->arch.cr0 |= vmcs_readl(GUEST_CR0) & cr0_guest_owned_bits;
4145 static void vmx_decache_cr3(struct kvm_vcpu *vcpu)
4147 if (enable_ept && is_paging(vcpu))
4148 vcpu->arch.cr3 = vmcs_readl(GUEST_CR3);
4149 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
4152 static void vmx_decache_cr4_guest_bits(struct kvm_vcpu *vcpu)
4154 ulong cr4_guest_owned_bits = vcpu->arch.cr4_guest_owned_bits;
4156 vcpu->arch.cr4 &= ~cr4_guest_owned_bits;
4157 vcpu->arch.cr4 |= vmcs_readl(GUEST_CR4) & cr4_guest_owned_bits;
4160 static void ept_load_pdptrs(struct kvm_vcpu *vcpu)
4162 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4164 if (!test_bit(VCPU_EXREG_PDPTR,
4165 (unsigned long *)&vcpu->arch.regs_dirty))
4168 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4169 vmcs_write64(GUEST_PDPTR0, mmu->pdptrs[0]);
4170 vmcs_write64(GUEST_PDPTR1, mmu->pdptrs[1]);
4171 vmcs_write64(GUEST_PDPTR2, mmu->pdptrs[2]);
4172 vmcs_write64(GUEST_PDPTR3, mmu->pdptrs[3]);
4176 static void ept_save_pdptrs(struct kvm_vcpu *vcpu)
4178 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
4180 if (is_paging(vcpu) && is_pae(vcpu) && !is_long_mode(vcpu)) {
4181 mmu->pdptrs[0] = vmcs_read64(GUEST_PDPTR0);
4182 mmu->pdptrs[1] = vmcs_read64(GUEST_PDPTR1);
4183 mmu->pdptrs[2] = vmcs_read64(GUEST_PDPTR2);
4184 mmu->pdptrs[3] = vmcs_read64(GUEST_PDPTR3);
4187 __set_bit(VCPU_EXREG_PDPTR,
4188 (unsigned long *)&vcpu->arch.regs_avail);
4189 __set_bit(VCPU_EXREG_PDPTR,
4190 (unsigned long *)&vcpu->arch.regs_dirty);
4193 static bool nested_guest_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4195 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4196 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4197 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
4199 if (to_vmx(vcpu)->nested.nested_vmx_secondary_ctls_high &
4200 SECONDARY_EXEC_UNRESTRICTED_GUEST &&
4201 nested_cpu_has2(vmcs12, SECONDARY_EXEC_UNRESTRICTED_GUEST))
4202 fixed0 &= ~(X86_CR0_PE | X86_CR0_PG);
4204 return fixed_bits_valid(val, fixed0, fixed1);
4207 static bool nested_host_cr0_valid(struct kvm_vcpu *vcpu, unsigned long val)
4209 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed0;
4210 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr0_fixed1;
4212 return fixed_bits_valid(val, fixed0, fixed1);
4215 static bool nested_cr4_valid(struct kvm_vcpu *vcpu, unsigned long val)
4217 u64 fixed0 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed0;
4218 u64 fixed1 = to_vmx(vcpu)->nested.nested_vmx_cr4_fixed1;
4220 return fixed_bits_valid(val, fixed0, fixed1);
4223 /* No difference in the restrictions on guest and host CR4 in VMX operation. */
4224 #define nested_guest_cr4_valid nested_cr4_valid
4225 #define nested_host_cr4_valid nested_cr4_valid
4227 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4);
4229 static void ept_update_paging_mode_cr0(unsigned long *hw_cr0,
4231 struct kvm_vcpu *vcpu)
4233 if (!test_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail))
4234 vmx_decache_cr3(vcpu);
4235 if (!(cr0 & X86_CR0_PG)) {
4236 /* From paging/starting to nonpaging */
4237 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4238 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) |
4239 (CPU_BASED_CR3_LOAD_EXITING |
4240 CPU_BASED_CR3_STORE_EXITING));
4241 vcpu->arch.cr0 = cr0;
4242 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4243 } else if (!is_paging(vcpu)) {
4244 /* From nonpaging to paging */
4245 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL,
4246 vmcs_read32(CPU_BASED_VM_EXEC_CONTROL) &
4247 ~(CPU_BASED_CR3_LOAD_EXITING |
4248 CPU_BASED_CR3_STORE_EXITING));
4249 vcpu->arch.cr0 = cr0;
4250 vmx_set_cr4(vcpu, kvm_read_cr4(vcpu));
4253 if (!(cr0 & X86_CR0_WP))
4254 *hw_cr0 &= ~X86_CR0_WP;
4257 static void vmx_set_cr0(struct kvm_vcpu *vcpu, unsigned long cr0)
4259 struct vcpu_vmx *vmx = to_vmx(vcpu);
4260 unsigned long hw_cr0;
4262 hw_cr0 = (cr0 & ~KVM_GUEST_CR0_MASK);
4263 if (enable_unrestricted_guest)
4264 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON_UNRESTRICTED_GUEST;
4266 hw_cr0 |= KVM_VM_CR0_ALWAYS_ON;
4268 if (vmx->rmode.vm86_active && (cr0 & X86_CR0_PE))
4271 if (!vmx->rmode.vm86_active && !(cr0 & X86_CR0_PE))
4275 #ifdef CONFIG_X86_64
4276 if (vcpu->arch.efer & EFER_LME) {
4277 if (!is_paging(vcpu) && (cr0 & X86_CR0_PG))
4279 if (is_paging(vcpu) && !(cr0 & X86_CR0_PG))
4285 ept_update_paging_mode_cr0(&hw_cr0, cr0, vcpu);
4287 vmcs_writel(CR0_READ_SHADOW, cr0);
4288 vmcs_writel(GUEST_CR0, hw_cr0);
4289 vcpu->arch.cr0 = cr0;
4291 /* depends on vcpu->arch.cr0 to be set to a new value */
4292 vmx->emulation_required = emulation_required(vcpu);
4295 static int get_ept_level(struct kvm_vcpu *vcpu)
4297 if (cpu_has_vmx_ept_5levels() && (cpuid_maxphyaddr(vcpu) > 48))
4302 static u64 construct_eptp(struct kvm_vcpu *vcpu, unsigned long root_hpa)
4304 u64 eptp = VMX_EPTP_MT_WB;
4306 eptp |= (get_ept_level(vcpu) == 5) ? VMX_EPTP_PWL_5 : VMX_EPTP_PWL_4;
4308 if (enable_ept_ad_bits &&
4309 (!is_guest_mode(vcpu) || nested_ept_ad_enabled(vcpu)))
4310 eptp |= VMX_EPTP_AD_ENABLE_BIT;
4311 eptp |= (root_hpa & PAGE_MASK);
4316 static void vmx_set_cr3(struct kvm_vcpu *vcpu, unsigned long cr3)
4318 unsigned long guest_cr3;
4323 eptp = construct_eptp(vcpu, cr3);
4324 vmcs_write64(EPT_POINTER, eptp);
4325 if (is_paging(vcpu) || is_guest_mode(vcpu))
4326 guest_cr3 = kvm_read_cr3(vcpu);
4328 guest_cr3 = vcpu->kvm->arch.ept_identity_map_addr;
4329 ept_load_pdptrs(vcpu);
4332 vmx_flush_tlb(vcpu);
4333 vmcs_writel(GUEST_CR3, guest_cr3);
4336 static int vmx_set_cr4(struct kvm_vcpu *vcpu, unsigned long cr4)
4339 * Pass through host's Machine Check Enable value to hw_cr4, which
4340 * is in force while we are in guest mode. Do not let guests control
4341 * this bit, even if host CR4.MCE == 0.
4343 unsigned long hw_cr4 =
4344 (cr4_read_shadow() & X86_CR4_MCE) |
4345 (cr4 & ~X86_CR4_MCE) |
4346 (to_vmx(vcpu)->rmode.vm86_active ?
4347 KVM_RMODE_VM_CR4_ALWAYS_ON : KVM_PMODE_VM_CR4_ALWAYS_ON);
4349 if (cr4 & X86_CR4_VMXE) {
4351 * To use VMXON (and later other VMX instructions), a guest
4352 * must first be able to turn on cr4.VMXE (see handle_vmon()).
4353 * So basically the check on whether to allow nested VMX
4356 if (!nested_vmx_allowed(vcpu))
4360 if (to_vmx(vcpu)->nested.vmxon && !nested_cr4_valid(vcpu, cr4))
4363 vcpu->arch.cr4 = cr4;
4365 if (!is_paging(vcpu)) {
4366 hw_cr4 &= ~X86_CR4_PAE;
4367 hw_cr4 |= X86_CR4_PSE;
4368 } else if (!(cr4 & X86_CR4_PAE)) {
4369 hw_cr4 &= ~X86_CR4_PAE;
4373 if (!enable_unrestricted_guest && !is_paging(vcpu))
4375 * SMEP/SMAP/PKU is disabled if CPU is in non-paging mode in
4376 * hardware. To emulate this behavior, SMEP/SMAP/PKU needs
4377 * to be manually disabled when guest switches to non-paging
4380 * If !enable_unrestricted_guest, the CPU is always running
4381 * with CR0.PG=1 and CR4 needs to be modified.
4382 * If enable_unrestricted_guest, the CPU automatically
4383 * disables SMEP/SMAP/PKU when the guest sets CR0.PG=0.
4385 hw_cr4 &= ~(X86_CR4_SMEP | X86_CR4_SMAP | X86_CR4_PKE);
4387 vmcs_writel(CR4_READ_SHADOW, cr4);
4388 vmcs_writel(GUEST_CR4, hw_cr4);
4392 static void vmx_get_segment(struct kvm_vcpu *vcpu,
4393 struct kvm_segment *var, int seg)
4395 struct vcpu_vmx *vmx = to_vmx(vcpu);
4398 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4399 *var = vmx->rmode.segs[seg];
4400 if (seg == VCPU_SREG_TR
4401 || var->selector == vmx_read_guest_seg_selector(vmx, seg))
4403 var->base = vmx_read_guest_seg_base(vmx, seg);
4404 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4407 var->base = vmx_read_guest_seg_base(vmx, seg);
4408 var->limit = vmx_read_guest_seg_limit(vmx, seg);
4409 var->selector = vmx_read_guest_seg_selector(vmx, seg);
4410 ar = vmx_read_guest_seg_ar(vmx, seg);
4411 var->unusable = (ar >> 16) & 1;
4412 var->type = ar & 15;
4413 var->s = (ar >> 4) & 1;
4414 var->dpl = (ar >> 5) & 3;
4416 * Some userspaces do not preserve unusable property. Since usable
4417 * segment has to be present according to VMX spec we can use present
4418 * property to amend userspace bug by making unusable segment always
4419 * nonpresent. vmx_segment_access_rights() already marks nonpresent
4420 * segment as unusable.
4422 var->present = !var->unusable;
4423 var->avl = (ar >> 12) & 1;
4424 var->l = (ar >> 13) & 1;
4425 var->db = (ar >> 14) & 1;
4426 var->g = (ar >> 15) & 1;
4429 static u64 vmx_get_segment_base(struct kvm_vcpu *vcpu, int seg)
4431 struct kvm_segment s;
4433 if (to_vmx(vcpu)->rmode.vm86_active) {
4434 vmx_get_segment(vcpu, &s, seg);
4437 return vmx_read_guest_seg_base(to_vmx(vcpu), seg);
4440 static int vmx_get_cpl(struct kvm_vcpu *vcpu)
4442 struct vcpu_vmx *vmx = to_vmx(vcpu);
4444 if (unlikely(vmx->rmode.vm86_active))
4447 int ar = vmx_read_guest_seg_ar(vmx, VCPU_SREG_SS);
4448 return VMX_AR_DPL(ar);
4452 static u32 vmx_segment_access_rights(struct kvm_segment *var)
4456 if (var->unusable || !var->present)
4459 ar = var->type & 15;
4460 ar |= (var->s & 1) << 4;
4461 ar |= (var->dpl & 3) << 5;
4462 ar |= (var->present & 1) << 7;
4463 ar |= (var->avl & 1) << 12;
4464 ar |= (var->l & 1) << 13;
4465 ar |= (var->db & 1) << 14;
4466 ar |= (var->g & 1) << 15;
4472 static void vmx_set_segment(struct kvm_vcpu *vcpu,
4473 struct kvm_segment *var, int seg)
4475 struct vcpu_vmx *vmx = to_vmx(vcpu);
4476 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4478 vmx_segment_cache_clear(vmx);
4480 if (vmx->rmode.vm86_active && seg != VCPU_SREG_LDTR) {
4481 vmx->rmode.segs[seg] = *var;
4482 if (seg == VCPU_SREG_TR)
4483 vmcs_write16(sf->selector, var->selector);
4485 fix_rmode_seg(seg, &vmx->rmode.segs[seg]);
4489 vmcs_writel(sf->base, var->base);
4490 vmcs_write32(sf->limit, var->limit);
4491 vmcs_write16(sf->selector, var->selector);
4494 * Fix the "Accessed" bit in AR field of segment registers for older
4496 * IA32 arch specifies that at the time of processor reset the
4497 * "Accessed" bit in the AR field of segment registers is 1. And qemu
4498 * is setting it to 0 in the userland code. This causes invalid guest
4499 * state vmexit when "unrestricted guest" mode is turned on.
4500 * Fix for this setup issue in cpu_reset is being pushed in the qemu
4501 * tree. Newer qemu binaries with that qemu fix would not need this
4504 if (enable_unrestricted_guest && (seg != VCPU_SREG_LDTR))
4505 var->type |= 0x1; /* Accessed */
4507 vmcs_write32(sf->ar_bytes, vmx_segment_access_rights(var));
4510 vmx->emulation_required = emulation_required(vcpu);
4513 static void vmx_get_cs_db_l_bits(struct kvm_vcpu *vcpu, int *db, int *l)
4515 u32 ar = vmx_read_guest_seg_ar(to_vmx(vcpu), VCPU_SREG_CS);
4517 *db = (ar >> 14) & 1;
4518 *l = (ar >> 13) & 1;
4521 static void vmx_get_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4523 dt->size = vmcs_read32(GUEST_IDTR_LIMIT);
4524 dt->address = vmcs_readl(GUEST_IDTR_BASE);
4527 static void vmx_set_idt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4529 vmcs_write32(GUEST_IDTR_LIMIT, dt->size);
4530 vmcs_writel(GUEST_IDTR_BASE, dt->address);
4533 static void vmx_get_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4535 dt->size = vmcs_read32(GUEST_GDTR_LIMIT);
4536 dt->address = vmcs_readl(GUEST_GDTR_BASE);
4539 static void vmx_set_gdt(struct kvm_vcpu *vcpu, struct desc_ptr *dt)
4541 vmcs_write32(GUEST_GDTR_LIMIT, dt->size);
4542 vmcs_writel(GUEST_GDTR_BASE, dt->address);
4545 static bool rmode_segment_valid(struct kvm_vcpu *vcpu, int seg)
4547 struct kvm_segment var;
4550 vmx_get_segment(vcpu, &var, seg);
4552 if (seg == VCPU_SREG_CS)
4554 ar = vmx_segment_access_rights(&var);
4556 if (var.base != (var.selector << 4))
4558 if (var.limit != 0xffff)
4566 static bool code_segment_valid(struct kvm_vcpu *vcpu)
4568 struct kvm_segment cs;
4569 unsigned int cs_rpl;
4571 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4572 cs_rpl = cs.selector & SEGMENT_RPL_MASK;
4576 if (~cs.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_ACCESSES_MASK))
4580 if (cs.type & VMX_AR_TYPE_WRITEABLE_MASK) {
4581 if (cs.dpl > cs_rpl)
4584 if (cs.dpl != cs_rpl)
4590 /* TODO: Add Reserved field check, this'll require a new member in the kvm_segment_field structure */
4594 static bool stack_segment_valid(struct kvm_vcpu *vcpu)
4596 struct kvm_segment ss;
4597 unsigned int ss_rpl;
4599 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4600 ss_rpl = ss.selector & SEGMENT_RPL_MASK;
4604 if (ss.type != 3 && ss.type != 7)
4608 if (ss.dpl != ss_rpl) /* DPL != RPL */
4616 static bool data_segment_valid(struct kvm_vcpu *vcpu, int seg)
4618 struct kvm_segment var;
4621 vmx_get_segment(vcpu, &var, seg);
4622 rpl = var.selector & SEGMENT_RPL_MASK;
4630 if (~var.type & (VMX_AR_TYPE_CODE_MASK|VMX_AR_TYPE_WRITEABLE_MASK)) {
4631 if (var.dpl < rpl) /* DPL < RPL */
4635 /* TODO: Add other members to kvm_segment_field to allow checking for other access
4641 static bool tr_valid(struct kvm_vcpu *vcpu)
4643 struct kvm_segment tr;
4645 vmx_get_segment(vcpu, &tr, VCPU_SREG_TR);
4649 if (tr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4651 if (tr.type != 3 && tr.type != 11) /* TODO: Check if guest is in IA32e mode */
4659 static bool ldtr_valid(struct kvm_vcpu *vcpu)
4661 struct kvm_segment ldtr;
4663 vmx_get_segment(vcpu, &ldtr, VCPU_SREG_LDTR);
4667 if (ldtr.selector & SEGMENT_TI_MASK) /* TI = 1 */
4677 static bool cs_ss_rpl_check(struct kvm_vcpu *vcpu)
4679 struct kvm_segment cs, ss;
4681 vmx_get_segment(vcpu, &cs, VCPU_SREG_CS);
4682 vmx_get_segment(vcpu, &ss, VCPU_SREG_SS);
4684 return ((cs.selector & SEGMENT_RPL_MASK) ==
4685 (ss.selector & SEGMENT_RPL_MASK));
4689 * Check if guest state is valid. Returns true if valid, false if
4691 * We assume that registers are always usable
4693 static bool guest_state_valid(struct kvm_vcpu *vcpu)
4695 if (enable_unrestricted_guest)
4698 /* real mode guest state checks */
4699 if (!is_protmode(vcpu) || (vmx_get_rflags(vcpu) & X86_EFLAGS_VM)) {
4700 if (!rmode_segment_valid(vcpu, VCPU_SREG_CS))
4702 if (!rmode_segment_valid(vcpu, VCPU_SREG_SS))
4704 if (!rmode_segment_valid(vcpu, VCPU_SREG_DS))
4706 if (!rmode_segment_valid(vcpu, VCPU_SREG_ES))
4708 if (!rmode_segment_valid(vcpu, VCPU_SREG_FS))
4710 if (!rmode_segment_valid(vcpu, VCPU_SREG_GS))
4713 /* protected mode guest state checks */
4714 if (!cs_ss_rpl_check(vcpu))
4716 if (!code_segment_valid(vcpu))
4718 if (!stack_segment_valid(vcpu))
4720 if (!data_segment_valid(vcpu, VCPU_SREG_DS))
4722 if (!data_segment_valid(vcpu, VCPU_SREG_ES))
4724 if (!data_segment_valid(vcpu, VCPU_SREG_FS))
4726 if (!data_segment_valid(vcpu, VCPU_SREG_GS))
4728 if (!tr_valid(vcpu))
4730 if (!ldtr_valid(vcpu))
4734 * - Add checks on RIP
4735 * - Add checks on RFLAGS
4741 static bool page_address_valid(struct kvm_vcpu *vcpu, gpa_t gpa)
4743 return PAGE_ALIGNED(gpa) && !(gpa >> cpuid_maxphyaddr(vcpu));
4746 static int init_rmode_tss(struct kvm *kvm)
4752 idx = srcu_read_lock(&kvm->srcu);
4753 fn = kvm->arch.tss_addr >> PAGE_SHIFT;
4754 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4757 data = TSS_BASE_SIZE + TSS_REDIRECTION_SIZE;
4758 r = kvm_write_guest_page(kvm, fn++, &data,
4759 TSS_IOPB_BASE_OFFSET, sizeof(u16));
4762 r = kvm_clear_guest_page(kvm, fn++, 0, PAGE_SIZE);
4765 r = kvm_clear_guest_page(kvm, fn, 0, PAGE_SIZE);
4769 r = kvm_write_guest_page(kvm, fn, &data,
4770 RMODE_TSS_SIZE - 2 * PAGE_SIZE - 1,
4773 srcu_read_unlock(&kvm->srcu, idx);
4777 static int init_rmode_identity_map(struct kvm *kvm)
4780 kvm_pfn_t identity_map_pfn;
4786 /* Protect kvm->arch.ept_identity_pagetable_done. */
4787 mutex_lock(&kvm->slots_lock);
4789 if (likely(kvm->arch.ept_identity_pagetable_done))
4792 identity_map_pfn = kvm->arch.ept_identity_map_addr >> PAGE_SHIFT;
4794 r = alloc_identity_pagetable(kvm);
4798 idx = srcu_read_lock(&kvm->srcu);
4799 r = kvm_clear_guest_page(kvm, identity_map_pfn, 0, PAGE_SIZE);
4802 /* Set up identity-mapping pagetable for EPT in real mode */
4803 for (i = 0; i < PT32_ENT_PER_PAGE; i++) {
4804 tmp = (i << 22) + (_PAGE_PRESENT | _PAGE_RW | _PAGE_USER |
4805 _PAGE_ACCESSED | _PAGE_DIRTY | _PAGE_PSE);
4806 r = kvm_write_guest_page(kvm, identity_map_pfn,
4807 &tmp, i * sizeof(tmp), sizeof(tmp));
4811 kvm->arch.ept_identity_pagetable_done = true;
4814 srcu_read_unlock(&kvm->srcu, idx);
4817 mutex_unlock(&kvm->slots_lock);
4821 static void seg_setup(int seg)
4823 const struct kvm_vmx_segment_field *sf = &kvm_vmx_segment_fields[seg];
4826 vmcs_write16(sf->selector, 0);
4827 vmcs_writel(sf->base, 0);
4828 vmcs_write32(sf->limit, 0xffff);
4830 if (seg == VCPU_SREG_CS)
4831 ar |= 0x08; /* code segment */
4833 vmcs_write32(sf->ar_bytes, ar);
4836 static int alloc_apic_access_page(struct kvm *kvm)
4841 mutex_lock(&kvm->slots_lock);
4842 if (kvm->arch.apic_access_page_done)
4844 r = __x86_set_memory_region(kvm, APIC_ACCESS_PAGE_PRIVATE_MEMSLOT,
4845 APIC_DEFAULT_PHYS_BASE, PAGE_SIZE);
4849 page = gfn_to_page(kvm, APIC_DEFAULT_PHYS_BASE >> PAGE_SHIFT);
4850 if (is_error_page(page)) {
4856 * Do not pin the page in memory, so that memory hot-unplug
4857 * is able to migrate it.
4860 kvm->arch.apic_access_page_done = true;
4862 mutex_unlock(&kvm->slots_lock);
4866 static int alloc_identity_pagetable(struct kvm *kvm)
4868 /* Called with kvm->slots_lock held. */
4872 BUG_ON(kvm->arch.ept_identity_pagetable_done);
4874 r = __x86_set_memory_region(kvm, IDENTITY_PAGETABLE_PRIVATE_MEMSLOT,
4875 kvm->arch.ept_identity_map_addr, PAGE_SIZE);
4880 static int allocate_vpid(void)
4886 spin_lock(&vmx_vpid_lock);
4887 vpid = find_first_zero_bit(vmx_vpid_bitmap, VMX_NR_VPIDS);
4888 if (vpid < VMX_NR_VPIDS)
4889 __set_bit(vpid, vmx_vpid_bitmap);
4892 spin_unlock(&vmx_vpid_lock);
4896 static void free_vpid(int vpid)
4898 if (!enable_vpid || vpid == 0)
4900 spin_lock(&vmx_vpid_lock);
4901 __clear_bit(vpid, vmx_vpid_bitmap);
4902 spin_unlock(&vmx_vpid_lock);
4905 #define MSR_TYPE_R 1
4906 #define MSR_TYPE_W 2
4907 static void __vmx_disable_intercept_for_msr(unsigned long *msr_bitmap,
4910 int f = sizeof(unsigned long);
4912 if (!cpu_has_vmx_msr_bitmap())
4916 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4917 * have the write-low and read-high bitmap offsets the wrong way round.
4918 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4920 if (msr <= 0x1fff) {
4921 if (type & MSR_TYPE_R)
4923 __clear_bit(msr, msr_bitmap + 0x000 / f);
4925 if (type & MSR_TYPE_W)
4927 __clear_bit(msr, msr_bitmap + 0x800 / f);
4929 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4931 if (type & MSR_TYPE_R)
4933 __clear_bit(msr, msr_bitmap + 0x400 / f);
4935 if (type & MSR_TYPE_W)
4937 __clear_bit(msr, msr_bitmap + 0xc00 / f);
4943 * If a msr is allowed by L0, we should check whether it is allowed by L1.
4944 * The corresponding bit will be cleared unless both of L0 and L1 allow it.
4946 static void nested_vmx_disable_intercept_for_msr(unsigned long *msr_bitmap_l1,
4947 unsigned long *msr_bitmap_nested,
4950 int f = sizeof(unsigned long);
4952 if (!cpu_has_vmx_msr_bitmap()) {
4958 * See Intel PRM Vol. 3, 20.6.9 (MSR-Bitmap Address). Early manuals
4959 * have the write-low and read-high bitmap offsets the wrong way round.
4960 * We can control MSRs 0x00000000-0x00001fff and 0xc0000000-0xc0001fff.
4962 if (msr <= 0x1fff) {
4963 if (type & MSR_TYPE_R &&
4964 !test_bit(msr, msr_bitmap_l1 + 0x000 / f))
4966 __clear_bit(msr, msr_bitmap_nested + 0x000 / f);
4968 if (type & MSR_TYPE_W &&
4969 !test_bit(msr, msr_bitmap_l1 + 0x800 / f))
4971 __clear_bit(msr, msr_bitmap_nested + 0x800 / f);
4973 } else if ((msr >= 0xc0000000) && (msr <= 0xc0001fff)) {
4975 if (type & MSR_TYPE_R &&
4976 !test_bit(msr, msr_bitmap_l1 + 0x400 / f))
4978 __clear_bit(msr, msr_bitmap_nested + 0x400 / f);
4980 if (type & MSR_TYPE_W &&
4981 !test_bit(msr, msr_bitmap_l1 + 0xc00 / f))
4983 __clear_bit(msr, msr_bitmap_nested + 0xc00 / f);
4988 static void vmx_disable_intercept_for_msr(u32 msr, bool longmode_only)
4991 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy,
4992 msr, MSR_TYPE_R | MSR_TYPE_W);
4993 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode,
4994 msr, MSR_TYPE_R | MSR_TYPE_W);
4997 static void vmx_disable_intercept_msr_x2apic(u32 msr, int type, bool apicv_active)
5000 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic_apicv,
5002 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic_apicv,
5005 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_legacy_x2apic,
5007 __vmx_disable_intercept_for_msr(vmx_msr_bitmap_longmode_x2apic,
5012 static bool vmx_get_enable_apicv(struct kvm_vcpu *vcpu)
5014 return enable_apicv;
5017 static void nested_mark_vmcs12_pages_dirty(struct kvm_vcpu *vcpu)
5019 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
5023 * Don't need to mark the APIC access page dirty; it is never
5024 * written to by the CPU during APIC virtualization.
5027 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
5028 gfn = vmcs12->virtual_apic_page_addr >> PAGE_SHIFT;
5029 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5032 if (nested_cpu_has_posted_intr(vmcs12)) {
5033 gfn = vmcs12->posted_intr_desc_addr >> PAGE_SHIFT;
5034 kvm_vcpu_mark_page_dirty(vcpu, gfn);
5039 static void vmx_complete_nested_posted_interrupt(struct kvm_vcpu *vcpu)
5041 struct vcpu_vmx *vmx = to_vmx(vcpu);
5046 if (!vmx->nested.pi_desc || !vmx->nested.pi_pending)
5049 vmx->nested.pi_pending = false;
5050 if (!pi_test_and_clear_on(vmx->nested.pi_desc))
5053 max_irr = find_last_bit((unsigned long *)vmx->nested.pi_desc->pir, 256);
5054 if (max_irr != 256) {
5055 vapic_page = kmap(vmx->nested.virtual_apic_page);
5056 __kvm_apic_update_irr(vmx->nested.pi_desc->pir, vapic_page);
5057 kunmap(vmx->nested.virtual_apic_page);
5059 status = vmcs_read16(GUEST_INTR_STATUS);
5060 if ((u8)max_irr > ((u8)status & 0xff)) {
5062 status |= (u8)max_irr;
5063 vmcs_write16(GUEST_INTR_STATUS, status);
5067 nested_mark_vmcs12_pages_dirty(vcpu);
5070 static inline bool kvm_vcpu_trigger_posted_interrupt(struct kvm_vcpu *vcpu,
5074 int pi_vec = nested ? POSTED_INTR_NESTED_VECTOR : POSTED_INTR_VECTOR;
5076 if (vcpu->mode == IN_GUEST_MODE) {
5078 * The vector of interrupt to be delivered to vcpu had
5079 * been set in PIR before this function.
5081 * Following cases will be reached in this block, and
5082 * we always send a notification event in all cases as
5085 * Case 1: vcpu keeps in non-root mode. Sending a
5086 * notification event posts the interrupt to vcpu.
5088 * Case 2: vcpu exits to root mode and is still
5089 * runnable. PIR will be synced to vIRR before the
5090 * next vcpu entry. Sending a notification event in
5091 * this case has no effect, as vcpu is not in root
5094 * Case 3: vcpu exits to root mode and is blocked.
5095 * vcpu_block() has already synced PIR to vIRR and
5096 * never blocks vcpu if vIRR is not cleared. Therefore,
5097 * a blocked vcpu here does not wait for any requested
5098 * interrupts in PIR, and sending a notification event
5099 * which has no effect is safe here.
5102 apic->send_IPI_mask(get_cpu_mask(vcpu->cpu), pi_vec);
5109 static int vmx_deliver_nested_posted_interrupt(struct kvm_vcpu *vcpu,
5112 struct vcpu_vmx *vmx = to_vmx(vcpu);
5114 if (is_guest_mode(vcpu) &&
5115 vector == vmx->nested.posted_intr_nv) {
5116 /* the PIR and ON have been set by L1. */
5117 kvm_vcpu_trigger_posted_interrupt(vcpu, true);
5119 * If a posted intr is not recognized by hardware,
5120 * we will accomplish it in the next vmentry.
5122 vmx->nested.pi_pending = true;
5123 kvm_make_request(KVM_REQ_EVENT, vcpu);
5129 * Send interrupt to vcpu via posted interrupt way.
5130 * 1. If target vcpu is running(non-root mode), send posted interrupt
5131 * notification to vcpu and hardware will sync PIR to vIRR atomically.
5132 * 2. If target vcpu isn't running(root mode), kick it to pick up the
5133 * interrupt from PIR in next vmentry.
5135 static void vmx_deliver_posted_interrupt(struct kvm_vcpu *vcpu, int vector)
5137 struct vcpu_vmx *vmx = to_vmx(vcpu);
5140 r = vmx_deliver_nested_posted_interrupt(vcpu, vector);
5144 if (pi_test_and_set_pir(vector, &vmx->pi_desc))
5147 /* If a previous notification has sent the IPI, nothing to do. */
5148 if (pi_test_and_set_on(&vmx->pi_desc))
5151 if (!kvm_vcpu_trigger_posted_interrupt(vcpu, false))
5152 kvm_vcpu_kick(vcpu);
5156 * Set up the vmcs's constant host-state fields, i.e., host-state fields that
5157 * will not change in the lifetime of the guest.
5158 * Note that host-state that does change is set elsewhere. E.g., host-state
5159 * that is set differently for each CPU is set in vmx_vcpu_load(), not here.
5161 static void vmx_set_constant_host_state(struct vcpu_vmx *vmx)
5166 unsigned long cr0, cr3, cr4;
5169 WARN_ON(cr0 & X86_CR0_TS);
5170 vmcs_writel(HOST_CR0, cr0); /* 22.2.3 */
5173 * Save the most likely value for this task's CR3 in the VMCS.
5174 * We can't use __get_current_cr3_fast() because we're not atomic.
5177 vmcs_writel(HOST_CR3, cr3); /* 22.2.3 FIXME: shadow tables */
5178 vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
5180 /* Save the most likely value for this task's CR4 in the VMCS. */
5181 cr4 = cr4_read_shadow();
5182 vmcs_writel(HOST_CR4, cr4); /* 22.2.3, 22.2.5 */
5183 vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
5185 vmcs_write16(HOST_CS_SELECTOR, __KERNEL_CS); /* 22.2.4 */
5186 #ifdef CONFIG_X86_64
5188 * Load null selectors, so we can avoid reloading them in
5189 * __vmx_load_host_state(), in case userspace uses the null selectors
5190 * too (the expected case).
5192 vmcs_write16(HOST_DS_SELECTOR, 0);
5193 vmcs_write16(HOST_ES_SELECTOR, 0);
5195 vmcs_write16(HOST_DS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5196 vmcs_write16(HOST_ES_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5198 vmcs_write16(HOST_SS_SELECTOR, __KERNEL_DS); /* 22.2.4 */
5199 vmcs_write16(HOST_TR_SELECTOR, GDT_ENTRY_TSS*8); /* 22.2.4 */
5202 vmcs_writel(HOST_IDTR_BASE, dt.address); /* 22.2.4 */
5203 vmx->host_idt_base = dt.address;
5205 vmcs_writel(HOST_RIP, vmx_return); /* 22.2.5 */
5207 rdmsr(MSR_IA32_SYSENTER_CS, low32, high32);
5208 vmcs_write32(HOST_IA32_SYSENTER_CS, low32);
5209 rdmsrl(MSR_IA32_SYSENTER_EIP, tmpl);
5210 vmcs_writel(HOST_IA32_SYSENTER_EIP, tmpl); /* 22.2.3 */
5212 if (vmcs_config.vmexit_ctrl & VM_EXIT_LOAD_IA32_PAT) {
5213 rdmsr(MSR_IA32_CR_PAT, low32, high32);
5214 vmcs_write64(HOST_IA32_PAT, low32 | ((u64) high32 << 32));
5218 static void set_cr4_guest_host_mask(struct vcpu_vmx *vmx)
5220 vmx->vcpu.arch.cr4_guest_owned_bits = KVM_CR4_GUEST_OWNED_BITS;
5222 vmx->vcpu.arch.cr4_guest_owned_bits |= X86_CR4_PGE;
5223 if (is_guest_mode(&vmx->vcpu))
5224 vmx->vcpu.arch.cr4_guest_owned_bits &=
5225 ~get_vmcs12(&vmx->vcpu)->cr4_guest_host_mask;
5226 vmcs_writel(CR4_GUEST_HOST_MASK, ~vmx->vcpu.arch.cr4_guest_owned_bits);
5229 static u32 vmx_pin_based_exec_ctrl(struct vcpu_vmx *vmx)
5231 u32 pin_based_exec_ctrl = vmcs_config.pin_based_exec_ctrl;
5233 if (!kvm_vcpu_apicv_active(&vmx->vcpu))
5234 pin_based_exec_ctrl &= ~PIN_BASED_POSTED_INTR;
5235 /* Enable the preemption timer dynamically */
5236 pin_based_exec_ctrl &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
5237 return pin_based_exec_ctrl;
5240 static void vmx_refresh_apicv_exec_ctrl(struct kvm_vcpu *vcpu)
5242 struct vcpu_vmx *vmx = to_vmx(vcpu);
5244 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5245 if (cpu_has_secondary_exec_ctrls()) {
5246 if (kvm_vcpu_apicv_active(vcpu))
5247 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
5248 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5249 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5251 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
5252 SECONDARY_EXEC_APIC_REGISTER_VIRT |
5253 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5256 if (cpu_has_vmx_msr_bitmap())
5257 vmx_set_msr_bitmap(vcpu);
5260 static u32 vmx_exec_control(struct vcpu_vmx *vmx)
5262 u32 exec_control = vmcs_config.cpu_based_exec_ctrl;
5264 if (vmx->vcpu.arch.switch_db_regs & KVM_DEBUGREG_WONT_EXIT)
5265 exec_control &= ~CPU_BASED_MOV_DR_EXITING;
5267 if (!cpu_need_tpr_shadow(&vmx->vcpu)) {
5268 exec_control &= ~CPU_BASED_TPR_SHADOW;
5269 #ifdef CONFIG_X86_64
5270 exec_control |= CPU_BASED_CR8_STORE_EXITING |
5271 CPU_BASED_CR8_LOAD_EXITING;
5275 exec_control |= CPU_BASED_CR3_STORE_EXITING |
5276 CPU_BASED_CR3_LOAD_EXITING |
5277 CPU_BASED_INVLPG_EXITING;
5278 return exec_control;
5281 static bool vmx_rdrand_supported(void)
5283 return vmcs_config.cpu_based_2nd_exec_ctrl &
5284 SECONDARY_EXEC_RDRAND;
5287 static bool vmx_rdseed_supported(void)
5289 return vmcs_config.cpu_based_2nd_exec_ctrl &
5290 SECONDARY_EXEC_RDSEED;
5293 static void vmx_compute_secondary_exec_control(struct vcpu_vmx *vmx)
5295 struct kvm_vcpu *vcpu = &vmx->vcpu;
5297 u32 exec_control = vmcs_config.cpu_based_2nd_exec_ctrl;
5298 if (!cpu_need_virtualize_apic_accesses(vcpu))
5299 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
5301 exec_control &= ~SECONDARY_EXEC_ENABLE_VPID;
5303 exec_control &= ~SECONDARY_EXEC_ENABLE_EPT;
5304 enable_unrestricted_guest = 0;
5305 /* Enable INVPCID for non-ept guests may cause performance regression. */
5306 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5308 if (!enable_unrestricted_guest)
5309 exec_control &= ~SECONDARY_EXEC_UNRESTRICTED_GUEST;
5311 exec_control &= ~SECONDARY_EXEC_PAUSE_LOOP_EXITING;
5312 if (!kvm_vcpu_apicv_active(vcpu))
5313 exec_control &= ~(SECONDARY_EXEC_APIC_REGISTER_VIRT |
5314 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY);
5315 exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
5316 /* SECONDARY_EXEC_SHADOW_VMCS is enabled when L1 executes VMPTRLD
5318 We can NOT enable shadow_vmcs here because we don't have yet
5321 exec_control &= ~SECONDARY_EXEC_SHADOW_VMCS;
5324 exec_control &= ~SECONDARY_EXEC_ENABLE_PML;
5326 if (vmx_xsaves_supported()) {
5327 /* Exposing XSAVES only when XSAVE is exposed */
5328 bool xsaves_enabled =
5329 guest_cpuid_has(vcpu, X86_FEATURE_XSAVE) &&
5330 guest_cpuid_has(vcpu, X86_FEATURE_XSAVES);
5332 if (!xsaves_enabled)
5333 exec_control &= ~SECONDARY_EXEC_XSAVES;
5337 vmx->nested.nested_vmx_secondary_ctls_high |=
5338 SECONDARY_EXEC_XSAVES;
5340 vmx->nested.nested_vmx_secondary_ctls_high &=
5341 ~SECONDARY_EXEC_XSAVES;
5345 if (vmx_rdtscp_supported()) {
5346 bool rdtscp_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP);
5347 if (!rdtscp_enabled)
5348 exec_control &= ~SECONDARY_EXEC_RDTSCP;
5352 vmx->nested.nested_vmx_secondary_ctls_high |=
5353 SECONDARY_EXEC_RDTSCP;
5355 vmx->nested.nested_vmx_secondary_ctls_high &=
5356 ~SECONDARY_EXEC_RDTSCP;
5360 if (vmx_invpcid_supported()) {
5361 /* Exposing INVPCID only when PCID is exposed */
5362 bool invpcid_enabled =
5363 guest_cpuid_has(vcpu, X86_FEATURE_INVPCID) &&
5364 guest_cpuid_has(vcpu, X86_FEATURE_PCID);
5366 if (!invpcid_enabled) {
5367 exec_control &= ~SECONDARY_EXEC_ENABLE_INVPCID;
5368 guest_cpuid_clear(vcpu, X86_FEATURE_INVPCID);
5372 if (invpcid_enabled)
5373 vmx->nested.nested_vmx_secondary_ctls_high |=
5374 SECONDARY_EXEC_ENABLE_INVPCID;
5376 vmx->nested.nested_vmx_secondary_ctls_high &=
5377 ~SECONDARY_EXEC_ENABLE_INVPCID;
5381 if (vmx_rdrand_supported()) {
5382 bool rdrand_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDRAND);
5384 exec_control &= ~SECONDARY_EXEC_RDRAND;
5388 vmx->nested.nested_vmx_secondary_ctls_high |=
5389 SECONDARY_EXEC_RDRAND;
5391 vmx->nested.nested_vmx_secondary_ctls_high &=
5392 ~SECONDARY_EXEC_RDRAND;
5396 if (vmx_rdseed_supported()) {
5397 bool rdseed_enabled = guest_cpuid_has(vcpu, X86_FEATURE_RDSEED);
5399 exec_control &= ~SECONDARY_EXEC_RDSEED;
5403 vmx->nested.nested_vmx_secondary_ctls_high |=
5404 SECONDARY_EXEC_RDSEED;
5406 vmx->nested.nested_vmx_secondary_ctls_high &=
5407 ~SECONDARY_EXEC_RDSEED;
5411 vmx->secondary_exec_control = exec_control;
5414 static void ept_set_mmio_spte_mask(void)
5417 * EPT Misconfigurations can be generated if the value of bits 2:0
5418 * of an EPT paging-structure entry is 110b (write/execute).
5420 kvm_mmu_set_mmio_spte_mask(VMX_EPT_RWX_MASK,
5421 VMX_EPT_MISCONFIG_WX_VALUE);
5424 #define VMX_XSS_EXIT_BITMAP 0
5426 * Sets up the vmcs for emulated real mode.
5428 static void vmx_vcpu_setup(struct vcpu_vmx *vmx)
5430 #ifdef CONFIG_X86_64
5436 vmcs_write64(IO_BITMAP_A, __pa(vmx_io_bitmap_a));
5437 vmcs_write64(IO_BITMAP_B, __pa(vmx_io_bitmap_b));
5439 if (enable_shadow_vmcs) {
5440 vmcs_write64(VMREAD_BITMAP, __pa(vmx_vmread_bitmap));
5441 vmcs_write64(VMWRITE_BITMAP, __pa(vmx_vmwrite_bitmap));
5443 if (cpu_has_vmx_msr_bitmap())
5444 vmcs_write64(MSR_BITMAP, __pa(vmx_msr_bitmap_legacy));
5446 vmcs_write64(VMCS_LINK_POINTER, -1ull); /* 22.3.1.5 */
5449 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, vmx_pin_based_exec_ctrl(vmx));
5450 vmx->hv_deadline_tsc = -1;
5452 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, vmx_exec_control(vmx));
5454 if (cpu_has_secondary_exec_ctrls()) {
5455 vmx_compute_secondary_exec_control(vmx);
5456 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
5457 vmx->secondary_exec_control);
5460 if (kvm_vcpu_apicv_active(&vmx->vcpu)) {
5461 vmcs_write64(EOI_EXIT_BITMAP0, 0);
5462 vmcs_write64(EOI_EXIT_BITMAP1, 0);
5463 vmcs_write64(EOI_EXIT_BITMAP2, 0);
5464 vmcs_write64(EOI_EXIT_BITMAP3, 0);
5466 vmcs_write16(GUEST_INTR_STATUS, 0);
5468 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
5469 vmcs_write64(POSTED_INTR_DESC_ADDR, __pa((&vmx->pi_desc)));
5473 vmcs_write32(PLE_GAP, ple_gap);
5474 vmx->ple_window = ple_window;
5475 vmx->ple_window_dirty = true;
5478 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK, 0);
5479 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH, 0);
5480 vmcs_write32(CR3_TARGET_COUNT, 0); /* 22.2.1 */
5482 vmcs_write16(HOST_FS_SELECTOR, 0); /* 22.2.4 */
5483 vmcs_write16(HOST_GS_SELECTOR, 0); /* 22.2.4 */
5484 vmx_set_constant_host_state(vmx);
5485 #ifdef CONFIG_X86_64
5486 rdmsrl(MSR_FS_BASE, a);
5487 vmcs_writel(HOST_FS_BASE, a); /* 22.2.4 */
5488 rdmsrl(MSR_GS_BASE, a);
5489 vmcs_writel(HOST_GS_BASE, a); /* 22.2.4 */
5491 vmcs_writel(HOST_FS_BASE, 0); /* 22.2.4 */
5492 vmcs_writel(HOST_GS_BASE, 0); /* 22.2.4 */
5495 if (cpu_has_vmx_vmfunc())
5496 vmcs_write64(VM_FUNCTION_CONTROL, 0);
5498 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
5499 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, 0);
5500 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
5501 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, 0);
5502 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
5504 if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT)
5505 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
5507 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i) {
5508 u32 index = vmx_msr_index[i];
5509 u32 data_low, data_high;
5512 if (rdmsr_safe(index, &data_low, &data_high) < 0)
5514 if (wrmsr_safe(index, data_low, data_high) < 0)
5516 vmx->guest_msrs[j].index = i;
5517 vmx->guest_msrs[j].data = 0;
5518 vmx->guest_msrs[j].mask = -1ull;
5523 vm_exit_controls_init(vmx, vmcs_config.vmexit_ctrl);
5525 /* 22.2.1, 20.8.1 */
5526 vm_entry_controls_init(vmx, vmcs_config.vmentry_ctrl);
5528 vmx->vcpu.arch.cr0_guest_owned_bits = X86_CR0_TS;
5529 vmcs_writel(CR0_GUEST_HOST_MASK, ~X86_CR0_TS);
5531 set_cr4_guest_host_mask(vmx);
5533 if (vmx_xsaves_supported())
5534 vmcs_write64(XSS_EXIT_BITMAP, VMX_XSS_EXIT_BITMAP);
5537 ASSERT(vmx->pml_pg);
5538 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
5539 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
5543 static void vmx_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event)
5545 struct vcpu_vmx *vmx = to_vmx(vcpu);
5546 struct msr_data apic_base_msr;
5549 vmx->rmode.vm86_active = 0;
5551 vmx->vcpu.arch.regs[VCPU_REGS_RDX] = get_rdx_init_val();
5552 kvm_set_cr8(vcpu, 0);
5555 apic_base_msr.data = APIC_DEFAULT_PHYS_BASE |
5556 MSR_IA32_APICBASE_ENABLE;
5557 if (kvm_vcpu_is_reset_bsp(vcpu))
5558 apic_base_msr.data |= MSR_IA32_APICBASE_BSP;
5559 apic_base_msr.host_initiated = true;
5560 kvm_set_apic_base(vcpu, &apic_base_msr);
5563 vmx_segment_cache_clear(vmx);
5565 seg_setup(VCPU_SREG_CS);
5566 vmcs_write16(GUEST_CS_SELECTOR, 0xf000);
5567 vmcs_writel(GUEST_CS_BASE, 0xffff0000ul);
5569 seg_setup(VCPU_SREG_DS);
5570 seg_setup(VCPU_SREG_ES);
5571 seg_setup(VCPU_SREG_FS);
5572 seg_setup(VCPU_SREG_GS);
5573 seg_setup(VCPU_SREG_SS);
5575 vmcs_write16(GUEST_TR_SELECTOR, 0);
5576 vmcs_writel(GUEST_TR_BASE, 0);
5577 vmcs_write32(GUEST_TR_LIMIT, 0xffff);
5578 vmcs_write32(GUEST_TR_AR_BYTES, 0x008b);
5580 vmcs_write16(GUEST_LDTR_SELECTOR, 0);
5581 vmcs_writel(GUEST_LDTR_BASE, 0);
5582 vmcs_write32(GUEST_LDTR_LIMIT, 0xffff);
5583 vmcs_write32(GUEST_LDTR_AR_BYTES, 0x00082);
5586 vmcs_write32(GUEST_SYSENTER_CS, 0);
5587 vmcs_writel(GUEST_SYSENTER_ESP, 0);
5588 vmcs_writel(GUEST_SYSENTER_EIP, 0);
5589 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
5592 vmcs_writel(GUEST_RFLAGS, 0x02);
5593 kvm_rip_write(vcpu, 0xfff0);
5595 vmcs_writel(GUEST_GDTR_BASE, 0);
5596 vmcs_write32(GUEST_GDTR_LIMIT, 0xffff);
5598 vmcs_writel(GUEST_IDTR_BASE, 0);
5599 vmcs_write32(GUEST_IDTR_LIMIT, 0xffff);
5601 vmcs_write32(GUEST_ACTIVITY_STATE, GUEST_ACTIVITY_ACTIVE);
5602 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO, 0);
5603 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS, 0);
5607 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0); /* 22.2.1 */
5609 if (cpu_has_vmx_tpr_shadow() && !init_event) {
5610 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, 0);
5611 if (cpu_need_tpr_shadow(vcpu))
5612 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR,
5613 __pa(vcpu->arch.apic->regs));
5614 vmcs_write32(TPR_THRESHOLD, 0);
5617 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
5619 if (kvm_vcpu_apicv_active(vcpu))
5620 memset(&vmx->pi_desc, 0, sizeof(struct pi_desc));
5623 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
5625 cr0 = X86_CR0_NW | X86_CR0_CD | X86_CR0_ET;
5626 vmx->vcpu.arch.cr0 = cr0;
5627 vmx_set_cr0(vcpu, cr0); /* enter rmode */
5628 vmx_set_cr4(vcpu, 0);
5629 vmx_set_efer(vcpu, 0);
5631 update_exception_bitmap(vcpu);
5633 vpid_sync_context(vmx->vpid);
5637 * In nested virtualization, check if L1 asked to exit on external interrupts.
5638 * For most existing hypervisors, this will always return true.
5640 static bool nested_exit_on_intr(struct kvm_vcpu *vcpu)
5642 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5643 PIN_BASED_EXT_INTR_MASK;
5647 * In nested virtualization, check if L1 has set
5648 * VM_EXIT_ACK_INTR_ON_EXIT
5650 static bool nested_exit_intr_ack_set(struct kvm_vcpu *vcpu)
5652 return get_vmcs12(vcpu)->vm_exit_controls &
5653 VM_EXIT_ACK_INTR_ON_EXIT;
5656 static bool nested_exit_on_nmi(struct kvm_vcpu *vcpu)
5658 return get_vmcs12(vcpu)->pin_based_vm_exec_control &
5659 PIN_BASED_NMI_EXITING;
5662 static void enable_irq_window(struct kvm_vcpu *vcpu)
5664 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5665 CPU_BASED_VIRTUAL_INTR_PENDING);
5668 static void enable_nmi_window(struct kvm_vcpu *vcpu)
5670 if (vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_STI) {
5671 enable_irq_window(vcpu);
5675 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL,
5676 CPU_BASED_VIRTUAL_NMI_PENDING);
5679 static void vmx_inject_irq(struct kvm_vcpu *vcpu)
5681 struct vcpu_vmx *vmx = to_vmx(vcpu);
5683 int irq = vcpu->arch.interrupt.nr;
5685 trace_kvm_inj_virq(irq);
5687 ++vcpu->stat.irq_injections;
5688 if (vmx->rmode.vm86_active) {
5690 if (vcpu->arch.interrupt.soft)
5691 inc_eip = vcpu->arch.event_exit_inst_len;
5692 if (kvm_inject_realmode_interrupt(vcpu, irq, inc_eip) != EMULATE_DONE)
5693 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5696 intr = irq | INTR_INFO_VALID_MASK;
5697 if (vcpu->arch.interrupt.soft) {
5698 intr |= INTR_TYPE_SOFT_INTR;
5699 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
5700 vmx->vcpu.arch.event_exit_inst_len);
5702 intr |= INTR_TYPE_EXT_INTR;
5703 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, intr);
5706 static void vmx_inject_nmi(struct kvm_vcpu *vcpu)
5708 struct vcpu_vmx *vmx = to_vmx(vcpu);
5710 ++vcpu->stat.nmi_injections;
5711 vmx->loaded_vmcs->nmi_known_unmasked = false;
5713 if (vmx->rmode.vm86_active) {
5714 if (kvm_inject_realmode_interrupt(vcpu, NMI_VECTOR, 0) != EMULATE_DONE)
5715 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
5719 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
5720 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR);
5723 static bool vmx_get_nmi_mask(struct kvm_vcpu *vcpu)
5725 struct vcpu_vmx *vmx = to_vmx(vcpu);
5728 if (vmx->loaded_vmcs->nmi_known_unmasked)
5730 masked = vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) & GUEST_INTR_STATE_NMI;
5731 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5735 static void vmx_set_nmi_mask(struct kvm_vcpu *vcpu, bool masked)
5737 struct vcpu_vmx *vmx = to_vmx(vcpu);
5739 vmx->loaded_vmcs->nmi_known_unmasked = !masked;
5741 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
5742 GUEST_INTR_STATE_NMI);
5744 vmcs_clear_bits(GUEST_INTERRUPTIBILITY_INFO,
5745 GUEST_INTR_STATE_NMI);
5748 static int vmx_nmi_allowed(struct kvm_vcpu *vcpu)
5750 if (to_vmx(vcpu)->nested.nested_run_pending)
5753 return !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5754 (GUEST_INTR_STATE_MOV_SS | GUEST_INTR_STATE_STI
5755 | GUEST_INTR_STATE_NMI));
5758 static int vmx_interrupt_allowed(struct kvm_vcpu *vcpu)
5760 return (!to_vmx(vcpu)->nested.nested_run_pending &&
5761 vmcs_readl(GUEST_RFLAGS) & X86_EFLAGS_IF) &&
5762 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO) &
5763 (GUEST_INTR_STATE_STI | GUEST_INTR_STATE_MOV_SS));
5766 static int vmx_set_tss_addr(struct kvm *kvm, unsigned int addr)
5770 ret = x86_set_memory_region(kvm, TSS_PRIVATE_MEMSLOT, addr,
5774 kvm->arch.tss_addr = addr;
5775 return init_rmode_tss(kvm);
5778 static bool rmode_exception(struct kvm_vcpu *vcpu, int vec)
5783 * Update instruction length as we may reinject the exception
5784 * from user space while in guest debugging mode.
5786 to_vmx(vcpu)->vcpu.arch.event_exit_inst_len =
5787 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5788 if (vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
5792 if (vcpu->guest_debug &
5793 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
5810 static int handle_rmode_exception(struct kvm_vcpu *vcpu,
5811 int vec, u32 err_code)
5814 * Instruction with address size override prefix opcode 0x67
5815 * Cause the #SS fault with 0 error code in VM86 mode.
5817 if (((vec == GP_VECTOR) || (vec == SS_VECTOR)) && err_code == 0) {
5818 if (emulate_instruction(vcpu, 0) == EMULATE_DONE) {
5819 if (vcpu->arch.halt_request) {
5820 vcpu->arch.halt_request = 0;
5821 return kvm_vcpu_halt(vcpu);
5829 * Forward all other exceptions that are valid in real mode.
5830 * FIXME: Breaks guest debugging in real mode, needs to be fixed with
5831 * the required debugging infrastructure rework.
5833 kvm_queue_exception(vcpu, vec);
5838 * Trigger machine check on the host. We assume all the MSRs are already set up
5839 * by the CPU and that we still run on the same CPU as the MCE occurred on.
5840 * We pass a fake environment to the machine check handler because we want
5841 * the guest to be always treated like user space, no matter what context
5842 * it used internally.
5844 static void kvm_machine_check(void)
5846 #if defined(CONFIG_X86_MCE) && defined(CONFIG_X86_64)
5847 struct pt_regs regs = {
5848 .cs = 3, /* Fake ring 3 no matter what the guest ran on */
5849 .flags = X86_EFLAGS_IF,
5852 do_machine_check(®s, 0);
5856 static int handle_machine_check(struct kvm_vcpu *vcpu)
5858 /* already handled by vcpu_run */
5862 static int handle_exception(struct kvm_vcpu *vcpu)
5864 struct vcpu_vmx *vmx = to_vmx(vcpu);
5865 struct kvm_run *kvm_run = vcpu->run;
5866 u32 intr_info, ex_no, error_code;
5867 unsigned long cr2, rip, dr6;
5869 enum emulation_result er;
5871 vect_info = vmx->idt_vectoring_info;
5872 intr_info = vmx->exit_intr_info;
5874 if (is_machine_check(intr_info))
5875 return handle_machine_check(vcpu);
5877 if (is_nmi(intr_info))
5878 return 1; /* already handled by vmx_vcpu_run() */
5880 if (is_invalid_opcode(intr_info)) {
5881 if (is_guest_mode(vcpu)) {
5882 kvm_queue_exception(vcpu, UD_VECTOR);
5885 er = emulate_instruction(vcpu, EMULTYPE_TRAP_UD);
5886 if (er != EMULATE_DONE)
5887 kvm_queue_exception(vcpu, UD_VECTOR);
5892 if (intr_info & INTR_INFO_DELIVER_CODE_MASK)
5893 error_code = vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
5896 * The #PF with PFEC.RSVD = 1 indicates the guest is accessing
5897 * MMIO, it is better to report an internal error.
5898 * See the comments in vmx_handle_exit.
5900 if ((vect_info & VECTORING_INFO_VALID_MASK) &&
5901 !(is_page_fault(intr_info) && !(error_code & PFERR_RSVD_MASK))) {
5902 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
5903 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_SIMUL_EX;
5904 vcpu->run->internal.ndata = 3;
5905 vcpu->run->internal.data[0] = vect_info;
5906 vcpu->run->internal.data[1] = intr_info;
5907 vcpu->run->internal.data[2] = error_code;
5911 if (is_page_fault(intr_info)) {
5912 cr2 = vmcs_readl(EXIT_QUALIFICATION);
5913 /* EPT won't cause page fault directly */
5914 WARN_ON_ONCE(!vcpu->arch.apf.host_apf_reason && enable_ept);
5915 return kvm_handle_page_fault(vcpu, error_code, cr2, NULL, 0,
5919 ex_no = intr_info & INTR_INFO_VECTOR_MASK;
5921 if (vmx->rmode.vm86_active && rmode_exception(vcpu, ex_no))
5922 return handle_rmode_exception(vcpu, ex_no, error_code);
5926 kvm_queue_exception_e(vcpu, AC_VECTOR, error_code);
5929 dr6 = vmcs_readl(EXIT_QUALIFICATION);
5930 if (!(vcpu->guest_debug &
5931 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))) {
5932 vcpu->arch.dr6 &= ~15;
5933 vcpu->arch.dr6 |= dr6 | DR6_RTM;
5934 if (!(dr6 & ~DR6_RESERVED)) /* icebp */
5935 skip_emulated_instruction(vcpu);
5937 kvm_queue_exception(vcpu, DB_VECTOR);
5940 kvm_run->debug.arch.dr6 = dr6 | DR6_FIXED_1;
5941 kvm_run->debug.arch.dr7 = vmcs_readl(GUEST_DR7);
5945 * Update instruction length as we may reinject #BP from
5946 * user space while in guest debugging mode. Reading it for
5947 * #DB as well causes no harm, it is not used in that case.
5949 vmx->vcpu.arch.event_exit_inst_len =
5950 vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
5951 kvm_run->exit_reason = KVM_EXIT_DEBUG;
5952 rip = kvm_rip_read(vcpu);
5953 kvm_run->debug.arch.pc = vmcs_readl(GUEST_CS_BASE) + rip;
5954 kvm_run->debug.arch.exception = ex_no;
5957 kvm_run->exit_reason = KVM_EXIT_EXCEPTION;
5958 kvm_run->ex.exception = ex_no;
5959 kvm_run->ex.error_code = error_code;
5965 static int handle_external_interrupt(struct kvm_vcpu *vcpu)
5967 ++vcpu->stat.irq_exits;
5971 static int handle_triple_fault(struct kvm_vcpu *vcpu)
5973 vcpu->run->exit_reason = KVM_EXIT_SHUTDOWN;
5974 vcpu->mmio_needed = 0;
5978 static int handle_io(struct kvm_vcpu *vcpu)
5980 unsigned long exit_qualification;
5981 int size, in, string, ret;
5984 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
5985 string = (exit_qualification & 16) != 0;
5986 in = (exit_qualification & 8) != 0;
5988 ++vcpu->stat.io_exits;
5991 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
5993 port = exit_qualification >> 16;
5994 size = (exit_qualification & 7) + 1;
5996 ret = kvm_skip_emulated_instruction(vcpu);
5999 * TODO: we might be squashing a KVM_GUESTDBG_SINGLESTEP-triggered
6000 * KVM_EXIT_DEBUG here.
6002 return kvm_fast_pio_out(vcpu, size, port) && ret;
6006 vmx_patch_hypercall(struct kvm_vcpu *vcpu, unsigned char *hypercall)
6009 * Patch in the VMCALL instruction:
6011 hypercall[0] = 0x0f;
6012 hypercall[1] = 0x01;
6013 hypercall[2] = 0xc1;
6016 /* called to set cr0 as appropriate for a mov-to-cr0 exit. */
6017 static int handle_set_cr0(struct kvm_vcpu *vcpu, unsigned long val)
6019 if (is_guest_mode(vcpu)) {
6020 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6021 unsigned long orig_val = val;
6024 * We get here when L2 changed cr0 in a way that did not change
6025 * any of L1's shadowed bits (see nested_vmx_exit_handled_cr),
6026 * but did change L0 shadowed bits. So we first calculate the
6027 * effective cr0 value that L1 would like to write into the
6028 * hardware. It consists of the L2-owned bits from the new
6029 * value combined with the L1-owned bits from L1's guest_cr0.
6031 val = (val & ~vmcs12->cr0_guest_host_mask) |
6032 (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask);
6034 if (!nested_guest_cr0_valid(vcpu, val))
6037 if (kvm_set_cr0(vcpu, val))
6039 vmcs_writel(CR0_READ_SHADOW, orig_val);
6042 if (to_vmx(vcpu)->nested.vmxon &&
6043 !nested_host_cr0_valid(vcpu, val))
6046 return kvm_set_cr0(vcpu, val);
6050 static int handle_set_cr4(struct kvm_vcpu *vcpu, unsigned long val)
6052 if (is_guest_mode(vcpu)) {
6053 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
6054 unsigned long orig_val = val;
6056 /* analogously to handle_set_cr0 */
6057 val = (val & ~vmcs12->cr4_guest_host_mask) |
6058 (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask);
6059 if (kvm_set_cr4(vcpu, val))
6061 vmcs_writel(CR4_READ_SHADOW, orig_val);
6064 return kvm_set_cr4(vcpu, val);
6067 static int handle_cr(struct kvm_vcpu *vcpu)
6069 unsigned long exit_qualification, val;
6075 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6076 cr = exit_qualification & 15;
6077 reg = (exit_qualification >> 8) & 15;
6078 switch ((exit_qualification >> 4) & 3) {
6079 case 0: /* mov to cr */
6080 val = kvm_register_readl(vcpu, reg);
6081 trace_kvm_cr_write(cr, val);
6084 err = handle_set_cr0(vcpu, val);
6085 return kvm_complete_insn_gp(vcpu, err);
6087 err = kvm_set_cr3(vcpu, val);
6088 return kvm_complete_insn_gp(vcpu, err);
6090 err = handle_set_cr4(vcpu, val);
6091 return kvm_complete_insn_gp(vcpu, err);
6093 u8 cr8_prev = kvm_get_cr8(vcpu);
6095 err = kvm_set_cr8(vcpu, cr8);
6096 ret = kvm_complete_insn_gp(vcpu, err);
6097 if (lapic_in_kernel(vcpu))
6099 if (cr8_prev <= cr8)
6102 * TODO: we might be squashing a
6103 * KVM_GUESTDBG_SINGLESTEP-triggered
6104 * KVM_EXIT_DEBUG here.
6106 vcpu->run->exit_reason = KVM_EXIT_SET_TPR;
6112 WARN_ONCE(1, "Guest should always own CR0.TS");
6113 vmx_set_cr0(vcpu, kvm_read_cr0_bits(vcpu, ~X86_CR0_TS));
6114 trace_kvm_cr_write(0, kvm_read_cr0(vcpu));
6115 return kvm_skip_emulated_instruction(vcpu);
6116 case 1: /*mov from cr*/
6119 val = kvm_read_cr3(vcpu);
6120 kvm_register_write(vcpu, reg, val);
6121 trace_kvm_cr_read(cr, val);
6122 return kvm_skip_emulated_instruction(vcpu);
6124 val = kvm_get_cr8(vcpu);
6125 kvm_register_write(vcpu, reg, val);
6126 trace_kvm_cr_read(cr, val);
6127 return kvm_skip_emulated_instruction(vcpu);
6131 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
6132 trace_kvm_cr_write(0, (kvm_read_cr0(vcpu) & ~0xful) | val);
6133 kvm_lmsw(vcpu, val);
6135 return kvm_skip_emulated_instruction(vcpu);
6139 vcpu->run->exit_reason = 0;
6140 vcpu_unimpl(vcpu, "unhandled control register: op %d cr %d\n",
6141 (int)(exit_qualification >> 4) & 3, cr);
6145 static int handle_dr(struct kvm_vcpu *vcpu)
6147 unsigned long exit_qualification;
6150 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6151 dr = exit_qualification & DEBUG_REG_ACCESS_NUM;
6153 /* First, if DR does not exist, trigger UD */
6154 if (!kvm_require_dr(vcpu, dr))
6157 /* Do not handle if the CPL > 0, will trigger GP on re-entry */
6158 if (!kvm_require_cpl(vcpu, 0))
6160 dr7 = vmcs_readl(GUEST_DR7);
6163 * As the vm-exit takes precedence over the debug trap, we
6164 * need to emulate the latter, either for the host or the
6165 * guest debugging itself.
6167 if (vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) {
6168 vcpu->run->debug.arch.dr6 = vcpu->arch.dr6;
6169 vcpu->run->debug.arch.dr7 = dr7;
6170 vcpu->run->debug.arch.pc = kvm_get_linear_rip(vcpu);
6171 vcpu->run->debug.arch.exception = DB_VECTOR;
6172 vcpu->run->exit_reason = KVM_EXIT_DEBUG;
6175 vcpu->arch.dr6 &= ~15;
6176 vcpu->arch.dr6 |= DR6_BD | DR6_RTM;
6177 kvm_queue_exception(vcpu, DB_VECTOR);
6182 if (vcpu->guest_debug == 0) {
6183 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6184 CPU_BASED_MOV_DR_EXITING);
6187 * No more DR vmexits; force a reload of the debug registers
6188 * and reenter on this instruction. The next vmexit will
6189 * retrieve the full state of the debug registers.
6191 vcpu->arch.switch_db_regs |= KVM_DEBUGREG_WONT_EXIT;
6195 reg = DEBUG_REG_ACCESS_REG(exit_qualification);
6196 if (exit_qualification & TYPE_MOV_FROM_DR) {
6199 if (kvm_get_dr(vcpu, dr, &val))
6201 kvm_register_write(vcpu, reg, val);
6203 if (kvm_set_dr(vcpu, dr, kvm_register_readl(vcpu, reg)))
6206 return kvm_skip_emulated_instruction(vcpu);
6209 static u64 vmx_get_dr6(struct kvm_vcpu *vcpu)
6211 return vcpu->arch.dr6;
6214 static void vmx_set_dr6(struct kvm_vcpu *vcpu, unsigned long val)
6218 static void vmx_sync_dirty_debug_regs(struct kvm_vcpu *vcpu)
6220 get_debugreg(vcpu->arch.db[0], 0);
6221 get_debugreg(vcpu->arch.db[1], 1);
6222 get_debugreg(vcpu->arch.db[2], 2);
6223 get_debugreg(vcpu->arch.db[3], 3);
6224 get_debugreg(vcpu->arch.dr6, 6);
6225 vcpu->arch.dr7 = vmcs_readl(GUEST_DR7);
6227 vcpu->arch.switch_db_regs &= ~KVM_DEBUGREG_WONT_EXIT;
6228 vmcs_set_bits(CPU_BASED_VM_EXEC_CONTROL, CPU_BASED_MOV_DR_EXITING);
6231 static void vmx_set_dr7(struct kvm_vcpu *vcpu, unsigned long val)
6233 vmcs_writel(GUEST_DR7, val);
6236 static int handle_cpuid(struct kvm_vcpu *vcpu)
6238 return kvm_emulate_cpuid(vcpu);
6241 static int handle_rdmsr(struct kvm_vcpu *vcpu)
6243 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6244 struct msr_data msr_info;
6246 msr_info.index = ecx;
6247 msr_info.host_initiated = false;
6248 if (vmx_get_msr(vcpu, &msr_info)) {
6249 trace_kvm_msr_read_ex(ecx);
6250 kvm_inject_gp(vcpu, 0);
6254 trace_kvm_msr_read(ecx, msr_info.data);
6256 /* FIXME: handling of bits 32:63 of rax, rdx */
6257 vcpu->arch.regs[VCPU_REGS_RAX] = msr_info.data & -1u;
6258 vcpu->arch.regs[VCPU_REGS_RDX] = (msr_info.data >> 32) & -1u;
6259 return kvm_skip_emulated_instruction(vcpu);
6262 static int handle_wrmsr(struct kvm_vcpu *vcpu)
6264 struct msr_data msr;
6265 u32 ecx = vcpu->arch.regs[VCPU_REGS_RCX];
6266 u64 data = (vcpu->arch.regs[VCPU_REGS_RAX] & -1u)
6267 | ((u64)(vcpu->arch.regs[VCPU_REGS_RDX] & -1u) << 32);
6271 msr.host_initiated = false;
6272 if (kvm_set_msr(vcpu, &msr) != 0) {
6273 trace_kvm_msr_write_ex(ecx, data);
6274 kvm_inject_gp(vcpu, 0);
6278 trace_kvm_msr_write(ecx, data);
6279 return kvm_skip_emulated_instruction(vcpu);
6282 static int handle_tpr_below_threshold(struct kvm_vcpu *vcpu)
6284 kvm_apic_update_ppr(vcpu);
6288 static int handle_interrupt_window(struct kvm_vcpu *vcpu)
6290 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6291 CPU_BASED_VIRTUAL_INTR_PENDING);
6293 kvm_make_request(KVM_REQ_EVENT, vcpu);
6295 ++vcpu->stat.irq_window_exits;
6299 static int handle_halt(struct kvm_vcpu *vcpu)
6301 return kvm_emulate_halt(vcpu);
6304 static int handle_vmcall(struct kvm_vcpu *vcpu)
6306 return kvm_emulate_hypercall(vcpu);
6309 static int handle_invd(struct kvm_vcpu *vcpu)
6311 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6314 static int handle_invlpg(struct kvm_vcpu *vcpu)
6316 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6318 kvm_mmu_invlpg(vcpu, exit_qualification);
6319 return kvm_skip_emulated_instruction(vcpu);
6322 static int handle_rdpmc(struct kvm_vcpu *vcpu)
6326 err = kvm_rdpmc(vcpu);
6327 return kvm_complete_insn_gp(vcpu, err);
6330 static int handle_wbinvd(struct kvm_vcpu *vcpu)
6332 return kvm_emulate_wbinvd(vcpu);
6335 static int handle_xsetbv(struct kvm_vcpu *vcpu)
6337 u64 new_bv = kvm_read_edx_eax(vcpu);
6338 u32 index = kvm_register_read(vcpu, VCPU_REGS_RCX);
6340 if (kvm_set_xcr(vcpu, index, new_bv) == 0)
6341 return kvm_skip_emulated_instruction(vcpu);
6345 static int handle_xsaves(struct kvm_vcpu *vcpu)
6347 kvm_skip_emulated_instruction(vcpu);
6348 WARN(1, "this should never happen\n");
6352 static int handle_xrstors(struct kvm_vcpu *vcpu)
6354 kvm_skip_emulated_instruction(vcpu);
6355 WARN(1, "this should never happen\n");
6359 static int handle_apic_access(struct kvm_vcpu *vcpu)
6361 if (likely(fasteoi)) {
6362 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6363 int access_type, offset;
6365 access_type = exit_qualification & APIC_ACCESS_TYPE;
6366 offset = exit_qualification & APIC_ACCESS_OFFSET;
6368 * Sane guest uses MOV to write EOI, with written value
6369 * not cared. So make a short-circuit here by avoiding
6370 * heavy instruction emulation.
6372 if ((access_type == TYPE_LINEAR_APIC_INST_WRITE) &&
6373 (offset == APIC_EOI)) {
6374 kvm_lapic_set_eoi(vcpu);
6375 return kvm_skip_emulated_instruction(vcpu);
6378 return emulate_instruction(vcpu, 0) == EMULATE_DONE;
6381 static int handle_apic_eoi_induced(struct kvm_vcpu *vcpu)
6383 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6384 int vector = exit_qualification & 0xff;
6386 /* EOI-induced VM exit is trap-like and thus no need to adjust IP */
6387 kvm_apic_set_eoi_accelerated(vcpu, vector);
6391 static int handle_apic_write(struct kvm_vcpu *vcpu)
6393 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6394 u32 offset = exit_qualification & 0xfff;
6396 /* APIC-write VM exit is trap-like and thus no need to adjust IP */
6397 kvm_apic_write_nodecode(vcpu, offset);
6401 static int handle_task_switch(struct kvm_vcpu *vcpu)
6403 struct vcpu_vmx *vmx = to_vmx(vcpu);
6404 unsigned long exit_qualification;
6405 bool has_error_code = false;
6408 int reason, type, idt_v, idt_index;
6410 idt_v = (vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK);
6411 idt_index = (vmx->idt_vectoring_info & VECTORING_INFO_VECTOR_MASK);
6412 type = (vmx->idt_vectoring_info & VECTORING_INFO_TYPE_MASK);
6414 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6416 reason = (u32)exit_qualification >> 30;
6417 if (reason == TASK_SWITCH_GATE && idt_v) {
6419 case INTR_TYPE_NMI_INTR:
6420 vcpu->arch.nmi_injected = false;
6421 vmx_set_nmi_mask(vcpu, true);
6423 case INTR_TYPE_EXT_INTR:
6424 case INTR_TYPE_SOFT_INTR:
6425 kvm_clear_interrupt_queue(vcpu);
6427 case INTR_TYPE_HARD_EXCEPTION:
6428 if (vmx->idt_vectoring_info &
6429 VECTORING_INFO_DELIVER_CODE_MASK) {
6430 has_error_code = true;
6432 vmcs_read32(IDT_VECTORING_ERROR_CODE);
6435 case INTR_TYPE_SOFT_EXCEPTION:
6436 kvm_clear_exception_queue(vcpu);
6442 tss_selector = exit_qualification;
6444 if (!idt_v || (type != INTR_TYPE_HARD_EXCEPTION &&
6445 type != INTR_TYPE_EXT_INTR &&
6446 type != INTR_TYPE_NMI_INTR))
6447 skip_emulated_instruction(vcpu);
6449 if (kvm_task_switch(vcpu, tss_selector,
6450 type == INTR_TYPE_SOFT_INTR ? idt_index : -1, reason,
6451 has_error_code, error_code) == EMULATE_FAIL) {
6452 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6453 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6454 vcpu->run->internal.ndata = 0;
6459 * TODO: What about debug traps on tss switch?
6460 * Are we supposed to inject them and update dr6?
6466 static int handle_ept_violation(struct kvm_vcpu *vcpu)
6468 unsigned long exit_qualification;
6472 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
6475 * EPT violation happened while executing iret from NMI,
6476 * "blocked by NMI" bit has to be set before next VM entry.
6477 * There are errata that may cause this bit to not be set:
6480 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
6481 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
6482 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO, GUEST_INTR_STATE_NMI);
6484 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6485 trace_kvm_page_fault(gpa, exit_qualification);
6487 /* Is it a read fault? */
6488 error_code = (exit_qualification & EPT_VIOLATION_ACC_READ)
6489 ? PFERR_USER_MASK : 0;
6490 /* Is it a write fault? */
6491 error_code |= (exit_qualification & EPT_VIOLATION_ACC_WRITE)
6492 ? PFERR_WRITE_MASK : 0;
6493 /* Is it a fetch fault? */
6494 error_code |= (exit_qualification & EPT_VIOLATION_ACC_INSTR)
6495 ? PFERR_FETCH_MASK : 0;
6496 /* ept page table entry is present? */
6497 error_code |= (exit_qualification &
6498 (EPT_VIOLATION_READABLE | EPT_VIOLATION_WRITABLE |
6499 EPT_VIOLATION_EXECUTABLE))
6500 ? PFERR_PRESENT_MASK : 0;
6502 error_code |= (exit_qualification & 0x100) != 0 ?
6503 PFERR_GUEST_FINAL_MASK : PFERR_GUEST_PAGE_MASK;
6505 vcpu->arch.exit_qualification = exit_qualification;
6506 return kvm_mmu_page_fault(vcpu, gpa, error_code, NULL, 0);
6509 static int handle_ept_misconfig(struct kvm_vcpu *vcpu)
6515 * A nested guest cannot optimize MMIO vmexits, because we have an
6516 * nGPA here instead of the required GPA.
6518 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS);
6519 if (!is_guest_mode(vcpu) &&
6520 !kvm_io_bus_write(vcpu, KVM_FAST_MMIO_BUS, gpa, 0, NULL)) {
6521 trace_kvm_fast_mmio(gpa);
6522 return kvm_skip_emulated_instruction(vcpu);
6525 ret = kvm_mmu_page_fault(vcpu, gpa, PFERR_RSVD_MASK, NULL, 0);
6529 /* It is the real ept misconfig */
6532 vcpu->run->exit_reason = KVM_EXIT_UNKNOWN;
6533 vcpu->run->hw.hardware_exit_reason = EXIT_REASON_EPT_MISCONFIG;
6538 static int handle_nmi_window(struct kvm_vcpu *vcpu)
6540 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
6541 CPU_BASED_VIRTUAL_NMI_PENDING);
6542 ++vcpu->stat.nmi_window_exits;
6543 kvm_make_request(KVM_REQ_EVENT, vcpu);
6548 static int handle_invalid_guest_state(struct kvm_vcpu *vcpu)
6550 struct vcpu_vmx *vmx = to_vmx(vcpu);
6551 enum emulation_result err = EMULATE_DONE;
6554 bool intr_window_requested;
6555 unsigned count = 130;
6557 cpu_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
6558 intr_window_requested = cpu_exec_ctrl & CPU_BASED_VIRTUAL_INTR_PENDING;
6560 while (vmx->emulation_required && count-- != 0) {
6561 if (intr_window_requested && vmx_interrupt_allowed(vcpu))
6562 return handle_interrupt_window(&vmx->vcpu);
6564 if (kvm_test_request(KVM_REQ_EVENT, vcpu))
6567 err = emulate_instruction(vcpu, EMULTYPE_NO_REEXECUTE);
6569 if (err == EMULATE_USER_EXIT) {
6570 ++vcpu->stat.mmio_exits;
6575 if (err != EMULATE_DONE) {
6576 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
6577 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_EMULATION;
6578 vcpu->run->internal.ndata = 0;
6582 if (vcpu->arch.halt_request) {
6583 vcpu->arch.halt_request = 0;
6584 ret = kvm_vcpu_halt(vcpu);
6588 if (signal_pending(current))
6598 static int __grow_ple_window(int val)
6600 if (ple_window_grow < 1)
6603 val = min(val, ple_window_actual_max);
6605 if (ple_window_grow < ple_window)
6606 val *= ple_window_grow;
6608 val += ple_window_grow;
6613 static int __shrink_ple_window(int val, int modifier, int minimum)
6618 if (modifier < ple_window)
6623 return max(val, minimum);
6626 static void grow_ple_window(struct kvm_vcpu *vcpu)
6628 struct vcpu_vmx *vmx = to_vmx(vcpu);
6629 int old = vmx->ple_window;
6631 vmx->ple_window = __grow_ple_window(old);
6633 if (vmx->ple_window != old)
6634 vmx->ple_window_dirty = true;
6636 trace_kvm_ple_window_grow(vcpu->vcpu_id, vmx->ple_window, old);
6639 static void shrink_ple_window(struct kvm_vcpu *vcpu)
6641 struct vcpu_vmx *vmx = to_vmx(vcpu);
6642 int old = vmx->ple_window;
6644 vmx->ple_window = __shrink_ple_window(old,
6645 ple_window_shrink, ple_window);
6647 if (vmx->ple_window != old)
6648 vmx->ple_window_dirty = true;
6650 trace_kvm_ple_window_shrink(vcpu->vcpu_id, vmx->ple_window, old);
6654 * ple_window_actual_max is computed to be one grow_ple_window() below
6655 * ple_window_max. (See __grow_ple_window for the reason.)
6656 * This prevents overflows, because ple_window_max is int.
6657 * ple_window_max effectively rounded down to a multiple of ple_window_grow in
6659 * ple_window_max is also prevented from setting vmx->ple_window < ple_window.
6661 static void update_ple_window_actual_max(void)
6663 ple_window_actual_max =
6664 __shrink_ple_window(max(ple_window_max, ple_window),
6665 ple_window_grow, INT_MIN);
6669 * Handler for POSTED_INTERRUPT_WAKEUP_VECTOR.
6671 static void wakeup_handler(void)
6673 struct kvm_vcpu *vcpu;
6674 int cpu = smp_processor_id();
6676 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6677 list_for_each_entry(vcpu, &per_cpu(blocked_vcpu_on_cpu, cpu),
6678 blocked_vcpu_list) {
6679 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
6681 if (pi_test_on(pi_desc) == 1)
6682 kvm_vcpu_kick(vcpu);
6684 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, cpu));
6687 void vmx_enable_tdp(void)
6689 kvm_mmu_set_mask_ptes(VMX_EPT_READABLE_MASK,
6690 enable_ept_ad_bits ? VMX_EPT_ACCESS_BIT : 0ull,
6691 enable_ept_ad_bits ? VMX_EPT_DIRTY_BIT : 0ull,
6692 0ull, VMX_EPT_EXECUTABLE_MASK,
6693 cpu_has_vmx_ept_execute_only() ? 0ull : VMX_EPT_READABLE_MASK,
6694 VMX_EPT_RWX_MASK, 0ull);
6696 ept_set_mmio_spte_mask();
6700 static __init int hardware_setup(void)
6702 int r = -ENOMEM, i, msr;
6704 rdmsrl_safe(MSR_EFER, &host_efer);
6706 for (i = 0; i < ARRAY_SIZE(vmx_msr_index); ++i)
6707 kvm_define_shared_msr(i, vmx_msr_index[i]);
6709 for (i = 0; i < VMX_BITMAP_NR; i++) {
6710 vmx_bitmap[i] = (unsigned long *)__get_free_page(GFP_KERNEL);
6715 vmx_io_bitmap_b = (unsigned long *)__get_free_page(GFP_KERNEL);
6716 memset(vmx_vmread_bitmap, 0xff, PAGE_SIZE);
6717 memset(vmx_vmwrite_bitmap, 0xff, PAGE_SIZE);
6720 * Allow direct access to the PC debug port (it is often used for I/O
6721 * delays, but the vmexits simply slow things down).
6723 memset(vmx_io_bitmap_a, 0xff, PAGE_SIZE);
6724 clear_bit(0x80, vmx_io_bitmap_a);
6726 memset(vmx_io_bitmap_b, 0xff, PAGE_SIZE);
6728 memset(vmx_msr_bitmap_legacy, 0xff, PAGE_SIZE);
6729 memset(vmx_msr_bitmap_longmode, 0xff, PAGE_SIZE);
6731 if (setup_vmcs_config(&vmcs_config) < 0) {
6736 if (boot_cpu_has(X86_FEATURE_NX))
6737 kvm_enable_efer_bits(EFER_NX);
6739 if (!cpu_has_vmx_vpid() || !cpu_has_vmx_invvpid() ||
6740 !(cpu_has_vmx_invvpid_single() || cpu_has_vmx_invvpid_global()))
6743 if (!cpu_has_vmx_shadow_vmcs())
6744 enable_shadow_vmcs = 0;
6745 if (enable_shadow_vmcs)
6746 init_vmcs_shadow_fields();
6748 if (!cpu_has_vmx_ept() ||
6749 !cpu_has_vmx_ept_4levels() ||
6750 !cpu_has_vmx_ept_mt_wb()) {
6752 enable_unrestricted_guest = 0;
6753 enable_ept_ad_bits = 0;
6756 if (!cpu_has_vmx_ept_ad_bits() || !enable_ept)
6757 enable_ept_ad_bits = 0;
6759 if (!cpu_has_vmx_unrestricted_guest())
6760 enable_unrestricted_guest = 0;
6762 if (!cpu_has_vmx_flexpriority())
6763 flexpriority_enabled = 0;
6766 * set_apic_access_page_addr() is used to reload apic access
6767 * page upon invalidation. No need to do anything if not
6768 * using the APIC_ACCESS_ADDR VMCS field.
6770 if (!flexpriority_enabled)
6771 kvm_x86_ops->set_apic_access_page_addr = NULL;
6773 if (!cpu_has_vmx_tpr_shadow())
6774 kvm_x86_ops->update_cr8_intercept = NULL;
6776 if (enable_ept && !cpu_has_vmx_ept_2m_page())
6777 kvm_disable_largepages();
6779 if (!cpu_has_vmx_ple()) {
6782 ple_window_grow = 0;
6784 ple_window_shrink = 0;
6787 if (!cpu_has_vmx_apicv()) {
6789 kvm_x86_ops->sync_pir_to_irr = NULL;
6792 if (cpu_has_vmx_tsc_scaling()) {
6793 kvm_has_tsc_control = true;
6794 kvm_max_tsc_scaling_ratio = KVM_VMX_TSC_MULTIPLIER_MAX;
6795 kvm_tsc_scaling_ratio_frac_bits = 48;
6798 vmx_disable_intercept_for_msr(MSR_FS_BASE, false);
6799 vmx_disable_intercept_for_msr(MSR_GS_BASE, false);
6800 vmx_disable_intercept_for_msr(MSR_KERNEL_GS_BASE, true);
6801 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_CS, false);
6802 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_ESP, false);
6803 vmx_disable_intercept_for_msr(MSR_IA32_SYSENTER_EIP, false);
6805 memcpy(vmx_msr_bitmap_legacy_x2apic_apicv,
6806 vmx_msr_bitmap_legacy, PAGE_SIZE);
6807 memcpy(vmx_msr_bitmap_longmode_x2apic_apicv,
6808 vmx_msr_bitmap_longmode, PAGE_SIZE);
6809 memcpy(vmx_msr_bitmap_legacy_x2apic,
6810 vmx_msr_bitmap_legacy, PAGE_SIZE);
6811 memcpy(vmx_msr_bitmap_longmode_x2apic,
6812 vmx_msr_bitmap_longmode, PAGE_SIZE);
6814 set_bit(0, vmx_vpid_bitmap); /* 0 is reserved for host */
6816 for (msr = 0x800; msr <= 0x8ff; msr++) {
6817 if (msr == 0x839 /* TMCCT */)
6819 vmx_disable_intercept_msr_x2apic(msr, MSR_TYPE_R, true);
6823 * TPR reads and writes can be virtualized even if virtual interrupt
6824 * delivery is not in use.
6826 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_W, true);
6827 vmx_disable_intercept_msr_x2apic(0x808, MSR_TYPE_R | MSR_TYPE_W, false);
6830 vmx_disable_intercept_msr_x2apic(0x80b, MSR_TYPE_W, true);
6832 vmx_disable_intercept_msr_x2apic(0x83f, MSR_TYPE_W, true);
6839 update_ple_window_actual_max();
6842 * Only enable PML when hardware supports PML feature, and both EPT
6843 * and EPT A/D bit features are enabled -- PML depends on them to work.
6845 if (!enable_ept || !enable_ept_ad_bits || !cpu_has_vmx_pml())
6849 kvm_x86_ops->slot_enable_log_dirty = NULL;
6850 kvm_x86_ops->slot_disable_log_dirty = NULL;
6851 kvm_x86_ops->flush_log_dirty = NULL;
6852 kvm_x86_ops->enable_log_dirty_pt_masked = NULL;
6855 if (cpu_has_vmx_preemption_timer() && enable_preemption_timer) {
6858 rdmsrl(MSR_IA32_VMX_MISC, vmx_msr);
6859 cpu_preemption_timer_multi =
6860 vmx_msr & VMX_MISC_PREEMPTION_TIMER_RATE_MASK;
6862 kvm_x86_ops->set_hv_timer = NULL;
6863 kvm_x86_ops->cancel_hv_timer = NULL;
6866 kvm_set_posted_intr_wakeup_handler(wakeup_handler);
6868 kvm_mce_cap_supported |= MCG_LMCE_P;
6870 return alloc_kvm_area();
6873 for (i = 0; i < VMX_BITMAP_NR; i++)
6874 free_page((unsigned long)vmx_bitmap[i]);
6879 static __exit void hardware_unsetup(void)
6883 for (i = 0; i < VMX_BITMAP_NR; i++)
6884 free_page((unsigned long)vmx_bitmap[i]);
6890 * Indicate a busy-waiting vcpu in spinlock. We do not enable the PAUSE
6891 * exiting, so only get here on cpu with PAUSE-Loop-Exiting.
6893 static int handle_pause(struct kvm_vcpu *vcpu)
6896 grow_ple_window(vcpu);
6899 * Intel sdm vol3 ch-25.1.3 says: The "PAUSE-loop exiting"
6900 * VM-execution control is ignored if CPL > 0. OTOH, KVM
6901 * never set PAUSE_EXITING and just set PLE if supported,
6902 * so the vcpu must be CPL=0 if it gets a PAUSE exit.
6904 kvm_vcpu_on_spin(vcpu, true);
6905 return kvm_skip_emulated_instruction(vcpu);
6908 static int handle_nop(struct kvm_vcpu *vcpu)
6910 return kvm_skip_emulated_instruction(vcpu);
6913 static int handle_mwait(struct kvm_vcpu *vcpu)
6915 printk_once(KERN_WARNING "kvm: MWAIT instruction emulated as NOP!\n");
6916 return handle_nop(vcpu);
6919 static int handle_invalid_op(struct kvm_vcpu *vcpu)
6921 kvm_queue_exception(vcpu, UD_VECTOR);
6925 static int handle_monitor_trap(struct kvm_vcpu *vcpu)
6930 static int handle_monitor(struct kvm_vcpu *vcpu)
6932 printk_once(KERN_WARNING "kvm: MONITOR instruction emulated as NOP!\n");
6933 return handle_nop(vcpu);
6937 * To run an L2 guest, we need a vmcs02 based on the L1-specified vmcs12.
6938 * We could reuse a single VMCS for all the L2 guests, but we also want the
6939 * option to allocate a separate vmcs02 for each separate loaded vmcs12 - this
6940 * allows keeping them loaded on the processor, and in the future will allow
6941 * optimizations where prepare_vmcs02 doesn't need to set all the fields on
6942 * every entry if they never change.
6943 * So we keep, in vmx->nested.vmcs02_pool, a cache of size VMCS02_POOL_SIZE
6944 * (>=0) with a vmcs02 for each recently loaded vmcs12s, most recent first.
6946 * The following functions allocate and free a vmcs02 in this pool.
6949 /* Get a VMCS from the pool to use as vmcs02 for the current vmcs12. */
6950 static struct loaded_vmcs *nested_get_current_vmcs02(struct vcpu_vmx *vmx)
6952 struct vmcs02_list *item;
6953 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6954 if (item->vmptr == vmx->nested.current_vmptr) {
6955 list_move(&item->list, &vmx->nested.vmcs02_pool);
6956 return &item->vmcs02;
6959 if (vmx->nested.vmcs02_num >= max(VMCS02_POOL_SIZE, 1)) {
6960 /* Recycle the least recently used VMCS. */
6961 item = list_last_entry(&vmx->nested.vmcs02_pool,
6962 struct vmcs02_list, list);
6963 item->vmptr = vmx->nested.current_vmptr;
6964 list_move(&item->list, &vmx->nested.vmcs02_pool);
6965 return &item->vmcs02;
6968 /* Create a new VMCS */
6969 item = kmalloc(sizeof(struct vmcs02_list), GFP_KERNEL);
6972 item->vmcs02.vmcs = alloc_vmcs();
6973 item->vmcs02.shadow_vmcs = NULL;
6974 if (!item->vmcs02.vmcs) {
6978 loaded_vmcs_init(&item->vmcs02);
6979 item->vmptr = vmx->nested.current_vmptr;
6980 list_add(&(item->list), &(vmx->nested.vmcs02_pool));
6981 vmx->nested.vmcs02_num++;
6982 return &item->vmcs02;
6985 /* Free and remove from pool a vmcs02 saved for a vmcs12 (if there is one) */
6986 static void nested_free_vmcs02(struct vcpu_vmx *vmx, gpa_t vmptr)
6988 struct vmcs02_list *item;
6989 list_for_each_entry(item, &vmx->nested.vmcs02_pool, list)
6990 if (item->vmptr == vmptr) {
6991 free_loaded_vmcs(&item->vmcs02);
6992 list_del(&item->list);
6994 vmx->nested.vmcs02_num--;
7000 * Free all VMCSs saved for this vcpu, except the one pointed by
7001 * vmx->loaded_vmcs. We must be running L1, so vmx->loaded_vmcs
7002 * must be &vmx->vmcs01.
7004 static void nested_free_all_saved_vmcss(struct vcpu_vmx *vmx)
7006 struct vmcs02_list *item, *n;
7008 WARN_ON(vmx->loaded_vmcs != &vmx->vmcs01);
7009 list_for_each_entry_safe(item, n, &vmx->nested.vmcs02_pool, list) {
7011 * Something will leak if the above WARN triggers. Better than
7014 if (vmx->loaded_vmcs == &item->vmcs02)
7017 free_loaded_vmcs(&item->vmcs02);
7018 list_del(&item->list);
7020 vmx->nested.vmcs02_num--;
7025 * The following 3 functions, nested_vmx_succeed()/failValid()/failInvalid(),
7026 * set the success or error code of an emulated VMX instruction, as specified
7027 * by Vol 2B, VMX Instruction Reference, "Conventions".
7029 static void nested_vmx_succeed(struct kvm_vcpu *vcpu)
7031 vmx_set_rflags(vcpu, vmx_get_rflags(vcpu)
7032 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
7033 X86_EFLAGS_ZF | X86_EFLAGS_SF | X86_EFLAGS_OF));
7036 static void nested_vmx_failInvalid(struct kvm_vcpu *vcpu)
7038 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
7039 & ~(X86_EFLAGS_PF | X86_EFLAGS_AF | X86_EFLAGS_ZF |
7040 X86_EFLAGS_SF | X86_EFLAGS_OF))
7044 static void nested_vmx_failValid(struct kvm_vcpu *vcpu,
7045 u32 vm_instruction_error)
7047 if (to_vmx(vcpu)->nested.current_vmptr == -1ull) {
7049 * failValid writes the error number to the current VMCS, which
7050 * can't be done there isn't a current VMCS.
7052 nested_vmx_failInvalid(vcpu);
7055 vmx_set_rflags(vcpu, (vmx_get_rflags(vcpu)
7056 & ~(X86_EFLAGS_CF | X86_EFLAGS_PF | X86_EFLAGS_AF |
7057 X86_EFLAGS_SF | X86_EFLAGS_OF))
7059 get_vmcs12(vcpu)->vm_instruction_error = vm_instruction_error;
7061 * We don't need to force a shadow sync because
7062 * VM_INSTRUCTION_ERROR is not shadowed
7066 static void nested_vmx_abort(struct kvm_vcpu *vcpu, u32 indicator)
7068 /* TODO: not to reset guest simply here. */
7069 kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
7070 pr_debug_ratelimited("kvm: nested vmx abort, indicator %d\n", indicator);
7073 static enum hrtimer_restart vmx_preemption_timer_fn(struct hrtimer *timer)
7075 struct vcpu_vmx *vmx =
7076 container_of(timer, struct vcpu_vmx, nested.preemption_timer);
7078 vmx->nested.preemption_timer_expired = true;
7079 kvm_make_request(KVM_REQ_EVENT, &vmx->vcpu);
7080 kvm_vcpu_kick(&vmx->vcpu);
7082 return HRTIMER_NORESTART;
7086 * Decode the memory-address operand of a vmx instruction, as recorded on an
7087 * exit caused by such an instruction (run by a guest hypervisor).
7088 * On success, returns 0. When the operand is invalid, returns 1 and throws
7091 static int get_vmx_mem_address(struct kvm_vcpu *vcpu,
7092 unsigned long exit_qualification,
7093 u32 vmx_instruction_info, bool wr, gva_t *ret)
7097 struct kvm_segment s;
7100 * According to Vol. 3B, "Information for VM Exits Due to Instruction
7101 * Execution", on an exit, vmx_instruction_info holds most of the
7102 * addressing components of the operand. Only the displacement part
7103 * is put in exit_qualification (see 3B, "Basic VM-Exit Information").
7104 * For how an actual address is calculated from all these components,
7105 * refer to Vol. 1, "Operand Addressing".
7107 int scaling = vmx_instruction_info & 3;
7108 int addr_size = (vmx_instruction_info >> 7) & 7;
7109 bool is_reg = vmx_instruction_info & (1u << 10);
7110 int seg_reg = (vmx_instruction_info >> 15) & 7;
7111 int index_reg = (vmx_instruction_info >> 18) & 0xf;
7112 bool index_is_valid = !(vmx_instruction_info & (1u << 22));
7113 int base_reg = (vmx_instruction_info >> 23) & 0xf;
7114 bool base_is_valid = !(vmx_instruction_info & (1u << 27));
7117 kvm_queue_exception(vcpu, UD_VECTOR);
7121 /* Addr = segment_base + offset */
7122 /* offset = base + [index * scale] + displacement */
7123 off = exit_qualification; /* holds the displacement */
7125 off += kvm_register_read(vcpu, base_reg);
7127 off += kvm_register_read(vcpu, index_reg)<<scaling;
7128 vmx_get_segment(vcpu, &s, seg_reg);
7129 *ret = s.base + off;
7131 if (addr_size == 1) /* 32 bit */
7134 /* Checks for #GP/#SS exceptions. */
7136 if (is_long_mode(vcpu)) {
7137 /* Long mode: #GP(0)/#SS(0) if the memory address is in a
7138 * non-canonical form. This is the only check on the memory
7139 * destination for long mode!
7141 exn = is_noncanonical_address(*ret, vcpu);
7142 } else if (is_protmode(vcpu)) {
7143 /* Protected mode: apply checks for segment validity in the
7145 * - segment type check (#GP(0) may be thrown)
7146 * - usability check (#GP(0)/#SS(0))
7147 * - limit check (#GP(0)/#SS(0))
7150 /* #GP(0) if the destination operand is located in a
7151 * read-only data segment or any code segment.
7153 exn = ((s.type & 0xa) == 0 || (s.type & 8));
7155 /* #GP(0) if the source operand is located in an
7156 * execute-only code segment
7158 exn = ((s.type & 0xa) == 8);
7160 kvm_queue_exception_e(vcpu, GP_VECTOR, 0);
7163 /* Protected mode: #GP(0)/#SS(0) if the segment is unusable.
7165 exn = (s.unusable != 0);
7166 /* Protected mode: #GP(0)/#SS(0) if the memory
7167 * operand is outside the segment limit.
7169 exn = exn || (off + sizeof(u64) > s.limit);
7172 kvm_queue_exception_e(vcpu,
7173 seg_reg == VCPU_SREG_SS ?
7174 SS_VECTOR : GP_VECTOR,
7182 static int nested_vmx_get_vmptr(struct kvm_vcpu *vcpu, gpa_t *vmpointer)
7185 struct x86_exception e;
7187 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7188 vmcs_read32(VMX_INSTRUCTION_INFO), false, &gva))
7191 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, vmpointer,
7192 sizeof(*vmpointer), &e)) {
7193 kvm_inject_page_fault(vcpu, &e);
7200 static int enter_vmx_operation(struct kvm_vcpu *vcpu)
7202 struct vcpu_vmx *vmx = to_vmx(vcpu);
7203 struct vmcs *shadow_vmcs;
7205 if (cpu_has_vmx_msr_bitmap()) {
7206 vmx->nested.msr_bitmap =
7207 (unsigned long *)__get_free_page(GFP_KERNEL);
7208 if (!vmx->nested.msr_bitmap)
7209 goto out_msr_bitmap;
7212 vmx->nested.cached_vmcs12 = kmalloc(VMCS12_SIZE, GFP_KERNEL);
7213 if (!vmx->nested.cached_vmcs12)
7214 goto out_cached_vmcs12;
7216 if (enable_shadow_vmcs) {
7217 shadow_vmcs = alloc_vmcs();
7219 goto out_shadow_vmcs;
7220 /* mark vmcs as shadow */
7221 shadow_vmcs->revision_id |= (1u << 31);
7222 /* init shadow vmcs */
7223 vmcs_clear(shadow_vmcs);
7224 vmx->vmcs01.shadow_vmcs = shadow_vmcs;
7227 INIT_LIST_HEAD(&(vmx->nested.vmcs02_pool));
7228 vmx->nested.vmcs02_num = 0;
7230 hrtimer_init(&vmx->nested.preemption_timer, CLOCK_MONOTONIC,
7231 HRTIMER_MODE_REL_PINNED);
7232 vmx->nested.preemption_timer.function = vmx_preemption_timer_fn;
7234 vmx->nested.vmxon = true;
7238 kfree(vmx->nested.cached_vmcs12);
7241 free_page((unsigned long)vmx->nested.msr_bitmap);
7248 * Emulate the VMXON instruction.
7249 * Currently, we just remember that VMX is active, and do not save or even
7250 * inspect the argument to VMXON (the so-called "VMXON pointer") because we
7251 * do not currently need to store anything in that guest-allocated memory
7252 * region. Consequently, VMCLEAR and VMPTRLD also do not verify that the their
7253 * argument is different from the VMXON pointer (which the spec says they do).
7255 static int handle_vmon(struct kvm_vcpu *vcpu)
7260 struct vcpu_vmx *vmx = to_vmx(vcpu);
7261 const u64 VMXON_NEEDED_FEATURES = FEATURE_CONTROL_LOCKED
7262 | FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
7265 * The Intel VMX Instruction Reference lists a bunch of bits that are
7266 * prerequisite to running VMXON, most notably cr4.VMXE must be set to
7267 * 1 (see vmx_set_cr4() for when we allow the guest to set this).
7268 * Otherwise, we should fail with #UD. But most faulting conditions
7269 * have already been checked by hardware, prior to the VM-exit for
7270 * VMXON. We do test guest cr4.VMXE because processor CR4 always has
7271 * that bit set to 1 in non-root mode.
7273 if (!kvm_read_cr4_bits(vcpu, X86_CR4_VMXE)) {
7274 kvm_queue_exception(vcpu, UD_VECTOR);
7278 if (vmx->nested.vmxon) {
7279 nested_vmx_failValid(vcpu, VMXERR_VMXON_IN_VMX_ROOT_OPERATION);
7280 return kvm_skip_emulated_instruction(vcpu);
7283 if ((vmx->msr_ia32_feature_control & VMXON_NEEDED_FEATURES)
7284 != VMXON_NEEDED_FEATURES) {
7285 kvm_inject_gp(vcpu, 0);
7289 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7294 * The first 4 bytes of VMXON region contain the supported
7295 * VMCS revision identifier
7297 * Note - IA32_VMX_BASIC[48] will never be 1 for the nested case;
7298 * which replaces physical address width with 32
7300 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7301 nested_vmx_failInvalid(vcpu);
7302 return kvm_skip_emulated_instruction(vcpu);
7305 page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
7306 if (is_error_page(page)) {
7307 nested_vmx_failInvalid(vcpu);
7308 return kvm_skip_emulated_instruction(vcpu);
7310 if (*(u32 *)kmap(page) != VMCS12_REVISION) {
7312 kvm_release_page_clean(page);
7313 nested_vmx_failInvalid(vcpu);
7314 return kvm_skip_emulated_instruction(vcpu);
7317 kvm_release_page_clean(page);
7319 vmx->nested.vmxon_ptr = vmptr;
7320 ret = enter_vmx_operation(vcpu);
7324 nested_vmx_succeed(vcpu);
7325 return kvm_skip_emulated_instruction(vcpu);
7329 * Intel's VMX Instruction Reference specifies a common set of prerequisites
7330 * for running VMX instructions (except VMXON, whose prerequisites are
7331 * slightly different). It also specifies what exception to inject otherwise.
7332 * Note that many of these exceptions have priority over VM exits, so they
7333 * don't have to be checked again here.
7335 static int nested_vmx_check_permission(struct kvm_vcpu *vcpu)
7337 if (!to_vmx(vcpu)->nested.vmxon) {
7338 kvm_queue_exception(vcpu, UD_VECTOR);
7344 static void vmx_disable_shadow_vmcs(struct vcpu_vmx *vmx)
7346 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL, SECONDARY_EXEC_SHADOW_VMCS);
7347 vmcs_write64(VMCS_LINK_POINTER, -1ull);
7350 static inline void nested_release_vmcs12(struct vcpu_vmx *vmx)
7352 if (vmx->nested.current_vmptr == -1ull)
7355 if (enable_shadow_vmcs) {
7356 /* copy to memory all shadowed fields in case
7357 they were modified */
7358 copy_shadow_to_vmcs12(vmx);
7359 vmx->nested.sync_shadow_vmcs = false;
7360 vmx_disable_shadow_vmcs(vmx);
7362 vmx->nested.posted_intr_nv = -1;
7364 /* Flush VMCS12 to guest memory */
7365 kvm_vcpu_write_guest_page(&vmx->vcpu,
7366 vmx->nested.current_vmptr >> PAGE_SHIFT,
7367 vmx->nested.cached_vmcs12, 0, VMCS12_SIZE);
7369 vmx->nested.current_vmptr = -1ull;
7373 * Free whatever needs to be freed from vmx->nested when L1 goes down, or
7374 * just stops using VMX.
7376 static void free_nested(struct vcpu_vmx *vmx)
7378 if (!vmx->nested.vmxon)
7381 vmx->nested.vmxon = false;
7382 free_vpid(vmx->nested.vpid02);
7383 vmx->nested.posted_intr_nv = -1;
7384 vmx->nested.current_vmptr = -1ull;
7385 if (vmx->nested.msr_bitmap) {
7386 free_page((unsigned long)vmx->nested.msr_bitmap);
7387 vmx->nested.msr_bitmap = NULL;
7389 if (enable_shadow_vmcs) {
7390 vmx_disable_shadow_vmcs(vmx);
7391 vmcs_clear(vmx->vmcs01.shadow_vmcs);
7392 free_vmcs(vmx->vmcs01.shadow_vmcs);
7393 vmx->vmcs01.shadow_vmcs = NULL;
7395 kfree(vmx->nested.cached_vmcs12);
7396 /* Unpin physical memory we referred to in current vmcs02 */
7397 if (vmx->nested.apic_access_page) {
7398 kvm_release_page_dirty(vmx->nested.apic_access_page);
7399 vmx->nested.apic_access_page = NULL;
7401 if (vmx->nested.virtual_apic_page) {
7402 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
7403 vmx->nested.virtual_apic_page = NULL;
7405 if (vmx->nested.pi_desc_page) {
7406 kunmap(vmx->nested.pi_desc_page);
7407 kvm_release_page_dirty(vmx->nested.pi_desc_page);
7408 vmx->nested.pi_desc_page = NULL;
7409 vmx->nested.pi_desc = NULL;
7412 nested_free_all_saved_vmcss(vmx);
7415 /* Emulate the VMXOFF instruction */
7416 static int handle_vmoff(struct kvm_vcpu *vcpu)
7418 if (!nested_vmx_check_permission(vcpu))
7420 free_nested(to_vmx(vcpu));
7421 nested_vmx_succeed(vcpu);
7422 return kvm_skip_emulated_instruction(vcpu);
7425 /* Emulate the VMCLEAR instruction */
7426 static int handle_vmclear(struct kvm_vcpu *vcpu)
7428 struct vcpu_vmx *vmx = to_vmx(vcpu);
7432 if (!nested_vmx_check_permission(vcpu))
7435 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7438 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7439 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_INVALID_ADDRESS);
7440 return kvm_skip_emulated_instruction(vcpu);
7443 if (vmptr == vmx->nested.vmxon_ptr) {
7444 nested_vmx_failValid(vcpu, VMXERR_VMCLEAR_VMXON_POINTER);
7445 return kvm_skip_emulated_instruction(vcpu);
7448 if (vmptr == vmx->nested.current_vmptr)
7449 nested_release_vmcs12(vmx);
7451 kvm_vcpu_write_guest(vcpu,
7452 vmptr + offsetof(struct vmcs12, launch_state),
7453 &zero, sizeof(zero));
7455 nested_free_vmcs02(vmx, vmptr);
7457 nested_vmx_succeed(vcpu);
7458 return kvm_skip_emulated_instruction(vcpu);
7461 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch);
7463 /* Emulate the VMLAUNCH instruction */
7464 static int handle_vmlaunch(struct kvm_vcpu *vcpu)
7466 return nested_vmx_run(vcpu, true);
7469 /* Emulate the VMRESUME instruction */
7470 static int handle_vmresume(struct kvm_vcpu *vcpu)
7473 return nested_vmx_run(vcpu, false);
7477 * Read a vmcs12 field. Since these can have varying lengths and we return
7478 * one type, we chose the biggest type (u64) and zero-extend the return value
7479 * to that size. Note that the caller, handle_vmread, might need to use only
7480 * some of the bits we return here (e.g., on 32-bit guests, only 32 bits of
7481 * 64-bit fields are to be returned).
7483 static inline int vmcs12_read_any(struct kvm_vcpu *vcpu,
7484 unsigned long field, u64 *ret)
7486 short offset = vmcs_field_to_offset(field);
7492 p = ((char *)(get_vmcs12(vcpu))) + offset;
7494 switch (vmcs_field_type(field)) {
7495 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7496 *ret = *((natural_width *)p);
7498 case VMCS_FIELD_TYPE_U16:
7501 case VMCS_FIELD_TYPE_U32:
7504 case VMCS_FIELD_TYPE_U64:
7514 static inline int vmcs12_write_any(struct kvm_vcpu *vcpu,
7515 unsigned long field, u64 field_value){
7516 short offset = vmcs_field_to_offset(field);
7517 char *p = ((char *) get_vmcs12(vcpu)) + offset;
7521 switch (vmcs_field_type(field)) {
7522 case VMCS_FIELD_TYPE_U16:
7523 *(u16 *)p = field_value;
7525 case VMCS_FIELD_TYPE_U32:
7526 *(u32 *)p = field_value;
7528 case VMCS_FIELD_TYPE_U64:
7529 *(u64 *)p = field_value;
7531 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7532 *(natural_width *)p = field_value;
7541 static void copy_shadow_to_vmcs12(struct vcpu_vmx *vmx)
7544 unsigned long field;
7546 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7547 const unsigned long *fields = shadow_read_write_fields;
7548 const int num_fields = max_shadow_read_write_fields;
7552 vmcs_load(shadow_vmcs);
7554 for (i = 0; i < num_fields; i++) {
7556 switch (vmcs_field_type(field)) {
7557 case VMCS_FIELD_TYPE_U16:
7558 field_value = vmcs_read16(field);
7560 case VMCS_FIELD_TYPE_U32:
7561 field_value = vmcs_read32(field);
7563 case VMCS_FIELD_TYPE_U64:
7564 field_value = vmcs_read64(field);
7566 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7567 field_value = vmcs_readl(field);
7573 vmcs12_write_any(&vmx->vcpu, field, field_value);
7576 vmcs_clear(shadow_vmcs);
7577 vmcs_load(vmx->loaded_vmcs->vmcs);
7582 static void copy_vmcs12_to_shadow(struct vcpu_vmx *vmx)
7584 const unsigned long *fields[] = {
7585 shadow_read_write_fields,
7586 shadow_read_only_fields
7588 const int max_fields[] = {
7589 max_shadow_read_write_fields,
7590 max_shadow_read_only_fields
7593 unsigned long field;
7594 u64 field_value = 0;
7595 struct vmcs *shadow_vmcs = vmx->vmcs01.shadow_vmcs;
7597 vmcs_load(shadow_vmcs);
7599 for (q = 0; q < ARRAY_SIZE(fields); q++) {
7600 for (i = 0; i < max_fields[q]; i++) {
7601 field = fields[q][i];
7602 vmcs12_read_any(&vmx->vcpu, field, &field_value);
7604 switch (vmcs_field_type(field)) {
7605 case VMCS_FIELD_TYPE_U16:
7606 vmcs_write16(field, (u16)field_value);
7608 case VMCS_FIELD_TYPE_U32:
7609 vmcs_write32(field, (u32)field_value);
7611 case VMCS_FIELD_TYPE_U64:
7612 vmcs_write64(field, (u64)field_value);
7614 case VMCS_FIELD_TYPE_NATURAL_WIDTH:
7615 vmcs_writel(field, (long)field_value);
7624 vmcs_clear(shadow_vmcs);
7625 vmcs_load(vmx->loaded_vmcs->vmcs);
7629 * VMX instructions which assume a current vmcs12 (i.e., that VMPTRLD was
7630 * used before) all generate the same failure when it is missing.
7632 static int nested_vmx_check_vmcs12(struct kvm_vcpu *vcpu)
7634 struct vcpu_vmx *vmx = to_vmx(vcpu);
7635 if (vmx->nested.current_vmptr == -1ull) {
7636 nested_vmx_failInvalid(vcpu);
7642 static int handle_vmread(struct kvm_vcpu *vcpu)
7644 unsigned long field;
7646 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7647 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7650 if (!nested_vmx_check_permission(vcpu))
7653 if (!nested_vmx_check_vmcs12(vcpu))
7654 return kvm_skip_emulated_instruction(vcpu);
7656 /* Decode instruction info and find the field to read */
7657 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7658 /* Read the field, zero-extended to a u64 field_value */
7659 if (vmcs12_read_any(vcpu, field, &field_value) < 0) {
7660 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7661 return kvm_skip_emulated_instruction(vcpu);
7664 * Now copy part of this value to register or memory, as requested.
7665 * Note that the number of bits actually copied is 32 or 64 depending
7666 * on the guest's mode (32 or 64 bit), not on the given field's length.
7668 if (vmx_instruction_info & (1u << 10)) {
7669 kvm_register_writel(vcpu, (((vmx_instruction_info) >> 3) & 0xf),
7672 if (get_vmx_mem_address(vcpu, exit_qualification,
7673 vmx_instruction_info, true, &gva))
7675 /* _system ok, as hardware has verified cpl=0 */
7676 kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, gva,
7677 &field_value, (is_long_mode(vcpu) ? 8 : 4), NULL);
7680 nested_vmx_succeed(vcpu);
7681 return kvm_skip_emulated_instruction(vcpu);
7685 static int handle_vmwrite(struct kvm_vcpu *vcpu)
7687 unsigned long field;
7689 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7690 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7691 /* The value to write might be 32 or 64 bits, depending on L1's long
7692 * mode, and eventually we need to write that into a field of several
7693 * possible lengths. The code below first zero-extends the value to 64
7694 * bit (field_value), and then copies only the appropriate number of
7695 * bits into the vmcs12 field.
7697 u64 field_value = 0;
7698 struct x86_exception e;
7700 if (!nested_vmx_check_permission(vcpu))
7703 if (!nested_vmx_check_vmcs12(vcpu))
7704 return kvm_skip_emulated_instruction(vcpu);
7706 if (vmx_instruction_info & (1u << 10))
7707 field_value = kvm_register_readl(vcpu,
7708 (((vmx_instruction_info) >> 3) & 0xf));
7710 if (get_vmx_mem_address(vcpu, exit_qualification,
7711 vmx_instruction_info, false, &gva))
7713 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva,
7714 &field_value, (is_64_bit_mode(vcpu) ? 8 : 4), &e)) {
7715 kvm_inject_page_fault(vcpu, &e);
7721 field = kvm_register_readl(vcpu, (((vmx_instruction_info) >> 28) & 0xf));
7722 if (vmcs_field_readonly(field)) {
7723 nested_vmx_failValid(vcpu,
7724 VMXERR_VMWRITE_READ_ONLY_VMCS_COMPONENT);
7725 return kvm_skip_emulated_instruction(vcpu);
7728 if (vmcs12_write_any(vcpu, field, field_value) < 0) {
7729 nested_vmx_failValid(vcpu, VMXERR_UNSUPPORTED_VMCS_COMPONENT);
7730 return kvm_skip_emulated_instruction(vcpu);
7733 nested_vmx_succeed(vcpu);
7734 return kvm_skip_emulated_instruction(vcpu);
7737 static void set_current_vmptr(struct vcpu_vmx *vmx, gpa_t vmptr)
7739 vmx->nested.current_vmptr = vmptr;
7740 if (enable_shadow_vmcs) {
7741 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
7742 SECONDARY_EXEC_SHADOW_VMCS);
7743 vmcs_write64(VMCS_LINK_POINTER,
7744 __pa(vmx->vmcs01.shadow_vmcs));
7745 vmx->nested.sync_shadow_vmcs = true;
7749 /* Emulate the VMPTRLD instruction */
7750 static int handle_vmptrld(struct kvm_vcpu *vcpu)
7752 struct vcpu_vmx *vmx = to_vmx(vcpu);
7755 if (!nested_vmx_check_permission(vcpu))
7758 if (nested_vmx_get_vmptr(vcpu, &vmptr))
7761 if (!PAGE_ALIGNED(vmptr) || (vmptr >> cpuid_maxphyaddr(vcpu))) {
7762 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_INVALID_ADDRESS);
7763 return kvm_skip_emulated_instruction(vcpu);
7766 if (vmptr == vmx->nested.vmxon_ptr) {
7767 nested_vmx_failValid(vcpu, VMXERR_VMPTRLD_VMXON_POINTER);
7768 return kvm_skip_emulated_instruction(vcpu);
7771 if (vmx->nested.current_vmptr != vmptr) {
7772 struct vmcs12 *new_vmcs12;
7774 page = kvm_vcpu_gpa_to_page(vcpu, vmptr);
7775 if (is_error_page(page)) {
7776 nested_vmx_failInvalid(vcpu);
7777 return kvm_skip_emulated_instruction(vcpu);
7779 new_vmcs12 = kmap(page);
7780 if (new_vmcs12->revision_id != VMCS12_REVISION) {
7782 kvm_release_page_clean(page);
7783 nested_vmx_failValid(vcpu,
7784 VMXERR_VMPTRLD_INCORRECT_VMCS_REVISION_ID);
7785 return kvm_skip_emulated_instruction(vcpu);
7788 nested_release_vmcs12(vmx);
7790 * Load VMCS12 from guest memory since it is not already
7793 memcpy(vmx->nested.cached_vmcs12, new_vmcs12, VMCS12_SIZE);
7795 kvm_release_page_clean(page);
7797 set_current_vmptr(vmx, vmptr);
7800 nested_vmx_succeed(vcpu);
7801 return kvm_skip_emulated_instruction(vcpu);
7804 /* Emulate the VMPTRST instruction */
7805 static int handle_vmptrst(struct kvm_vcpu *vcpu)
7807 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7808 u32 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7810 struct x86_exception e;
7812 if (!nested_vmx_check_permission(vcpu))
7815 if (get_vmx_mem_address(vcpu, exit_qualification,
7816 vmx_instruction_info, true, &vmcs_gva))
7818 /* ok to use *_system, as hardware has verified cpl=0 */
7819 if (kvm_write_guest_virt_system(&vcpu->arch.emulate_ctxt, vmcs_gva,
7820 (void *)&to_vmx(vcpu)->nested.current_vmptr,
7822 kvm_inject_page_fault(vcpu, &e);
7825 nested_vmx_succeed(vcpu);
7826 return kvm_skip_emulated_instruction(vcpu);
7829 /* Emulate the INVEPT instruction */
7830 static int handle_invept(struct kvm_vcpu *vcpu)
7832 struct vcpu_vmx *vmx = to_vmx(vcpu);
7833 u32 vmx_instruction_info, types;
7836 struct x86_exception e;
7841 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7842 SECONDARY_EXEC_ENABLE_EPT) ||
7843 !(vmx->nested.nested_vmx_ept_caps & VMX_EPT_INVEPT_BIT)) {
7844 kvm_queue_exception(vcpu, UD_VECTOR);
7848 if (!nested_vmx_check_permission(vcpu))
7851 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7852 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7854 types = (vmx->nested.nested_vmx_ept_caps >> VMX_EPT_EXTENT_SHIFT) & 6;
7856 if (type >= 32 || !(types & (1 << type))) {
7857 nested_vmx_failValid(vcpu,
7858 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7859 return kvm_skip_emulated_instruction(vcpu);
7862 /* According to the Intel VMX instruction reference, the memory
7863 * operand is read even if it isn't needed (e.g., for type==global)
7865 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7866 vmx_instruction_info, false, &gva))
7868 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7869 sizeof(operand), &e)) {
7870 kvm_inject_page_fault(vcpu, &e);
7875 case VMX_EPT_EXTENT_GLOBAL:
7877 * TODO: track mappings and invalidate
7878 * single context requests appropriately
7880 case VMX_EPT_EXTENT_CONTEXT:
7881 kvm_mmu_sync_roots(vcpu);
7882 kvm_make_request(KVM_REQ_TLB_FLUSH, vcpu);
7883 nested_vmx_succeed(vcpu);
7890 return kvm_skip_emulated_instruction(vcpu);
7893 static int handle_invvpid(struct kvm_vcpu *vcpu)
7895 struct vcpu_vmx *vmx = to_vmx(vcpu);
7896 u32 vmx_instruction_info;
7897 unsigned long type, types;
7899 struct x86_exception e;
7905 if (!(vmx->nested.nested_vmx_secondary_ctls_high &
7906 SECONDARY_EXEC_ENABLE_VPID) ||
7907 !(vmx->nested.nested_vmx_vpid_caps & VMX_VPID_INVVPID_BIT)) {
7908 kvm_queue_exception(vcpu, UD_VECTOR);
7912 if (!nested_vmx_check_permission(vcpu))
7915 vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
7916 type = kvm_register_readl(vcpu, (vmx_instruction_info >> 28) & 0xf);
7918 types = (vmx->nested.nested_vmx_vpid_caps &
7919 VMX_VPID_EXTENT_SUPPORTED_MASK) >> 8;
7921 if (type >= 32 || !(types & (1 << type))) {
7922 nested_vmx_failValid(vcpu,
7923 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7924 return kvm_skip_emulated_instruction(vcpu);
7927 /* according to the intel vmx instruction reference, the memory
7928 * operand is read even if it isn't needed (e.g., for type==global)
7930 if (get_vmx_mem_address(vcpu, vmcs_readl(EXIT_QUALIFICATION),
7931 vmx_instruction_info, false, &gva))
7933 if (kvm_read_guest_virt(&vcpu->arch.emulate_ctxt, gva, &operand,
7934 sizeof(operand), &e)) {
7935 kvm_inject_page_fault(vcpu, &e);
7938 if (operand.vpid >> 16) {
7939 nested_vmx_failValid(vcpu,
7940 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7941 return kvm_skip_emulated_instruction(vcpu);
7945 case VMX_VPID_EXTENT_INDIVIDUAL_ADDR:
7946 if (is_noncanonical_address(operand.gla, vcpu)) {
7947 nested_vmx_failValid(vcpu,
7948 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7949 return kvm_skip_emulated_instruction(vcpu);
7952 case VMX_VPID_EXTENT_SINGLE_CONTEXT:
7953 case VMX_VPID_EXTENT_SINGLE_NON_GLOBAL:
7954 if (!operand.vpid) {
7955 nested_vmx_failValid(vcpu,
7956 VMXERR_INVALID_OPERAND_TO_INVEPT_INVVPID);
7957 return kvm_skip_emulated_instruction(vcpu);
7960 case VMX_VPID_EXTENT_ALL_CONTEXT:
7964 return kvm_skip_emulated_instruction(vcpu);
7967 __vmx_flush_tlb(vcpu, vmx->nested.vpid02);
7968 nested_vmx_succeed(vcpu);
7970 return kvm_skip_emulated_instruction(vcpu);
7973 static int handle_pml_full(struct kvm_vcpu *vcpu)
7975 unsigned long exit_qualification;
7977 trace_kvm_pml_full(vcpu->vcpu_id);
7979 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
7982 * PML buffer FULL happened while executing iret from NMI,
7983 * "blocked by NMI" bit has to be set before next VM entry.
7985 if (!(to_vmx(vcpu)->idt_vectoring_info & VECTORING_INFO_VALID_MASK) &&
7986 (exit_qualification & INTR_INFO_UNBLOCK_NMI))
7987 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
7988 GUEST_INTR_STATE_NMI);
7991 * PML buffer already flushed at beginning of VMEXIT. Nothing to do
7992 * here.., and there's no userspace involvement needed for PML.
7997 static int handle_preemption_timer(struct kvm_vcpu *vcpu)
7999 kvm_lapic_expired_hv_timer(vcpu);
8003 static bool valid_ept_address(struct kvm_vcpu *vcpu, u64 address)
8005 struct vcpu_vmx *vmx = to_vmx(vcpu);
8006 int maxphyaddr = cpuid_maxphyaddr(vcpu);
8008 /* Check for memory type validity */
8009 switch (address & VMX_EPTP_MT_MASK) {
8010 case VMX_EPTP_MT_UC:
8011 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPTP_UC_BIT))
8014 case VMX_EPTP_MT_WB:
8015 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPTP_WB_BIT))
8022 /* only 4 levels page-walk length are valid */
8023 if ((address & VMX_EPTP_PWL_MASK) != VMX_EPTP_PWL_4)
8026 /* Reserved bits should not be set */
8027 if (address >> maxphyaddr || ((address >> 7) & 0x1f))
8030 /* AD, if set, should be supported */
8031 if (address & VMX_EPTP_AD_ENABLE_BIT) {
8032 if (!(vmx->nested.nested_vmx_ept_caps & VMX_EPT_AD_BIT))
8039 static int nested_vmx_eptp_switching(struct kvm_vcpu *vcpu,
8040 struct vmcs12 *vmcs12)
8042 u32 index = vcpu->arch.regs[VCPU_REGS_RCX];
8044 bool accessed_dirty;
8045 struct kvm_mmu *mmu = vcpu->arch.walk_mmu;
8047 if (!nested_cpu_has_eptp_switching(vmcs12) ||
8048 !nested_cpu_has_ept(vmcs12))
8051 if (index >= VMFUNC_EPTP_ENTRIES)
8055 if (kvm_vcpu_read_guest_page(vcpu, vmcs12->eptp_list_address >> PAGE_SHIFT,
8056 &address, index * 8, 8))
8059 accessed_dirty = !!(address & VMX_EPTP_AD_ENABLE_BIT);
8062 * If the (L2) guest does a vmfunc to the currently
8063 * active ept pointer, we don't have to do anything else
8065 if (vmcs12->ept_pointer != address) {
8066 if (!valid_ept_address(vcpu, address))
8069 kvm_mmu_unload(vcpu);
8070 mmu->ept_ad = accessed_dirty;
8071 mmu->base_role.ad_disabled = !accessed_dirty;
8072 vmcs12->ept_pointer = address;
8074 * TODO: Check what's the correct approach in case
8075 * mmu reload fails. Currently, we just let the next
8076 * reload potentially fail
8078 kvm_mmu_reload(vcpu);
8084 static int handle_vmfunc(struct kvm_vcpu *vcpu)
8086 struct vcpu_vmx *vmx = to_vmx(vcpu);
8087 struct vmcs12 *vmcs12;
8088 u32 function = vcpu->arch.regs[VCPU_REGS_RAX];
8091 * VMFUNC is only supported for nested guests, but we always enable the
8092 * secondary control for simplicity; for non-nested mode, fake that we
8093 * didn't by injecting #UD.
8095 if (!is_guest_mode(vcpu)) {
8096 kvm_queue_exception(vcpu, UD_VECTOR);
8100 vmcs12 = get_vmcs12(vcpu);
8101 if ((vmcs12->vm_function_control & (1 << function)) == 0)
8106 if (nested_vmx_eptp_switching(vcpu, vmcs12))
8112 return kvm_skip_emulated_instruction(vcpu);
8115 nested_vmx_vmexit(vcpu, vmx->exit_reason,
8116 vmcs_read32(VM_EXIT_INTR_INFO),
8117 vmcs_readl(EXIT_QUALIFICATION));
8122 * The exit handlers return 1 if the exit was handled fully and guest execution
8123 * may resume. Otherwise they set the kvm_run parameter to indicate what needs
8124 * to be done to userspace and return 0.
8126 static int (*const kvm_vmx_exit_handlers[])(struct kvm_vcpu *vcpu) = {
8127 [EXIT_REASON_EXCEPTION_NMI] = handle_exception,
8128 [EXIT_REASON_EXTERNAL_INTERRUPT] = handle_external_interrupt,
8129 [EXIT_REASON_TRIPLE_FAULT] = handle_triple_fault,
8130 [EXIT_REASON_NMI_WINDOW] = handle_nmi_window,
8131 [EXIT_REASON_IO_INSTRUCTION] = handle_io,
8132 [EXIT_REASON_CR_ACCESS] = handle_cr,
8133 [EXIT_REASON_DR_ACCESS] = handle_dr,
8134 [EXIT_REASON_CPUID] = handle_cpuid,
8135 [EXIT_REASON_MSR_READ] = handle_rdmsr,
8136 [EXIT_REASON_MSR_WRITE] = handle_wrmsr,
8137 [EXIT_REASON_PENDING_INTERRUPT] = handle_interrupt_window,
8138 [EXIT_REASON_HLT] = handle_halt,
8139 [EXIT_REASON_INVD] = handle_invd,
8140 [EXIT_REASON_INVLPG] = handle_invlpg,
8141 [EXIT_REASON_RDPMC] = handle_rdpmc,
8142 [EXIT_REASON_VMCALL] = handle_vmcall,
8143 [EXIT_REASON_VMCLEAR] = handle_vmclear,
8144 [EXIT_REASON_VMLAUNCH] = handle_vmlaunch,
8145 [EXIT_REASON_VMPTRLD] = handle_vmptrld,
8146 [EXIT_REASON_VMPTRST] = handle_vmptrst,
8147 [EXIT_REASON_VMREAD] = handle_vmread,
8148 [EXIT_REASON_VMRESUME] = handle_vmresume,
8149 [EXIT_REASON_VMWRITE] = handle_vmwrite,
8150 [EXIT_REASON_VMOFF] = handle_vmoff,
8151 [EXIT_REASON_VMON] = handle_vmon,
8152 [EXIT_REASON_TPR_BELOW_THRESHOLD] = handle_tpr_below_threshold,
8153 [EXIT_REASON_APIC_ACCESS] = handle_apic_access,
8154 [EXIT_REASON_APIC_WRITE] = handle_apic_write,
8155 [EXIT_REASON_EOI_INDUCED] = handle_apic_eoi_induced,
8156 [EXIT_REASON_WBINVD] = handle_wbinvd,
8157 [EXIT_REASON_XSETBV] = handle_xsetbv,
8158 [EXIT_REASON_TASK_SWITCH] = handle_task_switch,
8159 [EXIT_REASON_MCE_DURING_VMENTRY] = handle_machine_check,
8160 [EXIT_REASON_EPT_VIOLATION] = handle_ept_violation,
8161 [EXIT_REASON_EPT_MISCONFIG] = handle_ept_misconfig,
8162 [EXIT_REASON_PAUSE_INSTRUCTION] = handle_pause,
8163 [EXIT_REASON_MWAIT_INSTRUCTION] = handle_mwait,
8164 [EXIT_REASON_MONITOR_TRAP_FLAG] = handle_monitor_trap,
8165 [EXIT_REASON_MONITOR_INSTRUCTION] = handle_monitor,
8166 [EXIT_REASON_INVEPT] = handle_invept,
8167 [EXIT_REASON_INVVPID] = handle_invvpid,
8168 [EXIT_REASON_RDRAND] = handle_invalid_op,
8169 [EXIT_REASON_RDSEED] = handle_invalid_op,
8170 [EXIT_REASON_XSAVES] = handle_xsaves,
8171 [EXIT_REASON_XRSTORS] = handle_xrstors,
8172 [EXIT_REASON_PML_FULL] = handle_pml_full,
8173 [EXIT_REASON_VMFUNC] = handle_vmfunc,
8174 [EXIT_REASON_PREEMPTION_TIMER] = handle_preemption_timer,
8177 static const int kvm_vmx_max_exit_handlers =
8178 ARRAY_SIZE(kvm_vmx_exit_handlers);
8180 static bool nested_vmx_exit_handled_io(struct kvm_vcpu *vcpu,
8181 struct vmcs12 *vmcs12)
8183 unsigned long exit_qualification;
8184 gpa_t bitmap, last_bitmap;
8189 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
8190 return nested_cpu_has(vmcs12, CPU_BASED_UNCOND_IO_EXITING);
8192 exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8194 port = exit_qualification >> 16;
8195 size = (exit_qualification & 7) + 1;
8197 last_bitmap = (gpa_t)-1;
8202 bitmap = vmcs12->io_bitmap_a;
8203 else if (port < 0x10000)
8204 bitmap = vmcs12->io_bitmap_b;
8207 bitmap += (port & 0x7fff) / 8;
8209 if (last_bitmap != bitmap)
8210 if (kvm_vcpu_read_guest(vcpu, bitmap, &b, 1))
8212 if (b & (1 << (port & 7)))
8217 last_bitmap = bitmap;
8224 * Return 1 if we should exit from L2 to L1 to handle an MSR access access,
8225 * rather than handle it ourselves in L0. I.e., check whether L1 expressed
8226 * disinterest in the current event (read or write a specific MSR) by using an
8227 * MSR bitmap. This may be the case even when L0 doesn't use MSR bitmaps.
8229 static bool nested_vmx_exit_handled_msr(struct kvm_vcpu *vcpu,
8230 struct vmcs12 *vmcs12, u32 exit_reason)
8232 u32 msr_index = vcpu->arch.regs[VCPU_REGS_RCX];
8235 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
8239 * The MSR_BITMAP page is divided into four 1024-byte bitmaps,
8240 * for the four combinations of read/write and low/high MSR numbers.
8241 * First we need to figure out which of the four to use:
8243 bitmap = vmcs12->msr_bitmap;
8244 if (exit_reason == EXIT_REASON_MSR_WRITE)
8246 if (msr_index >= 0xc0000000) {
8247 msr_index -= 0xc0000000;
8251 /* Then read the msr_index'th bit from this bitmap: */
8252 if (msr_index < 1024*8) {
8254 if (kvm_vcpu_read_guest(vcpu, bitmap + msr_index/8, &b, 1))
8256 return 1 & (b >> (msr_index & 7));
8258 return true; /* let L1 handle the wrong parameter */
8262 * Return 1 if we should exit from L2 to L1 to handle a CR access exit,
8263 * rather than handle it ourselves in L0. I.e., check if L1 wanted to
8264 * intercept (via guest_host_mask etc.) the current event.
8266 static bool nested_vmx_exit_handled_cr(struct kvm_vcpu *vcpu,
8267 struct vmcs12 *vmcs12)
8269 unsigned long exit_qualification = vmcs_readl(EXIT_QUALIFICATION);
8270 int cr = exit_qualification & 15;
8274 switch ((exit_qualification >> 4) & 3) {
8275 case 0: /* mov to cr */
8276 reg = (exit_qualification >> 8) & 15;
8277 val = kvm_register_readl(vcpu, reg);
8280 if (vmcs12->cr0_guest_host_mask &
8281 (val ^ vmcs12->cr0_read_shadow))
8285 if ((vmcs12->cr3_target_count >= 1 &&
8286 vmcs12->cr3_target_value0 == val) ||
8287 (vmcs12->cr3_target_count >= 2 &&
8288 vmcs12->cr3_target_value1 == val) ||
8289 (vmcs12->cr3_target_count >= 3 &&
8290 vmcs12->cr3_target_value2 == val) ||
8291 (vmcs12->cr3_target_count >= 4 &&
8292 vmcs12->cr3_target_value3 == val))
8294 if (nested_cpu_has(vmcs12, CPU_BASED_CR3_LOAD_EXITING))
8298 if (vmcs12->cr4_guest_host_mask &
8299 (vmcs12->cr4_read_shadow ^ val))
8303 if (nested_cpu_has(vmcs12, CPU_BASED_CR8_LOAD_EXITING))
8309 if ((vmcs12->cr0_guest_host_mask & X86_CR0_TS) &&
8310 (vmcs12->cr0_read_shadow & X86_CR0_TS))
8313 case 1: /* mov from cr */
8316 if (vmcs12->cpu_based_vm_exec_control &
8317 CPU_BASED_CR3_STORE_EXITING)
8321 if (vmcs12->cpu_based_vm_exec_control &
8322 CPU_BASED_CR8_STORE_EXITING)
8329 * lmsw can change bits 1..3 of cr0, and only set bit 0 of
8330 * cr0. Other attempted changes are ignored, with no exit.
8332 val = (exit_qualification >> LMSW_SOURCE_DATA_SHIFT) & 0x0f;
8333 if (vmcs12->cr0_guest_host_mask & 0xe &
8334 (val ^ vmcs12->cr0_read_shadow))
8336 if ((vmcs12->cr0_guest_host_mask & 0x1) &&
8337 !(vmcs12->cr0_read_shadow & 0x1) &&
8346 * Return 1 if we should exit from L2 to L1 to handle an exit, or 0 if we
8347 * should handle it ourselves in L0 (and then continue L2). Only call this
8348 * when in is_guest_mode (L2).
8350 static bool nested_vmx_exit_reflected(struct kvm_vcpu *vcpu, u32 exit_reason)
8352 u32 intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8353 struct vcpu_vmx *vmx = to_vmx(vcpu);
8354 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8356 if (vmx->nested.nested_run_pending)
8359 if (unlikely(vmx->fail)) {
8360 pr_info_ratelimited("%s failed vm entry %x\n", __func__,
8361 vmcs_read32(VM_INSTRUCTION_ERROR));
8366 * The host physical addresses of some pages of guest memory
8367 * are loaded into VMCS02 (e.g. L1's Virtual APIC Page). The CPU
8368 * may write to these pages via their host physical address while
8369 * L2 is running, bypassing any address-translation-based dirty
8370 * tracking (e.g. EPT write protection).
8372 * Mark them dirty on every exit from L2 to prevent them from
8373 * getting out of sync with dirty tracking.
8375 nested_mark_vmcs12_pages_dirty(vcpu);
8377 trace_kvm_nested_vmexit(kvm_rip_read(vcpu), exit_reason,
8378 vmcs_readl(EXIT_QUALIFICATION),
8379 vmx->idt_vectoring_info,
8381 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8384 switch (exit_reason) {
8385 case EXIT_REASON_EXCEPTION_NMI:
8386 if (is_nmi(intr_info))
8388 else if (is_page_fault(intr_info))
8389 return !vmx->vcpu.arch.apf.host_apf_reason && enable_ept;
8390 else if (is_no_device(intr_info) &&
8391 !(vmcs12->guest_cr0 & X86_CR0_TS))
8393 else if (is_debug(intr_info) &&
8395 (KVM_GUESTDBG_SINGLESTEP | KVM_GUESTDBG_USE_HW_BP))
8397 else if (is_breakpoint(intr_info) &&
8398 vcpu->guest_debug & KVM_GUESTDBG_USE_SW_BP)
8400 return vmcs12->exception_bitmap &
8401 (1u << (intr_info & INTR_INFO_VECTOR_MASK));
8402 case EXIT_REASON_EXTERNAL_INTERRUPT:
8404 case EXIT_REASON_TRIPLE_FAULT:
8406 case EXIT_REASON_PENDING_INTERRUPT:
8407 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_INTR_PENDING);
8408 case EXIT_REASON_NMI_WINDOW:
8409 return nested_cpu_has(vmcs12, CPU_BASED_VIRTUAL_NMI_PENDING);
8410 case EXIT_REASON_TASK_SWITCH:
8412 case EXIT_REASON_CPUID:
8414 case EXIT_REASON_HLT:
8415 return nested_cpu_has(vmcs12, CPU_BASED_HLT_EXITING);
8416 case EXIT_REASON_INVD:
8418 case EXIT_REASON_INVLPG:
8419 return nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8420 case EXIT_REASON_RDPMC:
8421 return nested_cpu_has(vmcs12, CPU_BASED_RDPMC_EXITING);
8422 case EXIT_REASON_RDRAND:
8423 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDRAND);
8424 case EXIT_REASON_RDSEED:
8425 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_RDSEED);
8426 case EXIT_REASON_RDTSC: case EXIT_REASON_RDTSCP:
8427 return nested_cpu_has(vmcs12, CPU_BASED_RDTSC_EXITING);
8428 case EXIT_REASON_VMCALL: case EXIT_REASON_VMCLEAR:
8429 case EXIT_REASON_VMLAUNCH: case EXIT_REASON_VMPTRLD:
8430 case EXIT_REASON_VMPTRST: case EXIT_REASON_VMREAD:
8431 case EXIT_REASON_VMRESUME: case EXIT_REASON_VMWRITE:
8432 case EXIT_REASON_VMOFF: case EXIT_REASON_VMON:
8433 case EXIT_REASON_INVEPT: case EXIT_REASON_INVVPID:
8435 * VMX instructions trap unconditionally. This allows L1 to
8436 * emulate them for its L2 guest, i.e., allows 3-level nesting!
8439 case EXIT_REASON_CR_ACCESS:
8440 return nested_vmx_exit_handled_cr(vcpu, vmcs12);
8441 case EXIT_REASON_DR_ACCESS:
8442 return nested_cpu_has(vmcs12, CPU_BASED_MOV_DR_EXITING);
8443 case EXIT_REASON_IO_INSTRUCTION:
8444 return nested_vmx_exit_handled_io(vcpu, vmcs12);
8445 case EXIT_REASON_GDTR_IDTR: case EXIT_REASON_LDTR_TR:
8446 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_DESC);
8447 case EXIT_REASON_MSR_READ:
8448 case EXIT_REASON_MSR_WRITE:
8449 return nested_vmx_exit_handled_msr(vcpu, vmcs12, exit_reason);
8450 case EXIT_REASON_INVALID_STATE:
8452 case EXIT_REASON_MWAIT_INSTRUCTION:
8453 return nested_cpu_has(vmcs12, CPU_BASED_MWAIT_EXITING);
8454 case EXIT_REASON_MONITOR_TRAP_FLAG:
8455 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_TRAP_FLAG);
8456 case EXIT_REASON_MONITOR_INSTRUCTION:
8457 return nested_cpu_has(vmcs12, CPU_BASED_MONITOR_EXITING);
8458 case EXIT_REASON_PAUSE_INSTRUCTION:
8459 return nested_cpu_has(vmcs12, CPU_BASED_PAUSE_EXITING) ||
8460 nested_cpu_has2(vmcs12,
8461 SECONDARY_EXEC_PAUSE_LOOP_EXITING);
8462 case EXIT_REASON_MCE_DURING_VMENTRY:
8464 case EXIT_REASON_TPR_BELOW_THRESHOLD:
8465 return nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW);
8466 case EXIT_REASON_APIC_ACCESS:
8467 return nested_cpu_has2(vmcs12,
8468 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
8469 case EXIT_REASON_APIC_WRITE:
8470 case EXIT_REASON_EOI_INDUCED:
8471 /* apic_write and eoi_induced should exit unconditionally. */
8473 case EXIT_REASON_EPT_VIOLATION:
8475 * L0 always deals with the EPT violation. If nested EPT is
8476 * used, and the nested mmu code discovers that the address is
8477 * missing in the guest EPT table (EPT12), the EPT violation
8478 * will be injected with nested_ept_inject_page_fault()
8481 case EXIT_REASON_EPT_MISCONFIG:
8483 * L2 never uses directly L1's EPT, but rather L0's own EPT
8484 * table (shadow on EPT) or a merged EPT table that L0 built
8485 * (EPT on EPT). So any problems with the structure of the
8486 * table is L0's fault.
8489 case EXIT_REASON_INVPCID:
8491 nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_INVPCID) &&
8492 nested_cpu_has(vmcs12, CPU_BASED_INVLPG_EXITING);
8493 case EXIT_REASON_WBINVD:
8494 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_WBINVD_EXITING);
8495 case EXIT_REASON_XSETBV:
8497 case EXIT_REASON_XSAVES: case EXIT_REASON_XRSTORS:
8499 * This should never happen, since it is not possible to
8500 * set XSS to a non-zero value---neither in L1 nor in L2.
8501 * If if it were, XSS would have to be checked against
8502 * the XSS exit bitmap in vmcs12.
8504 return nested_cpu_has2(vmcs12, SECONDARY_EXEC_XSAVES);
8505 case EXIT_REASON_PREEMPTION_TIMER:
8507 case EXIT_REASON_PML_FULL:
8508 /* We emulate PML support to L1. */
8510 case EXIT_REASON_VMFUNC:
8511 /* VM functions are emulated through L2->L0 vmexits. */
8518 static int nested_vmx_reflect_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason)
8520 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
8523 * At this point, the exit interruption info in exit_intr_info
8524 * is only valid for EXCEPTION_NMI exits. For EXTERNAL_INTERRUPT
8525 * we need to query the in-kernel LAPIC.
8527 WARN_ON(exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT);
8528 if ((exit_intr_info &
8529 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) ==
8530 (INTR_INFO_VALID_MASK | INTR_INFO_DELIVER_CODE_MASK)) {
8531 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8532 vmcs12->vm_exit_intr_error_code =
8533 vmcs_read32(VM_EXIT_INTR_ERROR_CODE);
8536 nested_vmx_vmexit(vcpu, exit_reason, exit_intr_info,
8537 vmcs_readl(EXIT_QUALIFICATION));
8541 static void vmx_get_exit_info(struct kvm_vcpu *vcpu, u64 *info1, u64 *info2)
8543 *info1 = vmcs_readl(EXIT_QUALIFICATION);
8544 *info2 = vmcs_read32(VM_EXIT_INTR_INFO);
8547 static void vmx_destroy_pml_buffer(struct vcpu_vmx *vmx)
8550 __free_page(vmx->pml_pg);
8555 static void vmx_flush_pml_buffer(struct kvm_vcpu *vcpu)
8557 struct vcpu_vmx *vmx = to_vmx(vcpu);
8561 pml_idx = vmcs_read16(GUEST_PML_INDEX);
8563 /* Do nothing if PML buffer is empty */
8564 if (pml_idx == (PML_ENTITY_NUM - 1))
8567 /* PML index always points to next available PML buffer entity */
8568 if (pml_idx >= PML_ENTITY_NUM)
8573 pml_buf = page_address(vmx->pml_pg);
8574 for (; pml_idx < PML_ENTITY_NUM; pml_idx++) {
8577 gpa = pml_buf[pml_idx];
8578 WARN_ON(gpa & (PAGE_SIZE - 1));
8579 kvm_vcpu_mark_page_dirty(vcpu, gpa >> PAGE_SHIFT);
8582 /* reset PML index */
8583 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
8587 * Flush all vcpus' PML buffer and update logged GPAs to dirty_bitmap.
8588 * Called before reporting dirty_bitmap to userspace.
8590 static void kvm_flush_pml_buffers(struct kvm *kvm)
8593 struct kvm_vcpu *vcpu;
8595 * We only need to kick vcpu out of guest mode here, as PML buffer
8596 * is flushed at beginning of all VMEXITs, and it's obvious that only
8597 * vcpus running in guest are possible to have unflushed GPAs in PML
8600 kvm_for_each_vcpu(i, vcpu, kvm)
8601 kvm_vcpu_kick(vcpu);
8604 static void vmx_dump_sel(char *name, uint32_t sel)
8606 pr_err("%s sel=0x%04x, attr=0x%05x, limit=0x%08x, base=0x%016lx\n",
8607 name, vmcs_read16(sel),
8608 vmcs_read32(sel + GUEST_ES_AR_BYTES - GUEST_ES_SELECTOR),
8609 vmcs_read32(sel + GUEST_ES_LIMIT - GUEST_ES_SELECTOR),
8610 vmcs_readl(sel + GUEST_ES_BASE - GUEST_ES_SELECTOR));
8613 static void vmx_dump_dtsel(char *name, uint32_t limit)
8615 pr_err("%s limit=0x%08x, base=0x%016lx\n",
8616 name, vmcs_read32(limit),
8617 vmcs_readl(limit + GUEST_GDTR_BASE - GUEST_GDTR_LIMIT));
8620 static void dump_vmcs(void)
8622 u32 vmentry_ctl = vmcs_read32(VM_ENTRY_CONTROLS);
8623 u32 vmexit_ctl = vmcs_read32(VM_EXIT_CONTROLS);
8624 u32 cpu_based_exec_ctrl = vmcs_read32(CPU_BASED_VM_EXEC_CONTROL);
8625 u32 pin_based_exec_ctrl = vmcs_read32(PIN_BASED_VM_EXEC_CONTROL);
8626 u32 secondary_exec_control = 0;
8627 unsigned long cr4 = vmcs_readl(GUEST_CR4);
8628 u64 efer = vmcs_read64(GUEST_IA32_EFER);
8631 if (cpu_has_secondary_exec_ctrls())
8632 secondary_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8634 pr_err("*** Guest State ***\n");
8635 pr_err("CR0: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8636 vmcs_readl(GUEST_CR0), vmcs_readl(CR0_READ_SHADOW),
8637 vmcs_readl(CR0_GUEST_HOST_MASK));
8638 pr_err("CR4: actual=0x%016lx, shadow=0x%016lx, gh_mask=%016lx\n",
8639 cr4, vmcs_readl(CR4_READ_SHADOW), vmcs_readl(CR4_GUEST_HOST_MASK));
8640 pr_err("CR3 = 0x%016lx\n", vmcs_readl(GUEST_CR3));
8641 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT) &&
8642 (cr4 & X86_CR4_PAE) && !(efer & EFER_LMA))
8644 pr_err("PDPTR0 = 0x%016llx PDPTR1 = 0x%016llx\n",
8645 vmcs_read64(GUEST_PDPTR0), vmcs_read64(GUEST_PDPTR1));
8646 pr_err("PDPTR2 = 0x%016llx PDPTR3 = 0x%016llx\n",
8647 vmcs_read64(GUEST_PDPTR2), vmcs_read64(GUEST_PDPTR3));
8649 pr_err("RSP = 0x%016lx RIP = 0x%016lx\n",
8650 vmcs_readl(GUEST_RSP), vmcs_readl(GUEST_RIP));
8651 pr_err("RFLAGS=0x%08lx DR7 = 0x%016lx\n",
8652 vmcs_readl(GUEST_RFLAGS), vmcs_readl(GUEST_DR7));
8653 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8654 vmcs_readl(GUEST_SYSENTER_ESP),
8655 vmcs_read32(GUEST_SYSENTER_CS), vmcs_readl(GUEST_SYSENTER_EIP));
8656 vmx_dump_sel("CS: ", GUEST_CS_SELECTOR);
8657 vmx_dump_sel("DS: ", GUEST_DS_SELECTOR);
8658 vmx_dump_sel("SS: ", GUEST_SS_SELECTOR);
8659 vmx_dump_sel("ES: ", GUEST_ES_SELECTOR);
8660 vmx_dump_sel("FS: ", GUEST_FS_SELECTOR);
8661 vmx_dump_sel("GS: ", GUEST_GS_SELECTOR);
8662 vmx_dump_dtsel("GDTR:", GUEST_GDTR_LIMIT);
8663 vmx_dump_sel("LDTR:", GUEST_LDTR_SELECTOR);
8664 vmx_dump_dtsel("IDTR:", GUEST_IDTR_LIMIT);
8665 vmx_dump_sel("TR: ", GUEST_TR_SELECTOR);
8666 if ((vmexit_ctl & (VM_EXIT_SAVE_IA32_PAT | VM_EXIT_SAVE_IA32_EFER)) ||
8667 (vmentry_ctl & (VM_ENTRY_LOAD_IA32_PAT | VM_ENTRY_LOAD_IA32_EFER)))
8668 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8669 efer, vmcs_read64(GUEST_IA32_PAT));
8670 pr_err("DebugCtl = 0x%016llx DebugExceptions = 0x%016lx\n",
8671 vmcs_read64(GUEST_IA32_DEBUGCTL),
8672 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS));
8673 if (vmentry_ctl & VM_ENTRY_LOAD_IA32_PERF_GLOBAL_CTRL)
8674 pr_err("PerfGlobCtl = 0x%016llx\n",
8675 vmcs_read64(GUEST_IA32_PERF_GLOBAL_CTRL));
8676 if (vmentry_ctl & VM_ENTRY_LOAD_BNDCFGS)
8677 pr_err("BndCfgS = 0x%016llx\n", vmcs_read64(GUEST_BNDCFGS));
8678 pr_err("Interruptibility = %08x ActivityState = %08x\n",
8679 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO),
8680 vmcs_read32(GUEST_ACTIVITY_STATE));
8681 if (secondary_exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY)
8682 pr_err("InterruptStatus = %04x\n",
8683 vmcs_read16(GUEST_INTR_STATUS));
8685 pr_err("*** Host State ***\n");
8686 pr_err("RIP = 0x%016lx RSP = 0x%016lx\n",
8687 vmcs_readl(HOST_RIP), vmcs_readl(HOST_RSP));
8688 pr_err("CS=%04x SS=%04x DS=%04x ES=%04x FS=%04x GS=%04x TR=%04x\n",
8689 vmcs_read16(HOST_CS_SELECTOR), vmcs_read16(HOST_SS_SELECTOR),
8690 vmcs_read16(HOST_DS_SELECTOR), vmcs_read16(HOST_ES_SELECTOR),
8691 vmcs_read16(HOST_FS_SELECTOR), vmcs_read16(HOST_GS_SELECTOR),
8692 vmcs_read16(HOST_TR_SELECTOR));
8693 pr_err("FSBase=%016lx GSBase=%016lx TRBase=%016lx\n",
8694 vmcs_readl(HOST_FS_BASE), vmcs_readl(HOST_GS_BASE),
8695 vmcs_readl(HOST_TR_BASE));
8696 pr_err("GDTBase=%016lx IDTBase=%016lx\n",
8697 vmcs_readl(HOST_GDTR_BASE), vmcs_readl(HOST_IDTR_BASE));
8698 pr_err("CR0=%016lx CR3=%016lx CR4=%016lx\n",
8699 vmcs_readl(HOST_CR0), vmcs_readl(HOST_CR3),
8700 vmcs_readl(HOST_CR4));
8701 pr_err("Sysenter RSP=%016lx CS:RIP=%04x:%016lx\n",
8702 vmcs_readl(HOST_IA32_SYSENTER_ESP),
8703 vmcs_read32(HOST_IA32_SYSENTER_CS),
8704 vmcs_readl(HOST_IA32_SYSENTER_EIP));
8705 if (vmexit_ctl & (VM_EXIT_LOAD_IA32_PAT | VM_EXIT_LOAD_IA32_EFER))
8706 pr_err("EFER = 0x%016llx PAT = 0x%016llx\n",
8707 vmcs_read64(HOST_IA32_EFER),
8708 vmcs_read64(HOST_IA32_PAT));
8709 if (vmexit_ctl & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
8710 pr_err("PerfGlobCtl = 0x%016llx\n",
8711 vmcs_read64(HOST_IA32_PERF_GLOBAL_CTRL));
8713 pr_err("*** Control State ***\n");
8714 pr_err("PinBased=%08x CPUBased=%08x SecondaryExec=%08x\n",
8715 pin_based_exec_ctrl, cpu_based_exec_ctrl, secondary_exec_control);
8716 pr_err("EntryControls=%08x ExitControls=%08x\n", vmentry_ctl, vmexit_ctl);
8717 pr_err("ExceptionBitmap=%08x PFECmask=%08x PFECmatch=%08x\n",
8718 vmcs_read32(EXCEPTION_BITMAP),
8719 vmcs_read32(PAGE_FAULT_ERROR_CODE_MASK),
8720 vmcs_read32(PAGE_FAULT_ERROR_CODE_MATCH));
8721 pr_err("VMEntry: intr_info=%08x errcode=%08x ilen=%08x\n",
8722 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
8723 vmcs_read32(VM_ENTRY_EXCEPTION_ERROR_CODE),
8724 vmcs_read32(VM_ENTRY_INSTRUCTION_LEN));
8725 pr_err("VMExit: intr_info=%08x errcode=%08x ilen=%08x\n",
8726 vmcs_read32(VM_EXIT_INTR_INFO),
8727 vmcs_read32(VM_EXIT_INTR_ERROR_CODE),
8728 vmcs_read32(VM_EXIT_INSTRUCTION_LEN));
8729 pr_err(" reason=%08x qualification=%016lx\n",
8730 vmcs_read32(VM_EXIT_REASON), vmcs_readl(EXIT_QUALIFICATION));
8731 pr_err("IDTVectoring: info=%08x errcode=%08x\n",
8732 vmcs_read32(IDT_VECTORING_INFO_FIELD),
8733 vmcs_read32(IDT_VECTORING_ERROR_CODE));
8734 pr_err("TSC Offset = 0x%016llx\n", vmcs_read64(TSC_OFFSET));
8735 if (secondary_exec_control & SECONDARY_EXEC_TSC_SCALING)
8736 pr_err("TSC Multiplier = 0x%016llx\n",
8737 vmcs_read64(TSC_MULTIPLIER));
8738 if (cpu_based_exec_ctrl & CPU_BASED_TPR_SHADOW)
8739 pr_err("TPR Threshold = 0x%02x\n", vmcs_read32(TPR_THRESHOLD));
8740 if (pin_based_exec_ctrl & PIN_BASED_POSTED_INTR)
8741 pr_err("PostedIntrVec = 0x%02x\n", vmcs_read16(POSTED_INTR_NV));
8742 if ((secondary_exec_control & SECONDARY_EXEC_ENABLE_EPT))
8743 pr_err("EPT pointer = 0x%016llx\n", vmcs_read64(EPT_POINTER));
8744 n = vmcs_read32(CR3_TARGET_COUNT);
8745 for (i = 0; i + 1 < n; i += 4)
8746 pr_err("CR3 target%u=%016lx target%u=%016lx\n",
8747 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2),
8748 i + 1, vmcs_readl(CR3_TARGET_VALUE0 + i * 2 + 2));
8750 pr_err("CR3 target%u=%016lx\n",
8751 i, vmcs_readl(CR3_TARGET_VALUE0 + i * 2));
8752 if (secondary_exec_control & SECONDARY_EXEC_PAUSE_LOOP_EXITING)
8753 pr_err("PLE Gap=%08x Window=%08x\n",
8754 vmcs_read32(PLE_GAP), vmcs_read32(PLE_WINDOW));
8755 if (secondary_exec_control & SECONDARY_EXEC_ENABLE_VPID)
8756 pr_err("Virtual processor ID = 0x%04x\n",
8757 vmcs_read16(VIRTUAL_PROCESSOR_ID));
8761 * The guest has exited. See if we can fix it or if we need userspace
8764 static int vmx_handle_exit(struct kvm_vcpu *vcpu)
8766 struct vcpu_vmx *vmx = to_vmx(vcpu);
8767 u32 exit_reason = vmx->exit_reason;
8768 u32 vectoring_info = vmx->idt_vectoring_info;
8770 trace_kvm_exit(exit_reason, vcpu, KVM_ISA_VMX);
8773 * Flush logged GPAs PML buffer, this will make dirty_bitmap more
8774 * updated. Another good is, in kvm_vm_ioctl_get_dirty_log, before
8775 * querying dirty_bitmap, we only need to kick all vcpus out of guest
8776 * mode as if vcpus is in root mode, the PML buffer must has been
8780 vmx_flush_pml_buffer(vcpu);
8782 /* If guest state is invalid, start emulating */
8783 if (vmx->emulation_required)
8784 return handle_invalid_guest_state(vcpu);
8786 if (is_guest_mode(vcpu) && nested_vmx_exit_reflected(vcpu, exit_reason))
8787 return nested_vmx_reflect_vmexit(vcpu, exit_reason);
8789 if (exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY) {
8791 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8792 vcpu->run->fail_entry.hardware_entry_failure_reason
8797 if (unlikely(vmx->fail)) {
8798 vcpu->run->exit_reason = KVM_EXIT_FAIL_ENTRY;
8799 vcpu->run->fail_entry.hardware_entry_failure_reason
8800 = vmcs_read32(VM_INSTRUCTION_ERROR);
8806 * Do not try to fix EXIT_REASON_EPT_MISCONFIG if it caused by
8807 * delivery event since it indicates guest is accessing MMIO.
8808 * The vm-exit can be triggered again after return to guest that
8809 * will cause infinite loop.
8811 if ((vectoring_info & VECTORING_INFO_VALID_MASK) &&
8812 (exit_reason != EXIT_REASON_EXCEPTION_NMI &&
8813 exit_reason != EXIT_REASON_EPT_VIOLATION &&
8814 exit_reason != EXIT_REASON_PML_FULL &&
8815 exit_reason != EXIT_REASON_TASK_SWITCH)) {
8816 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
8817 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_DELIVERY_EV;
8818 vcpu->run->internal.ndata = 3;
8819 vcpu->run->internal.data[0] = vectoring_info;
8820 vcpu->run->internal.data[1] = exit_reason;
8821 vcpu->run->internal.data[2] = vcpu->arch.exit_qualification;
8822 if (exit_reason == EXIT_REASON_EPT_MISCONFIG) {
8823 vcpu->run->internal.ndata++;
8824 vcpu->run->internal.data[3] =
8825 vmcs_read64(GUEST_PHYSICAL_ADDRESS);
8830 if (exit_reason < kvm_vmx_max_exit_handlers
8831 && kvm_vmx_exit_handlers[exit_reason])
8832 return kvm_vmx_exit_handlers[exit_reason](vcpu);
8834 vcpu_unimpl(vcpu, "vmx: unexpected exit reason 0x%x\n",
8836 kvm_queue_exception(vcpu, UD_VECTOR);
8841 static void update_cr8_intercept(struct kvm_vcpu *vcpu, int tpr, int irr)
8843 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
8845 if (is_guest_mode(vcpu) &&
8846 nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
8849 if (irr == -1 || tpr < irr) {
8850 vmcs_write32(TPR_THRESHOLD, 0);
8854 vmcs_write32(TPR_THRESHOLD, irr);
8857 static void vmx_set_virtual_x2apic_mode(struct kvm_vcpu *vcpu, bool set)
8859 u32 sec_exec_control;
8861 /* Postpone execution until vmcs01 is the current VMCS. */
8862 if (is_guest_mode(vcpu)) {
8863 to_vmx(vcpu)->nested.change_vmcs01_virtual_x2apic_mode = true;
8867 if (!cpu_has_vmx_virtualize_x2apic_mode())
8870 if (!cpu_need_tpr_shadow(vcpu))
8873 sec_exec_control = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
8876 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8877 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8879 sec_exec_control &= ~SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE;
8880 sec_exec_control |= SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
8881 vmx_flush_tlb_ept_only(vcpu);
8883 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, sec_exec_control);
8885 vmx_set_msr_bitmap(vcpu);
8888 static void vmx_set_apic_access_page_addr(struct kvm_vcpu *vcpu, hpa_t hpa)
8890 struct vcpu_vmx *vmx = to_vmx(vcpu);
8893 * Currently we do not handle the nested case where L2 has an
8894 * APIC access page of its own; that page is still pinned.
8895 * Hence, we skip the case where the VCPU is in guest mode _and_
8896 * L1 prepared an APIC access page for L2.
8898 * For the case where L1 and L2 share the same APIC access page
8899 * (flexpriority=Y but SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES clear
8900 * in the vmcs12), this function will only update either the vmcs01
8901 * or the vmcs02. If the former, the vmcs02 will be updated by
8902 * prepare_vmcs02. If the latter, the vmcs01 will be updated in
8903 * the next L2->L1 exit.
8905 if (!is_guest_mode(vcpu) ||
8906 !nested_cpu_has2(get_vmcs12(&vmx->vcpu),
8907 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
8908 vmcs_write64(APIC_ACCESS_ADDR, hpa);
8909 vmx_flush_tlb_ept_only(vcpu);
8913 static void vmx_hwapic_isr_update(struct kvm_vcpu *vcpu, int max_isr)
8921 status = vmcs_read16(GUEST_INTR_STATUS);
8923 if (max_isr != old) {
8925 status |= max_isr << 8;
8926 vmcs_write16(GUEST_INTR_STATUS, status);
8930 static void vmx_set_rvi(int vector)
8938 status = vmcs_read16(GUEST_INTR_STATUS);
8939 old = (u8)status & 0xff;
8940 if ((u8)vector != old) {
8942 status |= (u8)vector;
8943 vmcs_write16(GUEST_INTR_STATUS, status);
8947 static void vmx_hwapic_irr_update(struct kvm_vcpu *vcpu, int max_irr)
8949 if (!is_guest_mode(vcpu)) {
8950 vmx_set_rvi(max_irr);
8958 * In guest mode. If a vmexit is needed, vmx_check_nested_events
8961 if (nested_exit_on_intr(vcpu))
8965 * Else, fall back to pre-APICv interrupt injection since L2
8966 * is run without virtual interrupt delivery.
8968 if (!kvm_event_needs_reinjection(vcpu) &&
8969 vmx_interrupt_allowed(vcpu)) {
8970 kvm_queue_interrupt(vcpu, max_irr, false);
8971 vmx_inject_irq(vcpu);
8975 static int vmx_sync_pir_to_irr(struct kvm_vcpu *vcpu)
8977 struct vcpu_vmx *vmx = to_vmx(vcpu);
8980 WARN_ON(!vcpu->arch.apicv_active);
8981 if (pi_test_on(&vmx->pi_desc)) {
8982 pi_clear_on(&vmx->pi_desc);
8984 * IOMMU can write to PIR.ON, so the barrier matters even on UP.
8985 * But on x86 this is just a compiler barrier anyway.
8987 smp_mb__after_atomic();
8988 max_irr = kvm_apic_update_irr(vcpu, vmx->pi_desc.pir);
8990 max_irr = kvm_lapic_find_highest_irr(vcpu);
8992 vmx_hwapic_irr_update(vcpu, max_irr);
8996 static void vmx_load_eoi_exitmap(struct kvm_vcpu *vcpu, u64 *eoi_exit_bitmap)
8998 if (!kvm_vcpu_apicv_active(vcpu))
9001 vmcs_write64(EOI_EXIT_BITMAP0, eoi_exit_bitmap[0]);
9002 vmcs_write64(EOI_EXIT_BITMAP1, eoi_exit_bitmap[1]);
9003 vmcs_write64(EOI_EXIT_BITMAP2, eoi_exit_bitmap[2]);
9004 vmcs_write64(EOI_EXIT_BITMAP3, eoi_exit_bitmap[3]);
9007 static void vmx_apicv_post_state_restore(struct kvm_vcpu *vcpu)
9009 struct vcpu_vmx *vmx = to_vmx(vcpu);
9011 pi_clear_on(&vmx->pi_desc);
9012 memset(vmx->pi_desc.pir, 0, sizeof(vmx->pi_desc.pir));
9015 static void vmx_complete_atomic_exit(struct vcpu_vmx *vmx)
9017 u32 exit_intr_info = 0;
9018 u16 basic_exit_reason = (u16)vmx->exit_reason;
9020 if (!(basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY
9021 || basic_exit_reason == EXIT_REASON_EXCEPTION_NMI))
9024 if (!(vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
9025 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9026 vmx->exit_intr_info = exit_intr_info;
9028 /* if exit due to PF check for async PF */
9029 if (is_page_fault(exit_intr_info))
9030 vmx->vcpu.arch.apf.host_apf_reason = kvm_read_and_reset_pf_reason();
9032 /* Handle machine checks before interrupts are enabled */
9033 if (basic_exit_reason == EXIT_REASON_MCE_DURING_VMENTRY ||
9034 is_machine_check(exit_intr_info))
9035 kvm_machine_check();
9037 /* We need to handle NMIs before interrupts are enabled */
9038 if (is_nmi(exit_intr_info)) {
9039 kvm_before_handle_nmi(&vmx->vcpu);
9041 kvm_after_handle_nmi(&vmx->vcpu);
9045 static void vmx_handle_external_intr(struct kvm_vcpu *vcpu)
9047 u32 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9049 if ((exit_intr_info & (INTR_INFO_VALID_MASK | INTR_INFO_INTR_TYPE_MASK))
9050 == (INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR)) {
9051 unsigned int vector;
9052 unsigned long entry;
9054 struct vcpu_vmx *vmx = to_vmx(vcpu);
9055 #ifdef CONFIG_X86_64
9059 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
9060 desc = (gate_desc *)vmx->host_idt_base + vector;
9061 entry = gate_offset(desc);
9063 #ifdef CONFIG_X86_64
9064 "mov %%" _ASM_SP ", %[sp]\n\t"
9065 "and $0xfffffffffffffff0, %%" _ASM_SP "\n\t"
9070 __ASM_SIZE(push) " $%c[cs]\n\t"
9071 "call *%[entry]\n\t"
9073 #ifdef CONFIG_X86_64
9079 [ss]"i"(__KERNEL_DS),
9080 [cs]"i"(__KERNEL_CS)
9084 STACK_FRAME_NON_STANDARD(vmx_handle_external_intr);
9086 static bool vmx_has_high_real_mode_segbase(void)
9088 return enable_unrestricted_guest || emulate_invalid_guest_state;
9091 static bool vmx_mpx_supported(void)
9093 return (vmcs_config.vmexit_ctrl & VM_EXIT_CLEAR_BNDCFGS) &&
9094 (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_BNDCFGS);
9097 static bool vmx_xsaves_supported(void)
9099 return vmcs_config.cpu_based_2nd_exec_ctrl &
9100 SECONDARY_EXEC_XSAVES;
9103 static void vmx_recover_nmi_blocking(struct vcpu_vmx *vmx)
9108 bool idtv_info_valid;
9110 idtv_info_valid = vmx->idt_vectoring_info & VECTORING_INFO_VALID_MASK;
9112 if (vmx->loaded_vmcs->nmi_known_unmasked)
9115 * Can't use vmx->exit_intr_info since we're not sure what
9116 * the exit reason is.
9118 exit_intr_info = vmcs_read32(VM_EXIT_INTR_INFO);
9119 unblock_nmi = (exit_intr_info & INTR_INFO_UNBLOCK_NMI) != 0;
9120 vector = exit_intr_info & INTR_INFO_VECTOR_MASK;
9122 * SDM 3: 27.7.1.2 (September 2008)
9123 * Re-set bit "block by NMI" before VM entry if vmexit caused by
9124 * a guest IRET fault.
9125 * SDM 3: 23.2.2 (September 2008)
9126 * Bit 12 is undefined in any of the following cases:
9127 * If the VM exit sets the valid bit in the IDT-vectoring
9128 * information field.
9129 * If the VM exit is due to a double fault.
9131 if ((exit_intr_info & INTR_INFO_VALID_MASK) && unblock_nmi &&
9132 vector != DF_VECTOR && !idtv_info_valid)
9133 vmcs_set_bits(GUEST_INTERRUPTIBILITY_INFO,
9134 GUEST_INTR_STATE_NMI);
9136 vmx->loaded_vmcs->nmi_known_unmasked =
9137 !(vmcs_read32(GUEST_INTERRUPTIBILITY_INFO)
9138 & GUEST_INTR_STATE_NMI);
9141 static void __vmx_complete_interrupts(struct kvm_vcpu *vcpu,
9142 u32 idt_vectoring_info,
9143 int instr_len_field,
9144 int error_code_field)
9148 bool idtv_info_valid;
9150 idtv_info_valid = idt_vectoring_info & VECTORING_INFO_VALID_MASK;
9152 vcpu->arch.nmi_injected = false;
9153 kvm_clear_exception_queue(vcpu);
9154 kvm_clear_interrupt_queue(vcpu);
9156 if (!idtv_info_valid)
9159 kvm_make_request(KVM_REQ_EVENT, vcpu);
9161 vector = idt_vectoring_info & VECTORING_INFO_VECTOR_MASK;
9162 type = idt_vectoring_info & VECTORING_INFO_TYPE_MASK;
9165 case INTR_TYPE_NMI_INTR:
9166 vcpu->arch.nmi_injected = true;
9168 * SDM 3: 27.7.1.2 (September 2008)
9169 * Clear bit "block by NMI" before VM entry if a NMI
9172 vmx_set_nmi_mask(vcpu, false);
9174 case INTR_TYPE_SOFT_EXCEPTION:
9175 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
9177 case INTR_TYPE_HARD_EXCEPTION:
9178 if (idt_vectoring_info & VECTORING_INFO_DELIVER_CODE_MASK) {
9179 u32 err = vmcs_read32(error_code_field);
9180 kvm_requeue_exception_e(vcpu, vector, err);
9182 kvm_requeue_exception(vcpu, vector);
9184 case INTR_TYPE_SOFT_INTR:
9185 vcpu->arch.event_exit_inst_len = vmcs_read32(instr_len_field);
9187 case INTR_TYPE_EXT_INTR:
9188 kvm_queue_interrupt(vcpu, vector, type == INTR_TYPE_SOFT_INTR);
9195 static void vmx_complete_interrupts(struct vcpu_vmx *vmx)
9197 __vmx_complete_interrupts(&vmx->vcpu, vmx->idt_vectoring_info,
9198 VM_EXIT_INSTRUCTION_LEN,
9199 IDT_VECTORING_ERROR_CODE);
9202 static void vmx_cancel_injection(struct kvm_vcpu *vcpu)
9204 __vmx_complete_interrupts(vcpu,
9205 vmcs_read32(VM_ENTRY_INTR_INFO_FIELD),
9206 VM_ENTRY_INSTRUCTION_LEN,
9207 VM_ENTRY_EXCEPTION_ERROR_CODE);
9209 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
9212 static void atomic_switch_perf_msrs(struct vcpu_vmx *vmx)
9215 struct perf_guest_switch_msr *msrs;
9217 msrs = perf_guest_get_msrs(&nr_msrs);
9222 for (i = 0; i < nr_msrs; i++)
9223 if (msrs[i].host == msrs[i].guest)
9224 clear_atomic_switch_msr(vmx, msrs[i].msr);
9226 add_atomic_switch_msr(vmx, msrs[i].msr, msrs[i].guest,
9230 static void vmx_arm_hv_timer(struct kvm_vcpu *vcpu)
9232 struct vcpu_vmx *vmx = to_vmx(vcpu);
9236 if (vmx->hv_deadline_tsc == -1)
9240 if (vmx->hv_deadline_tsc > tscl)
9241 /* sure to be 32 bit only because checked on set_hv_timer */
9242 delta_tsc = (u32)((vmx->hv_deadline_tsc - tscl) >>
9243 cpu_preemption_timer_multi);
9247 vmcs_write32(VMX_PREEMPTION_TIMER_VALUE, delta_tsc);
9250 static void __noclone vmx_vcpu_run(struct kvm_vcpu *vcpu)
9252 struct vcpu_vmx *vmx = to_vmx(vcpu);
9253 unsigned long debugctlmsr, cr3, cr4;
9255 /* Don't enter VMX if guest state is invalid, let the exit handler
9256 start emulation until we arrive back to a valid state */
9257 if (vmx->emulation_required)
9260 if (vmx->ple_window_dirty) {
9261 vmx->ple_window_dirty = false;
9262 vmcs_write32(PLE_WINDOW, vmx->ple_window);
9265 if (vmx->nested.sync_shadow_vmcs) {
9266 copy_vmcs12_to_shadow(vmx);
9267 vmx->nested.sync_shadow_vmcs = false;
9270 if (test_bit(VCPU_REGS_RSP, (unsigned long *)&vcpu->arch.regs_dirty))
9271 vmcs_writel(GUEST_RSP, vcpu->arch.regs[VCPU_REGS_RSP]);
9272 if (test_bit(VCPU_REGS_RIP, (unsigned long *)&vcpu->arch.regs_dirty))
9273 vmcs_writel(GUEST_RIP, vcpu->arch.regs[VCPU_REGS_RIP]);
9275 cr3 = __get_current_cr3_fast();
9276 if (unlikely(cr3 != vmx->loaded_vmcs->vmcs_host_cr3)) {
9277 vmcs_writel(HOST_CR3, cr3);
9278 vmx->loaded_vmcs->vmcs_host_cr3 = cr3;
9281 cr4 = cr4_read_shadow();
9282 if (unlikely(cr4 != vmx->loaded_vmcs->vmcs_host_cr4)) {
9283 vmcs_writel(HOST_CR4, cr4);
9284 vmx->loaded_vmcs->vmcs_host_cr4 = cr4;
9287 /* When single-stepping over STI and MOV SS, we must clear the
9288 * corresponding interruptibility bits in the guest state. Otherwise
9289 * vmentry fails as it then expects bit 14 (BS) in pending debug
9290 * exceptions being set, but that's not correct for the guest debugging
9292 if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)
9293 vmx_set_interrupt_shadow(vcpu, 0);
9295 if (static_cpu_has(X86_FEATURE_PKU) &&
9296 kvm_read_cr4_bits(vcpu, X86_CR4_PKE) &&
9297 vcpu->arch.pkru != vmx->host_pkru)
9298 __write_pkru(vcpu->arch.pkru);
9300 atomic_switch_perf_msrs(vmx);
9301 debugctlmsr = get_debugctlmsr();
9303 vmx_arm_hv_timer(vcpu);
9305 vmx->__launched = vmx->loaded_vmcs->launched;
9307 /* Store host registers */
9308 "push %%" _ASM_DX "; push %%" _ASM_BP ";"
9309 "push %%" _ASM_CX " \n\t" /* placeholder for guest rcx */
9310 "push %%" _ASM_CX " \n\t"
9311 "cmp %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9313 "mov %%" _ASM_SP ", %c[host_rsp](%0) \n\t"
9314 __ex(ASM_VMX_VMWRITE_RSP_RDX) "\n\t"
9316 /* Reload cr2 if changed */
9317 "mov %c[cr2](%0), %%" _ASM_AX " \n\t"
9318 "mov %%cr2, %%" _ASM_DX " \n\t"
9319 "cmp %%" _ASM_AX ", %%" _ASM_DX " \n\t"
9321 "mov %%" _ASM_AX", %%cr2 \n\t"
9323 /* Check if vmlaunch of vmresume is needed */
9324 "cmpl $0, %c[launched](%0) \n\t"
9325 /* Load guest registers. Don't clobber flags. */
9326 "mov %c[rax](%0), %%" _ASM_AX " \n\t"
9327 "mov %c[rbx](%0), %%" _ASM_BX " \n\t"
9328 "mov %c[rdx](%0), %%" _ASM_DX " \n\t"
9329 "mov %c[rsi](%0), %%" _ASM_SI " \n\t"
9330 "mov %c[rdi](%0), %%" _ASM_DI " \n\t"
9331 "mov %c[rbp](%0), %%" _ASM_BP " \n\t"
9332 #ifdef CONFIG_X86_64
9333 "mov %c[r8](%0), %%r8 \n\t"
9334 "mov %c[r9](%0), %%r9 \n\t"
9335 "mov %c[r10](%0), %%r10 \n\t"
9336 "mov %c[r11](%0), %%r11 \n\t"
9337 "mov %c[r12](%0), %%r12 \n\t"
9338 "mov %c[r13](%0), %%r13 \n\t"
9339 "mov %c[r14](%0), %%r14 \n\t"
9340 "mov %c[r15](%0), %%r15 \n\t"
9342 "mov %c[rcx](%0), %%" _ASM_CX " \n\t" /* kills %0 (ecx) */
9344 /* Enter guest mode */
9346 __ex(ASM_VMX_VMLAUNCH) "\n\t"
9348 "1: " __ex(ASM_VMX_VMRESUME) "\n\t"
9350 /* Save guest registers, load host registers, keep flags */
9351 "mov %0, %c[wordsize](%%" _ASM_SP ") \n\t"
9353 "mov %%" _ASM_AX ", %c[rax](%0) \n\t"
9354 "mov %%" _ASM_BX ", %c[rbx](%0) \n\t"
9355 __ASM_SIZE(pop) " %c[rcx](%0) \n\t"
9356 "mov %%" _ASM_DX ", %c[rdx](%0) \n\t"
9357 "mov %%" _ASM_SI ", %c[rsi](%0) \n\t"
9358 "mov %%" _ASM_DI ", %c[rdi](%0) \n\t"
9359 "mov %%" _ASM_BP ", %c[rbp](%0) \n\t"
9360 #ifdef CONFIG_X86_64
9361 "mov %%r8, %c[r8](%0) \n\t"
9362 "mov %%r9, %c[r9](%0) \n\t"
9363 "mov %%r10, %c[r10](%0) \n\t"
9364 "mov %%r11, %c[r11](%0) \n\t"
9365 "mov %%r12, %c[r12](%0) \n\t"
9366 "mov %%r13, %c[r13](%0) \n\t"
9367 "mov %%r14, %c[r14](%0) \n\t"
9368 "mov %%r15, %c[r15](%0) \n\t"
9370 "mov %%cr2, %%" _ASM_AX " \n\t"
9371 "mov %%" _ASM_AX ", %c[cr2](%0) \n\t"
9373 "pop %%" _ASM_BP "; pop %%" _ASM_DX " \n\t"
9374 "setbe %c[fail](%0) \n\t"
9375 ".pushsection .rodata \n\t"
9376 ".global vmx_return \n\t"
9377 "vmx_return: " _ASM_PTR " 2b \n\t"
9379 : : "c"(vmx), "d"((unsigned long)HOST_RSP),
9380 [launched]"i"(offsetof(struct vcpu_vmx, __launched)),
9381 [fail]"i"(offsetof(struct vcpu_vmx, fail)),
9382 [host_rsp]"i"(offsetof(struct vcpu_vmx, host_rsp)),
9383 [rax]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RAX])),
9384 [rbx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBX])),
9385 [rcx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RCX])),
9386 [rdx]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDX])),
9387 [rsi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RSI])),
9388 [rdi]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RDI])),
9389 [rbp]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_RBP])),
9390 #ifdef CONFIG_X86_64
9391 [r8]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R8])),
9392 [r9]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R9])),
9393 [r10]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R10])),
9394 [r11]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R11])),
9395 [r12]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R12])),
9396 [r13]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R13])),
9397 [r14]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R14])),
9398 [r15]"i"(offsetof(struct vcpu_vmx, vcpu.arch.regs[VCPU_REGS_R15])),
9400 [cr2]"i"(offsetof(struct vcpu_vmx, vcpu.arch.cr2)),
9401 [wordsize]"i"(sizeof(ulong))
9403 #ifdef CONFIG_X86_64
9404 , "rax", "rbx", "rdi", "rsi"
9405 , "r8", "r9", "r10", "r11", "r12", "r13", "r14", "r15"
9407 , "eax", "ebx", "edi", "esi"
9411 /* MSR_IA32_DEBUGCTLMSR is zeroed on vmexit. Restore it if needed */
9413 update_debugctlmsr(debugctlmsr);
9415 #ifndef CONFIG_X86_64
9417 * The sysexit path does not restore ds/es, so we must set them to
9418 * a reasonable value ourselves.
9420 * We can't defer this to vmx_load_host_state() since that function
9421 * may be executed in interrupt context, which saves and restore segments
9422 * around it, nullifying its effect.
9424 loadsegment(ds, __USER_DS);
9425 loadsegment(es, __USER_DS);
9428 vcpu->arch.regs_avail = ~((1 << VCPU_REGS_RIP) | (1 << VCPU_REGS_RSP)
9429 | (1 << VCPU_EXREG_RFLAGS)
9430 | (1 << VCPU_EXREG_PDPTR)
9431 | (1 << VCPU_EXREG_SEGMENTS)
9432 | (1 << VCPU_EXREG_CR3));
9433 vcpu->arch.regs_dirty = 0;
9436 * eager fpu is enabled if PKEY is supported and CR4 is switched
9437 * back on host, so it is safe to read guest PKRU from current
9440 if (static_cpu_has(X86_FEATURE_PKU) &&
9441 kvm_read_cr4_bits(vcpu, X86_CR4_PKE)) {
9442 vcpu->arch.pkru = __read_pkru();
9443 if (vcpu->arch.pkru != vmx->host_pkru)
9444 __write_pkru(vmx->host_pkru);
9448 * the KVM_REQ_EVENT optimization bit is only on for one entry, and if
9449 * we did not inject a still-pending event to L1 now because of
9450 * nested_run_pending, we need to re-enable this bit.
9452 if (vmx->nested.nested_run_pending)
9453 kvm_make_request(KVM_REQ_EVENT, vcpu);
9455 vmx->nested.nested_run_pending = 0;
9456 vmx->idt_vectoring_info = 0;
9458 vmx->exit_reason = vmx->fail ? 0xdead : vmcs_read32(VM_EXIT_REASON);
9459 if (vmx->fail || (vmx->exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY))
9462 vmx->loaded_vmcs->launched = 1;
9463 vmx->idt_vectoring_info = vmcs_read32(IDT_VECTORING_INFO_FIELD);
9465 vmx_complete_atomic_exit(vmx);
9466 vmx_recover_nmi_blocking(vmx);
9467 vmx_complete_interrupts(vmx);
9469 STACK_FRAME_NON_STANDARD(vmx_vcpu_run);
9471 static void vmx_switch_vmcs(struct kvm_vcpu *vcpu, struct loaded_vmcs *vmcs)
9473 struct vcpu_vmx *vmx = to_vmx(vcpu);
9476 if (vmx->loaded_vmcs == vmcs)
9480 vmx->loaded_vmcs = vmcs;
9482 vmx_vcpu_load(vcpu, cpu);
9488 * Ensure that the current vmcs of the logical processor is the
9489 * vmcs01 of the vcpu before calling free_nested().
9491 static void vmx_free_vcpu_nested(struct kvm_vcpu *vcpu)
9493 struct vcpu_vmx *vmx = to_vmx(vcpu);
9496 r = vcpu_load(vcpu);
9498 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
9503 static void vmx_free_vcpu(struct kvm_vcpu *vcpu)
9505 struct vcpu_vmx *vmx = to_vmx(vcpu);
9508 vmx_destroy_pml_buffer(vmx);
9509 free_vpid(vmx->vpid);
9510 leave_guest_mode(vcpu);
9511 vmx_free_vcpu_nested(vcpu);
9512 free_loaded_vmcs(vmx->loaded_vmcs);
9513 kfree(vmx->guest_msrs);
9514 kvm_vcpu_uninit(vcpu);
9515 kmem_cache_free(kvm_vcpu_cache, vmx);
9518 static struct kvm_vcpu *vmx_create_vcpu(struct kvm *kvm, unsigned int id)
9521 struct vcpu_vmx *vmx = kmem_cache_zalloc(kvm_vcpu_cache, GFP_KERNEL);
9525 return ERR_PTR(-ENOMEM);
9527 vmx->vpid = allocate_vpid();
9529 err = kvm_vcpu_init(&vmx->vcpu, kvm, id);
9536 * If PML is turned on, failure on enabling PML just results in failure
9537 * of creating the vcpu, therefore we can simplify PML logic (by
9538 * avoiding dealing with cases, such as enabling PML partially on vcpus
9539 * for the guest, etc.
9542 vmx->pml_pg = alloc_page(GFP_KERNEL | __GFP_ZERO);
9547 vmx->guest_msrs = kmalloc(PAGE_SIZE, GFP_KERNEL);
9548 BUILD_BUG_ON(ARRAY_SIZE(vmx_msr_index) * sizeof(vmx->guest_msrs[0])
9551 if (!vmx->guest_msrs)
9554 vmx->loaded_vmcs = &vmx->vmcs01;
9555 vmx->loaded_vmcs->vmcs = alloc_vmcs();
9556 vmx->loaded_vmcs->shadow_vmcs = NULL;
9557 if (!vmx->loaded_vmcs->vmcs)
9559 loaded_vmcs_init(vmx->loaded_vmcs);
9562 vmx_vcpu_load(&vmx->vcpu, cpu);
9563 vmx->vcpu.cpu = cpu;
9564 vmx_vcpu_setup(vmx);
9565 vmx_vcpu_put(&vmx->vcpu);
9567 if (cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9568 err = alloc_apic_access_page(kvm);
9574 if (!kvm->arch.ept_identity_map_addr)
9575 kvm->arch.ept_identity_map_addr =
9576 VMX_EPT_IDENTITY_PAGETABLE_ADDR;
9577 err = init_rmode_identity_map(kvm);
9583 nested_vmx_setup_ctls_msrs(vmx);
9584 vmx->nested.vpid02 = allocate_vpid();
9587 vmx->nested.posted_intr_nv = -1;
9588 vmx->nested.current_vmptr = -1ull;
9590 vmx->msr_ia32_feature_control_valid_bits = FEATURE_CONTROL_LOCKED;
9593 * Enforce invariant: pi_desc.nv is always either POSTED_INTR_VECTOR
9594 * or POSTED_INTR_WAKEUP_VECTOR.
9596 vmx->pi_desc.nv = POSTED_INTR_VECTOR;
9597 vmx->pi_desc.sn = 1;
9602 free_vpid(vmx->nested.vpid02);
9603 free_loaded_vmcs(vmx->loaded_vmcs);
9605 kfree(vmx->guest_msrs);
9607 vmx_destroy_pml_buffer(vmx);
9609 kvm_vcpu_uninit(&vmx->vcpu);
9611 free_vpid(vmx->vpid);
9612 kmem_cache_free(kvm_vcpu_cache, vmx);
9613 return ERR_PTR(err);
9616 static void __init vmx_check_processor_compat(void *rtn)
9618 struct vmcs_config vmcs_conf;
9621 if (setup_vmcs_config(&vmcs_conf) < 0)
9623 if (memcmp(&vmcs_config, &vmcs_conf, sizeof(struct vmcs_config)) != 0) {
9624 printk(KERN_ERR "kvm: CPU %d feature inconsistency!\n",
9625 smp_processor_id());
9630 static u64 vmx_get_mt_mask(struct kvm_vcpu *vcpu, gfn_t gfn, bool is_mmio)
9635 /* For VT-d and EPT combination
9636 * 1. MMIO: always map as UC
9638 * a. VT-d without snooping control feature: can't guarantee the
9639 * result, try to trust guest.
9640 * b. VT-d with snooping control feature: snooping control feature of
9641 * VT-d engine can guarantee the cache correctness. Just set it
9642 * to WB to keep consistent with host. So the same as item 3.
9643 * 3. EPT without VT-d: always map as WB and set IPAT=1 to keep
9644 * consistent with host MTRR
9647 cache = MTRR_TYPE_UNCACHABLE;
9651 if (!kvm_arch_has_noncoherent_dma(vcpu->kvm)) {
9652 ipat = VMX_EPT_IPAT_BIT;
9653 cache = MTRR_TYPE_WRBACK;
9657 if (kvm_read_cr0(vcpu) & X86_CR0_CD) {
9658 ipat = VMX_EPT_IPAT_BIT;
9659 if (kvm_check_has_quirk(vcpu->kvm, KVM_X86_QUIRK_CD_NW_CLEARED))
9660 cache = MTRR_TYPE_WRBACK;
9662 cache = MTRR_TYPE_UNCACHABLE;
9666 cache = kvm_mtrr_get_guest_memory_type(vcpu, gfn);
9669 return (cache << VMX_EPT_MT_EPTE_SHIFT) | ipat;
9672 static int vmx_get_lpage_level(void)
9674 if (enable_ept && !cpu_has_vmx_ept_1g_page())
9675 return PT_DIRECTORY_LEVEL;
9677 /* For shadow and EPT supported 1GB page */
9678 return PT_PDPE_LEVEL;
9681 static void vmcs_set_secondary_exec_control(u32 new_ctl)
9684 * These bits in the secondary execution controls field
9685 * are dynamic, the others are mostly based on the hypervisor
9686 * architecture and the guest's CPUID. Do not touch the
9690 SECONDARY_EXEC_SHADOW_VMCS |
9691 SECONDARY_EXEC_VIRTUALIZE_X2APIC_MODE |
9692 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES;
9694 u32 cur_ctl = vmcs_read32(SECONDARY_VM_EXEC_CONTROL);
9696 vmcs_write32(SECONDARY_VM_EXEC_CONTROL,
9697 (new_ctl & ~mask) | (cur_ctl & mask));
9701 * Generate MSR_IA32_VMX_CR{0,4}_FIXED1 according to CPUID. Only set bits
9702 * (indicating "allowed-1") if they are supported in the guest's CPUID.
9704 static void nested_vmx_cr_fixed1_bits_update(struct kvm_vcpu *vcpu)
9706 struct vcpu_vmx *vmx = to_vmx(vcpu);
9707 struct kvm_cpuid_entry2 *entry;
9709 vmx->nested.nested_vmx_cr0_fixed1 = 0xffffffff;
9710 vmx->nested.nested_vmx_cr4_fixed1 = X86_CR4_PCE;
9712 #define cr4_fixed1_update(_cr4_mask, _reg, _cpuid_mask) do { \
9713 if (entry && (entry->_reg & (_cpuid_mask))) \
9714 vmx->nested.nested_vmx_cr4_fixed1 |= (_cr4_mask); \
9717 entry = kvm_find_cpuid_entry(vcpu, 0x1, 0);
9718 cr4_fixed1_update(X86_CR4_VME, edx, bit(X86_FEATURE_VME));
9719 cr4_fixed1_update(X86_CR4_PVI, edx, bit(X86_FEATURE_VME));
9720 cr4_fixed1_update(X86_CR4_TSD, edx, bit(X86_FEATURE_TSC));
9721 cr4_fixed1_update(X86_CR4_DE, edx, bit(X86_FEATURE_DE));
9722 cr4_fixed1_update(X86_CR4_PSE, edx, bit(X86_FEATURE_PSE));
9723 cr4_fixed1_update(X86_CR4_PAE, edx, bit(X86_FEATURE_PAE));
9724 cr4_fixed1_update(X86_CR4_MCE, edx, bit(X86_FEATURE_MCE));
9725 cr4_fixed1_update(X86_CR4_PGE, edx, bit(X86_FEATURE_PGE));
9726 cr4_fixed1_update(X86_CR4_OSFXSR, edx, bit(X86_FEATURE_FXSR));
9727 cr4_fixed1_update(X86_CR4_OSXMMEXCPT, edx, bit(X86_FEATURE_XMM));
9728 cr4_fixed1_update(X86_CR4_VMXE, ecx, bit(X86_FEATURE_VMX));
9729 cr4_fixed1_update(X86_CR4_SMXE, ecx, bit(X86_FEATURE_SMX));
9730 cr4_fixed1_update(X86_CR4_PCIDE, ecx, bit(X86_FEATURE_PCID));
9731 cr4_fixed1_update(X86_CR4_OSXSAVE, ecx, bit(X86_FEATURE_XSAVE));
9733 entry = kvm_find_cpuid_entry(vcpu, 0x7, 0);
9734 cr4_fixed1_update(X86_CR4_FSGSBASE, ebx, bit(X86_FEATURE_FSGSBASE));
9735 cr4_fixed1_update(X86_CR4_SMEP, ebx, bit(X86_FEATURE_SMEP));
9736 cr4_fixed1_update(X86_CR4_SMAP, ebx, bit(X86_FEATURE_SMAP));
9737 cr4_fixed1_update(X86_CR4_PKE, ecx, bit(X86_FEATURE_PKU));
9738 /* TODO: Use X86_CR4_UMIP and X86_FEATURE_UMIP macros */
9739 cr4_fixed1_update(bit(11), ecx, bit(2));
9741 #undef cr4_fixed1_update
9744 static void vmx_cpuid_update(struct kvm_vcpu *vcpu)
9746 struct vcpu_vmx *vmx = to_vmx(vcpu);
9748 if (cpu_has_secondary_exec_ctrls()) {
9749 vmx_compute_secondary_exec_control(vmx);
9750 vmcs_set_secondary_exec_control(vmx->secondary_exec_control);
9753 if (nested_vmx_allowed(vcpu))
9754 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
9755 FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9757 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
9758 ~FEATURE_CONTROL_VMXON_ENABLED_OUTSIDE_SMX;
9760 if (nested_vmx_allowed(vcpu))
9761 nested_vmx_cr_fixed1_bits_update(vcpu);
9764 static void vmx_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
9766 if (func == 1 && nested)
9767 entry->ecx |= bit(X86_FEATURE_VMX);
9770 static void nested_ept_inject_page_fault(struct kvm_vcpu *vcpu,
9771 struct x86_exception *fault)
9773 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9774 struct vcpu_vmx *vmx = to_vmx(vcpu);
9776 unsigned long exit_qualification = vcpu->arch.exit_qualification;
9778 if (vmx->nested.pml_full) {
9779 exit_reason = EXIT_REASON_PML_FULL;
9780 vmx->nested.pml_full = false;
9781 exit_qualification &= INTR_INFO_UNBLOCK_NMI;
9782 } else if (fault->error_code & PFERR_RSVD_MASK)
9783 exit_reason = EXIT_REASON_EPT_MISCONFIG;
9785 exit_reason = EXIT_REASON_EPT_VIOLATION;
9787 nested_vmx_vmexit(vcpu, exit_reason, 0, exit_qualification);
9788 vmcs12->guest_physical_address = fault->address;
9791 static bool nested_ept_ad_enabled(struct kvm_vcpu *vcpu)
9793 return nested_ept_get_cr3(vcpu) & VMX_EPTP_AD_ENABLE_BIT;
9796 /* Callbacks for nested_ept_init_mmu_context: */
9798 static unsigned long nested_ept_get_cr3(struct kvm_vcpu *vcpu)
9800 /* return the page table to be shadowed - in our case, EPT12 */
9801 return get_vmcs12(vcpu)->ept_pointer;
9804 static int nested_ept_init_mmu_context(struct kvm_vcpu *vcpu)
9806 WARN_ON(mmu_is_nested(vcpu));
9807 if (!valid_ept_address(vcpu, nested_ept_get_cr3(vcpu)))
9810 kvm_mmu_unload(vcpu);
9811 kvm_init_shadow_ept_mmu(vcpu,
9812 to_vmx(vcpu)->nested.nested_vmx_ept_caps &
9813 VMX_EPT_EXECUTE_ONLY_BIT,
9814 nested_ept_ad_enabled(vcpu));
9815 vcpu->arch.mmu.set_cr3 = vmx_set_cr3;
9816 vcpu->arch.mmu.get_cr3 = nested_ept_get_cr3;
9817 vcpu->arch.mmu.inject_page_fault = nested_ept_inject_page_fault;
9819 vcpu->arch.walk_mmu = &vcpu->arch.nested_mmu;
9823 static void nested_ept_uninit_mmu_context(struct kvm_vcpu *vcpu)
9825 vcpu->arch.walk_mmu = &vcpu->arch.mmu;
9828 static bool nested_vmx_is_page_fault_vmexit(struct vmcs12 *vmcs12,
9831 bool inequality, bit;
9833 bit = (vmcs12->exception_bitmap & (1u << PF_VECTOR)) != 0;
9835 (error_code & vmcs12->page_fault_error_code_mask) !=
9836 vmcs12->page_fault_error_code_match;
9837 return inequality ^ bit;
9840 static void vmx_inject_page_fault_nested(struct kvm_vcpu *vcpu,
9841 struct x86_exception *fault)
9843 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
9845 WARN_ON(!is_guest_mode(vcpu));
9847 if (nested_vmx_is_page_fault_vmexit(vmcs12, fault->error_code) &&
9848 !to_vmx(vcpu)->nested.nested_run_pending) {
9849 vmcs12->vm_exit_intr_error_code = fault->error_code;
9850 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
9851 PF_VECTOR | INTR_TYPE_HARD_EXCEPTION |
9852 INTR_INFO_DELIVER_CODE_MASK | INTR_INFO_VALID_MASK,
9855 kvm_inject_page_fault(vcpu, fault);
9859 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
9860 struct vmcs12 *vmcs12);
9862 static void nested_get_vmcs12_pages(struct kvm_vcpu *vcpu,
9863 struct vmcs12 *vmcs12)
9865 struct vcpu_vmx *vmx = to_vmx(vcpu);
9869 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
9871 * Translate L1 physical address to host physical
9872 * address for vmcs02. Keep the page pinned, so this
9873 * physical address remains valid. We keep a reference
9874 * to it so we can release it later.
9876 if (vmx->nested.apic_access_page) { /* shouldn't happen */
9877 kvm_release_page_dirty(vmx->nested.apic_access_page);
9878 vmx->nested.apic_access_page = NULL;
9880 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->apic_access_addr);
9882 * If translation failed, no matter: This feature asks
9883 * to exit when accessing the given address, and if it
9884 * can never be accessed, this feature won't do
9887 if (!is_error_page(page)) {
9888 vmx->nested.apic_access_page = page;
9889 hpa = page_to_phys(vmx->nested.apic_access_page);
9890 vmcs_write64(APIC_ACCESS_ADDR, hpa);
9892 vmcs_clear_bits(SECONDARY_VM_EXEC_CONTROL,
9893 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9895 } else if (!(nested_cpu_has_virt_x2apic_mode(vmcs12)) &&
9896 cpu_need_virtualize_apic_accesses(&vmx->vcpu)) {
9897 vmcs_set_bits(SECONDARY_VM_EXEC_CONTROL,
9898 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES);
9899 kvm_vcpu_reload_apic_access_page(vcpu);
9902 if (nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW)) {
9903 if (vmx->nested.virtual_apic_page) { /* shouldn't happen */
9904 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
9905 vmx->nested.virtual_apic_page = NULL;
9907 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->virtual_apic_page_addr);
9910 * If translation failed, VM entry will fail because
9911 * prepare_vmcs02 set VIRTUAL_APIC_PAGE_ADDR to -1ull.
9912 * Failing the vm entry is _not_ what the processor
9913 * does but it's basically the only possibility we
9914 * have. We could still enter the guest if CR8 load
9915 * exits are enabled, CR8 store exits are enabled, and
9916 * virtualize APIC access is disabled; in this case
9917 * the processor would never use the TPR shadow and we
9918 * could simply clear the bit from the execution
9919 * control. But such a configuration is useless, so
9920 * let's keep the code simple.
9922 if (!is_error_page(page)) {
9923 vmx->nested.virtual_apic_page = page;
9924 hpa = page_to_phys(vmx->nested.virtual_apic_page);
9925 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, hpa);
9929 if (nested_cpu_has_posted_intr(vmcs12)) {
9930 if (vmx->nested.pi_desc_page) { /* shouldn't happen */
9931 kunmap(vmx->nested.pi_desc_page);
9932 kvm_release_page_dirty(vmx->nested.pi_desc_page);
9933 vmx->nested.pi_desc_page = NULL;
9935 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->posted_intr_desc_addr);
9936 if (is_error_page(page))
9938 vmx->nested.pi_desc_page = page;
9939 vmx->nested.pi_desc = kmap(vmx->nested.pi_desc_page);
9940 vmx->nested.pi_desc =
9941 (struct pi_desc *)((void *)vmx->nested.pi_desc +
9942 (unsigned long)(vmcs12->posted_intr_desc_addr &
9944 vmcs_write64(POSTED_INTR_DESC_ADDR,
9945 page_to_phys(vmx->nested.pi_desc_page) +
9946 (unsigned long)(vmcs12->posted_intr_desc_addr &
9949 if (cpu_has_vmx_msr_bitmap() &&
9950 nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS) &&
9951 nested_vmx_merge_msr_bitmap(vcpu, vmcs12))
9954 vmcs_clear_bits(CPU_BASED_VM_EXEC_CONTROL,
9955 CPU_BASED_USE_MSR_BITMAPS);
9958 static void vmx_start_preemption_timer(struct kvm_vcpu *vcpu)
9960 u64 preemption_timeout = get_vmcs12(vcpu)->vmx_preemption_timer_value;
9961 struct vcpu_vmx *vmx = to_vmx(vcpu);
9963 if (vcpu->arch.virtual_tsc_khz == 0)
9966 /* Make sure short timeouts reliably trigger an immediate vmexit.
9967 * hrtimer_start does not guarantee this. */
9968 if (preemption_timeout <= 1) {
9969 vmx_preemption_timer_fn(&vmx->nested.preemption_timer);
9973 preemption_timeout <<= VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
9974 preemption_timeout *= 1000000;
9975 do_div(preemption_timeout, vcpu->arch.virtual_tsc_khz);
9976 hrtimer_start(&vmx->nested.preemption_timer,
9977 ns_to_ktime(preemption_timeout), HRTIMER_MODE_REL);
9980 static int nested_vmx_check_io_bitmap_controls(struct kvm_vcpu *vcpu,
9981 struct vmcs12 *vmcs12)
9983 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_IO_BITMAPS))
9986 if (!page_address_valid(vcpu, vmcs12->io_bitmap_a) ||
9987 !page_address_valid(vcpu, vmcs12->io_bitmap_b))
9993 static int nested_vmx_check_msr_bitmap_controls(struct kvm_vcpu *vcpu,
9994 struct vmcs12 *vmcs12)
9996 if (!nested_cpu_has(vmcs12, CPU_BASED_USE_MSR_BITMAPS))
9999 if (!page_address_valid(vcpu, vmcs12->msr_bitmap))
10005 static int nested_vmx_check_tpr_shadow_controls(struct kvm_vcpu *vcpu,
10006 struct vmcs12 *vmcs12)
10008 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
10011 if (!page_address_valid(vcpu, vmcs12->virtual_apic_page_addr))
10018 * Merge L0's and L1's MSR bitmap, return false to indicate that
10019 * we do not use the hardware.
10021 static inline bool nested_vmx_merge_msr_bitmap(struct kvm_vcpu *vcpu,
10022 struct vmcs12 *vmcs12)
10026 unsigned long *msr_bitmap_l1;
10027 unsigned long *msr_bitmap_l0 = to_vmx(vcpu)->nested.msr_bitmap;
10029 /* This shortcut is ok because we support only x2APIC MSRs so far. */
10030 if (!nested_cpu_has_virt_x2apic_mode(vmcs12))
10033 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->msr_bitmap);
10034 if (is_error_page(page))
10036 msr_bitmap_l1 = (unsigned long *)kmap(page);
10038 memset(msr_bitmap_l0, 0xff, PAGE_SIZE);
10040 if (nested_cpu_has_virt_x2apic_mode(vmcs12)) {
10041 if (nested_cpu_has_apic_reg_virt(vmcs12))
10042 for (msr = 0x800; msr <= 0x8ff; msr++)
10043 nested_vmx_disable_intercept_for_msr(
10044 msr_bitmap_l1, msr_bitmap_l0,
10047 nested_vmx_disable_intercept_for_msr(
10048 msr_bitmap_l1, msr_bitmap_l0,
10049 APIC_BASE_MSR + (APIC_TASKPRI >> 4),
10050 MSR_TYPE_R | MSR_TYPE_W);
10052 if (nested_cpu_has_vid(vmcs12)) {
10053 nested_vmx_disable_intercept_for_msr(
10054 msr_bitmap_l1, msr_bitmap_l0,
10055 APIC_BASE_MSR + (APIC_EOI >> 4),
10057 nested_vmx_disable_intercept_for_msr(
10058 msr_bitmap_l1, msr_bitmap_l0,
10059 APIC_BASE_MSR + (APIC_SELF_IPI >> 4),
10064 kvm_release_page_clean(page);
10069 static int nested_vmx_check_apicv_controls(struct kvm_vcpu *vcpu,
10070 struct vmcs12 *vmcs12)
10072 if (!nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10073 !nested_cpu_has_apic_reg_virt(vmcs12) &&
10074 !nested_cpu_has_vid(vmcs12) &&
10075 !nested_cpu_has_posted_intr(vmcs12))
10079 * If virtualize x2apic mode is enabled,
10080 * virtualize apic access must be disabled.
10082 if (nested_cpu_has_virt_x2apic_mode(vmcs12) &&
10083 nested_cpu_has2(vmcs12, SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES))
10087 * If virtual interrupt delivery is enabled,
10088 * we must exit on external interrupts.
10090 if (nested_cpu_has_vid(vmcs12) &&
10091 !nested_exit_on_intr(vcpu))
10095 * bits 15:8 should be zero in posted_intr_nv,
10096 * the descriptor address has been already checked
10097 * in nested_get_vmcs12_pages.
10099 if (nested_cpu_has_posted_intr(vmcs12) &&
10100 (!nested_cpu_has_vid(vmcs12) ||
10101 !nested_exit_intr_ack_set(vcpu) ||
10102 vmcs12->posted_intr_nv & 0xff00))
10105 /* tpr shadow is needed by all apicv features. */
10106 if (!nested_cpu_has(vmcs12, CPU_BASED_TPR_SHADOW))
10112 static int nested_vmx_check_msr_switch(struct kvm_vcpu *vcpu,
10113 unsigned long count_field,
10114 unsigned long addr_field)
10119 if (vmcs12_read_any(vcpu, count_field, &count) ||
10120 vmcs12_read_any(vcpu, addr_field, &addr)) {
10126 maxphyaddr = cpuid_maxphyaddr(vcpu);
10127 if (!IS_ALIGNED(addr, 16) || addr >> maxphyaddr ||
10128 (addr + count * sizeof(struct vmx_msr_entry) - 1) >> maxphyaddr) {
10129 pr_debug_ratelimited(
10130 "nVMX: invalid MSR switch (0x%lx, %d, %llu, 0x%08llx)",
10131 addr_field, maxphyaddr, count, addr);
10137 static int nested_vmx_check_msr_switch_controls(struct kvm_vcpu *vcpu,
10138 struct vmcs12 *vmcs12)
10140 if (vmcs12->vm_exit_msr_load_count == 0 &&
10141 vmcs12->vm_exit_msr_store_count == 0 &&
10142 vmcs12->vm_entry_msr_load_count == 0)
10143 return 0; /* Fast path */
10144 if (nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_LOAD_COUNT,
10145 VM_EXIT_MSR_LOAD_ADDR) ||
10146 nested_vmx_check_msr_switch(vcpu, VM_EXIT_MSR_STORE_COUNT,
10147 VM_EXIT_MSR_STORE_ADDR) ||
10148 nested_vmx_check_msr_switch(vcpu, VM_ENTRY_MSR_LOAD_COUNT,
10149 VM_ENTRY_MSR_LOAD_ADDR))
10154 static int nested_vmx_check_pml_controls(struct kvm_vcpu *vcpu,
10155 struct vmcs12 *vmcs12)
10157 u64 address = vmcs12->pml_address;
10158 int maxphyaddr = cpuid_maxphyaddr(vcpu);
10160 if (nested_cpu_has2(vmcs12, SECONDARY_EXEC_ENABLE_PML)) {
10161 if (!nested_cpu_has_ept(vmcs12) ||
10162 !IS_ALIGNED(address, 4096) ||
10163 address >> maxphyaddr)
10170 static int nested_vmx_msr_check_common(struct kvm_vcpu *vcpu,
10171 struct vmx_msr_entry *e)
10173 /* x2APIC MSR accesses are not allowed */
10174 if (vcpu->arch.apic_base & X2APIC_ENABLE && e->index >> 8 == 0x8)
10176 if (e->index == MSR_IA32_UCODE_WRITE || /* SDM Table 35-2 */
10177 e->index == MSR_IA32_UCODE_REV)
10179 if (e->reserved != 0)
10184 static int nested_vmx_load_msr_check(struct kvm_vcpu *vcpu,
10185 struct vmx_msr_entry *e)
10187 if (e->index == MSR_FS_BASE ||
10188 e->index == MSR_GS_BASE ||
10189 e->index == MSR_IA32_SMM_MONITOR_CTL || /* SMM is not supported */
10190 nested_vmx_msr_check_common(vcpu, e))
10195 static int nested_vmx_store_msr_check(struct kvm_vcpu *vcpu,
10196 struct vmx_msr_entry *e)
10198 if (e->index == MSR_IA32_SMBASE || /* SMM is not supported */
10199 nested_vmx_msr_check_common(vcpu, e))
10205 * Load guest's/host's msr at nested entry/exit.
10206 * return 0 for success, entry index for failure.
10208 static u32 nested_vmx_load_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
10211 struct vmx_msr_entry e;
10212 struct msr_data msr;
10214 msr.host_initiated = false;
10215 for (i = 0; i < count; i++) {
10216 if (kvm_vcpu_read_guest(vcpu, gpa + i * sizeof(e),
10218 pr_debug_ratelimited(
10219 "%s cannot read MSR entry (%u, 0x%08llx)\n",
10220 __func__, i, gpa + i * sizeof(e));
10223 if (nested_vmx_load_msr_check(vcpu, &e)) {
10224 pr_debug_ratelimited(
10225 "%s check failed (%u, 0x%x, 0x%x)\n",
10226 __func__, i, e.index, e.reserved);
10229 msr.index = e.index;
10230 msr.data = e.value;
10231 if (kvm_set_msr(vcpu, &msr)) {
10232 pr_debug_ratelimited(
10233 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
10234 __func__, i, e.index, e.value);
10243 static int nested_vmx_store_msr(struct kvm_vcpu *vcpu, u64 gpa, u32 count)
10246 struct vmx_msr_entry e;
10248 for (i = 0; i < count; i++) {
10249 struct msr_data msr_info;
10250 if (kvm_vcpu_read_guest(vcpu,
10251 gpa + i * sizeof(e),
10252 &e, 2 * sizeof(u32))) {
10253 pr_debug_ratelimited(
10254 "%s cannot read MSR entry (%u, 0x%08llx)\n",
10255 __func__, i, gpa + i * sizeof(e));
10258 if (nested_vmx_store_msr_check(vcpu, &e)) {
10259 pr_debug_ratelimited(
10260 "%s check failed (%u, 0x%x, 0x%x)\n",
10261 __func__, i, e.index, e.reserved);
10264 msr_info.host_initiated = false;
10265 msr_info.index = e.index;
10266 if (kvm_get_msr(vcpu, &msr_info)) {
10267 pr_debug_ratelimited(
10268 "%s cannot read MSR (%u, 0x%x)\n",
10269 __func__, i, e.index);
10272 if (kvm_vcpu_write_guest(vcpu,
10273 gpa + i * sizeof(e) +
10274 offsetof(struct vmx_msr_entry, value),
10275 &msr_info.data, sizeof(msr_info.data))) {
10276 pr_debug_ratelimited(
10277 "%s cannot write MSR (%u, 0x%x, 0x%llx)\n",
10278 __func__, i, e.index, msr_info.data);
10285 static bool nested_cr3_valid(struct kvm_vcpu *vcpu, unsigned long val)
10287 unsigned long invalid_mask;
10289 invalid_mask = (~0ULL) << cpuid_maxphyaddr(vcpu);
10290 return (val & invalid_mask) == 0;
10294 * Load guest's/host's cr3 at nested entry/exit. nested_ept is true if we are
10295 * emulating VM entry into a guest with EPT enabled.
10296 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10297 * is assigned to entry_failure_code on failure.
10299 static int nested_vmx_load_cr3(struct kvm_vcpu *vcpu, unsigned long cr3, bool nested_ept,
10300 u32 *entry_failure_code)
10302 if (cr3 != kvm_read_cr3(vcpu) || (!nested_ept && pdptrs_changed(vcpu))) {
10303 if (!nested_cr3_valid(vcpu, cr3)) {
10304 *entry_failure_code = ENTRY_FAIL_DEFAULT;
10309 * If PAE paging and EPT are both on, CR3 is not used by the CPU and
10310 * must not be dereferenced.
10312 if (!is_long_mode(vcpu) && is_pae(vcpu) && is_paging(vcpu) &&
10314 if (!load_pdptrs(vcpu, vcpu->arch.walk_mmu, cr3)) {
10315 *entry_failure_code = ENTRY_FAIL_PDPTE;
10320 vcpu->arch.cr3 = cr3;
10321 __set_bit(VCPU_EXREG_CR3, (ulong *)&vcpu->arch.regs_avail);
10324 kvm_mmu_reset_context(vcpu);
10329 * prepare_vmcs02 is called when the L1 guest hypervisor runs its nested
10330 * L2 guest. L1 has a vmcs for L2 (vmcs12), and this function "merges" it
10331 * with L0's requirements for its guest (a.k.a. vmcs01), so we can run the L2
10332 * guest in a way that will both be appropriate to L1's requests, and our
10333 * needs. In addition to modifying the active vmcs (which is vmcs02), this
10334 * function also has additional necessary side-effects, like setting various
10335 * vcpu->arch fields.
10336 * Returns 0 on success, 1 on failure. Invalid state exit qualification code
10337 * is assigned to entry_failure_code on failure.
10339 static int prepare_vmcs02(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10340 bool from_vmentry, u32 *entry_failure_code)
10342 struct vcpu_vmx *vmx = to_vmx(vcpu);
10343 u32 exec_control, vmcs12_exec_ctrl;
10345 vmcs_write16(GUEST_ES_SELECTOR, vmcs12->guest_es_selector);
10346 vmcs_write16(GUEST_CS_SELECTOR, vmcs12->guest_cs_selector);
10347 vmcs_write16(GUEST_SS_SELECTOR, vmcs12->guest_ss_selector);
10348 vmcs_write16(GUEST_DS_SELECTOR, vmcs12->guest_ds_selector);
10349 vmcs_write16(GUEST_FS_SELECTOR, vmcs12->guest_fs_selector);
10350 vmcs_write16(GUEST_GS_SELECTOR, vmcs12->guest_gs_selector);
10351 vmcs_write16(GUEST_LDTR_SELECTOR, vmcs12->guest_ldtr_selector);
10352 vmcs_write16(GUEST_TR_SELECTOR, vmcs12->guest_tr_selector);
10353 vmcs_write32(GUEST_ES_LIMIT, vmcs12->guest_es_limit);
10354 vmcs_write32(GUEST_CS_LIMIT, vmcs12->guest_cs_limit);
10355 vmcs_write32(GUEST_SS_LIMIT, vmcs12->guest_ss_limit);
10356 vmcs_write32(GUEST_DS_LIMIT, vmcs12->guest_ds_limit);
10357 vmcs_write32(GUEST_FS_LIMIT, vmcs12->guest_fs_limit);
10358 vmcs_write32(GUEST_GS_LIMIT, vmcs12->guest_gs_limit);
10359 vmcs_write32(GUEST_LDTR_LIMIT, vmcs12->guest_ldtr_limit);
10360 vmcs_write32(GUEST_TR_LIMIT, vmcs12->guest_tr_limit);
10361 vmcs_write32(GUEST_GDTR_LIMIT, vmcs12->guest_gdtr_limit);
10362 vmcs_write32(GUEST_IDTR_LIMIT, vmcs12->guest_idtr_limit);
10363 vmcs_write32(GUEST_ES_AR_BYTES, vmcs12->guest_es_ar_bytes);
10364 vmcs_write32(GUEST_CS_AR_BYTES, vmcs12->guest_cs_ar_bytes);
10365 vmcs_write32(GUEST_SS_AR_BYTES, vmcs12->guest_ss_ar_bytes);
10366 vmcs_write32(GUEST_DS_AR_BYTES, vmcs12->guest_ds_ar_bytes);
10367 vmcs_write32(GUEST_FS_AR_BYTES, vmcs12->guest_fs_ar_bytes);
10368 vmcs_write32(GUEST_GS_AR_BYTES, vmcs12->guest_gs_ar_bytes);
10369 vmcs_write32(GUEST_LDTR_AR_BYTES, vmcs12->guest_ldtr_ar_bytes);
10370 vmcs_write32(GUEST_TR_AR_BYTES, vmcs12->guest_tr_ar_bytes);
10371 vmcs_writel(GUEST_ES_BASE, vmcs12->guest_es_base);
10372 vmcs_writel(GUEST_CS_BASE, vmcs12->guest_cs_base);
10373 vmcs_writel(GUEST_SS_BASE, vmcs12->guest_ss_base);
10374 vmcs_writel(GUEST_DS_BASE, vmcs12->guest_ds_base);
10375 vmcs_writel(GUEST_FS_BASE, vmcs12->guest_fs_base);
10376 vmcs_writel(GUEST_GS_BASE, vmcs12->guest_gs_base);
10377 vmcs_writel(GUEST_LDTR_BASE, vmcs12->guest_ldtr_base);
10378 vmcs_writel(GUEST_TR_BASE, vmcs12->guest_tr_base);
10379 vmcs_writel(GUEST_GDTR_BASE, vmcs12->guest_gdtr_base);
10380 vmcs_writel(GUEST_IDTR_BASE, vmcs12->guest_idtr_base);
10382 if (from_vmentry &&
10383 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS)) {
10384 kvm_set_dr(vcpu, 7, vmcs12->guest_dr7);
10385 vmcs_write64(GUEST_IA32_DEBUGCTL, vmcs12->guest_ia32_debugctl);
10387 kvm_set_dr(vcpu, 7, vcpu->arch.dr7);
10388 vmcs_write64(GUEST_IA32_DEBUGCTL, vmx->nested.vmcs01_debugctl);
10390 if (from_vmentry) {
10391 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD,
10392 vmcs12->vm_entry_intr_info_field);
10393 vmcs_write32(VM_ENTRY_EXCEPTION_ERROR_CODE,
10394 vmcs12->vm_entry_exception_error_code);
10395 vmcs_write32(VM_ENTRY_INSTRUCTION_LEN,
10396 vmcs12->vm_entry_instruction_len);
10397 vmcs_write32(GUEST_INTERRUPTIBILITY_INFO,
10398 vmcs12->guest_interruptibility_info);
10399 vmx->loaded_vmcs->nmi_known_unmasked =
10400 !(vmcs12->guest_interruptibility_info & GUEST_INTR_STATE_NMI);
10402 vmcs_write32(VM_ENTRY_INTR_INFO_FIELD, 0);
10404 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->guest_sysenter_cs);
10405 vmx_set_rflags(vcpu, vmcs12->guest_rflags);
10406 vmcs_writel(GUEST_PENDING_DBG_EXCEPTIONS,
10407 vmcs12->guest_pending_dbg_exceptions);
10408 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->guest_sysenter_esp);
10409 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->guest_sysenter_eip);
10411 if (nested_cpu_has_xsaves(vmcs12))
10412 vmcs_write64(XSS_EXIT_BITMAP, vmcs12->xss_exit_bitmap);
10413 vmcs_write64(VMCS_LINK_POINTER, -1ull);
10415 exec_control = vmcs12->pin_based_vm_exec_control;
10417 /* Preemption timer setting is only taken from vmcs01. */
10418 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10419 exec_control |= vmcs_config.pin_based_exec_ctrl;
10420 if (vmx->hv_deadline_tsc == -1)
10421 exec_control &= ~PIN_BASED_VMX_PREEMPTION_TIMER;
10423 /* Posted interrupts setting is only taken from vmcs12. */
10424 if (nested_cpu_has_posted_intr(vmcs12)) {
10425 vmx->nested.posted_intr_nv = vmcs12->posted_intr_nv;
10426 vmx->nested.pi_pending = false;
10427 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_NESTED_VECTOR);
10429 exec_control &= ~PIN_BASED_POSTED_INTR;
10432 vmcs_write32(PIN_BASED_VM_EXEC_CONTROL, exec_control);
10434 vmx->nested.preemption_timer_expired = false;
10435 if (nested_cpu_has_preemption_timer(vmcs12))
10436 vmx_start_preemption_timer(vcpu);
10439 * Whether page-faults are trapped is determined by a combination of
10440 * 3 settings: PFEC_MASK, PFEC_MATCH and EXCEPTION_BITMAP.PF.
10441 * If enable_ept, L0 doesn't care about page faults and we should
10442 * set all of these to L1's desires. However, if !enable_ept, L0 does
10443 * care about (at least some) page faults, and because it is not easy
10444 * (if at all possible?) to merge L0 and L1's desires, we simply ask
10445 * to exit on each and every L2 page fault. This is done by setting
10446 * MASK=MATCH=0 and (see below) EB.PF=1.
10447 * Note that below we don't need special code to set EB.PF beyond the
10448 * "or"ing of the EB of vmcs01 and vmcs12, because when enable_ept,
10449 * vmcs01's EB.PF is 0 so the "or" will take vmcs12's value, and when
10450 * !enable_ept, EB.PF is 1, so the "or" will always be 1.
10452 vmcs_write32(PAGE_FAULT_ERROR_CODE_MASK,
10453 enable_ept ? vmcs12->page_fault_error_code_mask : 0);
10454 vmcs_write32(PAGE_FAULT_ERROR_CODE_MATCH,
10455 enable_ept ? vmcs12->page_fault_error_code_match : 0);
10457 if (cpu_has_secondary_exec_ctrls()) {
10458 exec_control = vmx->secondary_exec_control;
10460 /* Take the following fields only from vmcs12 */
10461 exec_control &= ~(SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES |
10462 SECONDARY_EXEC_ENABLE_INVPCID |
10463 SECONDARY_EXEC_RDTSCP |
10464 SECONDARY_EXEC_XSAVES |
10465 SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY |
10466 SECONDARY_EXEC_APIC_REGISTER_VIRT |
10467 SECONDARY_EXEC_ENABLE_VMFUNC);
10468 if (nested_cpu_has(vmcs12,
10469 CPU_BASED_ACTIVATE_SECONDARY_CONTROLS)) {
10470 vmcs12_exec_ctrl = vmcs12->secondary_vm_exec_control &
10471 ~SECONDARY_EXEC_ENABLE_PML;
10472 exec_control |= vmcs12_exec_ctrl;
10475 /* All VMFUNCs are currently emulated through L0 vmexits. */
10476 if (exec_control & SECONDARY_EXEC_ENABLE_VMFUNC)
10477 vmcs_write64(VM_FUNCTION_CONTROL, 0);
10479 if (exec_control & SECONDARY_EXEC_VIRTUAL_INTR_DELIVERY) {
10480 vmcs_write64(EOI_EXIT_BITMAP0,
10481 vmcs12->eoi_exit_bitmap0);
10482 vmcs_write64(EOI_EXIT_BITMAP1,
10483 vmcs12->eoi_exit_bitmap1);
10484 vmcs_write64(EOI_EXIT_BITMAP2,
10485 vmcs12->eoi_exit_bitmap2);
10486 vmcs_write64(EOI_EXIT_BITMAP3,
10487 vmcs12->eoi_exit_bitmap3);
10488 vmcs_write16(GUEST_INTR_STATUS,
10489 vmcs12->guest_intr_status);
10493 * Write an illegal value to APIC_ACCESS_ADDR. Later,
10494 * nested_get_vmcs12_pages will either fix it up or
10495 * remove the VM execution control.
10497 if (exec_control & SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)
10498 vmcs_write64(APIC_ACCESS_ADDR, -1ull);
10500 vmcs_write32(SECONDARY_VM_EXEC_CONTROL, exec_control);
10505 * Set host-state according to L0's settings (vmcs12 is irrelevant here)
10506 * Some constant fields are set here by vmx_set_constant_host_state().
10507 * Other fields are different per CPU, and will be set later when
10508 * vmx_vcpu_load() is called, and when vmx_save_host_state() is called.
10510 vmx_set_constant_host_state(vmx);
10513 * Set the MSR load/store lists to match L0's settings.
10515 vmcs_write32(VM_EXIT_MSR_STORE_COUNT, 0);
10516 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10517 vmcs_write64(VM_EXIT_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.host));
10518 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
10519 vmcs_write64(VM_ENTRY_MSR_LOAD_ADDR, __pa(vmx->msr_autoload.guest));
10522 * HOST_RSP is normally set correctly in vmx_vcpu_run() just before
10523 * entry, but only if the current (host) sp changed from the value
10524 * we wrote last (vmx->host_rsp). This cache is no longer relevant
10525 * if we switch vmcs, and rather than hold a separate cache per vmcs,
10526 * here we just force the write to happen on entry.
10530 exec_control = vmx_exec_control(vmx); /* L0's desires */
10531 exec_control &= ~CPU_BASED_VIRTUAL_INTR_PENDING;
10532 exec_control &= ~CPU_BASED_VIRTUAL_NMI_PENDING;
10533 exec_control &= ~CPU_BASED_TPR_SHADOW;
10534 exec_control |= vmcs12->cpu_based_vm_exec_control;
10537 * Write an illegal value to VIRTUAL_APIC_PAGE_ADDR. Later, if
10538 * nested_get_vmcs12_pages can't fix it up, the illegal value
10539 * will result in a VM entry failure.
10541 if (exec_control & CPU_BASED_TPR_SHADOW) {
10542 vmcs_write64(VIRTUAL_APIC_PAGE_ADDR, -1ull);
10543 vmcs_write32(TPR_THRESHOLD, vmcs12->tpr_threshold);
10545 #ifdef CONFIG_X86_64
10546 exec_control |= CPU_BASED_CR8_LOAD_EXITING |
10547 CPU_BASED_CR8_STORE_EXITING;
10552 * Merging of IO bitmap not currently supported.
10553 * Rather, exit every time.
10555 exec_control &= ~CPU_BASED_USE_IO_BITMAPS;
10556 exec_control |= CPU_BASED_UNCOND_IO_EXITING;
10558 vmcs_write32(CPU_BASED_VM_EXEC_CONTROL, exec_control);
10560 /* EXCEPTION_BITMAP and CR0_GUEST_HOST_MASK should basically be the
10561 * bitwise-or of what L1 wants to trap for L2, and what we want to
10562 * trap. Note that CR0.TS also needs updating - we do this later.
10564 update_exception_bitmap(vcpu);
10565 vcpu->arch.cr0_guest_owned_bits &= ~vmcs12->cr0_guest_host_mask;
10566 vmcs_writel(CR0_GUEST_HOST_MASK, ~vcpu->arch.cr0_guest_owned_bits);
10568 /* L2->L1 exit controls are emulated - the hardware exit is to L0 so
10569 * we should use its exit controls. Note that VM_EXIT_LOAD_IA32_EFER
10570 * bits are further modified by vmx_set_efer() below.
10572 vmcs_write32(VM_EXIT_CONTROLS, vmcs_config.vmexit_ctrl);
10574 /* vmcs12's VM_ENTRY_LOAD_IA32_EFER and VM_ENTRY_IA32E_MODE are
10575 * emulated by vmx_set_efer(), below.
10577 vm_entry_controls_init(vmx,
10578 (vmcs12->vm_entry_controls & ~VM_ENTRY_LOAD_IA32_EFER &
10579 ~VM_ENTRY_IA32E_MODE) |
10580 (vmcs_config.vmentry_ctrl & ~VM_ENTRY_IA32E_MODE));
10582 if (from_vmentry &&
10583 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_PAT)) {
10584 vmcs_write64(GUEST_IA32_PAT, vmcs12->guest_ia32_pat);
10585 vcpu->arch.pat = vmcs12->guest_ia32_pat;
10586 } else if (vmcs_config.vmentry_ctrl & VM_ENTRY_LOAD_IA32_PAT) {
10587 vmcs_write64(GUEST_IA32_PAT, vmx->vcpu.arch.pat);
10590 set_cr4_guest_host_mask(vmx);
10592 if (from_vmentry &&
10593 vmcs12->vm_entry_controls & VM_ENTRY_LOAD_BNDCFGS)
10594 vmcs_write64(GUEST_BNDCFGS, vmcs12->guest_bndcfgs);
10596 if (vmcs12->cpu_based_vm_exec_control & CPU_BASED_USE_TSC_OFFSETING)
10597 vmcs_write64(TSC_OFFSET,
10598 vcpu->arch.tsc_offset + vmcs12->tsc_offset);
10600 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
10601 if (kvm_has_tsc_control)
10602 decache_tsc_multiplier(vmx);
10606 * There is no direct mapping between vpid02 and vpid12, the
10607 * vpid02 is per-vCPU for L0 and reused while the value of
10608 * vpid12 is changed w/ one invvpid during nested vmentry.
10609 * The vpid12 is allocated by L1 for L2, so it will not
10610 * influence global bitmap(for vpid01 and vpid02 allocation)
10611 * even if spawn a lot of nested vCPUs.
10613 if (nested_cpu_has_vpid(vmcs12) && vmx->nested.vpid02) {
10614 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->nested.vpid02);
10615 if (vmcs12->virtual_processor_id != vmx->nested.last_vpid) {
10616 vmx->nested.last_vpid = vmcs12->virtual_processor_id;
10617 __vmx_flush_tlb(vcpu, to_vmx(vcpu)->nested.vpid02);
10620 vmcs_write16(VIRTUAL_PROCESSOR_ID, vmx->vpid);
10621 vmx_flush_tlb(vcpu);
10628 * Conceptually we want to copy the PML address and index from
10629 * vmcs01 here, and then back to vmcs01 on nested vmexit. But,
10630 * since we always flush the log on each vmexit, this happens
10631 * to be equivalent to simply resetting the fields in vmcs02.
10633 ASSERT(vmx->pml_pg);
10634 vmcs_write64(PML_ADDRESS, page_to_phys(vmx->pml_pg));
10635 vmcs_write16(GUEST_PML_INDEX, PML_ENTITY_NUM - 1);
10638 if (nested_cpu_has_ept(vmcs12)) {
10639 if (nested_ept_init_mmu_context(vcpu)) {
10640 *entry_failure_code = ENTRY_FAIL_DEFAULT;
10643 } else if (nested_cpu_has2(vmcs12,
10644 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
10645 vmx_flush_tlb_ept_only(vcpu);
10649 * This sets GUEST_CR0 to vmcs12->guest_cr0, possibly modifying those
10650 * bits which we consider mandatory enabled.
10651 * The CR0_READ_SHADOW is what L2 should have expected to read given
10652 * the specifications by L1; It's not enough to take
10653 * vmcs12->cr0_read_shadow because on our cr0_guest_host_mask we we
10654 * have more bits than L1 expected.
10656 vmx_set_cr0(vcpu, vmcs12->guest_cr0);
10657 vmcs_writel(CR0_READ_SHADOW, nested_read_cr0(vmcs12));
10659 vmx_set_cr4(vcpu, vmcs12->guest_cr4);
10660 vmcs_writel(CR4_READ_SHADOW, nested_read_cr4(vmcs12));
10662 if (from_vmentry &&
10663 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER))
10664 vcpu->arch.efer = vmcs12->guest_ia32_efer;
10665 else if (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE)
10666 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
10668 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
10669 /* Note: modifies VM_ENTRY/EXIT_CONTROLS and GUEST/HOST_IA32_EFER */
10670 vmx_set_efer(vcpu, vcpu->arch.efer);
10672 /* Shadow page tables on either EPT or shadow page tables. */
10673 if (nested_vmx_load_cr3(vcpu, vmcs12->guest_cr3, nested_cpu_has_ept(vmcs12),
10674 entry_failure_code))
10678 vcpu->arch.walk_mmu->inject_page_fault = vmx_inject_page_fault_nested;
10681 * L1 may access the L2's PDPTR, so save them to construct vmcs12
10684 vmcs_write64(GUEST_PDPTR0, vmcs12->guest_pdptr0);
10685 vmcs_write64(GUEST_PDPTR1, vmcs12->guest_pdptr1);
10686 vmcs_write64(GUEST_PDPTR2, vmcs12->guest_pdptr2);
10687 vmcs_write64(GUEST_PDPTR3, vmcs12->guest_pdptr3);
10690 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->guest_rsp);
10691 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->guest_rip);
10695 static int check_vmentry_prereqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10697 struct vcpu_vmx *vmx = to_vmx(vcpu);
10699 if (vmcs12->guest_activity_state != GUEST_ACTIVITY_ACTIVE &&
10700 vmcs12->guest_activity_state != GUEST_ACTIVITY_HLT)
10701 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10703 if (nested_vmx_check_io_bitmap_controls(vcpu, vmcs12))
10704 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10706 if (nested_vmx_check_msr_bitmap_controls(vcpu, vmcs12))
10707 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10709 if (nested_vmx_check_tpr_shadow_controls(vcpu, vmcs12))
10710 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10712 if (nested_vmx_check_apicv_controls(vcpu, vmcs12))
10713 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10715 if (nested_vmx_check_msr_switch_controls(vcpu, vmcs12))
10716 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10718 if (nested_vmx_check_pml_controls(vcpu, vmcs12))
10719 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10721 if (!vmx_control_verify(vmcs12->cpu_based_vm_exec_control,
10722 vmx->nested.nested_vmx_procbased_ctls_low,
10723 vmx->nested.nested_vmx_procbased_ctls_high) ||
10724 (nested_cpu_has(vmcs12, CPU_BASED_ACTIVATE_SECONDARY_CONTROLS) &&
10725 !vmx_control_verify(vmcs12->secondary_vm_exec_control,
10726 vmx->nested.nested_vmx_secondary_ctls_low,
10727 vmx->nested.nested_vmx_secondary_ctls_high)) ||
10728 !vmx_control_verify(vmcs12->pin_based_vm_exec_control,
10729 vmx->nested.nested_vmx_pinbased_ctls_low,
10730 vmx->nested.nested_vmx_pinbased_ctls_high) ||
10731 !vmx_control_verify(vmcs12->vm_exit_controls,
10732 vmx->nested.nested_vmx_exit_ctls_low,
10733 vmx->nested.nested_vmx_exit_ctls_high) ||
10734 !vmx_control_verify(vmcs12->vm_entry_controls,
10735 vmx->nested.nested_vmx_entry_ctls_low,
10736 vmx->nested.nested_vmx_entry_ctls_high))
10737 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10739 if (nested_cpu_has_vmfunc(vmcs12)) {
10740 if (vmcs12->vm_function_control &
10741 ~vmx->nested.nested_vmx_vmfunc_controls)
10742 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10744 if (nested_cpu_has_eptp_switching(vmcs12)) {
10745 if (!nested_cpu_has_ept(vmcs12) ||
10746 !page_address_valid(vcpu, vmcs12->eptp_list_address))
10747 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10751 if (vmcs12->cr3_target_count > nested_cpu_vmx_misc_cr3_count(vcpu))
10752 return VMXERR_ENTRY_INVALID_CONTROL_FIELD;
10754 if (!nested_host_cr0_valid(vcpu, vmcs12->host_cr0) ||
10755 !nested_host_cr4_valid(vcpu, vmcs12->host_cr4) ||
10756 !nested_cr3_valid(vcpu, vmcs12->host_cr3))
10757 return VMXERR_ENTRY_INVALID_HOST_STATE_FIELD;
10762 static int check_vmentry_postreqs(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
10767 *exit_qual = ENTRY_FAIL_DEFAULT;
10769 if (!nested_guest_cr0_valid(vcpu, vmcs12->guest_cr0) ||
10770 !nested_guest_cr4_valid(vcpu, vmcs12->guest_cr4))
10773 if (!nested_cpu_has2(vmcs12, SECONDARY_EXEC_SHADOW_VMCS) &&
10774 vmcs12->vmcs_link_pointer != -1ull) {
10775 *exit_qual = ENTRY_FAIL_VMCS_LINK_PTR;
10780 * If the load IA32_EFER VM-entry control is 1, the following checks
10781 * are performed on the field for the IA32_EFER MSR:
10782 * - Bits reserved in the IA32_EFER MSR must be 0.
10783 * - Bit 10 (corresponding to IA32_EFER.LMA) must equal the value of
10784 * the IA-32e mode guest VM-exit control. It must also be identical
10785 * to bit 8 (LME) if bit 31 in the CR0 field (corresponding to
10788 if (to_vmx(vcpu)->nested.nested_run_pending &&
10789 (vmcs12->vm_entry_controls & VM_ENTRY_LOAD_IA32_EFER)) {
10790 ia32e = (vmcs12->vm_entry_controls & VM_ENTRY_IA32E_MODE) != 0;
10791 if (!kvm_valid_efer(vcpu, vmcs12->guest_ia32_efer) ||
10792 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LMA) ||
10793 ((vmcs12->guest_cr0 & X86_CR0_PG) &&
10794 ia32e != !!(vmcs12->guest_ia32_efer & EFER_LME)))
10799 * If the load IA32_EFER VM-exit control is 1, bits reserved in the
10800 * IA32_EFER MSR must be 0 in the field for that register. In addition,
10801 * the values of the LMA and LME bits in the field must each be that of
10802 * the host address-space size VM-exit control.
10804 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER) {
10805 ia32e = (vmcs12->vm_exit_controls &
10806 VM_EXIT_HOST_ADDR_SPACE_SIZE) != 0;
10807 if (!kvm_valid_efer(vcpu, vmcs12->host_ia32_efer) ||
10808 ia32e != !!(vmcs12->host_ia32_efer & EFER_LMA) ||
10809 ia32e != !!(vmcs12->host_ia32_efer & EFER_LME))
10816 static int enter_vmx_non_root_mode(struct kvm_vcpu *vcpu, bool from_vmentry)
10818 struct vcpu_vmx *vmx = to_vmx(vcpu);
10819 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
10820 struct loaded_vmcs *vmcs02;
10824 vmcs02 = nested_get_current_vmcs02(vmx);
10828 enter_guest_mode(vcpu);
10830 if (!(vmcs12->vm_entry_controls & VM_ENTRY_LOAD_DEBUG_CONTROLS))
10831 vmx->nested.vmcs01_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
10833 vmx_switch_vmcs(vcpu, vmcs02);
10834 vmx_segment_cache_clear(vmx);
10836 if (prepare_vmcs02(vcpu, vmcs12, from_vmentry, &exit_qual)) {
10837 leave_guest_mode(vcpu);
10838 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10839 nested_vmx_entry_failure(vcpu, vmcs12,
10840 EXIT_REASON_INVALID_STATE, exit_qual);
10844 nested_get_vmcs12_pages(vcpu, vmcs12);
10846 msr_entry_idx = nested_vmx_load_msr(vcpu,
10847 vmcs12->vm_entry_msr_load_addr,
10848 vmcs12->vm_entry_msr_load_count);
10849 if (msr_entry_idx) {
10850 leave_guest_mode(vcpu);
10851 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
10852 nested_vmx_entry_failure(vcpu, vmcs12,
10853 EXIT_REASON_MSR_LOAD_FAIL, msr_entry_idx);
10858 * Note no nested_vmx_succeed or nested_vmx_fail here. At this point
10859 * we are no longer running L1, and VMLAUNCH/VMRESUME has not yet
10860 * returned as far as L1 is concerned. It will only return (and set
10861 * the success flag) when L2 exits (see nested_vmx_vmexit()).
10867 * nested_vmx_run() handles a nested entry, i.e., a VMLAUNCH or VMRESUME on L1
10868 * for running an L2 nested guest.
10870 static int nested_vmx_run(struct kvm_vcpu *vcpu, bool launch)
10872 struct vmcs12 *vmcs12;
10873 struct vcpu_vmx *vmx = to_vmx(vcpu);
10874 u32 interrupt_shadow = vmx_get_interrupt_shadow(vcpu);
10878 if (!nested_vmx_check_permission(vcpu))
10881 if (!nested_vmx_check_vmcs12(vcpu))
10884 vmcs12 = get_vmcs12(vcpu);
10886 if (enable_shadow_vmcs)
10887 copy_shadow_to_vmcs12(vmx);
10890 * The nested entry process starts with enforcing various prerequisites
10891 * on vmcs12 as required by the Intel SDM, and act appropriately when
10892 * they fail: As the SDM explains, some conditions should cause the
10893 * instruction to fail, while others will cause the instruction to seem
10894 * to succeed, but return an EXIT_REASON_INVALID_STATE.
10895 * To speed up the normal (success) code path, we should avoid checking
10896 * for misconfigurations which will anyway be caught by the processor
10897 * when using the merged vmcs02.
10899 if (interrupt_shadow & KVM_X86_SHADOW_INT_MOV_SS) {
10900 nested_vmx_failValid(vcpu,
10901 VMXERR_ENTRY_EVENTS_BLOCKED_BY_MOV_SS);
10905 if (vmcs12->launch_state == launch) {
10906 nested_vmx_failValid(vcpu,
10907 launch ? VMXERR_VMLAUNCH_NONCLEAR_VMCS
10908 : VMXERR_VMRESUME_NONLAUNCHED_VMCS);
10912 ret = check_vmentry_prereqs(vcpu, vmcs12);
10914 nested_vmx_failValid(vcpu, ret);
10919 * After this point, the trap flag no longer triggers a singlestep trap
10920 * on the vm entry instructions; don't call kvm_skip_emulated_instruction.
10921 * This is not 100% correct; for performance reasons, we delegate most
10922 * of the checks on host state to the processor. If those fail,
10923 * the singlestep trap is missed.
10925 skip_emulated_instruction(vcpu);
10927 ret = check_vmentry_postreqs(vcpu, vmcs12, &exit_qual);
10929 nested_vmx_entry_failure(vcpu, vmcs12,
10930 EXIT_REASON_INVALID_STATE, exit_qual);
10935 * We're finally done with prerequisite checking, and can start with
10936 * the nested entry.
10939 ret = enter_vmx_non_root_mode(vcpu, true);
10943 if (vmcs12->guest_activity_state == GUEST_ACTIVITY_HLT)
10944 return kvm_vcpu_halt(vcpu);
10946 vmx->nested.nested_run_pending = 1;
10951 return kvm_skip_emulated_instruction(vcpu);
10955 * On a nested exit from L2 to L1, vmcs12.guest_cr0 might not be up-to-date
10956 * because L2 may have changed some cr0 bits directly (CRO_GUEST_HOST_MASK).
10957 * This function returns the new value we should put in vmcs12.guest_cr0.
10958 * It's not enough to just return the vmcs02 GUEST_CR0. Rather,
10959 * 1. Bits that neither L0 nor L1 trapped, were set directly by L2 and are now
10960 * available in vmcs02 GUEST_CR0. (Note: It's enough to check that L0
10961 * didn't trap the bit, because if L1 did, so would L0).
10962 * 2. Bits that L1 asked to trap (and therefore L0 also did) could not have
10963 * been modified by L2, and L1 knows it. So just leave the old value of
10964 * the bit from vmcs12.guest_cr0. Note that the bit from vmcs02 GUEST_CR0
10965 * isn't relevant, because if L0 traps this bit it can set it to anything.
10966 * 3. Bits that L1 didn't trap, but L0 did. L1 believes the guest could have
10967 * changed these bits, and therefore they need to be updated, but L0
10968 * didn't necessarily allow them to be changed in GUEST_CR0 - and rather
10969 * put them in vmcs02 CR0_READ_SHADOW. So take these bits from there.
10971 static inline unsigned long
10972 vmcs12_guest_cr0(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10975 /*1*/ (vmcs_readl(GUEST_CR0) & vcpu->arch.cr0_guest_owned_bits) |
10976 /*2*/ (vmcs12->guest_cr0 & vmcs12->cr0_guest_host_mask) |
10977 /*3*/ (vmcs_readl(CR0_READ_SHADOW) & ~(vmcs12->cr0_guest_host_mask |
10978 vcpu->arch.cr0_guest_owned_bits));
10981 static inline unsigned long
10982 vmcs12_guest_cr4(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
10985 /*1*/ (vmcs_readl(GUEST_CR4) & vcpu->arch.cr4_guest_owned_bits) |
10986 /*2*/ (vmcs12->guest_cr4 & vmcs12->cr4_guest_host_mask) |
10987 /*3*/ (vmcs_readl(CR4_READ_SHADOW) & ~(vmcs12->cr4_guest_host_mask |
10988 vcpu->arch.cr4_guest_owned_bits));
10991 static void vmcs12_save_pending_event(struct kvm_vcpu *vcpu,
10992 struct vmcs12 *vmcs12)
10997 if (vcpu->arch.exception.injected) {
10998 nr = vcpu->arch.exception.nr;
10999 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
11001 if (kvm_exception_is_soft(nr)) {
11002 vmcs12->vm_exit_instruction_len =
11003 vcpu->arch.event_exit_inst_len;
11004 idt_vectoring |= INTR_TYPE_SOFT_EXCEPTION;
11006 idt_vectoring |= INTR_TYPE_HARD_EXCEPTION;
11008 if (vcpu->arch.exception.has_error_code) {
11009 idt_vectoring |= VECTORING_INFO_DELIVER_CODE_MASK;
11010 vmcs12->idt_vectoring_error_code =
11011 vcpu->arch.exception.error_code;
11014 vmcs12->idt_vectoring_info_field = idt_vectoring;
11015 } else if (vcpu->arch.nmi_injected) {
11016 vmcs12->idt_vectoring_info_field =
11017 INTR_TYPE_NMI_INTR | INTR_INFO_VALID_MASK | NMI_VECTOR;
11018 } else if (vcpu->arch.interrupt.pending) {
11019 nr = vcpu->arch.interrupt.nr;
11020 idt_vectoring = nr | VECTORING_INFO_VALID_MASK;
11022 if (vcpu->arch.interrupt.soft) {
11023 idt_vectoring |= INTR_TYPE_SOFT_INTR;
11024 vmcs12->vm_entry_instruction_len =
11025 vcpu->arch.event_exit_inst_len;
11027 idt_vectoring |= INTR_TYPE_EXT_INTR;
11029 vmcs12->idt_vectoring_info_field = idt_vectoring;
11033 static int vmx_check_nested_events(struct kvm_vcpu *vcpu, bool external_intr)
11035 struct vcpu_vmx *vmx = to_vmx(vcpu);
11036 unsigned long exit_qual;
11038 if (kvm_event_needs_reinjection(vcpu))
11041 if (vcpu->arch.exception.pending &&
11042 nested_vmx_check_exception(vcpu, &exit_qual)) {
11043 if (vmx->nested.nested_run_pending)
11045 nested_vmx_inject_exception_vmexit(vcpu, exit_qual);
11046 vcpu->arch.exception.pending = false;
11050 if (nested_cpu_has_preemption_timer(get_vmcs12(vcpu)) &&
11051 vmx->nested.preemption_timer_expired) {
11052 if (vmx->nested.nested_run_pending)
11054 nested_vmx_vmexit(vcpu, EXIT_REASON_PREEMPTION_TIMER, 0, 0);
11058 if (vcpu->arch.nmi_pending && nested_exit_on_nmi(vcpu)) {
11059 if (vmx->nested.nested_run_pending)
11061 nested_vmx_vmexit(vcpu, EXIT_REASON_EXCEPTION_NMI,
11062 NMI_VECTOR | INTR_TYPE_NMI_INTR |
11063 INTR_INFO_VALID_MASK, 0);
11065 * The NMI-triggered VM exit counts as injection:
11066 * clear this one and block further NMIs.
11068 vcpu->arch.nmi_pending = 0;
11069 vmx_set_nmi_mask(vcpu, true);
11073 if ((kvm_cpu_has_interrupt(vcpu) || external_intr) &&
11074 nested_exit_on_intr(vcpu)) {
11075 if (vmx->nested.nested_run_pending)
11077 nested_vmx_vmexit(vcpu, EXIT_REASON_EXTERNAL_INTERRUPT, 0, 0);
11081 vmx_complete_nested_posted_interrupt(vcpu);
11085 static u32 vmx_get_preemption_timer_value(struct kvm_vcpu *vcpu)
11087 ktime_t remaining =
11088 hrtimer_get_remaining(&to_vmx(vcpu)->nested.preemption_timer);
11091 if (ktime_to_ns(remaining) <= 0)
11094 value = ktime_to_ns(remaining) * vcpu->arch.virtual_tsc_khz;
11095 do_div(value, 1000000);
11096 return value >> VMX_MISC_EMULATED_PREEMPTION_TIMER_RATE;
11100 * Update the guest state fields of vmcs12 to reflect changes that
11101 * occurred while L2 was running. (The "IA-32e mode guest" bit of the
11102 * VM-entry controls is also updated, since this is really a guest
11105 static void sync_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12)
11107 vmcs12->guest_cr0 = vmcs12_guest_cr0(vcpu, vmcs12);
11108 vmcs12->guest_cr4 = vmcs12_guest_cr4(vcpu, vmcs12);
11110 vmcs12->guest_rsp = kvm_register_read(vcpu, VCPU_REGS_RSP);
11111 vmcs12->guest_rip = kvm_register_read(vcpu, VCPU_REGS_RIP);
11112 vmcs12->guest_rflags = vmcs_readl(GUEST_RFLAGS);
11114 vmcs12->guest_es_selector = vmcs_read16(GUEST_ES_SELECTOR);
11115 vmcs12->guest_cs_selector = vmcs_read16(GUEST_CS_SELECTOR);
11116 vmcs12->guest_ss_selector = vmcs_read16(GUEST_SS_SELECTOR);
11117 vmcs12->guest_ds_selector = vmcs_read16(GUEST_DS_SELECTOR);
11118 vmcs12->guest_fs_selector = vmcs_read16(GUEST_FS_SELECTOR);
11119 vmcs12->guest_gs_selector = vmcs_read16(GUEST_GS_SELECTOR);
11120 vmcs12->guest_ldtr_selector = vmcs_read16(GUEST_LDTR_SELECTOR);
11121 vmcs12->guest_tr_selector = vmcs_read16(GUEST_TR_SELECTOR);
11122 vmcs12->guest_es_limit = vmcs_read32(GUEST_ES_LIMIT);
11123 vmcs12->guest_cs_limit = vmcs_read32(GUEST_CS_LIMIT);
11124 vmcs12->guest_ss_limit = vmcs_read32(GUEST_SS_LIMIT);
11125 vmcs12->guest_ds_limit = vmcs_read32(GUEST_DS_LIMIT);
11126 vmcs12->guest_fs_limit = vmcs_read32(GUEST_FS_LIMIT);
11127 vmcs12->guest_gs_limit = vmcs_read32(GUEST_GS_LIMIT);
11128 vmcs12->guest_ldtr_limit = vmcs_read32(GUEST_LDTR_LIMIT);
11129 vmcs12->guest_tr_limit = vmcs_read32(GUEST_TR_LIMIT);
11130 vmcs12->guest_gdtr_limit = vmcs_read32(GUEST_GDTR_LIMIT);
11131 vmcs12->guest_idtr_limit = vmcs_read32(GUEST_IDTR_LIMIT);
11132 vmcs12->guest_es_ar_bytes = vmcs_read32(GUEST_ES_AR_BYTES);
11133 vmcs12->guest_cs_ar_bytes = vmcs_read32(GUEST_CS_AR_BYTES);
11134 vmcs12->guest_ss_ar_bytes = vmcs_read32(GUEST_SS_AR_BYTES);
11135 vmcs12->guest_ds_ar_bytes = vmcs_read32(GUEST_DS_AR_BYTES);
11136 vmcs12->guest_fs_ar_bytes = vmcs_read32(GUEST_FS_AR_BYTES);
11137 vmcs12->guest_gs_ar_bytes = vmcs_read32(GUEST_GS_AR_BYTES);
11138 vmcs12->guest_ldtr_ar_bytes = vmcs_read32(GUEST_LDTR_AR_BYTES);
11139 vmcs12->guest_tr_ar_bytes = vmcs_read32(GUEST_TR_AR_BYTES);
11140 vmcs12->guest_es_base = vmcs_readl(GUEST_ES_BASE);
11141 vmcs12->guest_cs_base = vmcs_readl(GUEST_CS_BASE);
11142 vmcs12->guest_ss_base = vmcs_readl(GUEST_SS_BASE);
11143 vmcs12->guest_ds_base = vmcs_readl(GUEST_DS_BASE);
11144 vmcs12->guest_fs_base = vmcs_readl(GUEST_FS_BASE);
11145 vmcs12->guest_gs_base = vmcs_readl(GUEST_GS_BASE);
11146 vmcs12->guest_ldtr_base = vmcs_readl(GUEST_LDTR_BASE);
11147 vmcs12->guest_tr_base = vmcs_readl(GUEST_TR_BASE);
11148 vmcs12->guest_gdtr_base = vmcs_readl(GUEST_GDTR_BASE);
11149 vmcs12->guest_idtr_base = vmcs_readl(GUEST_IDTR_BASE);
11151 vmcs12->guest_interruptibility_info =
11152 vmcs_read32(GUEST_INTERRUPTIBILITY_INFO);
11153 vmcs12->guest_pending_dbg_exceptions =
11154 vmcs_readl(GUEST_PENDING_DBG_EXCEPTIONS);
11155 if (vcpu->arch.mp_state == KVM_MP_STATE_HALTED)
11156 vmcs12->guest_activity_state = GUEST_ACTIVITY_HLT;
11158 vmcs12->guest_activity_state = GUEST_ACTIVITY_ACTIVE;
11160 if (nested_cpu_has_preemption_timer(vmcs12)) {
11161 if (vmcs12->vm_exit_controls &
11162 VM_EXIT_SAVE_VMX_PREEMPTION_TIMER)
11163 vmcs12->vmx_preemption_timer_value =
11164 vmx_get_preemption_timer_value(vcpu);
11165 hrtimer_cancel(&to_vmx(vcpu)->nested.preemption_timer);
11169 * In some cases (usually, nested EPT), L2 is allowed to change its
11170 * own CR3 without exiting. If it has changed it, we must keep it.
11171 * Of course, if L0 is using shadow page tables, GUEST_CR3 was defined
11172 * by L0, not L1 or L2, so we mustn't unconditionally copy it to vmcs12.
11174 * Additionally, restore L2's PDPTR to vmcs12.
11177 vmcs12->guest_cr3 = vmcs_readl(GUEST_CR3);
11178 vmcs12->guest_pdptr0 = vmcs_read64(GUEST_PDPTR0);
11179 vmcs12->guest_pdptr1 = vmcs_read64(GUEST_PDPTR1);
11180 vmcs12->guest_pdptr2 = vmcs_read64(GUEST_PDPTR2);
11181 vmcs12->guest_pdptr3 = vmcs_read64(GUEST_PDPTR3);
11184 vmcs12->guest_linear_address = vmcs_readl(GUEST_LINEAR_ADDRESS);
11186 if (nested_cpu_has_vid(vmcs12))
11187 vmcs12->guest_intr_status = vmcs_read16(GUEST_INTR_STATUS);
11189 vmcs12->vm_entry_controls =
11190 (vmcs12->vm_entry_controls & ~VM_ENTRY_IA32E_MODE) |
11191 (vm_entry_controls_get(to_vmx(vcpu)) & VM_ENTRY_IA32E_MODE);
11193 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_DEBUG_CONTROLS) {
11194 kvm_get_dr(vcpu, 7, (unsigned long *)&vmcs12->guest_dr7);
11195 vmcs12->guest_ia32_debugctl = vmcs_read64(GUEST_IA32_DEBUGCTL);
11198 /* TODO: These cannot have changed unless we have MSR bitmaps and
11199 * the relevant bit asks not to trap the change */
11200 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_PAT)
11201 vmcs12->guest_ia32_pat = vmcs_read64(GUEST_IA32_PAT);
11202 if (vmcs12->vm_exit_controls & VM_EXIT_SAVE_IA32_EFER)
11203 vmcs12->guest_ia32_efer = vcpu->arch.efer;
11204 vmcs12->guest_sysenter_cs = vmcs_read32(GUEST_SYSENTER_CS);
11205 vmcs12->guest_sysenter_esp = vmcs_readl(GUEST_SYSENTER_ESP);
11206 vmcs12->guest_sysenter_eip = vmcs_readl(GUEST_SYSENTER_EIP);
11207 if (kvm_mpx_supported())
11208 vmcs12->guest_bndcfgs = vmcs_read64(GUEST_BNDCFGS);
11212 * prepare_vmcs12 is part of what we need to do when the nested L2 guest exits
11213 * and we want to prepare to run its L1 parent. L1 keeps a vmcs for L2 (vmcs12),
11214 * and this function updates it to reflect the changes to the guest state while
11215 * L2 was running (and perhaps made some exits which were handled directly by L0
11216 * without going back to L1), and to reflect the exit reason.
11217 * Note that we do not have to copy here all VMCS fields, just those that
11218 * could have changed by the L2 guest or the exit - i.e., the guest-state and
11219 * exit-information fields only. Other fields are modified by L1 with VMWRITE,
11220 * which already writes to vmcs12 directly.
11222 static void prepare_vmcs12(struct kvm_vcpu *vcpu, struct vmcs12 *vmcs12,
11223 u32 exit_reason, u32 exit_intr_info,
11224 unsigned long exit_qualification)
11226 /* update guest state fields: */
11227 sync_vmcs12(vcpu, vmcs12);
11229 /* update exit information fields: */
11231 vmcs12->vm_exit_reason = exit_reason;
11232 vmcs12->exit_qualification = exit_qualification;
11233 vmcs12->vm_exit_intr_info = exit_intr_info;
11235 vmcs12->idt_vectoring_info_field = 0;
11236 vmcs12->vm_exit_instruction_len = vmcs_read32(VM_EXIT_INSTRUCTION_LEN);
11237 vmcs12->vmx_instruction_info = vmcs_read32(VMX_INSTRUCTION_INFO);
11239 if (!(vmcs12->vm_exit_reason & VMX_EXIT_REASONS_FAILED_VMENTRY)) {
11240 vmcs12->launch_state = 1;
11242 /* vm_entry_intr_info_field is cleared on exit. Emulate this
11243 * instead of reading the real value. */
11244 vmcs12->vm_entry_intr_info_field &= ~INTR_INFO_VALID_MASK;
11247 * Transfer the event that L0 or L1 may wanted to inject into
11248 * L2 to IDT_VECTORING_INFO_FIELD.
11250 vmcs12_save_pending_event(vcpu, vmcs12);
11254 * Drop what we picked up for L2 via vmx_complete_interrupts. It is
11255 * preserved above and would only end up incorrectly in L1.
11257 vcpu->arch.nmi_injected = false;
11258 kvm_clear_exception_queue(vcpu);
11259 kvm_clear_interrupt_queue(vcpu);
11263 * A part of what we need to when the nested L2 guest exits and we want to
11264 * run its L1 parent, is to reset L1's guest state to the host state specified
11266 * This function is to be called not only on normal nested exit, but also on
11267 * a nested entry failure, as explained in Intel's spec, 3B.23.7 ("VM-Entry
11268 * Failures During or After Loading Guest State").
11269 * This function should be called when the active VMCS is L1's (vmcs01).
11271 static void load_vmcs12_host_state(struct kvm_vcpu *vcpu,
11272 struct vmcs12 *vmcs12)
11274 struct kvm_segment seg;
11275 u32 entry_failure_code;
11277 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_EFER)
11278 vcpu->arch.efer = vmcs12->host_ia32_efer;
11279 else if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
11280 vcpu->arch.efer |= (EFER_LMA | EFER_LME);
11282 vcpu->arch.efer &= ~(EFER_LMA | EFER_LME);
11283 vmx_set_efer(vcpu, vcpu->arch.efer);
11285 kvm_register_write(vcpu, VCPU_REGS_RSP, vmcs12->host_rsp);
11286 kvm_register_write(vcpu, VCPU_REGS_RIP, vmcs12->host_rip);
11287 vmx_set_rflags(vcpu, X86_EFLAGS_FIXED);
11289 * Note that calling vmx_set_cr0 is important, even if cr0 hasn't
11290 * actually changed, because vmx_set_cr0 refers to efer set above.
11292 * CR0_GUEST_HOST_MASK is already set in the original vmcs01
11293 * (KVM doesn't change it);
11295 vcpu->arch.cr0_guest_owned_bits = X86_CR0_TS;
11296 vmx_set_cr0(vcpu, vmcs12->host_cr0);
11298 /* Same as above - no reason to call set_cr4_guest_host_mask(). */
11299 vcpu->arch.cr4_guest_owned_bits = ~vmcs_readl(CR4_GUEST_HOST_MASK);
11300 vmx_set_cr4(vcpu, vmcs12->host_cr4);
11302 nested_ept_uninit_mmu_context(vcpu);
11305 * Only PDPTE load can fail as the value of cr3 was checked on entry and
11306 * couldn't have changed.
11308 if (nested_vmx_load_cr3(vcpu, vmcs12->host_cr3, false, &entry_failure_code))
11309 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_PDPTE_FAIL);
11312 vcpu->arch.walk_mmu->inject_page_fault = kvm_inject_page_fault;
11316 * Trivially support vpid by letting L2s share their parent
11317 * L1's vpid. TODO: move to a more elaborate solution, giving
11318 * each L2 its own vpid and exposing the vpid feature to L1.
11320 vmx_flush_tlb(vcpu);
11322 /* Restore posted intr vector. */
11323 if (nested_cpu_has_posted_intr(vmcs12))
11324 vmcs_write16(POSTED_INTR_NV, POSTED_INTR_VECTOR);
11326 vmcs_write32(GUEST_SYSENTER_CS, vmcs12->host_ia32_sysenter_cs);
11327 vmcs_writel(GUEST_SYSENTER_ESP, vmcs12->host_ia32_sysenter_esp);
11328 vmcs_writel(GUEST_SYSENTER_EIP, vmcs12->host_ia32_sysenter_eip);
11329 vmcs_writel(GUEST_IDTR_BASE, vmcs12->host_idtr_base);
11330 vmcs_writel(GUEST_GDTR_BASE, vmcs12->host_gdtr_base);
11332 /* If not VM_EXIT_CLEAR_BNDCFGS, the L2 value propagates to L1. */
11333 if (vmcs12->vm_exit_controls & VM_EXIT_CLEAR_BNDCFGS)
11334 vmcs_write64(GUEST_BNDCFGS, 0);
11336 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PAT) {
11337 vmcs_write64(GUEST_IA32_PAT, vmcs12->host_ia32_pat);
11338 vcpu->arch.pat = vmcs12->host_ia32_pat;
11340 if (vmcs12->vm_exit_controls & VM_EXIT_LOAD_IA32_PERF_GLOBAL_CTRL)
11341 vmcs_write64(GUEST_IA32_PERF_GLOBAL_CTRL,
11342 vmcs12->host_ia32_perf_global_ctrl);
11344 /* Set L1 segment info according to Intel SDM
11345 27.5.2 Loading Host Segment and Descriptor-Table Registers */
11346 seg = (struct kvm_segment) {
11348 .limit = 0xFFFFFFFF,
11349 .selector = vmcs12->host_cs_selector,
11355 if (vmcs12->vm_exit_controls & VM_EXIT_HOST_ADDR_SPACE_SIZE)
11359 vmx_set_segment(vcpu, &seg, VCPU_SREG_CS);
11360 seg = (struct kvm_segment) {
11362 .limit = 0xFFFFFFFF,
11369 seg.selector = vmcs12->host_ds_selector;
11370 vmx_set_segment(vcpu, &seg, VCPU_SREG_DS);
11371 seg.selector = vmcs12->host_es_selector;
11372 vmx_set_segment(vcpu, &seg, VCPU_SREG_ES);
11373 seg.selector = vmcs12->host_ss_selector;
11374 vmx_set_segment(vcpu, &seg, VCPU_SREG_SS);
11375 seg.selector = vmcs12->host_fs_selector;
11376 seg.base = vmcs12->host_fs_base;
11377 vmx_set_segment(vcpu, &seg, VCPU_SREG_FS);
11378 seg.selector = vmcs12->host_gs_selector;
11379 seg.base = vmcs12->host_gs_base;
11380 vmx_set_segment(vcpu, &seg, VCPU_SREG_GS);
11381 seg = (struct kvm_segment) {
11382 .base = vmcs12->host_tr_base,
11384 .selector = vmcs12->host_tr_selector,
11388 vmx_set_segment(vcpu, &seg, VCPU_SREG_TR);
11390 kvm_set_dr(vcpu, 7, 0x400);
11391 vmcs_write64(GUEST_IA32_DEBUGCTL, 0);
11393 if (cpu_has_vmx_msr_bitmap())
11394 vmx_set_msr_bitmap(vcpu);
11396 if (nested_vmx_load_msr(vcpu, vmcs12->vm_exit_msr_load_addr,
11397 vmcs12->vm_exit_msr_load_count))
11398 nested_vmx_abort(vcpu, VMX_ABORT_LOAD_HOST_MSR_FAIL);
11402 * Emulate an exit from nested guest (L2) to L1, i.e., prepare to run L1
11403 * and modify vmcs12 to make it see what it would expect to see there if
11404 * L2 was its real guest. Must only be called when in L2 (is_guest_mode())
11406 static void nested_vmx_vmexit(struct kvm_vcpu *vcpu, u32 exit_reason,
11407 u32 exit_intr_info,
11408 unsigned long exit_qualification)
11410 struct vcpu_vmx *vmx = to_vmx(vcpu);
11411 struct vmcs12 *vmcs12 = get_vmcs12(vcpu);
11413 /* trying to cancel vmlaunch/vmresume is a bug */
11414 WARN_ON_ONCE(vmx->nested.nested_run_pending);
11417 * The only expected VM-instruction error is "VM entry with
11418 * invalid control field(s)." Anything else indicates a
11421 WARN_ON_ONCE(vmx->fail && (vmcs_read32(VM_INSTRUCTION_ERROR) !=
11422 VMXERR_ENTRY_INVALID_CONTROL_FIELD));
11424 leave_guest_mode(vcpu);
11426 if (likely(!vmx->fail)) {
11427 prepare_vmcs12(vcpu, vmcs12, exit_reason, exit_intr_info,
11428 exit_qualification);
11430 if (nested_vmx_store_msr(vcpu, vmcs12->vm_exit_msr_store_addr,
11431 vmcs12->vm_exit_msr_store_count))
11432 nested_vmx_abort(vcpu, VMX_ABORT_SAVE_GUEST_MSR_FAIL);
11435 vmx_switch_vmcs(vcpu, &vmx->vmcs01);
11436 vm_entry_controls_reset_shadow(vmx);
11437 vm_exit_controls_reset_shadow(vmx);
11438 vmx_segment_cache_clear(vmx);
11440 /* if no vmcs02 cache requested, remove the one we used */
11441 if (VMCS02_POOL_SIZE == 0)
11442 nested_free_vmcs02(vmx, vmx->nested.current_vmptr);
11444 /* Update any VMCS fields that might have changed while L2 ran */
11445 vmcs_write32(VM_EXIT_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11446 vmcs_write32(VM_ENTRY_MSR_LOAD_COUNT, vmx->msr_autoload.nr);
11447 vmcs_write64(TSC_OFFSET, vcpu->arch.tsc_offset);
11448 if (vmx->hv_deadline_tsc == -1)
11449 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11450 PIN_BASED_VMX_PREEMPTION_TIMER);
11452 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11453 PIN_BASED_VMX_PREEMPTION_TIMER);
11454 if (kvm_has_tsc_control)
11455 decache_tsc_multiplier(vmx);
11457 if (vmx->nested.change_vmcs01_virtual_x2apic_mode) {
11458 vmx->nested.change_vmcs01_virtual_x2apic_mode = false;
11459 vmx_set_virtual_x2apic_mode(vcpu,
11460 vcpu->arch.apic_base & X2APIC_ENABLE);
11461 } else if (!nested_cpu_has_ept(vmcs12) &&
11462 nested_cpu_has2(vmcs12,
11463 SECONDARY_EXEC_VIRTUALIZE_APIC_ACCESSES)) {
11464 vmx_flush_tlb_ept_only(vcpu);
11467 /* This is needed for same reason as it was needed in prepare_vmcs02 */
11470 /* Unpin physical memory we referred to in vmcs02 */
11471 if (vmx->nested.apic_access_page) {
11472 kvm_release_page_dirty(vmx->nested.apic_access_page);
11473 vmx->nested.apic_access_page = NULL;
11475 if (vmx->nested.virtual_apic_page) {
11476 kvm_release_page_dirty(vmx->nested.virtual_apic_page);
11477 vmx->nested.virtual_apic_page = NULL;
11479 if (vmx->nested.pi_desc_page) {
11480 kunmap(vmx->nested.pi_desc_page);
11481 kvm_release_page_dirty(vmx->nested.pi_desc_page);
11482 vmx->nested.pi_desc_page = NULL;
11483 vmx->nested.pi_desc = NULL;
11487 * We are now running in L2, mmu_notifier will force to reload the
11488 * page's hpa for L2 vmcs. Need to reload it for L1 before entering L1.
11490 kvm_make_request(KVM_REQ_APIC_PAGE_RELOAD, vcpu);
11492 if (enable_shadow_vmcs)
11493 vmx->nested.sync_shadow_vmcs = true;
11495 /* in case we halted in L2 */
11496 vcpu->arch.mp_state = KVM_MP_STATE_RUNNABLE;
11498 if (likely(!vmx->fail)) {
11500 * TODO: SDM says that with acknowledge interrupt on
11501 * exit, bit 31 of the VM-exit interrupt information
11502 * (valid interrupt) is always set to 1 on
11503 * EXIT_REASON_EXTERNAL_INTERRUPT, so we shouldn't
11504 * need kvm_cpu_has_interrupt(). See the commit
11505 * message for details.
11507 if (nested_exit_intr_ack_set(vcpu) &&
11508 exit_reason == EXIT_REASON_EXTERNAL_INTERRUPT &&
11509 kvm_cpu_has_interrupt(vcpu)) {
11510 int irq = kvm_cpu_get_interrupt(vcpu);
11512 vmcs12->vm_exit_intr_info = irq |
11513 INTR_INFO_VALID_MASK | INTR_TYPE_EXT_INTR;
11516 trace_kvm_nested_vmexit_inject(vmcs12->vm_exit_reason,
11517 vmcs12->exit_qualification,
11518 vmcs12->idt_vectoring_info_field,
11519 vmcs12->vm_exit_intr_info,
11520 vmcs12->vm_exit_intr_error_code,
11523 load_vmcs12_host_state(vcpu, vmcs12);
11529 * After an early L2 VM-entry failure, we're now back
11530 * in L1 which thinks it just finished a VMLAUNCH or
11531 * VMRESUME instruction, so we need to set the failure
11532 * flag and the VM-instruction error field of the VMCS
11535 nested_vmx_failValid(vcpu, VMXERR_ENTRY_INVALID_CONTROL_FIELD);
11537 * The emulated instruction was already skipped in
11538 * nested_vmx_run, but the updated RIP was never
11539 * written back to the vmcs01.
11541 skip_emulated_instruction(vcpu);
11546 * Forcibly leave nested mode in order to be able to reset the VCPU later on.
11548 static void vmx_leave_nested(struct kvm_vcpu *vcpu)
11550 if (is_guest_mode(vcpu)) {
11551 to_vmx(vcpu)->nested.nested_run_pending = 0;
11552 nested_vmx_vmexit(vcpu, -1, 0, 0);
11554 free_nested(to_vmx(vcpu));
11558 * L1's failure to enter L2 is a subset of a normal exit, as explained in
11559 * 23.7 "VM-entry failures during or after loading guest state" (this also
11560 * lists the acceptable exit-reason and exit-qualification parameters).
11561 * It should only be called before L2 actually succeeded to run, and when
11562 * vmcs01 is current (it doesn't leave_guest_mode() or switch vmcss).
11564 static void nested_vmx_entry_failure(struct kvm_vcpu *vcpu,
11565 struct vmcs12 *vmcs12,
11566 u32 reason, unsigned long qualification)
11568 load_vmcs12_host_state(vcpu, vmcs12);
11569 vmcs12->vm_exit_reason = reason | VMX_EXIT_REASONS_FAILED_VMENTRY;
11570 vmcs12->exit_qualification = qualification;
11571 nested_vmx_succeed(vcpu);
11572 if (enable_shadow_vmcs)
11573 to_vmx(vcpu)->nested.sync_shadow_vmcs = true;
11576 static int vmx_check_intercept(struct kvm_vcpu *vcpu,
11577 struct x86_instruction_info *info,
11578 enum x86_intercept_stage stage)
11580 return X86EMUL_CONTINUE;
11583 #ifdef CONFIG_X86_64
11584 /* (a << shift) / divisor, return 1 if overflow otherwise 0 */
11585 static inline int u64_shl_div_u64(u64 a, unsigned int shift,
11586 u64 divisor, u64 *result)
11588 u64 low = a << shift, high = a >> (64 - shift);
11590 /* To avoid the overflow on divq */
11591 if (high >= divisor)
11594 /* Low hold the result, high hold rem which is discarded */
11595 asm("divq %2\n\t" : "=a" (low), "=d" (high) :
11596 "rm" (divisor), "0" (low), "1" (high));
11602 static int vmx_set_hv_timer(struct kvm_vcpu *vcpu, u64 guest_deadline_tsc)
11604 struct vcpu_vmx *vmx = to_vmx(vcpu);
11605 u64 tscl = rdtsc();
11606 u64 guest_tscl = kvm_read_l1_tsc(vcpu, tscl);
11607 u64 delta_tsc = max(guest_deadline_tsc, guest_tscl) - guest_tscl;
11609 /* Convert to host delta tsc if tsc scaling is enabled */
11610 if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio &&
11611 u64_shl_div_u64(delta_tsc,
11612 kvm_tsc_scaling_ratio_frac_bits,
11613 vcpu->arch.tsc_scaling_ratio,
11618 * If the delta tsc can't fit in the 32 bit after the multi shift,
11619 * we can't use the preemption timer.
11620 * It's possible that it fits on later vmentries, but checking
11621 * on every vmentry is costly so we just use an hrtimer.
11623 if (delta_tsc >> (cpu_preemption_timer_multi + 32))
11626 vmx->hv_deadline_tsc = tscl + delta_tsc;
11627 vmcs_set_bits(PIN_BASED_VM_EXEC_CONTROL,
11628 PIN_BASED_VMX_PREEMPTION_TIMER);
11630 return delta_tsc == 0;
11633 static void vmx_cancel_hv_timer(struct kvm_vcpu *vcpu)
11635 struct vcpu_vmx *vmx = to_vmx(vcpu);
11636 vmx->hv_deadline_tsc = -1;
11637 vmcs_clear_bits(PIN_BASED_VM_EXEC_CONTROL,
11638 PIN_BASED_VMX_PREEMPTION_TIMER);
11642 static void vmx_sched_in(struct kvm_vcpu *vcpu, int cpu)
11645 shrink_ple_window(vcpu);
11648 static void vmx_slot_enable_log_dirty(struct kvm *kvm,
11649 struct kvm_memory_slot *slot)
11651 kvm_mmu_slot_leaf_clear_dirty(kvm, slot);
11652 kvm_mmu_slot_largepage_remove_write_access(kvm, slot);
11655 static void vmx_slot_disable_log_dirty(struct kvm *kvm,
11656 struct kvm_memory_slot *slot)
11658 kvm_mmu_slot_set_dirty(kvm, slot);
11661 static void vmx_flush_log_dirty(struct kvm *kvm)
11663 kvm_flush_pml_buffers(kvm);
11666 static int vmx_write_pml_buffer(struct kvm_vcpu *vcpu)
11668 struct vmcs12 *vmcs12;
11669 struct vcpu_vmx *vmx = to_vmx(vcpu);
11671 struct page *page = NULL;
11674 if (is_guest_mode(vcpu)) {
11675 WARN_ON_ONCE(vmx->nested.pml_full);
11678 * Check if PML is enabled for the nested guest.
11679 * Whether eptp bit 6 is set is already checked
11680 * as part of A/D emulation.
11682 vmcs12 = get_vmcs12(vcpu);
11683 if (!nested_cpu_has_pml(vmcs12))
11686 if (vmcs12->guest_pml_index >= PML_ENTITY_NUM) {
11687 vmx->nested.pml_full = true;
11691 gpa = vmcs_read64(GUEST_PHYSICAL_ADDRESS) & ~0xFFFull;
11693 page = kvm_vcpu_gpa_to_page(vcpu, vmcs12->pml_address);
11694 if (is_error_page(page))
11697 pml_address = kmap(page);
11698 pml_address[vmcs12->guest_pml_index--] = gpa;
11700 kvm_release_page_clean(page);
11706 static void vmx_enable_log_dirty_pt_masked(struct kvm *kvm,
11707 struct kvm_memory_slot *memslot,
11708 gfn_t offset, unsigned long mask)
11710 kvm_mmu_clear_dirty_pt_masked(kvm, memslot, offset, mask);
11713 static void __pi_post_block(struct kvm_vcpu *vcpu)
11715 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11716 struct pi_desc old, new;
11720 old.control = new.control = pi_desc->control;
11721 WARN(old.nv != POSTED_INTR_WAKEUP_VECTOR,
11722 "Wakeup handler not enabled while the VCPU is blocked\n");
11724 dest = cpu_physical_id(vcpu->cpu);
11726 if (x2apic_enabled())
11729 new.ndst = (dest << 8) & 0xFF00;
11731 /* set 'NV' to 'notification vector' */
11732 new.nv = POSTED_INTR_VECTOR;
11733 } while (cmpxchg64(&pi_desc->control, old.control,
11734 new.control) != old.control);
11736 if (!WARN_ON_ONCE(vcpu->pre_pcpu == -1)) {
11737 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
11738 list_del(&vcpu->blocked_vcpu_list);
11739 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
11740 vcpu->pre_pcpu = -1;
11745 * This routine does the following things for vCPU which is going
11746 * to be blocked if VT-d PI is enabled.
11747 * - Store the vCPU to the wakeup list, so when interrupts happen
11748 * we can find the right vCPU to wake up.
11749 * - Change the Posted-interrupt descriptor as below:
11750 * 'NDST' <-- vcpu->pre_pcpu
11751 * 'NV' <-- POSTED_INTR_WAKEUP_VECTOR
11752 * - If 'ON' is set during this process, which means at least one
11753 * interrupt is posted for this vCPU, we cannot block it, in
11754 * this case, return 1, otherwise, return 0.
11757 static int pi_pre_block(struct kvm_vcpu *vcpu)
11760 struct pi_desc old, new;
11761 struct pi_desc *pi_desc = vcpu_to_pi_desc(vcpu);
11763 if (!kvm_arch_has_assigned_device(vcpu->kvm) ||
11764 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11765 !kvm_vcpu_apicv_active(vcpu))
11768 WARN_ON(irqs_disabled());
11769 local_irq_disable();
11770 if (!WARN_ON_ONCE(vcpu->pre_pcpu != -1)) {
11771 vcpu->pre_pcpu = vcpu->cpu;
11772 spin_lock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
11773 list_add_tail(&vcpu->blocked_vcpu_list,
11774 &per_cpu(blocked_vcpu_on_cpu,
11776 spin_unlock(&per_cpu(blocked_vcpu_on_cpu_lock, vcpu->pre_pcpu));
11780 old.control = new.control = pi_desc->control;
11782 WARN((pi_desc->sn == 1),
11783 "Warning: SN field of posted-interrupts "
11784 "is set before blocking\n");
11787 * Since vCPU can be preempted during this process,
11788 * vcpu->cpu could be different with pre_pcpu, we
11789 * need to set pre_pcpu as the destination of wakeup
11790 * notification event, then we can find the right vCPU
11791 * to wakeup in wakeup handler if interrupts happen
11792 * when the vCPU is in blocked state.
11794 dest = cpu_physical_id(vcpu->pre_pcpu);
11796 if (x2apic_enabled())
11799 new.ndst = (dest << 8) & 0xFF00;
11801 /* set 'NV' to 'wakeup vector' */
11802 new.nv = POSTED_INTR_WAKEUP_VECTOR;
11803 } while (cmpxchg64(&pi_desc->control, old.control,
11804 new.control) != old.control);
11806 /* We should not block the vCPU if an interrupt is posted for it. */
11807 if (pi_test_on(pi_desc) == 1)
11808 __pi_post_block(vcpu);
11810 local_irq_enable();
11811 return (vcpu->pre_pcpu == -1);
11814 static int vmx_pre_block(struct kvm_vcpu *vcpu)
11816 if (pi_pre_block(vcpu))
11819 if (kvm_lapic_hv_timer_in_use(vcpu))
11820 kvm_lapic_switch_to_sw_timer(vcpu);
11825 static void pi_post_block(struct kvm_vcpu *vcpu)
11827 if (vcpu->pre_pcpu == -1)
11830 WARN_ON(irqs_disabled());
11831 local_irq_disable();
11832 __pi_post_block(vcpu);
11833 local_irq_enable();
11836 static void vmx_post_block(struct kvm_vcpu *vcpu)
11838 if (kvm_x86_ops->set_hv_timer)
11839 kvm_lapic_switch_to_hv_timer(vcpu);
11841 pi_post_block(vcpu);
11845 * vmx_update_pi_irte - set IRTE for Posted-Interrupts
11848 * @host_irq: host irq of the interrupt
11849 * @guest_irq: gsi of the interrupt
11850 * @set: set or unset PI
11851 * returns 0 on success, < 0 on failure
11853 static int vmx_update_pi_irte(struct kvm *kvm, unsigned int host_irq,
11854 uint32_t guest_irq, bool set)
11856 struct kvm_kernel_irq_routing_entry *e;
11857 struct kvm_irq_routing_table *irq_rt;
11858 struct kvm_lapic_irq irq;
11859 struct kvm_vcpu *vcpu;
11860 struct vcpu_data vcpu_info;
11863 if (!kvm_arch_has_assigned_device(kvm) ||
11864 !irq_remapping_cap(IRQ_POSTING_CAP) ||
11865 !kvm_vcpu_apicv_active(kvm->vcpus[0]))
11868 idx = srcu_read_lock(&kvm->irq_srcu);
11869 irq_rt = srcu_dereference(kvm->irq_routing, &kvm->irq_srcu);
11870 if (guest_irq >= irq_rt->nr_rt_entries ||
11871 hlist_empty(&irq_rt->map[guest_irq])) {
11872 pr_warn_once("no route for guest_irq %u/%u (broken user space?)\n",
11873 guest_irq, irq_rt->nr_rt_entries);
11877 hlist_for_each_entry(e, &irq_rt->map[guest_irq], link) {
11878 if (e->type != KVM_IRQ_ROUTING_MSI)
11881 * VT-d PI cannot support posting multicast/broadcast
11882 * interrupts to a vCPU, we still use interrupt remapping
11883 * for these kind of interrupts.
11885 * For lowest-priority interrupts, we only support
11886 * those with single CPU as the destination, e.g. user
11887 * configures the interrupts via /proc/irq or uses
11888 * irqbalance to make the interrupts single-CPU.
11890 * We will support full lowest-priority interrupt later.
11893 kvm_set_msi_irq(kvm, e, &irq);
11894 if (!kvm_intr_is_single_vcpu(kvm, &irq, &vcpu)) {
11896 * Make sure the IRTE is in remapped mode if
11897 * we don't handle it in posted mode.
11899 ret = irq_set_vcpu_affinity(host_irq, NULL);
11902 "failed to back to remapped mode, irq: %u\n",
11910 vcpu_info.pi_desc_addr = __pa(vcpu_to_pi_desc(vcpu));
11911 vcpu_info.vector = irq.vector;
11913 trace_kvm_pi_irte_update(vcpu->vcpu_id, host_irq, e->gsi,
11914 vcpu_info.vector, vcpu_info.pi_desc_addr, set);
11917 ret = irq_set_vcpu_affinity(host_irq, &vcpu_info);
11919 ret = irq_set_vcpu_affinity(host_irq, NULL);
11922 printk(KERN_INFO "%s: failed to update PI IRTE\n",
11930 srcu_read_unlock(&kvm->irq_srcu, idx);
11934 static void vmx_setup_mce(struct kvm_vcpu *vcpu)
11936 if (vcpu->arch.mcg_cap & MCG_LMCE_P)
11937 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits |=
11938 FEATURE_CONTROL_LMCE;
11940 to_vmx(vcpu)->msr_ia32_feature_control_valid_bits &=
11941 ~FEATURE_CONTROL_LMCE;
11944 static struct kvm_x86_ops vmx_x86_ops __ro_after_init = {
11945 .cpu_has_kvm_support = cpu_has_kvm_support,
11946 .disabled_by_bios = vmx_disabled_by_bios,
11947 .hardware_setup = hardware_setup,
11948 .hardware_unsetup = hardware_unsetup,
11949 .check_processor_compatibility = vmx_check_processor_compat,
11950 .hardware_enable = hardware_enable,
11951 .hardware_disable = hardware_disable,
11952 .cpu_has_accelerated_tpr = report_flexpriority,
11953 .cpu_has_high_real_mode_segbase = vmx_has_high_real_mode_segbase,
11955 .vcpu_create = vmx_create_vcpu,
11956 .vcpu_free = vmx_free_vcpu,
11957 .vcpu_reset = vmx_vcpu_reset,
11959 .prepare_guest_switch = vmx_save_host_state,
11960 .vcpu_load = vmx_vcpu_load,
11961 .vcpu_put = vmx_vcpu_put,
11963 .update_bp_intercept = update_exception_bitmap,
11964 .get_msr = vmx_get_msr,
11965 .set_msr = vmx_set_msr,
11966 .get_segment_base = vmx_get_segment_base,
11967 .get_segment = vmx_get_segment,
11968 .set_segment = vmx_set_segment,
11969 .get_cpl = vmx_get_cpl,
11970 .get_cs_db_l_bits = vmx_get_cs_db_l_bits,
11971 .decache_cr0_guest_bits = vmx_decache_cr0_guest_bits,
11972 .decache_cr3 = vmx_decache_cr3,
11973 .decache_cr4_guest_bits = vmx_decache_cr4_guest_bits,
11974 .set_cr0 = vmx_set_cr0,
11975 .set_cr3 = vmx_set_cr3,
11976 .set_cr4 = vmx_set_cr4,
11977 .set_efer = vmx_set_efer,
11978 .get_idt = vmx_get_idt,
11979 .set_idt = vmx_set_idt,
11980 .get_gdt = vmx_get_gdt,
11981 .set_gdt = vmx_set_gdt,
11982 .get_dr6 = vmx_get_dr6,
11983 .set_dr6 = vmx_set_dr6,
11984 .set_dr7 = vmx_set_dr7,
11985 .sync_dirty_debug_regs = vmx_sync_dirty_debug_regs,
11986 .cache_reg = vmx_cache_reg,
11987 .get_rflags = vmx_get_rflags,
11988 .set_rflags = vmx_set_rflags,
11990 .tlb_flush = vmx_flush_tlb,
11992 .run = vmx_vcpu_run,
11993 .handle_exit = vmx_handle_exit,
11994 .skip_emulated_instruction = skip_emulated_instruction,
11995 .set_interrupt_shadow = vmx_set_interrupt_shadow,
11996 .get_interrupt_shadow = vmx_get_interrupt_shadow,
11997 .patch_hypercall = vmx_patch_hypercall,
11998 .set_irq = vmx_inject_irq,
11999 .set_nmi = vmx_inject_nmi,
12000 .queue_exception = vmx_queue_exception,
12001 .cancel_injection = vmx_cancel_injection,
12002 .interrupt_allowed = vmx_interrupt_allowed,
12003 .nmi_allowed = vmx_nmi_allowed,
12004 .get_nmi_mask = vmx_get_nmi_mask,
12005 .set_nmi_mask = vmx_set_nmi_mask,
12006 .enable_nmi_window = enable_nmi_window,
12007 .enable_irq_window = enable_irq_window,
12008 .update_cr8_intercept = update_cr8_intercept,
12009 .set_virtual_x2apic_mode = vmx_set_virtual_x2apic_mode,
12010 .set_apic_access_page_addr = vmx_set_apic_access_page_addr,
12011 .get_enable_apicv = vmx_get_enable_apicv,
12012 .refresh_apicv_exec_ctrl = vmx_refresh_apicv_exec_ctrl,
12013 .load_eoi_exitmap = vmx_load_eoi_exitmap,
12014 .apicv_post_state_restore = vmx_apicv_post_state_restore,
12015 .hwapic_irr_update = vmx_hwapic_irr_update,
12016 .hwapic_isr_update = vmx_hwapic_isr_update,
12017 .sync_pir_to_irr = vmx_sync_pir_to_irr,
12018 .deliver_posted_interrupt = vmx_deliver_posted_interrupt,
12020 .set_tss_addr = vmx_set_tss_addr,
12021 .get_tdp_level = get_ept_level,
12022 .get_mt_mask = vmx_get_mt_mask,
12024 .get_exit_info = vmx_get_exit_info,
12026 .get_lpage_level = vmx_get_lpage_level,
12028 .cpuid_update = vmx_cpuid_update,
12030 .rdtscp_supported = vmx_rdtscp_supported,
12031 .invpcid_supported = vmx_invpcid_supported,
12033 .set_supported_cpuid = vmx_set_supported_cpuid,
12035 .has_wbinvd_exit = cpu_has_vmx_wbinvd_exit,
12037 .write_tsc_offset = vmx_write_tsc_offset,
12039 .set_tdp_cr3 = vmx_set_cr3,
12041 .check_intercept = vmx_check_intercept,
12042 .handle_external_intr = vmx_handle_external_intr,
12043 .mpx_supported = vmx_mpx_supported,
12044 .xsaves_supported = vmx_xsaves_supported,
12046 .check_nested_events = vmx_check_nested_events,
12048 .sched_in = vmx_sched_in,
12050 .slot_enable_log_dirty = vmx_slot_enable_log_dirty,
12051 .slot_disable_log_dirty = vmx_slot_disable_log_dirty,
12052 .flush_log_dirty = vmx_flush_log_dirty,
12053 .enable_log_dirty_pt_masked = vmx_enable_log_dirty_pt_masked,
12054 .write_log_dirty = vmx_write_pml_buffer,
12056 .pre_block = vmx_pre_block,
12057 .post_block = vmx_post_block,
12059 .pmu_ops = &intel_pmu_ops,
12061 .update_pi_irte = vmx_update_pi_irte,
12063 #ifdef CONFIG_X86_64
12064 .set_hv_timer = vmx_set_hv_timer,
12065 .cancel_hv_timer = vmx_cancel_hv_timer,
12068 .setup_mce = vmx_setup_mce,
12071 static int __init vmx_init(void)
12073 int r = kvm_init(&vmx_x86_ops, sizeof(struct vcpu_vmx),
12074 __alignof__(struct vcpu_vmx), THIS_MODULE);
12078 #ifdef CONFIG_KEXEC_CORE
12079 rcu_assign_pointer(crash_vmclear_loaded_vmcss,
12080 crash_vmclear_local_loaded_vmcss);
12086 static void __exit vmx_exit(void)
12088 #ifdef CONFIG_KEXEC_CORE
12089 RCU_INIT_POINTER(crash_vmclear_loaded_vmcss, NULL);
12096 module_init(vmx_init)
12097 module_exit(vmx_exit)